Sup nerds

Are we ready to get going? Are we ready to get going? Let's do it. Sup, nerds? It's been like Let's do it. Sup, nerds? It's been like Let's do it. Sup, nerds? It's been like over a week or so, right? Like, it's over a week or so, right? Like, it's over a week or so, right? Like, it's been a while. How are y'all? been a while. How are y'all? been a while. How are y'all? I I I have a lot going on. Actually, I don't have a lot going on. Actually, I don't have a lot going on. Actually, I don't know if I know if I know if I did all this right. Give me a sec. be. This is where I want to be. be. This is where I want to be. Pop out. Pop out. Pop out. Cool. There we go. There we go. What's up? These are always funny. Is it What's up? These are always funny. Is it What's up? These are always funny. Is it true he worked at Twitch? Like half of true he worked at Twitch? Like half of true he worked at Twitch? Like half of people say I brag about my previous jobs people say I brag about my previous jobs people say I brag about my previous jobs too much. The other half say, "Wait, he too much. The other half say, "Wait, he too much. The other half say, "Wait, he worked at Twitch? Like worked at Twitch? Like worked at Twitch? Like what do I do? was even actually I'll do what do I do? was even actually I'll do what do I do? was even actually I'll do a funnier one. I I'm genuinely curious. a funnier one. I I'm genuinely curious. a funnier one. I I'm genuinely curious. We're going to do a poll. We're going to do a poll. We're going to do a poll. Did you know I am a Y Combinator Did you know I am a Y Combinator Did you know I am a Y Combinator founder? As in I went through a Y founder? As in I went through a Y founder? As in I went through a Y Combinator batch.

Okay. Yeah, that's what I thought. I I Okay. Yeah, that's what I thought. I I am happy to see you guys actually like am happy to see you guys actually like am happy to see you guys actually like know my [ __ ] Does this count as a work know my [ __ ] Does this count as a work know my [ __ ] Does this count as a work expense? expense? expense? The cyclic thing is annoying, but that The cyclic thing is annoying, but that The cyclic thing is annoying, but that is a good reminder that I owe you a is a good reminder that I owe you a is a good reminder that I owe you a company card, too, though. Yeah, I the void zero thing will Yeah, I the void zero thing will definitely be covered. Uh, I have not definitely be covered. Uh, I have not definitely be covered. Uh, I have not gotten the cursor handle yet. I hope I gotten the cursor handle yet. I hope I gotten the cursor handle yet. I hope I can. I Twitch at work is actually a [ __ ] I Twitch at work is actually a [ __ ] hilarious message. hilarious message. hilarious message. How much of my income do I invest the How much of my income do I invest the How much of my income do I invest the vast majority of my profit at least vast majority of my profit at least vast majority of my profit at least because like like my my businesses have because like like my my businesses have because like like my my businesses have revenue and then I have to pay employees revenue and then I have to pay employees revenue and then I have to pay employees and whatnot. and whatnot. and whatnot. Uh I only usually do credit cards for Uh I only usually do credit cards for Uh I only usually do credit cards for full-time US-based employees because full-time US-based employees because full-time US-based employees because giving cards overseas is a whole mess.

giving cards overseas is a whole mess. giving cards overseas is a whole mess. like I went through it with Julius to like I went through it with Julius to like I went through it with Julius to get him a card and it ended up being get him a card and it ended up being get him a card and it ended up being easier to just give him numbers for my easier to just give him numbers for my easier to just give him numbers for my own. But uh if there are business own. But uh if there are business own. But uh if there are business expenses you guys need covered, you expenses you guys need covered, you expenses you guys need covered, you obviously know you can hit me up or obviously know you can hit me up or obviously know you can hit me up or Alyssa honestly is probably better and Alyssa honestly is probably better and Alyssa honestly is probably better and we'll get that handled ASAP. If there we'll get that handled ASAP. If there we'll get that handled ASAP. If there are actual like things you could have are actual like things you could have are actual like things you could have money spent on that would make it easier money spent on that would make it easier money spent on that would make it easier to do your jobs and make our content to do your jobs and make our content to do your jobs and make our content better, obviously we're down to spend better, obviously we're down to spend better, obviously we're down to spend the money. the money. the money. I am also and I'm probably going to talk I am also and I'm probably going to talk I am also and I'm probably going to talk about this in videos. I talked about it about this in videos. I talked about it about this in videos. I talked about it with the founder of Cognition yesterday. with the founder of Cognition yesterday. with the founder of Cognition yesterday. The fact that Cognition realized they The fact that Cognition realized they The fact that Cognition realized they can get my attention by giving Shiv and can get my attention by giving Shiv and can get my attention by giving Shiv and Maria access to [ __ ] for free is very Maria access to [ __ ] for free is very Maria access to [ __ ] for free is very smart of them. smart of them. smart of them. Why have I not given you a mod badge Why have I not given you a mod badge Why have I not given you a mod badge yet? Yeah, it's actually like one of the best Yeah, it's actually like one of the best ways to get me to learn more about a ways to get me to learn more about a ways to get me to learn more about a product and be more likely to cover it product and be more likely to cover it product and be more likely to cover it is to convince Shiv and Maria to use it. is to convince Shiv and Maria to use it. is to convince Shiv and Maria to use it. If you can do that, the likelihood it If you can do that, the likelihood it If you can do that, the likelihood it makes it to me and I start makes it to me and I start makes it to me and I start hearing about the thing more, the hearing about the thing more, the hearing about the thing more, the likelihood I actually like use the thing likelihood I actually like use the thing likelihood I actually like use the thing goes up exponentially.

goes up exponentially. goes up exponentially. What do you got for me here? the droid guys are cool. Actually, the the droid guys are cool. Actually, the droid guys themselves are so cool and so droid guys themselves are so cool and so droid guys themselves are so cool and so chill that it makes me want to try the chill that it makes me want to try the chill that it makes me want to try the product more. product more. product more. Oh yeah, thank you for that reminder, Oh yeah, thank you for that reminder, Oh yeah, thank you for that reminder, Es. I'm in the middle of overhauling the Es. I'm in the middle of overhauling the Es. I'm in the middle of overhauling the data layer, but I'll take a look at that data layer, but I'll take a look at that data layer, but I'll take a look at that hopefully early next week. I should do hopefully early next week. I should do hopefully early next week. I should do like an actual to-do list for stream like an actual to-do list for stream like an actual to-do list for stream because there's so much random [ __ ] I because there's so much random [ __ ] I because there's so much random [ __ ] I want to like think about and talk about. The the cloud platforms have talked The the cloud platforms have talked about a little bit in this what are the about a little bit in this what are the about a little bit in this what are the videos I'm doing today. Let me thank all the subs we've gotten Let me thank all the subs we've gotten so far because you guys went crazy with so far because you guys went crazy with so far because you guys went crazy with the gifts when I was getting started. the gifts when I was getting started. the gifts when I was getting started. Snowy throwing a shitload of gifts. God Snowy throwing a shitload of gifts. God Snowy throwing a shitload of gifts. God damn. I'll properly thank you in a sec damn. I'll properly thank you in a sec damn. I'll properly thank you in a sec once we get to the start. Appears to be once we get to the start. Appears to be once we get to the start. Appears to be Edge TR with the five months of support. Edge TR with the five months of support. Edge TR with the five months of support. Then SJ John continuing the gift from Then SJ John continuing the gift from Then SJ John continuing the gift from NMG. Dead Zen here with the four months NMG. Dead Zen here with the four months NMG. Dead Zen here with the four months of Prime. Pretty good trip overall. I am of Prime. Pretty good trip overall. I am of Prime. Pretty good trip overall. I am actually really proud of the talk I did actually really proud of the talk I did actually really proud of the talk I did at Cascadia. I can't wait for the videos at Cascadia. I can't wait for the videos at Cascadia. I can't wait for the videos to come out. I think I did a very good to come out. I think I did a very good to come out. I think I did a very good job with that.

job with that. job with that. Thank you, Snowy, for the 10 bomb. Thank you, Snowy, for the 10 bomb. Thank you, Snowy, for the 10 bomb. Additional gift to Grayson Gaming Zone Additional gift to Grayson Gaming Zone Additional gift to Grayson Gaming Zone and then another 10 bomb after that. God and then another 10 bomb after that. God and then another 10 bomb after that. God damn. Hopefully the the opportunities we damn. Hopefully the the opportunities we damn. Hopefully the the opportunities we were talking about came through because were talking about came through because were talking about came through because you should not be spending that much you should not be spending that much you should not be spending that much otherwise. otherwise. otherwise. We got Oso with the 11 months, but they We got Oso with the 11 months, but they We got Oso with the 11 months, but they hit that one year. It always good to see hit that one year. It always good to see hit that one year. It always good to see you, man. Going to be a year soon. you, man. Going to be a year soon. you, man. Going to be a year soon. Should you prep for the Perf review? Should you prep for the Perf review? Should you prep for the Perf review? Yeah, as long as you keep dropping good Yeah, as long as you keep dropping good Yeah, as long as you keep dropping good links, you'll be just fine, I'm sure. links, you'll be just fine, I'm sure. links, you'll be just fine, I'm sure. While Elantro with the nine months of While Elantro with the nine months of While Elantro with the nine months of support. Yeah, I I'm happy you support. Yeah, I I'm happy you support. Yeah, I I'm happy you understand that the support from subs is understand that the support from subs is understand that the support from subs is not where most of our income comes. Like not where most of our income comes. Like not where most of our income comes. Like I could not afford to keep almost any of I could not afford to keep almost any of I could not afford to keep almost any of our staff if I was just making money off our staff if I was just making money off our staff if I was just making money off subs and YouTube ads. The sponsors are subs and YouTube ads. The sponsors are subs and YouTube ads. The sponsors are what make this all work. what make this all work. what make this all work. Google's like shift of search becoming Google's like shift of search becoming Google's like shift of search becoming more and more AI. more and more AI. more and more AI. The the reset of SEO is interesting, but The the reset of SEO is interesting, but The the reset of SEO is interesting, but I want to see more of like the impact I want to see more of like the impact I want to see more of like the impact and like experience it more myself and like experience it more myself and like experience it more myself before I lean too hard on that. But it before I lean too hard on that. But it before I lean too hard on that. But it is very very interesting. I agree. No is very very interesting. I agree. No is very very interesting. I agree. No good Nick with the 12 months of tier good Nick with the 12 months of tier good Nick with the 12 months of tier three, 47 months total. That's insane.

three, 47 months total. That's insane. three, 47 months total. That's insane. Thank you so much. We got Justy42 here Thank you so much. We got Justy42 here Thank you so much. We got Justy42 here with the prime. We got an anonymous 10 with the prime. We got an anonymous 10 with the prime. We got an anonymous 10 bomb. God damn. X Cervantes X with the 35 months. Co 6 X Cervantes X with the 35 months. Co 6 BXL with the tier one. Prunch here with BXL with the tier one. Prunch here with BXL with the tier one. Prunch here with the nine months of support. Continuing the nine months of support. Continuing the nine months of support. Continuing the gift from NMG Ardum with the 14 the gift from NMG Ardum with the 14 the gift from NMG Ardum with the 14 months. months. months. What career would I pursue if software What career would I pursue if software What career would I pursue if software dev and YouTube were not an option? Is dev and YouTube were not an option? Is dev and YouTube were not an option? Is this considering like this is a tough this considering like this is a tough this considering like this is a tough question cuz it's like if they were question cuz it's like if they were question cuz it's like if they were never an option so I like lived my life never an option so I like lived my life never an option so I like lived my life different or somebody snap their fingers different or somebody snap their fingers different or somebody snap their fingers and those two opportunities are gone and those two opportunities are gone and those two opportunities are gone where I'm at right now. What would I do? where I'm at right now. What would I do? where I'm at right now. What would I do? Cuz those are very different. I if it was right now I would go into I if it was right now I would go into like marketing, social media, business like marketing, social media, business like marketing, social media, business planning stuff or maybe just VC. But if planning stuff or maybe just VC. But if planning stuff or maybe just VC. But if like software dev and YouTube are not like software dev and YouTube are not like software dev and YouTube are not options, I don't know how much those are options, I don't know how much those are options, I don't know how much those are still options too. It's a it's a hard still options too. It's a it's a hard still options too. It's a it's a hard theoretical for me to consider.

Admouse here with the nine months of Admouse here with the nine months of support. Thank you for that. NMG with support. Thank you for that. NMG with support. Thank you for that. NMG with the gifted space doodle. Good to see the gifted space doodle. Good to see the gifted space doodle. Good to see you, man. Monday's work expenses. Prom you, man. Monday's work expenses. Prom you, man. Monday's work expenses. Prom TTV with the nine months. Dallas with TTV with the nine months. Dallas with TTV with the nine months. Dallas with the three months. the three months. the three months. Igno with the prime. Satchin with the Igno with the prime. Satchin with the Igno with the prime. Satchin with the prime. Mr. and Mrs. Llama with the prime prime. Mr. and Mrs. Llama with the prime prime. Mr. and Mrs. Llama with the prime or the yeah prime for two months. Sammy or the yeah prime for two months. Sammy or the yeah prime for two months. Sammy with the three months. And Bad Mike with with the three months. And Bad Mike with with the three months. And Bad Mike with the eight. I I I don't know how I don't know how I don't know how like let's just look through I I I don't like let's just look through I I I don't like let's just look through I I I don't want to one guy myself with the your want to one guy myself with the your want to one guy myself with the your videos feel like paid ads but like let's videos feel like paid ads but like let's videos feel like paid ads but like let's look at my recent videos. look at my recent videos. look at my recent videos. How are any of these three ads like one where I like one where I talk about my experience with really talk about my experience with really talk about my experience with really shitty engineers, one where I talk about shitty engineers, one where I talk about shitty engineers, one where I talk about how prompts are tech debt and like you how prompts are tech debt and like you how prompts are tech debt and like you need to keep your prompts more up to need to keep your prompts more up to need to keep your prompts more up to date, and one where I talk all about how date, and one where I talk all about how date, and one where I talk all about how like lazy programmers were better than like lazy programmers were better than like lazy programmers were better than all of these more traditional like vibe all of these more traditional like vibe all of these more traditional like vibe coding people that are popping up now.

coding people that are popping up now. coding people that are popping up now. Like I just don't see how any of those Like I just don't see how any of those Like I just don't see how any of those three could even be vaguely considered three could even be vaguely considered three could even be vaguely considered an ad. Yeah. I just I don't see it for most of Yeah. I just I don't see it for most of these. like how any of these could be these. like how any of these could be these. like how any of these could be seen as an ad, especially like these seen as an ad, especially like these seen as an ad, especially like these three. The GitHub hack, Google can't be three. The GitHub hack, Google can't be three. The GitHub hack, Google can't be trusted, why Copilot's billing had to trusted, why Copilot's billing had to trusted, why Copilot's billing had to change. Like change. Like change. Like I I don't see it. I don't get it. I I don't see it. I don't get it. I I don't see it. I don't get it. There's another question in here I There's another question in here I There's another question in here I wanted to mention though. Um or is it? It was something about the or is it? It was something about the YouTube algorithm changing. Yeah. How YouTube algorithm changing. Yeah. How YouTube algorithm changing. Yeah. How did last year's YouTube algo change did last year's YouTube algo change did last year's YouTube algo change affect my channel? I didn't notice a affect my channel? I didn't notice a affect my channel? I didn't notice a YouTube algo change and it didn't seem YouTube algo change and it didn't seem YouTube algo change and it didn't seem to affect my channel last year. I the I to affect my channel last year. I the I to affect my channel last year. I the I find people fixate on the algorithm too find people fixate on the algorithm too find people fixate on the algorithm too much when the way the algorithm works is much when the way the algorithm works is much when the way the algorithm works is pretty simple. It's just a diffing pretty simple. It's just a diffing pretty simple. It's just a diffing algorithm. It just checks like if three algorithm. It just checks like if three algorithm. It just checks like if three people watched nine of the same videos people watched nine of the same videos people watched nine of the same videos and two of them watched a tenth video, and two of them watched a tenth video, and two of them watched a tenth video, the third person will just be spammed the third person will just be spammed the third person will just be spammed that extra video over and over again that extra video over and over again that extra video over and over again until hopefully they maybe click it.

until hopefully they maybe click it. until hopefully they maybe click it. Like the algorithm is super simple. I've not played with Codex being able to I've not played with Codex being able to automate Codex itself yet. automate Codex itself yet. automate Codex itself yet. Why was I troubleshooting Ben's network? Why was I troubleshooting Ben's network? Why was I troubleshooting Ben's network? That's his story to tell and we'll That's his story to tell and we'll That's his story to tell and we'll probably tell it on the podcast. Why don't I see old streams in Twitch? Why don't I see old streams in Twitch? Cuz Twitch uh archives them and like Cuz Twitch uh archives them and like Cuz Twitch uh archives them and like hides them after a certain amount of hides them after a certain amount of hides them after a certain amount of time. Feel like the videos and two sponsors Feel like the videos and two sponsors are a bit much. I don't love that we are a bit much. I don't love that we are a bit much. I don't love that we have to, but it's like we lose so much have to, but it's like we lose so much have to, but it's like we lose so much money if we don't. Like it's it's double money if we don't. Like it's it's double money if we don't. Like it's it's double the revenue and allows us to like the revenue and allows us to like the revenue and allows us to like continue growing the team and do more continue growing the team and do more continue growing the team and do more better content. Like we wouldn't be able better content. Like we wouldn't be able better content. Like we wouldn't be able to bring somebody like Mondays on if we to bring somebody like Mondays on if we to bring somebody like Mondays on if we weren't increasing weren't increasing weren't increasing stuff.

stuff. stuff. Uh what's this responding to? Uh I'll Uh what's this responding to? Uh I'll Uh what's this responding to? Uh I'll allow it. Yeah, the recursive cell allow it. Yeah, the recursive cell allow it. Yeah, the recursive cell improvement stuff's interesting. Nothing on that, but got a good new Nothing on that, but got a good new remote roll. Interesting. Hope that goes remote roll. Interesting. Hope that goes remote roll. Interesting. Hope that goes well for you, Snowy. well for you, Snowy. well for you, Snowy. Yeah, keep an eye for secondary Yeah, keep an eye for secondary Yeah, keep an eye for secondary transitions for sure. I don't know how transitions for sure. I don't know how transitions for sure. I don't know how long today's videos are going to be. I'm long today's videos are going to be. I'm long today's videos are going to be. I'm hoping most of them will be shorter, but hoping most of them will be shorter, but hoping most of them will be shorter, but I have a bad feeling one or two are I have a bad feeling one or two are I have a bad feeling one or two are going to go long. So, the the two I have like relatively So, the the two I have like relatively high priority on are these two. high priority on are these two. high priority on are these two. Confirm they're all going to be at least 90 plus they're all going to be at least 90 plus minutes. Need at least one long one. minutes. Need at least one long one. minutes. Need at least one long one. I'll I'll get you at least one, Ally. I'll I'll get you at least one, Ally. I'll I'll get you at least one, Ally. Don't worry.

I don't mind sponsor block and I don't I don't mind sponsor block and I don't really have any way to see it in the really have any way to see it in the really have any way to see it in the analytics. What I mind is people analytics. What I mind is people analytics. What I mind is people constantly like constantly like constantly like like pushing it and how do I put it? like pushing it and how do I put it? like pushing it and how do I put it? Like if sponsor blocks default was a Like if sponsor blocks default was a Like if sponsor blocks default was a one-click skip, I would be a huge one-click skip, I would be a huge one-click skip, I would be a huge supporter of it and would tell everyone supporter of it and would tell everyone supporter of it and would tell everyone to use it. I am fully okay with people to use it. I am fully okay with people to use it. I am fully okay with people skipping my ads if they choose to. What skipping my ads if they choose to. What skipping my ads if they choose to. What I'm less okay with is the idea that I'm less okay with is the idea that I'm less okay with is the idea that somebody could have a like somebody could have a like somebody could have a like they watch a video on some other they watch a video on some other they watch a video on some other channel, channel, channel, they get a bunch of ads for shitty VPNs they get a bunch of ads for shitty VPNs they get a bunch of ads for shitty VPNs and stuff. They're annoyed with it. They and stuff. They're annoyed with it. They and stuff. They're annoyed with it. They install sponsor block and set it up to install sponsor block and set it up to install sponsor block and set it up to autos skip everything forever. and then they never see our ads for and then they never see our ads for brands that are actually cool and brands that are actually cool and brands that are actually cool and related to the [ __ ] that we talk about. related to the [ __ ] that we talk about. related to the [ __ ] that we talk about. And I really try to lean into that. Like And I really try to lean into that. Like And I really try to lean into that. Like I don't want to do another set of I don't want to do another set of I don't want to do another set of [ __ ] VPN ads. And honestly, like if [ __ ] VPN ads. And honestly, like if [ __ ] VPN ads. And honestly, like if you've seen enough work OS ads that you've seen enough work OS ads that you've seen enough work OS ads that you're tired of seeing them again and you're tired of seeing them again and you're tired of seeing them again and you just skip, go ahead. I don't care. I you just skip, go ahead. I don't care. I you just skip, go ahead. I don't care. I will never be upset with people skipping will never be upset with people skipping will never be upset with people skipping ads. The only thing that can get ads. The only thing that can get ads. The only thing that can get slightly frustrating for me is if slightly frustrating for me is if slightly frustrating for me is if somebody somebody somebody because of other channels not putting because of other channels not putting because of other channels not putting effort into their ads have set up a effort into their ads have set up a effort into their ads have set up a system where they autoskip all ads going system where they autoskip all ads going system where they autoskip all ads going forward. And now the extra work we are forward. And now the extra work we are forward. And now the extra work we are putting in that is used to like fund our putting in that is used to like fund our putting in that is used to like fund our team, donate to open source and do all team, donate to open source and do all team, donate to open source and do all the other cool [ __ ] we do here. We get the other cool [ __ ] we do here. We get the other cool [ __ ] we do here. We get less resources because brands see less

less resources because brands see less less resources because brands see less conversion because you are missing out conversion because you are missing out conversion because you are missing out on a company that you might be on a company that you might be on a company that you might be interested in. I appreciate the deal. We put a lot of I appreciate the deal. We put a lot of effort into that. I'm the only person who people will go, I'm the only person who people will go, "Oh, cool. New ad." That's really cool "Oh, cool. New ad." That's really cool "Oh, cool. New ad." That's really cool to hear. Like especially, to hear. Like especially, to hear. Like especially, we could just like ramp up the amount of we could just like ramp up the amount of we could just like ramp up the amount of ads we do for given brands that we work ads we do for given brands that we work ads we do for given brands that we work with, but I like having the variety. It with, but I like having the variety. It with, but I like having the variety. It makes a shitload of work, but the makes a shitload of work, but the makes a shitload of work, but the variety is really nice. And I get really variety is really nice. And I get really variety is really nice. And I get really hyped when we bring on a new brand that hyped when we bring on a new brand that hyped when we bring on a new brand that I actually like the product of and can I actually like the product of and can I actually like the product of and can show you guys more about. Yeah, my team has actually been Yeah, my team has actually been pressuring me to reuse ads instead of pressuring me to reuse ads instead of pressuring me to reuse ads instead of filming a new one for every single filming a new one for every single filming a new one for every single video, but I continuously do it because video, but I continuously do it because video, but I continuously do it because of who I am as a person. It's also a lot of who I am as a person. It's also a lot of who I am as a person. It's also a lot more work for FaZe, too. So, shout out more work for FaZe, too. So, shout out more work for FaZe, too. So, shout out to him for putting up with all of it.

How am I making money? I mean, the How am I making money? I mean, the majority of my personal income comes majority of my personal income comes majority of my personal income comes from content at this point, but I also from content at this point, but I also from content at this point, but I also run a company that sells a service, T3. run a company that sells a service, T3. run a company that sells a service, T3. Which is the best AI chat ever made, and Which is the best AI chat ever made, and Which is the best AI chat ever made, and T3 Chat is profitable. T3 Chat is profitable. T3 Chat is profitable. Check Discord. I know I'm so behind on Check Discord. I know I'm so behind on Check Discord. I know I'm so behind on DMs. You guys have no idea. It's like DMs. You guys have no idea. It's like DMs. You guys have no idea. It's like bad. And I cleared out like half of them bad. And I cleared out like half of them bad. And I cleared out like half of them yesterday, and there's still so much yesterday, and there's still so much yesterday, and there's still so much more. and Twitter DMs are even worse. more. and Twitter DMs are even worse. more. and Twitter DMs are even worse. So, okay. Yeah, we will hang soon. I So, okay. Yeah, we will hang soon. I So, okay. Yeah, we will hang soon. I promise. promise. promise. Late next week, early the week after is Late next week, early the week after is Late next week, early the week after is probably going to be best for me. Thank you, Energy, for the cheer. I hope Thank you, Energy, for the cheer. I hope you enjoy the circus finale. I'll get you enjoy the circus finale. I'll get you enjoy the circus finale. I'll get you Lake Bed access soon. Not sure if you Lake Bed access soon. Not sure if you Lake Bed access soon. Not sure if you're still here, but appreciate you as you're still here, but appreciate you as you're still here, but appreciate you as always, man. Got Gamergirl with the always, man. Got Gamergirl with the always, man. Got Gamergirl with the cheer. Please add Kubernetes to Lakebed. cheer. Please add Kubernetes to Lakebed. cheer. Please add Kubernetes to Lakebed. [ __ ] no. Never. We got sex dev with the [ __ ] no. Never. We got sex dev with the [ __ ] no. Never. We got sex dev with the four months. Uh, we were trying out Devon a while Uh, we were trying out Devon a while ago, but you definitely inspired Ben to ago, but you definitely inspired Ben to ago, but you definitely inspired Ben to use it a bit more.

can make more money by putting out a can make more money by putting out a course on how to make ads that don't course on how to make ads that don't course on how to make ads that don't work with a $500 version for crypto work with a $500 version for crypto work with a $500 version for crypto Bros. You're not wrong. Like I could do Bros. You're not wrong. Like I could do Bros. You're not wrong. Like I could do a video like I could do a course on how a video like I could do a course on how a video like I could do a course on how to make money making mater like like I don't want to make my money like like I don't want to make my money scamming y'all and scamming audience in scamming y'all and scamming audience in scamming y'all and scamming audience in general. I really want to make general. I really want to make general. I really want to make money from brands that I'm aligned with. You'll figure this out soon. So, CR will You'll figure this out soon. So, CR will hang and it'll get a lot easier from hang and it'll get a lot easier from hang and it'll get a lot easier from there. Once you just beat the crew, there. Once you just beat the crew, there. Once you just beat the crew, you'll be pulled in fast. And once you'll be pulled in fast. And once you'll be pulled in fast. And once you've been invited to like two or three you've been invited to like two or three you've been invited to like two or three events on Luma, you'll get spammed with events on Luma, you'll get spammed with events on Luma, you'll get spammed with everything and there's always too much everything and there's always too much everything and there's always too much to do. Not worried at all, Pedro. I I I was Not worried at all, Pedro. I I I was trying to say that as much of your as trying to say that as much of your as trying to say that as much of your as much as what you said is a joke, it is much as what you said is a joke, it is much as what you said is a joke, it is also true.

also true. also true. I'll be talking about the Void Zero I'll be talking about the Void Zero I'll be talking about the Void Zero stuff. I have time to see movies. I I did watch I have time to see movies. I I did watch the Iron Lung movie finally recently, the Iron Lung movie finally recently, the Iron Lung movie finally recently, but that's like the only movie I've had but that's like the only movie I've had but that's like the only movie I've had time for. A 10 bomb on YouTube. God time for. A 10 bomb on YouTube. God time for. A 10 bomb on YouTube. God damn. Thanks, Tantalus Complex, for damn. Thanks, Tantalus Complex, for damn. Thanks, Tantalus Complex, for that. Appreciate you a ton. Thanks for that. Appreciate you a ton. Thanks for that. Appreciate you a ton. Thanks for the generosity. Thank you. No, for the prime, Xerox 622 Thank you. No, for the prime, Xerox 622 for the cheer, and JD PA for the 10 for the cheer, and JD PA for the 10 for the cheer, and JD PA for the 10 months. Cool. Let's figure out what we're Cool. Let's figure out what we're covering first. The environment cash issue was The environment cash issue was hilarious. hilarious. hilarious. Uh, Uh, Uh, did I see that Brody got a sponsorship did I see that Brody got a sponsorship did I see that Brody got a sponsorship in one of his recent videos? Lots more info coming out about that Lots more info coming out about that soon. Uh, let's just say I'm very aware, soon. Uh, let's just say I'm very aware, soon. Uh, let's just say I'm very aware, extraordinarily aware. Love you, Brody.

extraordinarily aware. Love you, Brody. extraordinarily aware. Love you, Brody. Thank you. I'm so happy you could Thank you. I'm so happy you could Thank you. I'm so happy you could finally make money now. Brody's live right now. Oh, [ __ ] yeah. Brody's live right now. Oh, [ __ ] yeah. And did I like Iron Lung? Yes, I quite And did I like Iron Lung? Yes, I quite And did I like Iron Lung? Yes, I quite enjoyed it. It was a little longer than enjoyed it. It was a little longer than enjoyed it. It was a little longer than it should have been. I think it's kind it should have been. I think it's kind it should have been. I think it's kind of the point, but like I really feel of the point, but like I really feel of the point, but like I really feel like they could have cut like half an like they could have cut like half an like they could have cut like half an hour out and made the movie better, but hour out and made the movie better, but hour out and made the movie better, but it was pretty good. as a fan of the it was pretty good. as a fan of the it was pretty good. as a fan of the game. game. game. Did your bachelor networks and telecom Did your bachelor networks and telecom Did your bachelor networks and telecom engineer graduated recently? Been engineer graduated recently? Been engineer graduated recently? Been self-learning, doing programming for self-learning, doing programming for self-learning, doing programming for seven plus years. Get good opportunities seven plus years. Get good opportunities seven plus years. Get good opportunities to software dev, but it's hard to get to software dev, but it's hard to get to software dev, but it's hard to get but it's very hard to get for the real but it's very hard to get for the real but it's very hard to get for the real network engineering. Making you feel network engineering. Making you feel network engineering. Making you feel uneasy, bad that you have the resource uneasy, bad that you have the resource uneasy, bad that you have the resource knowledge that you're not using. Um, knowledge that you're not using. Um, knowledge that you're not using. Um, I didn't use my audio edge degree for I didn't use my audio edge degree for I didn't use my audio edge degree for like 6 years after graduating and then like 6 years after graduating and then like 6 years after graduating and then it came somewhat useful for making it came somewhat useful for making it came somewhat useful for making video. video. video. I wouldn't worry too much about it. Like I wouldn't worry too much about it. Like I wouldn't worry too much about it. Like do whatever is the highest impact right do whatever is the highest impact right do whatever is the highest impact right now that surrounds you with the highest now that surrounds you with the highest now that surrounds you with the highest impact people right now. Whatever impact people right now. Whatever impact people right now. Whatever whatever path surrounds you with the whatever path surrounds you with the whatever path surrounds you with the types of people that you'd want to like types of people that you'd want to like types of people that you'd want to like have at your wedding is the path to take have at your wedding is the path to take have at your wedding is the path to take and whatever crazy knowledge you have on and whatever crazy knowledge you have on and whatever crazy knowledge you have on various things will become useful various things will become useful various things will become useful eventually.

eventually. eventually. If that network engineering knowledge is If that network engineering knowledge is If that network engineering knowledge is useful for you on that journey useful for you on that journey useful for you on that journey that will happen almost certainly that will happen almost certainly that will happen almost certainly regardless. regardless. regardless. And the way I think of this is more so And the way I think of this is more so And the way I think of this is more so like how do I put this? Once you've just like as an example of Once you've just like as an example of how this could play out that would work how this could play out that would work how this could play out that would work really well is you do normal software really well is you do normal software really well is you do normal software dev jobs. You make a lot of connections dev jobs. You make a lot of connections dev jobs. You make a lot of connections with awesome people, most of which do with awesome people, most of which do with awesome people, most of which do quite well at the companies they're at. quite well at the companies they're at. quite well at the companies they're at. You realize that all the places you've You realize that all the places you've You realize that all the places you've been have similar issues around been have similar issues around been have similar issues around networking stuff and that you could networking stuff and that you could networking stuff and that you could build a solution to it. So you quit and build a solution to it. So you quit and build a solution to it. So you quit and go build that. And now you can tap all go build that. And now you can tap all go build that. And now you can tap all the people you worked with in the past the people you worked with in the past the people you worked with in the past to potentially join your team, to to potentially join your team, to to potentially join your team, to potentially invest, to potentially be potentially invest, to potentially be potentially invest, to potentially be early customers, to bring your solution early customers, to bring your solution early customers, to bring your solution to other businesses. All of the to other businesses. All of the to other businesses. All of the connections and relationships you build connections and relationships you build connections and relationships you build over the next few years of your career over the next few years of your career over the next few years of your career will be so much more valuable than any will be so much more valuable than any will be so much more valuable than any one degree would be or any one piece of one degree would be or any one piece of one degree would be or any one piece of knowledge could be. But when you combine knowledge could be. But when you combine knowledge could be. But when you combine those things together, you put yourself those things together, you put yourself those things together, you put yourself in a really good spot.

Chris has been around for [ __ ] ever. Chris has been around for [ __ ] ever. Damn. Good to see you, man. Hope you're Damn. Good to see you, man. Hope you're Damn. Good to see you, man. Hope you're doing well. Yeah, Brody is Brody doing well. Yeah, Brody is Brody doing well. Yeah, Brody is Brody Robertson. One of my favorite tech Robertson. One of my favorite tech Robertson. One of my favorite tech YouTube channels. My favorite Linux guy. YouTube channels. My favorite Linux guy. YouTube channels. My favorite Linux guy. Like, if you're wondering why I'm so Like, if you're wondering why I'm so Like, if you're wondering why I'm so near pill and why I'm pushing hard to near pill and why I'm pushing hard to near pill and why I'm pushing hard to like get more into Linux, it's entirely like get more into Linux, it's entirely like get more into Linux, it's entirely Brody. The fundamental difference between Lakeb The fundamental difference between Lakeb is doing. Don't worry, we'll talk about is doing. Don't worry, we'll talk about is doing. Don't worry, we'll talk about that a lot. Begging to get Pi working in T3 code. Begging to get Pi working in T3 code. Yeah, we really do need to. The The Yeah, we really do need to. The The Yeah, we really do need to. The The reason we haven't isn't because reason we haven't isn't because reason we haven't isn't because like we have something against Pi. We like we have something against Pi. We like we have something against Pi. We actually do love Pi. I want to make sure actually do love Pi. I want to make sure actually do love Pi. I want to make sure the Pi integration is special. that the Pi integration is special. that the Pi integration is special. that we're not just another generic ACP we're not just another generic ACP we're not just another generic ACP [ __ ] integration, but that the magic [ __ ] integration, but that the magic [ __ ] integration, but that the magic of PI and it's like self building, of PI and it's like self building, of PI and it's like self building, self-healing, customizing capabilities self-healing, customizing capabilities self-healing, customizing capabilities are properly encoded in the T3 code are properly encoded in the T3 code are properly encoded in the T3 code bindings. Like I want to make sure we bindings. Like I want to make sure we bindings. Like I want to make sure we get PI right.

I appreciate you a lot, Lazy Coder. I'm I appreciate you a lot, Lazy Coder. I'm so behind on email. Alyssa has a whole so behind on email. Alyssa has a whole so behind on email. Alyssa has a whole list of things I need to get through, list of things I need to get through, list of things I need to get through, but I will hopefully this weekend, early but I will hopefully this weekend, early but I will hopefully this weekend, early next week. next week. next week. Been mod for longer than you've been a Been mod for longer than you've been a Been mod for longer than you've been a sub. No worries, Chris. You were in sub. No worries, Chris. You were in sub. No worries, Chris. You were in college. Thank you for the cheerleader. I Thank you for the cheerleader. I appreciate that. [ __ ] My benchmark video is what got [ __ ] My benchmark video is what got your company to actually consider your company to actually consider your company to actually consider allowing and paying for AI at your allowing and paying for AI at your allowing and paying for AI at your company. That is awesome to hear, D. company. That is awesome to hear, D. company. That is awesome to hear, D. Thank you for sharing that. Did Enthropic actually like talk about a Did Enthropic actually like talk about a global AI pause in that blog post? If global AI pause in that blog post? If global AI pause in that blog post? If so, then I do have to [ __ ] cover it. so, then I do have to [ __ ] cover it. so, then I do have to [ __ ] cover it. I haven't read it yet. I haven't read it yet. I haven't read it yet. Just build [ __ ] Ron. That's all that Just build [ __ ] Ron. That's all that Just build [ __ ] Ron. That's all that matters. Like, build things that are matters. Like, build things that are matters. Like, build things that are cool and useful to you. Seattle was cool and useful to you. Seattle was cool and useful to you. Seattle was fine. It's a good city. I just didn't fine. It's a good city. I just didn't fine. It's a good city. I just didn't like plan my trip thoroughly enough and like plan my trip thoroughly enough and like plan my trip thoroughly enough and had a lot of [ __ ] I had to do. and the had a lot of [ __ ] I had to do. and the had a lot of [ __ ] I had to do. and the internet was awful. So, that was really internet was awful. So, that was really internet was awful. So, that was really annoying. Made the most of it though. It annoying. Made the most of it though. It annoying. Made the most of it though. It was good to be there.

Yeah, I know that this blog is people Yeah, I know that this blog is people are freaking out about. I just didn't are freaking out about. I just didn't are freaking out about. I just didn't know that part of this was like pause. know that part of this was like pause. know that part of this was like pause. Yeah. Yeah. You know what? Yeah. You know what? It gets in I was worried I wouldn't have enough I was worried I wouldn't have enough topics for the day. topics for the day. topics for the day. Been working on paper about how AI is Been working on paper about how AI is Been working on paper about how AI is driving up the cost of living in SF. Uh, driving up the cost of living in SF. Uh, driving up the cost of living in SF. Uh, I am down. But the one thing I would I am down. But the one thing I would I am down. But the one thing I would highly recommend, let me see if I can highly recommend, let me see if I can highly recommend, let me see if I can find the post. I have it bookmarked. Yeah, I have gro hunting for it, but I Yeah, I have gro hunting for it, but I saw a post saw a post saw a post where um how do I explain it? Words are where um how do I explain it? Words are where um how do I explain it? Words are hard. The the post was about how much hard. The the post was about how much hard. The the post was about how much housing has been made in SF housing has been made in SF housing has been made in SF year-over-year.

year-over-year. year-over-year. And the problem is that we're just not And the problem is that we're just not And the problem is that we're just not making anywhere near enough housing. So, making anywhere near enough housing. So, making anywhere near enough housing. So, how many people live in SF? how many people live in SF? how many people live in SF? We currently have almost a million We currently have almost a million We currently have almost a million people living in SF. 826,000. How many new units of housing do you How many new units of housing do you think were made? So, like new residences think were made? So, like new residences think were made? So, like new residences where like one person could live. How where like one person could live. How where like one person could live. How many units of housing do you guys think many units of housing do you guys think many units of housing do you guys think have been made in SF this year so far? have been made in SF this year so far? have been made in SF this year so far? Not looking for a percent. I'm looking Not looking for a percent. I'm looking Not looking for a percent. I'm looking for a number of housing units. People for a number of housing units. People for a number of housing units. People saying like 20K, 25K, 10K, 5,000, saying like 20K, 25K, 10K, 5,000, saying like 20K, 25K, 10K, 5,000, 30,000. 30,000. 30,000. Those all are numbers that make sense. Um, Um, does this link to the post? It does not. It does not. The post had a chart in it. Okay. I'm The post had a chart in it. Okay. I'm The post had a chart in it. Okay. I'm trying to find the post, but the numbers trying to find the post, but the numbers trying to find the post, but the numbers I saw Did you find that already for me?

I saw Did you find that already for me? I saw Did you find that already for me? Is this the what I was That's not what Is this the what I was That's not what Is this the what I was That's not what Yep. Less than 380 new housing units Yep. Less than 380 new housing units Yep. Less than 380 new housing units were built this year. were built this year. were built this year. 380 new units. Do you understand? Do you understand? This is the actual reason rent's getting This is the actual reason rent's getting This is the actual reason rent's getting more expensive. This is the one I was looking for. Thank This is the one I was looking for. Thank you. you. you. Housing units completed in San Francisco Housing units completed in San Francisco Housing units completed in San Francisco per year over the last 20 years. Best per year over the last 20 years. Best per year over the last 20 years. Best years, we would see like 4,000ish. years, we would see like 4,000ish. years, we would see like 4,000ish. Thus far this year, we're at 377. I've ordered more pizza than SF has I've ordered more pizza than SF has built houses. Like, yeah, it's it's that built houses. Like, yeah, it's it's that built houses. Like, yeah, it's it's that bad. It's really bad.

bad. It's really bad. bad. It's really bad. This is the problem. There is just so This is the problem. There is just so This is the problem. There is just so much [ __ ] restriction on housing much [ __ ] restriction on housing much [ __ ] restriction on housing development. development. development. The problem is that there is so much The problem is that there is so much The problem is that there is so much like [ __ ] that prevents housing from like [ __ ] that prevents housing from like [ __ ] that prevents housing from being built. Is there a founder that 15K for a Is there a founder that 15K for a two-bedroom seems within normal these two-bedroom seems within normal these two-bedroom seems within normal these days? Yeah, it's really bad. It's days? Yeah, it's really bad. It's days? Yeah, it's really bad. It's really, really bad. Like we need to be building thousands of Like we need to be building thousands of units a year, not hundreds. units a year, not hundreds. units a year, not hundreds. And there's a lot of layers. I could I And there's a lot of layers. I could I And there's a lot of layers. I could I could [ __ ] about this forever. I I'm a could [ __ ] about this forever. I I'm a could [ __ ] about this forever. I I'm a like Yimi extremist at this point. It's like Yimi extremist at this point. It's like Yimi extremist at this point. It's so pathetic that the city is not doing so pathetic that the city is not doing so pathetic that the city is not doing so much more to make [ __ ] built and so much more to make [ __ ] built and so much more to make [ __ ] built and happen here. It's It's so fixable. They need to Okay, first thing they need They need to Okay, first thing they need to do, massive taxes on vacancy. If a to do, massive taxes on vacancy. If a to do, massive taxes on vacancy. If a house is sitting empty, it should be house is sitting empty, it should be house is sitting empty, it should be taxed accordingly.

taxed accordingly. taxed accordingly. You should lose a shitload of money for You should lose a shitload of money for You should lose a shitload of money for letting housing sit empty. Second thing, letting housing sit empty. Second thing, letting housing sit empty. Second thing, massive tax breaks for new housing. If massive tax breaks for new housing. If massive tax breaks for new housing. If new construction increases the number of new construction increases the number of new construction increases the number of units on a plot, so if you have a plot units on a plot, so if you have a plot units on a plot, so if you have a plot of land that currently has a house that of land that currently has a house that of land that currently has a house that can fit maybe one family in it and you can fit maybe one family in it and you can fit maybe one family in it and you destroy it and rebuild it to fit four destroy it and rebuild it to fit four destroy it and rebuild it to fit four families in it with more square footage, families in it with more square footage, families in it with more square footage, you should get great tax breaks for you should get great tax breaks for you should get great tax breaks for that. There should be lots of incentive. that. There should be lots of incentive. that. There should be lots of incentive. Most importantly, we need a fast track Most importantly, we need a fast track Most importantly, we need a fast track for companies and developers that have a for companies and developers that have a for companies and developers that have a good track record, having built good track record, having built good track record, having built successfully lots of different housing successfully lots of different housing successfully lots of different housing projects in the city. They should have a projects in the city. They should have a projects in the city. They should have a fast track to be approved more fast track to be approved more fast track to be approved more efficiently because right now they're efficiently because right now they're efficiently because right now they're caught up in so much bureaucratic caught up in so much bureaucratic caught up in so much bureaucratic [ __ ] from rich [ __ ] who own a [ __ ] from rich [ __ ] who own a [ __ ] from rich [ __ ] who own a bunch of property that they're letting bunch of property that they're letting bunch of property that they're letting sit rot. Do nothing. They're just sit rot. Do nothing. They're just sit rot. Do nothing. They're just letting this property sit empty because letting this property sit empty because letting this property sit empty because they have crazy tax breaks because they they have crazy tax breaks because they they have crazy tax breaks because they bought them in the 80s. They let them bought them in the 80s. They let them bought them in the 80s. They let them sit empty. They want the property value sit empty. They want the property value sit empty. They want the property value to keep going up. And if it and if more to keep going up. And if it and if more to keep going up. And if it and if more units get built, their property value units get built, their property value units get built, their property value doesn't keep going up. So they're doesn't keep going up. So they're doesn't keep going up. So they're jumping at every opportunity and making jumping at every opportunity and making jumping at every opportunity and making all these like fake heritage groups and all these like fake heritage groups and all these like fake heritage groups and [ __ ] just to prevent the development [ __ ] just to prevent the development [ __ ] just to prevent the development of more [ __ ] housing. It's insane. So of more [ __ ] housing. It's insane. So of more [ __ ] housing. It's insane. So yeah, like the the reason SF rent is yeah, like the the reason SF rent is yeah, like the the reason SF rent is high has nothing to do with AI. It has high has nothing to do with AI. It has high has nothing to do with AI. It has everything to do with the city refusing everything to do with the city refusing everything to do with the city refusing to let [ __ ] building happen.

Yeah, we need more housing. That that's Yeah, we need more housing. That that's that that is the solution to most of that that is the solution to most of that that is the solution to most of these problems. So, uh to the YouTube chatter, uh MC So, uh to the YouTube chatter, uh MC Rich, hopefully this is a useful Rich, hopefully this is a useful Rich, hopefully this is a useful perspective for you in your article perspective for you in your article perspective for you in your article before it goes live or your paper. Yeah, before it goes live or your paper. Yeah, before it goes live or your paper. Yeah, thanks for the inspiration to go on one thanks for the inspiration to go on one thanks for the inspiration to go on one of my favorite rants. Yeah, we we need of my favorite rants. Yeah, we we need of my favorite rants. Yeah, we we need more incentive to build, more speed more incentive to build, more speed more incentive to build, more speed running of building, more opportunity by running of building, more opportunity by running of building, more opportunity by far. Yeah. Like like people being able far. Yeah. Like like people being able far. Yeah. Like like people being able to afford to live in SF and there being to afford to live in SF and there being to afford to live in SF and there being enough housing for all the people who enough housing for all the people who enough housing for all the people who want to live in SF. This shouldn't be want to live in SF. This shouldn't be want to live in SF. This shouldn't be political. political. political. Like this really should not be Like this really should not be Like this really should not be political. This should just be as simple political. This should just be as simple political. This should just be as simple as people want to live here. There as people want to live here. There as people want to live here. There should be more housing so more people should be more housing so more people should be more housing so more people can live here. uh for a decent place in a SF. It uh for a decent place in a SF. It depends on what you're going for. If depends on what you're going for. If depends on what you're going for. If you're doing like a small studio, 3K you you're doing like a small studio, 3K you you're doing like a small studio, 3K you can probably still get away with. Bigger can probably still get away with. Bigger can probably still get away with. Bigger studio or a like onebedroom, you're studio or a like onebedroom, you're studio or a like onebedroom, you're probably in the 4K range. A twobedroom probably in the 4K range. A twobedroom probably in the 4K range. A twobedroom that's like,000 foot or less, you're that's like,000 foot or less, you're that's like,000 foot or less, you're probably in the 6K range or so now.

talking about SF proper, but the Bay talking about SF proper, but the Bay Area is similarly restricted. Like if you look at any chart of like Like if you look at any chart of like rent prices over time. Let me thank all the other subs that Let me thank all the other subs that were going to start. Yeah, I saw the Valdi update actually Yeah, I saw the Valdi update actually looked really compelling, but I'm looked really compelling, but I'm looked really compelling, but I'm obviously on Helium forever now. obviously on Helium forever now. obviously on Helium forever now. Cool. Thank you JD Grage both for the Cool. Thank you JD Grage both for the Cool. Thank you JD Grage both for the prime subs. 45 months. God damn lazy prime subs. 45 months. God damn lazy prime subs. 45 months. God damn lazy coder launch your product. Congrats Kish coder launch your product. Congrats Kish coder launch your product. Congrats Kish and mod longer. Okay, cool. We had all and mod longer. Okay, cool. We had all and mod longer. Okay, cool. We had all those rotating phase with the 31 months. those rotating phase with the 31 months. those rotating phase with the 31 months. Thought about your Azure issue. Not sure Thought about your Azure issue. Not sure Thought about your Azure issue. Not sure if it regressed again. And with open if it regressed again. And with open if it regressed again. And with open router, you could use the smart route router, you could use the smart route router, you could use the smart route and bring your own K to make use of the and bring your own K to make use of the and bring your own K to make use of the 1M. We're already starting to do that 1M. We're already starting to do that 1M. We're already starting to do that suggested remote work killing jobs suggested remote work killing jobs suggested remote work killing jobs instead of AI. Interesting.

instead of AI. Interesting. instead of AI. Interesting. Oh, I have to subscribe to FT to read Oh, I have to subscribe to FT to read Oh, I have to subscribe to FT to read that. That's a good thing to know exists that. That's a good thing to know exists that. That's a good thing to know exists though. I'll save it so I can try and though. I'll save it so I can try and though. I'll save it so I can try and get into that later. Melo yellow here get into that later. Melo yellow here get into that later. Melo yellow here with the prime for five months. with the prime for five months. with the prime for five months. Personally don't mind the ads because I Personally don't mind the ads because I Personally don't mind the ads because I only allow sponsors that I approve of only allow sponsors that I approve of only allow sponsors that I approve of and that means something. Of course. and that means something. Of course. and that means something. Of course. Yeah. I if I wouldn't honestly recommend Yeah. I if I wouldn't honestly recommend Yeah. I if I wouldn't honestly recommend the thing at lunch or dinner, then I the thing at lunch or dinner, then I the thing at lunch or dinner, then I won't won't won't like, how do I put this? Like if I like, how do I put this? Like if I like, how do I put this? Like if I wouldn't actually recommend this at wouldn't actually recommend this at wouldn't actually recommend this at lunch or dinner, then I won't let them lunch or dinner, then I won't let them lunch or dinner, then I won't let them advertise on my channel. Like it needs advertise on my channel. Like it needs advertise on my channel. Like it needs to be a thing I actually genuinely would to be a thing I actually genuinely would to be a thing I actually genuinely would recommend. Yeah, you send me that later, recommend. Yeah, you send me that later, recommend. Yeah, you send me that later, Faser. I don't know when I'll get to it, Faser. I don't know when I'll get to it, Faser. I don't know when I'll get to it, but it's definitely something I'm but it's definitely something I'm but it's definitely something I'm interested in reading. We got Gruns in interested in reading. We got Gruns in interested in reading. We got Gruns in here with the 13 months. God damn. here with the 13 months. God damn. here with the 13 months. God damn. Jasper with the prime and Brandon with Jasper with the prime and Brandon with Jasper with the prime and Brandon with the prime as well. Appreciate all of you the prime as well. Appreciate all of you the prime as well. Appreciate all of you guys for the support. WWDC rumors seem chill but good. WWDC rumors seem chill but good. Excited. Excited. Excited. Oh, is Ladybird not going to be open Oh, is Ladybird not going to be open Oh, is Ladybird not going to be open source? Oh, no. Sensors is just contribution. I Oh, no. Sensors is just contribution. I think that's fair.

think that's fair. think that's fair. I need to skate more, Lauren. I'm hoping I need to skate more, Lauren. I'm hoping I need to skate more, Lauren. I'm hoping I can soon. Renon to to answer this, I I will ask a Renon to to answer this, I I will ask a silly sounding question, but hear me silly sounding question, but hear me silly sounding question, but hear me out. What problem does Vim solve? The reason T3 Code exists isn't because The reason T3 Code exists isn't because like Codex app and cursor don't exist. like Codex app and cursor don't exist. like Codex app and cursor don't exist. The reason T3 code exists is because we The reason T3 code exists is because we The reason T3 code exists is because we wanted a good open- source solution that wanted a good open- source solution that wanted a good open- source solution that works with all the different harnesses works with all the different harnesses works with all the different harnesses that is customizable, reliable, that is customizable, reliable, that is customizable, reliable, powerful, works good, remote, powerful, works good, remote, powerful, works good, remote, extensible, and of course open source. extensible, and of course open source. extensible, and of course open source. So you can do whatever the [ __ ] you want So you can do whatever the [ __ ] you want So you can do whatever the [ __ ] you want to it.

interesting. I will play with this in interesting. I will play with this in the future. Sammy, good work. Congrats the future. Sammy, good work. Congrats the future. Sammy, good work. Congrats on the project. Oh [ __ ] Jamie with the on the project. Oh [ __ ] Jamie with the on the project. Oh [ __ ] Jamie with the 10 bump. Thank you so much for the 10 bump. Thank you so much for the 10 bump. Thank you so much for the support. Always good to see you, man. support. Always good to see you, man. support. Always good to see you, man. Hope you're doing well. Hope we get some Hope you're doing well. Hope we get some Hope you're doing well. Hope we get some fun news from y'all in the near future. >> You need to come up with your own >> You need to come up with your own projects, man. I'm not here to just like projects, man. I'm not here to just like projects, man. I'm not here to just like like the one thing you can bring to the like the one thing you can bring to the like the one thing you can bring to the table now is like new ideas and things table now is like new ideas and things table now is like new ideas and things like solve problems you have. Thank you Brandon for the prime, Rodrigo Thank you Brandon for the prime, Rodrigo for the prime, Jasper for the prime, and for the prime, Jasper for the prime, and for the prime, Jasper for the prime, and Grunzen for the prime. Appreciate all Grunzen for the prime. Appreciate all Grunzen for the prime. Appreciate all you guys. So displacing non- tech communities not So displacing non- tech communities not cost of rent. Uh cost of rent. Uh cost of rent. Uh I still think that the like rent problem I still think that the like rent problem I still think that the like rent problem so Mc Rich this is my issue with that is so Mc Rich this is my issue with that is so Mc Rich this is my issue with that is a significant portion of the a significant portion of the a significant portion of the displacement is because of rent displacement is because of rent displacement is because of rent increases increases increases and rent increases are a problem

and rent increases are a problem and rent increases are a problem specifically because we're not building specifically because we're not building specifically because we're not building enough housing. enough housing. enough housing. The displacement is, in my opinion, The displacement is, in my opinion, The displacement is, in my opinion, largely driven by the lack of additional largely driven by the lack of additional largely driven by the lack of additional housing being built. Cities that do a better job keeping up Cities that do a better job keeping up with the increasing in demand for with the increasing in demand for with the increasing in demand for housing do a much better job of not housing do a much better job of not housing do a much better job of not displacing people. SF failed to address displacing people. SF failed to address displacing people. SF failed to address the fact that like there is so much the fact that like there is so much the fact that like there is so much increase in demand for housing and that increase in demand for housing and that increase in demand for housing and that failure to address the increase in failure to address the increase in failure to address the increase in demand results in prices going up which demand results in prices going up which demand results in prices going up which results in the displacement that is results in the displacement that is results in the displacement that is occurring. Friend was telling you about a video Friend was telling you about a video that I made today which is hilarious that I made today which is hilarious that I made today which is hilarious because he isn't a dev and only heard because he isn't a dev and only heard because he isn't a dev and only heard about me from you. So telling him about about me from you. So telling him about about me from you. So telling him about my so him telling you about my latest my so him telling you about my latest my so him telling you about my latest vid was hilarious. You miss when vid was hilarious. You miss when vid was hilarious. You miss when programmers are lazy video. Interesting. programmers are lazy video. Interesting. programmers are lazy video. Interesting. That one kind of bombed. So, it's good That one kind of bombed. So, it's good That one kind of bombed. So, it's good to hear that it's making it out of my to hear that it's making it out of my to hear that it's making it out of my usual circle. I I think that there is a huge bug that I I think that there is a huge bug that caused a lot of people to get caused a lot of people to get caused a lot of people to get erroneously banned because they they had erroneously banned because they they had erroneously banned because they they had a bug where they were miscounting usage a bug where they were miscounting usage a bug where they were miscounting usage and I think that means that they weren't and I think that means that they weren't and I think that means that they weren't limiting you properly. Some users would limiting you properly. Some users would limiting you properly. Some users would go over the limit because they weren't go over the limit because they weren't go over the limit because they weren't enforcing limit properly. So, anybody enforcing limit properly. So, anybody enforcing limit properly. So, anybody who accidentally went over got banned.

who accidentally went over got banned. who accidentally went over got banned. That is my guess. I have no inside info That is my guess. I have no inside info That is my guess. I have no inside info there. It's funny to say that you're you think It's funny to say that you're you think I always say things confident as fact I always say things confident as fact I always say things confident as fact when I'm literally in the process of when I'm literally in the process of when I'm literally in the process of saying something that I have no info on saying something that I have no info on saying something that I have no info on that is my belief. that is my belief. that is my belief. What's this interesting? I've grown to like Hermes interesting? I've grown to like Hermes quite a bit. I'll probably talk about quite a bit. I'll probably talk about quite a bit. I'll probably talk about that in soon or videos in the near that in soon or videos in the near that in soon or videos in the near future. future. future. Thank you again for the cheer, Mike P. Thank you again for the cheer, Mike P. Thank you again for the cheer, Mike P. Thank you, Louis, for the prime. I know Thank you, Louis, for the prime. I know Thank you, Louis, for the prime. I know we just got the hype train, but we got we just got the hype train, but we got we just got the hype train, but we got to get ready to start filming real to get ready to start filming real to get ready to start filming real videos. So, what's this? Um, not sure what this is. not sure what this is. One less YouTube donation. Thank you. One less YouTube donation. Thank you. One less YouTube donation. Thank you. Came here wondering how everyone feels Came here wondering how everyone feels Came here wondering how everyone feels about GitHub Copilot's new billing about GitHub Copilot's new billing about GitHub Copilot's new billing model. Happy to talk about affordable model. Happy to talk about affordable model. Happy to talk about affordable housing.

housing. housing. Yeah, I I don't know if like like Yeah, I I don't know if like like Yeah, I I don't know if like like I I wish housing was just a money I I wish housing was just a money I I wish housing was just a money problem because if so, a lot of people problem because if so, a lot of people problem because if so, a lot of people would put the money in and solve it. The would put the money in and solve it. The would put the money in and solve it. The problem is that there is a bunch of problem is that there is a bunch of problem is that there is a bunch of restrictions preventing new housing from restrictions preventing new housing from restrictions preventing new housing from being made. And I know a lot of people, being made. And I know a lot of people, being made. And I know a lot of people, myself included, that would put down a myself included, that would put down a myself included, that would put down a lot of money to incentivize the building lot of money to incentivize the building lot of money to incentivize the building of additional housing to keep people of additional housing to keep people of additional housing to keep people from having to be [ __ ] displaced in from having to be [ __ ] displaced in from having to be [ __ ] displaced in the city. I lost a lot of skate friends the city. I lost a lot of skate friends the city. I lost a lot of skate friends to the city because they can't afford to to the city because they can't afford to to the city because they can't afford to live here anymore. Thank you, Matt, as live here anymore. Thank you, Matt, as live here anymore. Thank you, Matt, as well for the 20 bucks. Shouting out well for the 20 bucks. Shouting out well for the 20 bucks. Shouting out Jensen. What are you up to? What are you up to? A secure agent harness runtime.

I like where you're going with this. I I like where you're going with this. I like where you're going with this a lot. I will do a deeper dive on this later I will do a deeper dive on this later for sure. Good work. for sure. Good work. for sure. Good work. definitely thing I was thinking about, definitely thing I was thinking about, definitely thing I was thinking about, but yeah. It's just a habit at this point, Snowy. It's just a habit at this point, Snowy. But also, I do like Zen's appearance and But also, I do like Zen's appearance and But also, I do like Zen's appearance and whole vibe. It it is. It It feels nicer whole vibe. It it is. It It feels nicer whole vibe. It it is. It It feels nicer minus the Firefox parts. I remote a lot. minus the Firefox parts. I remote a lot. minus the Firefox parts. I remote a lot. I I do a lot of remote. I'll talk about I I do a lot of remote. I'll talk about I I do a lot of remote. I'll talk about that in some of the videos we're doing that in some of the videos we're doing that in some of the videos we're doing today. Let's figure out what we're today. Let's figure out what we're today. Let's figure out what we're filming. Hop over to Notion. Okay. So, we have to Hop over to Notion. Okay. So, we have to do the Cloudflare stuff. The anthropic do the Cloudflare stuff. The anthropic do the Cloudflare stuff. The anthropic pausing AI stuff. Compute Crunch. I have pausing AI stuff. Compute Crunch. I have pausing AI stuff. Compute Crunch. I have a lot of thoughts on.

a lot of thoughts on. a lot of thoughts on. I have been using Cloud Code more. So, I I have been using Cloud Code more. So, I I have been using Cloud Code more. So, I have thoughts here. What did these two have thoughts here. What did these two have thoughts here. What did these two add as well? Modern engineering values add as well? Modern engineering values add as well? Modern engineering values is Oh, CPO post. Cool. The dangerously is Oh, CPO post. Cool. The dangerously is Oh, CPO post. Cool. The dangerously skip is a really compelling one. Myarchy skip is a really compelling one. Myarchy skip is a really compelling one. Myarchy crash out. I I'm scared to do, but I am crash out. I I'm scared to do, but I am crash out. I I'm scared to do, but I am tempted. Anti-AII nostalgia. Ooh, it's a Sean Go. Anti-AII nostalgia. Ooh, it's a Sean Go. I love his articles. What do humans need from docs? This is What do humans need from docs? This is an interesting concept. I have been using tail scale a bit, but I have been using tail scale a bit, but other fun plans coming soon. That is a good start. Happy that that That is a good start. Happy that that happened. Happy that you're enjoying happened. Happy that you're enjoying happened. Happy that you're enjoying 9500s, Maria. They're very comfy and 9500s, Maria. They're very comfy and 9500s, Maria. They're very comfy and very useful.

Thank you, Damon Twink, for the Prime Thank you, Damon Twink, for the Prime and Shoe Boom. You used to get anxiety and Shoe Boom. You used to get anxiety and Shoe Boom. You used to get anxiety working on product ideas when you saw working on product ideas when you saw working on product ideas when you saw somebody else working on something somebody else working on something somebody else working on something similar and give up and see them still similar and give up and see them still similar and give up and see them still fumbling it months later. A lot of my fumbling it months later. A lot of my fumbling it months later. A lot of my streams give you the confidence to keep streams give you the confidence to keep streams give you the confidence to keep grinding. That's really cool to hear. I grinding. That's really cool to hear. I grinding. That's really cool to hear. I I think you'll like the talk I gave at I think you'll like the talk I gave at I think you'll like the talk I gave at Cascadia a lot. I I might just redo it Cascadia a lot. I I might just redo it Cascadia a lot. I I might just redo it as a video, but yeah, the now is the as a video, but yeah, the now is the as a video, but yeah, the now is the time to go further with your ideas. time to go further with your ideas. time to go further with your ideas. There's no reason to not. There's no reason to not. There's no reason to not. I've seen flu. I got early access to it. I've seen flu. I got early access to it. I've seen flu. I got early access to it. I haven't had a chance to actually build I haven't had a chance to actually build I haven't had a chance to actually build it or anything with it or play with it, it or anything with it or play with it, it or anything with it or play with it, but it seems pretty solid. but it seems pretty solid. but it seems pretty solid. Thank you, Al, for the sub, by the way. doing videos of my talks in person on doing videos of my talks in person on the channel outright. No edits. I did the channel outright. No edits. I did the channel outright. No edits. I did that once before, but probably not too m that once before, but probably not too m that once before, but probably not too m much. Like much. Like much. Like edits make the videos much more edits make the videos much more edits make the videos much more watchable, and the more watchable they watchable, and the more watchable they watchable, and the more watchable they are, the better they perform.

are, the better they perform. are, the better they perform. Ooh, here we go. This sounds very fun. I Ooh, here we go. This sounds very fun. I Ooh, here we go. This sounds very fun. I cannot wait to hear more about this. All cannot wait to hear more about this. All cannot wait to hear more about this. All my favorite things Oh, keep an eye on what you're cooking, Oh, keep an eye on what you're cooking, Michaela. You seem to be right in the Michaela. You seem to be right in the Michaela. You seem to be right in the the right spot. We need to hang more the right spot. We need to hang more the right spot. We need to hang more soon, by the way. Let's make sure next soon, by the way. Let's make sure next soon, by the way. Let's make sure next time you're in the city, we chill. Any more YouTube misses? Cool. We're Any more YouTube misses? Cool. We're good. Let's start. good. Let's start. good. Let's start. I think I'm going to start with the I think I'm going to start with the I think I'm going to start with the Cloudflare stuff because I have a lot of Cloudflare stuff because I have a lot of Cloudflare stuff because I have a lot of thoughts here.

Going default set Going default set proper. I'll look at that later. I'll look at that later. We got to work on this though. V and Cloudflare. V and Cloudflare. Thank you. Atomic for the tier one as Thank you. Atomic for the tier one as Thank you. Atomic for the tier one as well as Vi for the tier ones. Three, six well as Vi for the tier ones. Three, six well as Vi for the tier ones. Three, six months and four months respectively.

Let's do this. It's time Let's do this. It's time be a fun one. One sec. I got to double check One sec. I got to double check something. Cool. Yeah, we're ready to go. Nice. Cool. Yeah, we're ready to go. Nice. Let's start. Let's start. Let's start. Oh boy. I want the opening to be some comment on I want the opening to be some comment on how important V is. I'm trying to figure how important V is. I'm trying to figure how important V is. I'm trying to figure out the right way to to do that. Um, I don't think it's particularly I don't think it's particularly controversial to say that Vit's one of controversial to say that Vit's one of controversial to say that Vit's one of the most important tools for webdev the most important tools for webdev the most important tools for webdev nowadays. I don't like building projects nowadays. I don't like building projects nowadays. I don't like building projects without Vit as one of the core without Vit as one of the core without Vit as one of the core dependencies. It just makes building so dependencies. It just makes building so dependencies. It just makes building so much easier and more reliable and most much easier and more reliable and most much easier and more reliable and most importantly fast. Vit took over for a importantly fast. Vit took over for a importantly fast. Vit took over for a reason. There's a big reason. There's a big reason. There's a big V took over for a reason and it's mostly V took over for a reason and it's mostly V took over for a reason and it's mostly just because it's good. It's a really just because it's good. It's a really just because it's good. It's a really reliable primitive that we can build on

reliable primitive that we can build on reliable primitive that we can build on top of and the ecosystem that's been top of and the ecosystem that's been top of and the ecosystem that's been built around it through the void zero built around it through the void zero built around it through the void zero team is incredible. Things like Vest, roll down, oxe, vit Things like Vest, roll down, oxe, vit plus as of recent and more have made the plus as of recent and more have made the plus as of recent and more have made the vit ecosystem so compelling. In fact, it's become so compelling that In fact, it's become so compelling that re in fact it's become so compelling re in fact it's become so compelling re in fact it's become so compelling that another big company noticed. that another big company noticed. that another big company noticed. Cloudflare is now the owners of Void Cloudflare is now the owners of Void Cloudflare is now the owners of Void Zero, the company that Evan Yu, the Zero, the company that Evan Yu, the Zero, the company that Evan Yu, the creator of you invited to maintain all creator of you invited to maintain all creator of you invited to maintain all of these projects of these projects of these projects and as such and as such and as such and my guess is that their plan is to and my guess is that their plan is to and my guess is that their plan is to build a new cloud. I've been keeping an build a new cloud. I've been keeping an build a new cloud. I've been keeping an eye on the relationship between Void eye on the relationship between Void eye on the relationship between Void Zero and Cloudflare for a while. I think Zero and Cloudflare for a while. I think Zero and Cloudflare for a while. I think this acquisition makes a ton of sense, this acquisition makes a ton of sense, this acquisition makes a ton of sense, but in order for it to make sense, we but in order for it to make sense, we but in order for it to make sense, we need to think far beyond what VT itself need to think far beyond what VT itself need to think far beyond what VT itself is as a bundler and layer for building is as a bundler and layer for building is as a bundler and layer for building apps and think more about how agents are apps and think more about how agents are apps and think more about how agents are building things for the future and what building things for the future and what building things for the future and what the cloud will look like in a world the cloud will look like in a world the cloud will look like in a world where we're not the ones configuring it where we're not the ones configuring it where we're not the ones configuring it ourselves.

This also means that void zero is now a This also means that void zero is now a much scarier competitor to some of the much scarier competitor to some of the much scarier competitor to some of the fun things that I've been working on. fun things that I've been working on. fun things that I've been working on. And I'm very excited to lay out the And I'm very excited to lay out the And I'm very excited to lay out the This also means Void Zero is suddenly a This also means Void Zero is suddenly a This also means Void Zero is suddenly a much more compelling competitor to some much more compelling competitor to some much more compelling competitor to some of the fun things I've been building of the fun things I've been building of the fun things I've been building with my cloud recently, which also helps with my cloud recently, which also helps with my cloud recently, which also helps give me an interesting perspective on give me an interesting perspective on give me an interesting perspective on this whole change. I'm going to do my best to lay out what I'm going to do my best to lay out what Void Zero is. I'm going to do my best to Void Zero is. I'm going to do my best to Void Zero is. I'm going to do my best to lay out what the future looks like, not lay out what the future looks like, not lay out what the future looks like, not just for Void Zero and Cloudflare, but just for Void Zero and Cloudflare, but just for Void Zero and Cloudflare, but for building with agents on the cloud as for building with agents on the cloud as for building with agents on the cloud as a whole. But if I just lost a Cloud a whole. But if I just lost a Cloud a whole. But if I just lost a Cloud Flare, I'm going to need to make money Flare, I'm going to need to make money Flare, I'm going to need to make money somehow. And we're going to start with somehow. And we're going to start with somehow. And we're going to start with today's sponsor. Oh no. Let me figure out what's going on Oh no. Let me figure out what's going on there. Cool. In order to understand this void Cool. In order to understand this void zero v cloudflare merger. In order to zero v cloudflare merger. In order to zero v cloudflare merger. In order to understand the void zero v cloudflare understand the void zero v cloudflare understand the void zero v cloudflare merger we first need to understand a bit merger we first need to understand a bit merger we first need to understand a bit more about void zero and the direction more about void zero and the direction more about void zero and the direction it was going in.

As I mentioned before, void as I As I mentioned before, void as I mentioned before, void is the set of mentioned before, void is the set of mentioned before, void is the set of tools that ah while void zero was very well regarded while void zero was very well regarded for tools like oxc, vit and everything for tools like oxc, vit and everything for tools like oxc, vit and everything around it, their goals were much further around it, their goals were much further around it, their goals were much further down the line. Their hope was to build down the line. Their hope was to build down the line. Their hope was to build their new platform void as the place to their new platform void as the place to their new platform void as the place to deploy all the apps that you build with deploy all the apps that you build with deploy all the apps that you build with things like vit. They wanted it to be things like vit. They wanted it to be things like vit. They wanted it to be way easier to deploy an app. way easier to deploy an app. way easier to deploy an app. And I very much sympathize with this And I very much sympathize with this And I very much sympathize with this goal. I have a question for y'all. How goal. I have a question for y'all. How goal. I have a question for y'all. How many apps do you have sitting on your many apps do you have sitting on your many apps do you have sitting on your computer that are mostly working? Like computer that are mostly working? Like computer that are mostly working? Like they work fine on your machine that you they work fine on your machine that you they work fine on your machine that you just haven't deployed because that part just haven't deployed because that part just haven't deployed because that part takes too much effort. So the void employee says zero because So the void employee says zero because they're all on void. So many many six or seven all my So many many six or seven all my projects so far 22. You don't want to projects so far 22. You don't want to projects so far 22. You don't want to know at least 10 like 10. Yeah. Yeah. I know at least 10 like 10. Yeah. Yeah. I know at least 10 like 10. Yeah. Yeah. I get it. And there's a reason for this.

I'm going to grab some slides from my I'm going to grab some slides from my Cascadia talk because I talk about this Cascadia talk because I talk about this Cascadia talk because I talk about this in it. in it. in it. Here we are. Here we are. Here we are. In a very general sense, building a side In a very general sense, building a side In a very general sense, building a side project used to be a in a very very project used to be a in a very very project used to be a in a very very general sense. For me, building a side general sense. For me, building a side general sense. For me, building a side project, used to take like 40 to 100 project, used to take like 40 to 100 project, used to take like 40 to 100 hours to like spin it up and get it hours to like spin it up and get it hours to like spin it up and get it working how I wanted. Things like Marker working how I wanted. Things like Marker working how I wanted. Things like Marker Thing, the service that I built for Thing, the service that I built for Thing, the service that I built for tracking all of my markers when I'm tracking all of my markers when I'm tracking all of my markers when I'm streaming to make it easier to grab the streaming to make it easier to grab the streaming to make it easier to grab the clips from the stream and turn those clips from the stream and turn those clips from the stream and turn those into videos. This took me 30 to 40 hours into videos. This took me 30 to 40 hours into videos. This took me 30 to 40 hours to build and get working how I wanted it to build and get working how I wanted it to build and get working how I wanted it to. And I did that 5 years ago and it's to. And I did that 5 years ago and it's to. And I did that 5 years ago and it's worked perfectly every time we stream worked perfectly every time we stream worked perfectly every time we stream since. Super useful tool. might have since. Super useful tool. might have since. Super useful tool. might have taken me those 30 so or might have taken taken me those 30 so or might have taken taken me those 30 so or might have taken 30 or so hours to build but since I 30 or so hours to build but since I 30 or so hours to build but since I built that it has saved me way more time built that it has saved me way more time built that it has saved me way more time since since since building it probably took 30 or 40 hours building it probably took 30 or 40 hours building it probably took 30 or 40 hours deploying it probably took two or three deploying it probably took two or three deploying it probably took two or three like getting the cloud set up getting it like getting the cloud set up getting it like getting the cloud set up getting it plugged into ver getting the cloud set plugged into ver getting the cloud set plugged into ver getting the cloud set up getting it plugged into versel up getting it plugged into versel up getting it plugged into versel properly setting up the database setting properly setting up the database setting properly setting up the database setting up the o layer with clerk none of it was up the o layer with clerk none of it was up the o layer with clerk none of it was hard there was just a lot of those hard there was just a lot of those hard there was just a lot of those pieces it didn't feel too bad because it pieces it didn't feel too bad because it pieces it didn't feel too bad because it was like less than a tenth or so of the was like less than a tenth or so of the was like less than a tenth or so of the time but this was in an era before AI time but this was in an era before AI time but this was in an era before AI could write all the code. Now building could write all the code. Now building could write all the code. Now building has become way easier.

has become way easier. has become way easier. This lines up with my experience in a This lines up with my experience in a This lines up with my experience in a lot of projects now where actually lot of projects now where actually lot of projects now where actually building it and getting the code working building it and getting the code working building it and getting the code working how I want only takes a couple minutes, how I want only takes a couple minutes, how I want only takes a couple minutes, 30 or so at best and then deploying 30 or so at best and then deploying 30 or so at best and then deploying still takes the same 3 plus hours. still takes the same 3 plus hours. still takes the same 3 plus hours. Previously that was acceptical or Previously that was acceptical or Previously that was acceptical or previously that was acceptable because previously that was acceptable because previously that was acceptable because it was less than 10% of the time was it was less than 10% of the time was it was less than 10% of the time was spent doing the actual deployment stuff. spent doing the actual deployment stuff. spent doing the actual deployment stuff. Now it's a lot less acceptable because Now it's a lot less acceptable because Now it's a lot less acceptable because the deploying feels so much worse when the deploying feels so much worse when the deploying feels so much worse when everything else was way quicker to do. And that's why void zero or and that's And that's why void zero or and that's why the void platform existed. It was why the void platform existed. It was why the void platform existed. It was the goal of the team to build the the goal of the team to build the the goal of the team to build the fastest way to deploy your apps on real fastest way to deploy your apps on real fastest way to deploy your apps on real infrastructure. So once you build infrastructure. So once you build infrastructure. So once you build something with vit, you can run a simple something with vit, you can run a simple something with vit, you can run a simple command and get it out there for anyone command and get it out there for anyone command and get it out there for anyone to use.

They made it as simple as possible to They made it as simple as possible to use. You install the void plugin. You use. You install the void plugin. You use. You install the void plugin. You You install the void plugin. You add You install the void plugin. You add You install the void plugin. You add this as a plugin in your V config. And this as a plugin in your V config. And this as a plugin in your V config. And now you can run void deploy. And your now you can run void deploy. And your now you can run void deploy. And your app is now live. If you've been around app is now live. If you've been around app is now live. If you've been around for a bit, this might seem a little for a bit, this might seem a little for a bit, this might seem a little familiar. With most stacks, your app is with most With most stacks, your app is with most stacks. Your app and its platform do not stacks. Your app and its platform do not stacks. Your app and its platform do not talk to each other directly. You end up talk to each other directly. You end up talk to each other directly. You end up stitching them together with config stitching them together with config stitching them together with config files, environment setup, resource files, environment setup, resource files, environment setup, resource provisioning, and deployment scripts. provisioning, and deployment scripts. provisioning, and deployment scripts. Also, a lot of time in weird dashboards. Also, a lot of time in weird dashboards. Also, a lot of time in weird dashboards. Once the app is live, caching, scaling, Once the app is live, caching, scaling, Once the app is live, caching, scaling, and limits often live in separate and limits often live in separate and limits often live in separate control planes as well. Void closes the control planes as well. Void closes the control planes as well. Void closes the gap. It connects your app directly to gap. It connects your app directly to gap. It connects your app directly to the platform through Vit. You import the the platform through Vit. You import the the platform through Vit. You import the DB from void/DB and now you have a DB from void/DB and now you have a DB from void/DB and now you have a database in local dev as well as in the database in local dev as well as in the database in local dev as well as in the matching production resources on matching production resources on matching production resources on deployment. The same idea applies for deployment. The same idea applies for deployment. The same idea applies for KV, storage, cues, and AI. In most KV, storage, cues, and AI. In most KV, storage, cues, and AI. In most cases, the code already describes what cases, the code already describes what cases, the code already describes what the platform needs to provision. I absolutely agree with this mindset I absolutely agree with this mindset here. This is the code should be enough here. This is the code should be enough here. This is the code should be enough to tell the infrastructure what it needs to tell the infrastructure what it needs to tell the infrastructure what it needs to provision. And the fact that it isn't to provision. And the fact that it isn't to provision. And the fact that it isn't is insane. And it's time to to do this, is insane. And it's time to to do this, is insane. And it's time to to do this, right?

right? right? And it seems like people are confused And it seems like people are confused And it seems like people are confused about the 3-hour thing I'm saying here. about the 3-hour thing I'm saying here. about the 3-hour thing I'm saying here. I'm not saying that I've set up the I'm not saying that I've set up the I'm not saying that I've set up the deployment system and then I hit deploy deployment system and then I hit deploy deployment system and then I hit deploy and it takes three hours to do. If you and it takes three hours to do. If you and it takes three hours to do. If you think that's what I'm saying, I have think that's what I'm saying, I have think that's what I'm saying, I have questions about how good you are at questions about how good you are at questions about how good you are at listening. listening. listening. Regardless, the thing I'm trying to say Regardless, the thing I'm trying to say Regardless, the thing I'm trying to say here is when you start a new project, here is when you start a new project, here is when you start a new project, getting it built and working is easier getting it built and working is easier getting it built and working is easier than ever thanks to AI. Getting it than ever thanks to AI. Getting it than ever thanks to AI. Getting it deployed has not gotten easier thanks to deployed has not gotten easier thanks to deployed has not gotten easier thanks to AI. Once you're building with all the AI. Once you're building with all the AI. Once you're building with all the tools AI likes and it loves things like tools AI likes and it loves things like tools AI likes and it loves things like Convex and Versell, Convex and Versell, Convex and Versell, getting all of those parts linked getting all of those parts linked getting all of those parts linked together can take up to three hours to together can take up to three hours to together can take up to three hours to set up. set up. set up. Not that once you hit go, it takes 3 Not that once you hit go, it takes 3 Not that once you hit go, it takes 3 hours to deploy. Come on, guys. Let's be hours to deploy. Come on, guys. Let's be hours to deploy. Come on, guys. Let's be realistic here, but that's what Void is trying to solve but that's what Void is trying to solve as well. They want the code to describe as well. They want the code to describe as well. They want the code to describe all the things that are needed so that all the things that are needed so that all the things that are needed so that you don't have to go set up Terraform you don't have to go set up Terraform you don't have to go set up Terraform and configure a bunch of AWS dashboards and configure a bunch of AWS dashboards and configure a bunch of AWS dashboards and do all those pieces. And I think and do all those pieces. And I think and do all those pieces. And I think this is the right way to go. this is the right way to go. this is the right way to go. There were some interesting choices that There were some interesting choices that There were some interesting choices that the void team made though largely around the void team made though largely around the void team made though largely around how they integrate standards and what how they integrate standards and what how they integrate standards and what clouds they were building on top of. For clouds they were building on top of. For clouds they were building on top of. For example, when you define things for your example, when you define things for your example, when you define things for your database, you do it through your drizzle database, you do it through your drizzle database, you do it through your drizzle schema which defines database types.

And when you deploy Cloudflare worker or And when you deploy Cloudflare worker or And when you deploy void, it deploys to And when you deploy void, it deploys to And when you deploy void, it deploys to Cloudflare workers. Your server code Cloudflare workers. Your server code Cloudflare workers. Your server code runs at the edge close to users and runs at the edge close to users and runs at the edge close to users and scales without extra per platform work scales without extra per platform work scales without extra per platform work on your side. I understand why they did on your side. I understand why they did on your side. I understand why they did this. This makes a lot of sense, this. This makes a lot of sense, this. This makes a lot of sense, especially for people who already have especially for people who already have especially for people who already have workers set up or do it through void. workers set up or do it through void. workers set up or do it through void. The worker platform scales great for The worker platform scales great for The worker platform scales great for these types of workloads. It does have a these types of workloads. It does have a these types of workloads. It does have a lot of catches as well and we'll lot of catches as well and we'll lot of catches as well and we'll certainly talk about those. But I do certainly talk about those. But I do certainly talk about those. But I do think it was the right decision for Void think it was the right decision for Void think it was the right decision for Void to build this way, especially now that to build this way, especially now that to build this way, especially now that this especially now this acquisition has this especially now this acquisition has this especially now this acquisition has went through. I also just realized I forgot to hit I also just realized I forgot to hit record on the [ __ ] camera cuz I'm record on the [ __ ] camera cuz I'm record on the [ __ ] camera cuz I'm stupid. stupid. stupid. Sorry FaZe, you're not getting a camera Sorry FaZe, you're not getting a camera Sorry FaZe, you're not getting a camera footage for this one. I'll do the offset footage for this one. I'll do the offset footage for this one. I'll do the offset after. That's annoying. after. That's annoying. after. That's annoying. That's going to make all the asset That's going to make all the asset That's going to make all the asset management post stream really annoying. management post stream really annoying. management post stream really annoying. [ __ ] me. Void Zero is joining Cloudflare. Void Zero is joining Cloudflare. I'll just read through here. I'll just read through here. I'll just read through here. Now that we have some context, it's Now that we have some context, it's Now that we have some context, it's probably time to actually read the probably time to actually read the probably time to actually read the announcement. I'm sure there'll be lots announcement. I'm sure there'll be lots announcement. I'm sure there'll be lots of good info here. It's wild to think of good info here. It's wild to think of good info here. It's wild to think Evanu, the creator of Vue, is now an Evanu, the creator of Vue, is now an Evanu, the creator of Vue, is now an employee at Cloudflare, but it does make employee at Cloudflare, but it does make employee at Cloudflare, but it does make sense.

sense. sense. Voidzero, the company behind Vit, Vest, Voidzero, the company behind Vit, Vest, Voidzero, the company behind Vit, Vest, Rollown, OXC, and Vit Plus, is joining Rollown, OXC, and Vit Plus, is joining Rollown, OXC, and Vit Plus, is joining Cloudflare. As part of the change, all Cloudflare. As part of the change, all Cloudflare. As part of the change, all team members of Void Zero are joining team members of Void Zero are joining team members of Void Zero are joining Cloudflare 2. This is really good. This Cloudflare 2. This is really good. This Cloudflare 2. This is really good. This isn't like a traditional aqua hire. This isn't like a traditional aqua hire. This isn't like a traditional aqua hire. This is the business being folded in. is the business being folded in. is the business being folded in. Before saying anything else, we want to Before saying anything else, we want to Before saying anything else, we want to make the most important thing clear. make the most important thing clear. make the most important thing clear. Vit, V test, roll down, oxe, and vit Vit, V test, roll down, oxe, and vit Vit, V test, roll down, oxe, and vit plus will all stay open source, vendor plus will all stay open source, vendor plus will all stay open source, vendor agnostic, and communitydriven. Nothing agnostic, and communitydriven. Nothing agnostic, and communitydriven. Nothing about that changes. Cloudflare's mission is to help better. Cloudflare's mission is to help better. Cloudflare's mission is to help build a Cloudflare's mission is to help build a Cloudflare's mission is to help build a better internet. And a better internet better internet. And a better internet better internet. And a better internet is an open internet. Developers need is an open internet. Developers need is an open internet. Developers need choice. Frameworks need a neutral choice. Frameworks need a neutral choice. Frameworks need a neutral foundation and applications need to be foundation and applications need to be foundation and applications need to be portable. It is not reasonable to expect portable. It is not reasonable to expect portable. It is not reasonable to expect the entire web ecosystem to build around the entire web ecosystem to build around the entire web ecosystem to build around a single vendor. The most important a single vendor. The most important a single vendor. The most important tools and frameworks are portable by tools and frameworks are portable by tools and frameworks are portable by design. design. design. I want to talk about this for a sec I want to talk about this for a sec I want to talk about this for a sec before we keep reading cuz I have a hot before we keep reading cuz I have a hot before we keep reading cuz I have a hot take.

Every Cloudflare app is more tailored to Every Cloudflare app is more tailored to Cloudflare than Next.js apps are Cloudflare than Next.js apps are Cloudflare than Next.js apps are tailored to Verscell. I already can see tailored to Verscell. I already can see tailored to Verscell. I already can see the comment section from this. For some the comment section from this. For some the comment section from this. For some reason, whenever I say anything nice reason, whenever I say anything nice reason, whenever I say anything nice about Versel in contrast to Cloudflare, about Versel in contrast to Cloudflare, about Versel in contrast to Cloudflare, I get called a lot of homophobic slurs. I get called a lot of homophobic slurs. I get called a lot of homophobic slurs. So, uh, whoever's doing that in the So, uh, whoever's doing that in the So, uh, whoever's doing that in the comment section now, know that I see you comment section now, know that I see you comment section now, know that I see you and that nobody else does because I've and that nobody else does because I've and that nobody else does because I've already hidden you from the channel. already hidden you from the channel. already hidden you from the channel. Anyways, the point I'm trying to make Anyways, the point I'm trying to make Anyways, the point I'm trying to make here with this bold statement is that here with this bold statement is that here with this bold statement is that Nex.js apps don't have any platform Nex.js apps don't have any platform Nex.js apps don't have any platform specific code in them. specific code in them. specific code in them. Nex.js apps have Nex.js code and Nex.js apps have Nex.js code and Nex.js apps have Nex.js code and TypeScript code in them. TypeScript code in them. TypeScript code in them. Cloudflare apps need some configuration Cloudflare apps need some configuration Cloudflare apps need some configuration that is Cloudflare specific. I don't think I've ever had success I don't think I've ever had success deploying on Cloudflare without at least deploying on Cloudflare without at least deploying on Cloudflare without at least one annoying YAML file describing a one annoying YAML file describing a one annoying YAML file describing a bunch of [ __ ] as well as running bunch of [ __ ] as well as running bunch of [ __ ] as well as running something like Wrangler to pull something like Wrangler to pull something like Wrangler to pull everything together. everything together. everything together. When I'm building on when I'm building When I'm building on when I'm building When I'm building on when I'm building on Cloudflare, I'm not running the Vde on Cloudflare, I'm not running the Vde on Cloudflare, I'm not running the Vde dev command. I'm running the Wrangler dev command. I'm running the Wrangler dev command. I'm running the Wrangler dev command. When I am adding features dev command. When I am adding features dev command. When I am adding features to my apps on workers, I'm not doing it to my apps on workers, I'm not doing it to my apps on workers, I'm not doing it just by updating the code. I'm doing it just by updating the code. I'm doing it just by updating the code. I'm doing it by updating the configuration that is by updating the configuration that is by updating the configuration that is specific to Cloudflare.

This is one of the biggest weaknesses This is one of the biggest weaknesses for Cloudflare in my opinion. When I was for Cloudflare in my opinion. When I was for Cloudflare in my opinion. When I was trying to port T3 chat over to trying to port T3 chat over to trying to port T3 chat over to Cloudflare last year just for the Cloudflare last year just for the Cloudflare last year just for the differences in compute costs before differences in compute costs before differences in compute costs before Versell had fluid. By the way, when I Versell had fluid. By the way, when I Versell had fluid. By the way, when I was working on that port, I had to spend was working on that port, I had to spend was working on that port, I had to spend so much time simply trying to get an app so much time simply trying to get an app so much time simply trying to get an app that would serve static content by that would serve static content by that would serve static content by default for all routes except for the default for all routes except for the default for all routes except for the ones that are being served through the ones that are being served through the ones that are being served through the API. So I could have one project and one API. So I could have one project and one API. So I could have one project and one instance of a worker as well as the instance of a worker as well as the instance of a worker as well as the static asset host from Cloudflare sites. static asset host from Cloudflare sites. static asset host from Cloudflare sites. So everything would just be this one So everything would just be this one So everything would just be this one repo deployed in one simple way to repo deployed in one simple way to repo deployed in one simple way to Cloudflare. Cloudflare. Cloudflare. I had to go through like 15 layers of I had to go through like 15 layers of I had to go through like 15 layers of [ __ ] hacks and multiple calls with [ __ ] hacks and multiple calls with [ __ ] hacks and multiple calls with my friends at Cloudflare to figure out my friends at Cloudflare to figure out my friends at Cloudflare to figure out how to work around all the [ __ ] how to work around all the [ __ ] how to work around all the [ __ ] There ended up being like three There ended up being like three There ended up being like three overrides that conflicted with each overrides that conflicted with each overrides that conflicted with each other that I had to stack on top of each other that I had to stack on top of each other that I had to stack on top of each other just to get a basic V react app other just to get a basic V react app other just to get a basic V react app that was serving Hano endpoints for the that was serving Hano endpoints for the that was serving Hano endpoints for the SL API routes and serving an actual SL API routes and serving an actual SL API routes and serving an actual JavaScript bundle and HTML page for the JavaScript bundle and HTML page for the JavaScript bundle and HTML page for the rest. It was so bad that Cloudflare rest. It was so bad that Cloudflare rest. It was so bad that Cloudflare actually went out and built better actually went out and built better actually went out and built better templates to showcase how to do this templates to showcase how to do this templates to showcase how to do this because it was impossible to get right because it was impossible to get right because it was impossible to get right at the time without a lot of work and at the time without a lot of work and at the time without a lot of work and effort and getting through docs that effort and getting through docs that effort and getting through docs that were contradicting with each other. You were contradicting with each other. You were contradicting with each other. You know how I do this on Verscell know how I do this on Verscell know how I do this on Verscell MPX create next app. I go to Verscell. I MPX create next app. I go to Verscell. I MPX create next app. I go to Verscell. I click the GitHub repo and now it's on click the GitHub repo and now it's on click the GitHub repo and now it's on Verscell. Cloudflare's infrastructure is Verscell. Cloudflare's infrastructure is Verscell. Cloudflare's infrastructure is better in so many ways.

better in so many ways. better in so many ways. The problem with Cloudflare was never The problem with Cloudflare was never The problem with Cloudflare was never the capabilities of the infra. Okay, the capabilities of the infra. Okay, the capabilities of the infra. Okay, there's some issue there with like not there's some issue there with like not there's some issue there with like not having proper node support because having proper node support because having proper node support because they're running in isolates, but it's they're running in isolates, but it's they're running in isolates, but it's not that big a deal for a lot of not that big a deal for a lot of not that big a deal for a lot of workloads. The biggest problem I and workloads. The biggest problem I and workloads. The biggest problem I and many others have had with Cloudflare is many others have had with Cloudflare is many others have had with Cloudflare is that it is a different thing to get that it is a different thing to get that it is a different thing to get deployed on Cloudflare than to build deployed on Cloudflare than to build deployed on Cloudflare than to build your apps. The gap between I built this your apps. The gap between I built this your apps. The gap between I built this thing and it works on my computer and thing and it works on my computer and thing and it works on my computer and it's now running on Cloudflare is far it's now running on Cloudflare is far it's now running on Cloudflare is far too big and involves you basically too big and involves you basically too big and involves you basically running Cloudflare's infra on your running Cloudflare's infra on your running Cloudflare's infra on your machine through tools like Wrangler. That was bad. That was bad. And the it was so bad that Cloudflare And the it was so bad that Cloudflare And the it was so bad that Cloudflare wasn't even building their Terraform wasn't even building their Terraform wasn't even building their Terraform configs themselves. We know that because configs themselves. We know that because configs themselves. We know that because we were using Cloudflare for upload we were using Cloudflare for upload we were using Cloudflare for upload thing and getting the most recent thing and getting the most recent thing and getting the most recent features we needed out of Cloudflare on features we needed out of Cloudflare on features we needed out of Cloudflare on upload thing was impossible for us upload thing was impossible for us upload thing was impossible for us because their V6 Terraform config had because their V6 Terraform config had because their V6 Terraform config had workers working and routes working but workers working and routes working but workers working and routes working but didn't have the new pooling feature we didn't have the new pooling feature we didn't have the new pooling feature we were using working. V7 had the pooling were using working. V7 had the pooling were using working. V7 had the pooling feature as well as durable objects we feature as well as durable objects we feature as well as durable objects we wanted to use, but it didn't have routes wanted to use, but it didn't have routes wanted to use, but it didn't have routes or workers in it because these because or workers in it because these because or workers in it because these because the Terraform configs and the Terraform the Terraform configs and the Terraform the Terraform configs and the Terraform plugin was being built by a third party plugin was being built by a third party plugin was being built by a third party company called Stainless I have a lot of company called Stainless I have a lot of company called Stainless I have a lot of grudges with.

Speaking of which, a company so Speaking of which, a company so incompetent as to provide a Terraform incompetent as to provide a Terraform incompetent as to provide a Terraform config to Cloudflare that's missing most config to Cloudflare that's missing most config to Cloudflare that's missing most of Cloudflare's core features clearly of Cloudflare's core features clearly of Cloudflare's core features clearly doesn't know infrastructure really well, doesn't know infrastructure really well, doesn't know infrastructure really well, right? right? right? Seems like they found a good home with Seems like they found a good home with Seems like they found a good home with Anthropic. Anyways, the point I'm trying Anthropic. Anyways, the point I'm trying Anthropic. Anyways, the point I'm trying to make is that Cloudflare is so bad at to make is that Cloudflare is so bad at to make is that Cloudflare is so bad at dev tooling, so egregiously bad at dev dev tooling, so egregiously bad at dev dev tooling, so egregiously bad at dev tooling tooling tooling that there was no world in which they that there was no world in which they that there was no world in which they could catch up by themselves. they just could catch up by themselves. they just could catch up by themselves. they just couldn't. couldn't. couldn't. If you don't think Cloudflare's If you don't think Cloudflare's If you don't think Cloudflare's developer experience is [ __ ] you developer experience is [ __ ] you developer experience is [ __ ] you haven't had a good developer experience haven't had a good developer experience haven't had a good developer experience yet. There's a reason that they poach yet. There's a reason that they poach yet. There's a reason that they poach all these employees from Verscell. all these employees from Verscell. all these employees from Verscell. There's a reason that they're acquiring There's a reason that they're acquiring There's a reason that they're acquiring companies like Astro and now Void Zero. companies like Astro and now Void Zero. companies like Astro and now Void Zero. They know they can't build good DX, but They know they can't build good DX, but They know they can't build good DX, but they also know they can build they also know they can build they also know they can build unbelievably powerful infrastructure and unbelievably powerful infrastructure and unbelievably powerful infrastructure and they're looking for the help to get they're looking for the help to get they're looking for the help to get their DX closer to where something like their DX closer to where something like their DX closer to where something like Versel is or something like Railway is Versel is or something like Railway is Versel is or something like Railway is because right now deploying on because right now deploying on because right now deploying on Cloudflare requires knowing a lot about Cloudflare requires knowing a lot about Cloudflare requires knowing a lot about Cloudflare and doing a lot of Cloudflare Cloudflare and doing a lot of Cloudflare Cloudflare and doing a lot of Cloudflare for specific [ __ ] So going back to void, what here is So going back to void, what here is Cloudflare specific.

Cloudflare specific. Cloudflare specific. This is how you set up void in order to This is how you set up void in order to This is how you set up void in order to use the void ecosystem stuff. And as use the void ecosystem stuff. And as use the void ecosystem stuff. And as Alex mentioned in chat earlier, you can Alex mentioned in chat earlier, you can Alex mentioned in chat earlier, you can deploy on void without doing any of this deploy on void without doing any of this deploy on void without doing any of this if you're just deploying a single page if you're just deploying a single page if you're just deploying a single page app. app. app. Because again, they're trying to solve Because again, they're trying to solve Because again, they're trying to solve the problem of code just being code and the problem of code just being code and the problem of code just being code and deployment just being deployment and not deployment just being deployment and not deployment just being deployment and not having to do all of this crazy work to having to do all of this crazy work to having to do all of this crazy work to rebuild your app and its different rebuild your app and its different rebuild your app and its different bundles and pieces, different places bundles and pieces, different places bundles and pieces, different places just to get stuff working properly on just to get stuff working properly on just to get stuff working properly on Cloudflare. Cloudflare. Cloudflare. Void has been building this better Void has been building this better Void has been building this better abstraction on top of Cloudflare's abstraction on top of Cloudflare's abstraction on top of Cloudflare's infrastructure to make it as easy to infrastructure to make it as easy to infrastructure to make it as easy to deploy on as a platform like Verscell deploy on as a platform like Verscell deploy on as a platform like Verscell is. There is an issue with versel though is. There is an issue with versel though is. There is an issue with versel though and this is why this makes so much and this is why this makes so much and this is why this makes so much sense. In order to describe this I need a In order to describe this I need a spectrum. This is a this is meant to be spectrum. This is a this is meant to be spectrum. This is a this is meant to be the breath of concerns that a given the breath of concerns that a given the breath of concerns that a given application development platform would application development platform would application development platform would have on this spectrum. You have stuff like on this spectrum. You have stuff like CSS libraries.

You have stuff like You have stuff like You have stuff like You have stuff like front-end frameworks. You have stuff like compute You have stuff like compute in the sense of like the thing that your in the sense of like the thing that your in the sense of like the thing that your code runs on on the server side. And you code runs on on the server side. And you code runs on on the server side. And you have things like your database. have things like your database. have things like your database. This is a really rough idea of like the This is a really rough idea of like the This is a really rough idea of like the spectrum of things you have to think spectrum of things you have to think spectrum of things you have to think about as a platform providing full stack about as a platform providing full stack about as a platform providing full stack application hosting. application hosting. application hosting. Companies like Verscell focused in from Companies like Verscell focused in from Companies like Verscell focused in from here to here roughly and more to here here to here roughly and more to here here to here roughly and more to here where Verscell saw issues with people where Verscell saw issues with people where Verscell saw issues with people who were trying to build apps with React who were trying to build apps with React who were trying to build apps with React and they realized they could make a and they realized they could make a and they realized they could make a better bundler, a better framework better bundler, a better framework better bundler, a better framework around React with Nex.js and solve a lot around React with Nex.js and solve a lot around React with Nex.js and solve a lot of those problems to make it easier to of those problems to make it easier to of those problems to make it easier to build apps with React. But then they build apps with React. But then they build apps with React. But then they realized that deploying the apps became realized that deploying the apps became realized that deploying the apps became the hard part. They went a little the hard part. They went a little the hard part. They went a little further introducing CDNs and compute further introducing CDNs and compute further introducing CDNs and compute stuff with a company that we might stuff with a company that we might stuff with a company that we might remember if you were around in the days remember if you were around in the days remember if you were around in the days of of of it wasn't Gist, it was um Zeit. Yeah, if it wasn't Gist, it was um Zeit. Yeah, if it wasn't Gist, it was um Zeit. Yeah, if you were around for the days of Zeites, you were around for the days of Zeites, you were around for the days of Zeites, I hope you have good life insurance I hope you have good life insurance I hope you have good life insurance because we're getting old now guys.

because we're getting old now guys. because we're getting old now guys. But the point of Versel was to extend But the point of Versel was to extend But the point of Versel was to extend from this. from this. from this. Let me do a Let me do a Let me do a But the goal of our cell was to extend But the goal of our cell was to extend But the goal of our cell was to extend this range to take from where React went this range to take from where React went this range to take from where React went and expand further to the right until and expand further to the right until and expand further to the right until enough was covered that you had a good enough was covered that you had a good enough was covered that you had a good way to deploy your web framework. There's a problem here though, and the There's a problem here though, and the problem isn't that they're not going far problem isn't that they're not going far problem isn't that they're not going far enough into the CSS side. That's enough into the CSS side. That's enough into the CSS side. That's probably why they bought Chad CNN. The probably why they bought Chad CNN. The probably why they bought Chad CNN. The problem is the other direction. What problem is the other direction. What problem is the other direction. What happens when Verscell tries to go happens when Verscell tries to go happens when Verscell tries to go further, right? Because I don't have a further, right? Because I don't have a further, right? Because I don't have a good way to set up O when I deploy on good way to set up O when I deploy on good way to set up O when I deploy on Verscell or to set up a database when I Verscell or to set up a database when I Verscell or to set up a database when I deploy on Versel. I need to go to deploy on Versel. I need to go to deploy on Versel. I need to go to companies like Clerk or Work OS to get O companies like Clerk or Work OS to get O companies like Clerk or Work OS to get O working, right? And I need to go to working, right? And I need to go to working, right? And I need to go to companies like Convex, Planet Scale or companies like Convex, Planet Scale or companies like Convex, Planet Scale or others in order to get the database all others in order to get the database all others in order to get the database all set up. And linking all the pieces set up. And linking all the pieces set up. And linking all the pieces together quickly becomes the most together quickly becomes the most together quickly becomes the most annoying part when you're building lots annoying part when you're building lots annoying part when you're building lots of different apps. of different apps. of different apps. While Verscell did an incredible job While Verscell did an incredible job While Verscell did an incredible job solving from framework down to compute, solving from framework down to compute, solving from framework down to compute, they stopped there and their focus was they stopped there and their focus was they stopped there and their focus was integrations from this point forward.

integrations from this point forward. integrations from this point forward. They did try a little bit going or they They did try a little bit going or they They did try a little bit going or they did try to go further for a little bit. did try to go further for a little bit. did try to go further for a little bit. Things that I'm sure we all love like Things that I'm sure we all love like Things that I'm sure we all love like Verscell Postgress Verscell Postgress Verscell Postgress that they have since deprecated. If that they have since deprecated. If that they have since deprecated. If you're not familiar, Verscell tried you're not familiar, Verscell tried you're not familiar, Verscell tried white labeling. If you're not familiar, white labeling. If you're not familiar, white labeling. If you're not familiar, Verscell back in the day tried white Verscell back in the day tried white Verscell back in the day tried white labeling Neon as a Verscell Postgress. labeling Neon as a Verscell Postgress. labeling Neon as a Verscell Postgress. didn't go great for them and now they didn't go great for them and now they didn't go great for them and now they just lean into the marketplace where you just lean into the marketplace where you just lean into the marketplace where you can bring in other cloud platforms for can bring in other cloud platforms for can bring in other cloud platforms for your database for your off etc. your database for your off etc. your database for your off etc. But Versel is pretty squarely between But Versel is pretty squarely between But Versel is pretty squarely between this range. Cloudflare is an interesting company Cloudflare is an interesting company because their range because their range because their range is strongest between the CDN and is strongest between the CDN and is strongest between the CDN and compute. like they are really really compute. like they are really really compute. like they are really really good from here to here, but they're good from here to here, but they're good from here to here, but they're starting to expand into databases with starting to expand into databases with starting to expand into databases with platforms like D1, which are admittedly platforms like D1, which are admittedly platforms like D1, which are admittedly not as strong as other dedicated not as strong as other dedicated not as strong as other dedicated database offerings, but they work. database offerings, but they work. database offerings, but they work. They're decent. I know people who use They're decent. I know people who use They're decent. I know people who use them. They don't love them, but they're them. They don't love them, but they're them. They don't love them, but they're fine with them. Cloudflare has extended fine with them. Cloudflare has extended fine with them. Cloudflare has extended past where Versel went by going into past where Versel went by going into past where Versel went by going into things like databases, but they also things like databases, but they also things like databases, but they also don't have much the other direction.

don't have much the other direction. don't have much the other direction. They don't have bundlers that work great They don't have bundlers that work great They don't have bundlers that work great on Cloudflare. kind of have to finagle on Cloudflare. kind of have to finagle on Cloudflare. kind of have to finagle your bundler to work right for your bundler to work right for your bundler to work right for Cloudflare. They don't have a framework Cloudflare. They don't have a framework Cloudflare. They don't have a framework and I'm sure they've considered building and I'm sure they've considered building and I'm sure they've considered building something like there was never really a something like there was never really a something like there was never really a Rails for Cloudflare the way that Rails for Cloudflare the way that Rails for Cloudflare the way that there's Nex.js on Verscell. there's Nex.js on Verscell. there's Nex.js on Verscell. They acquired Hano the minimal HTTP They acquired Hano the minimal HTTP They acquired Hano the minimal HTTP routing library that's great for routing library that's great for routing library that's great for spinning up servers to help on that side spinning up servers to help on that side spinning up servers to help on that side a bit, but it didn't really solve the a bit, but it didn't really solve the a bit, but it didn't really solve the problem of building full stack apps. So, problem of building full stack apps. So, problem of building full stack apps. So, they made Cloudflare sites which kind of they made Cloudflare sites which kind of they made Cloudflare sites which kind of worked, but they've since deprecated. worked, but they've since deprecated. worked, but they've since deprecated. Sorry, Cloudflare pages, not sites. I Sorry, Cloudflare pages, not sites. I Sorry, Cloudflare pages, not sites. I need to get the name right. It doesn't need to get the name right. It doesn't need to get the name right. It doesn't really matter cuz it's dead, but you get really matter cuz it's dead, but you get really matter cuz it's dead, but you get the idea. Cloudflare is much further the idea. Cloudflare is much further the idea. Cloudflare is much further skewed right here, which means that all skewed right here, which means that all skewed right here, which means that all of the tools have to be customized to of the tools have to be customized to of the tools have to be customized to work on Cloudflare. Versel owns enough work on Cloudflare. Versel owns enough work on Cloudflare. Versel owns enough tooling and enough of the compute layer tooling and enough of the compute layer tooling and enough of the compute layer to make a good experience here.

What would it look like to cover the What would it look like to cover the whole range here? What would it look whole range here? What would it look whole range here? What would it look like to build everything end to end so like to build everything end to end so like to build everything end to end so that you get to own it all? that you get to own it all? that you get to own it all? This seems to be what void zero is going This seems to be what void zero is going This seems to be what void zero is going for now as part of Cloudflare. for now as part of Cloudflare. for now as part of Cloudflare. But I have a lot of insight here because But I have a lot of insight here because But I have a lot of insight here because of a project I've been working on for a of a project I've been working on for a of a project I've been working on for a bit. bit. bit. You all might know it of you all might You all might know it of you all might You all might know it of you all might have heard about the T3 cloud, but have heard about the T3 cloud, but have heard about the T3 cloud, but that's not what it's going to be called that's not what it's going to be called that's not what it's going to be called because now it's time to talk a little because now it's time to talk a little because now it's time to talk a little Now it's time to talk a little bit about Now it's time to talk a little bit about Now it's time to talk a little bit about Lakebed. Well, it will be time in just a Lakebed. Well, it will be time in just a Lakebed. Well, it will be time in just a sec because Lake Bed might be dead as a sec because Lake Bed might be dead as a sec because Lake Bed might be dead as a result of this acquisition and I got result of this acquisition and I got result of this acquisition and I got bills to pay. So, quick sponsor break bills to pay. So, quick sponsor break bills to pay. So, quick sponsor break and then we'll talk about Lake Bed. Okay. Apparently, Cloud didn't acquire Okay. Apparently, Cloud didn't acquire Hana. So, I'm going to delete this and Hana. So, I'm going to delete this and Hana. So, I'm going to delete this and then when I'm talking about that phase, then when I'm talking about that phase, then when I'm talking about that phase, just insert just insert just insert and delete that and do the arrow back and delete that and do the arrow back and delete that and do the arrow back here. Okay, so this is roughly where we here. Okay, so this is roughly where we here. Okay, so this is roughly where we were when that happened. So, uh I'm sure were when that happened. So, uh I'm sure were when that happened. So, uh I'm sure you can cut this naturally phase when I you can cut this naturally phase when I you can cut this naturally phase when I talk about Hano. Okay, they technically talk about Hano. Okay, they technically talk about Hano. Okay, they technically don't own Hano. They employ the creator don't own Hano. They employ the creator don't own Hano. They employ the creator of Hano and he works closely with of Hano and he works closely with of Hano and he works closely with Cloudflare to make Hano work great. So, Cloudflare to make Hano work great. So, Cloudflare to make Hano work great. So, not the same thing, but worth noting.

not the same thing, but worth noting. not the same thing, but worth noting. Anyways, Anyways, Anyways, cool. It's time to talk a little bit about It's time to talk a little bit about Lake Bed. I want to be clear, Lake Bed's Lake Bed. I want to be clear, Lake Bed's Lake Bed. I want to be clear, Lake Bed's not real yet. It's not out yet, but it's a good lens to understand these but it's a good lens to understand these things through. things through. things through. There are two truths we need to agree There are two truths we need to agree There are two truths we need to agree upon before going further in this upon before going further in this upon before going further in this conversation makes sense. Truth. One is that agents are better at Truth. One is that agents are better at writing code than they are at navigating writing code than they are at navigating writing code than they are at navigating bad dashboards. I hope we can all agree bad dashboards. I hope we can all agree bad dashboards. I hope we can all agree on this. Whether or not you like AI on this. Whether or not you like AI on this. Whether or not you like AI code, I don't care. They're pretty code, I don't care. They're pretty code, I don't care. They're pretty decent at it.

decent at it. decent at it. But when it comes to actually navigating But when it comes to actually navigating But when it comes to actually navigating a dashboard like Cloudflare or GCP to go a dashboard like Cloudflare or GCP to go a dashboard like Cloudflare or GCP to go configure things for you, nope. Not configure things for you, nope. Not configure things for you, nope. Not great. great. great. Hopefully, we all agree on that. There's a lot of big projects that were There's a lot of big projects that were not worth building before AI. not worth building before AI. not worth building before AI. There's two aspects to this one. There's two aspects to this one. There's two aspects to this one. The first aspect is that building The The first aspect is that building The The first aspect is that building The first aspect is that the reasons to first aspect is that the reasons to first aspect is that the reasons to build these things before AI were not as build these things before AI were not as build these things before AI were not as big because the benefits to a dev not big because the benefits to a dev not big because the benefits to a dev not having to leave their editor are smaller having to leave their editor are smaller having to leave their editor are smaller than an agent not having to traverse the than an agent not having to traverse the than an agent not having to traverse the network and do screen capture and try to network and do screen capture and try to network and do screen capture and try to click buttons for you. When a human just click buttons for you. When a human just click buttons for you. When a human just has to command tab between when a human has to command tab between when a human has to command tab between when a human just had to command tab from VS Code just had to command tab from VS Code just had to command tab from VS Code over to the browser to click a couple over to the browser to click a couple over to the browser to click a couple buttons, it wasn't too big a deal. I buttons, it wasn't too big a deal. I buttons, it wasn't too big a deal. I hated it. I always hated that. That's a hated it. I always hated that. That's a hated it. I always hated that. That's a big part of why I like platforms like big part of why I like platforms like big part of why I like platforms like Verscell. I liked doing everything Verscell. I liked doing everything Verscell. I liked doing everything through code.

through code. through code. So there was much less incentive to do So there was much less incentive to do So there was much less incentive to do this when devs just had to tab out of this when devs just had to tab out of this when devs just had to tab out of their editor. There's more incentive now their editor. There's more incentive now their editor. There's more incentive now that AI is going to struggle a lot more that AI is going to struggle a lot more that AI is going to struggle a lot more to do that. But the bigger reason, and to do that. But the bigger reason, and to do that. But the bigger reason, and this is the super controversial one, it's time to boil the ocean. it's time to boil the ocean. The phrase don't boil the ocean has the The phrase don't boil the ocean has the The phrase don't boil the ocean has the phrase don't boil the ocean is always phrase don't boil the ocean is always phrase don't boil the ocean is always meant to indicate that there are some meant to indicate that there are some meant to indicate that there are some things that are just too big to be worth things that are just too big to be worth things that are just too big to be worth doing and it's not worth putting the doing and it's not worth putting the doing and it's not worth putting the effort in because it's silly and effort in because it's silly and effort in because it's silly and excessive. Suddenly it makes sense to shave the Suddenly it makes sense to shave the yak. It makes sense to boil the ocean yak. It makes sense to boil the ocean yak. It makes sense to boil the ocean because agents can do the tedious because agents can do the tedious because agents can do the tedious [ __ ] Projects that didn't make [ __ ] Projects that didn't make [ __ ] Projects that didn't make sense because they were too big before sense because they were too big before sense because they were too big before suddenly make way more sense. So a suddenly make way more sense. So a suddenly make way more sense. So a project as bold as rebuilding all of the project as bold as rebuilding all of the project as bold as rebuilding all of the features Cloudflare offers in a new SDK features Cloudflare offers in a new SDK features Cloudflare offers in a new SDK that is hidden to your agents so they that is hidden to your agents so they that is hidden to your agents so they can just call it the way they call can just call it the way they call can just call it the way they call anything in a TypeScript project would anything in a TypeScript project would anything in a TypeScript project would not have made a lot of sense before.

not have made a lot of sense before. not have made a lot of sense before. But now that there's incentive for this But now that there's incentive for this But now that there's incentive for this because agents benefit from it and because agents benefit from it and because agents benefit from it and there's a reason to or and there's the there's a reason to or and there's the there's a reason to or and there's the ability to write infinite code to go ability to write infinite code to go ability to write infinite code to go retake on that bit. Now that we have retake on that bit. Now that we have retake on that bit. Now that we have both this new customer which is agents both this new customer which is agents both this new customer which is agents that benefit greatly from these that benefit greatly from these that benefit greatly from these workloads as well as the ability to workloads as well as the ability to workloads as well as the ability to write way more code than ever using write way more code than ever using write way more code than ever using agents. The incentive to build something agents. The incentive to build something agents. The incentive to build something like that that has to integrate with like that that has to integrate with like that that has to integrate with everything ever has gone up a ton. everything ever has gone up a ton. everything ever has gone up a ton. But you can go further. And I can't But you can go further. And I can't But you can go further. And I can't believe I'm saying this to Void Zero, believe I'm saying this to Void Zero, believe I'm saying this to Void Zero, the company that literally built the company that literally built the company that literally built everything that we rely on nowadays with everything that we rely on nowadays with everything that we rely on nowadays with V, Vit Plus, Vest, Roll Down, OXC, and V, Vit Plus, Vest, Roll Down, OXC, and V, Vit Plus, Vest, Roll Down, OXC, and Evanu, the creator of Vue and the whole Evanu, the creator of Vue and the whole Evanu, the creator of Vue and the whole Vue ecosystem. Like, for those people to Vue ecosystem. Like, for those people to Vue ecosystem. Like, for those people to be told they're not getting bold enough be told they're not getting bold enough be told they're not getting bold enough is insane and stupid. But I don't mind is insane and stupid. But I don't mind is insane and stupid. But I don't mind being insane and stupid. So, let me be being insane and stupid. So, let me be being insane and stupid. So, let me be insane and stupid for a minute insane and stupid for a minute insane and stupid for a minute because Lakebed is all of these things. because Lakebed is all of these things. because Lakebed is all of these things. To be clear, I am leaning on existing To be clear, I am leaning on existing To be clear, I am leaning on existing solutions for a lot of them. Like my CSS solutions for a lot of them. Like my CSS solutions for a lot of them. Like my CSS library for now is just Tailwind using library for now is just Tailwind using library for now is just Tailwind using the Tailwind CDN host. I'm going to the Tailwind CDN host. I'm going to the Tailwind CDN host. I'm going to centralize that in the future, but I centralize that in the future, but I centralize that in the future, but I haven't yet. The framework is a very haven't yet. The framework is a very haven't yet. The framework is a very light fork of Pact. I'm probably going light fork of Pact. I'm probably going light fork of Pact. I'm probably going to fork it more heavily in the future.

to fork it more heavily in the future. to fork it more heavily in the future. The bundler right now is ESB build, but The bundler right now is ESB build, but The bundler right now is ESB build, but that's going to be the thing I target that's going to be the thing I target that's going to be the thing I target the most aggressively. I'm already the most aggressively. I'm already the most aggressively. I'm already starting to make some changes there. starting to make some changes there. starting to make some changes there. The CDN, comput, and database are all The CDN, comput, and database are all The CDN, comput, and database are all shims on top of other solutions, too. shims on top of other solutions, too. shims on top of other solutions, too. But that could change at any point. The But that could change at any point. The But that could change at any point. The reason I'm building this isn't because I reason I'm building this isn't because I reason I'm building this isn't because I think CDNs are bad and need to be think CDNs are bad and need to be think CDNs are bad and need to be reinvented. It's that I want an ex or is reinvented. It's that I want an ex or is reinvented. It's that I want an ex or is because I want a abstraction, a layer, because I want a abstraction, a layer, because I want a abstraction, a layer, an SDK, so to speak, where everything an SDK, so to speak, where everything an SDK, so to speak, where everything can be done by an agent in a project can be done by an agent in a project can be done by an agent in a project without having to run a CLI or npm without having to run a CLI or npm without having to run a CLI or npm install 100 [ __ ] things. And that's install 100 [ __ ] things. And that's install 100 [ __ ] things. And that's why I built Lake Bed Let me come up with a good name for Let me come up with a good name for this. I'll do a quick demo because this is I'll do a quick demo because this is hard to understand otherwise.

hard to understand otherwise. hard to understand otherwise. Actually, I have one more thing I want Actually, I have one more thing I want Actually, I have one more thing I want to do before MPX Lake bed off log out. to do before MPX Lake bed off log out. to do before MPX Lake bed off log out. Cool. This is hard to explain. It's much This is hard to explain. It's much easier to just show. So, I'm going to easier to just show. So, I'm going to easier to just show. So, I'm going to just show just show just show I'm running the command npx lake bed I'm running the command npx lake bed I'm running the command npx lake bed new. I'm going to make a new project. new. I'm going to make a new project. new. I'm going to make a new project. We're going to call it if only cuz if We're going to call it if only cuz if We're going to call it if only cuz if only the cloud was this easy to use. only the cloud was this easy to use. only the cloud was this easy to use. Just ran. Now the project exists. We'll Just ran. Now the project exists. We'll Just ran. Now the project exists. We'll cd in. And I want to be clear before I cd in. And I want to be clear before I cd in. And I want to be clear before I do this next step. I don't have anything do this next step. I don't have anything do this next step. I don't have anything special configured on this machine. This special configured on this machine. This special configured on this machine. This would work in a stock node box in the would work in a stock node box in the would work in a stock node box in the cloud with no off or anything. cloud with no off or anything. cloud with no off or anything. MPX Lake deploy. MPX Lake deploy. MPX Lake deploy. This is now a real app. Oops. This is now a real app. Oops. This is now a real app. Oops. Why is that doing that? Why is that opening like that? That's Why is that opening like that? That's annoying. This is now a real app that is annoying. This is now a real app that is annoying. This is now a real app that is live live live with a real database backing it. Real with a real database backing it. Real with a real database backing it. Real off that works. And and of course you off that works. And and of course you off that works. And and of course you guys know me. Sync works too. I am what guys know me. Sync works too. I am what guys know me. Sync works too. I am what I am. People already coming in from I am. People already coming in from I am. People already coming in from chat.

This is a real project that you can do This is a real project that you can do whatever you want to with queries, whatever you want to with queries, whatever you want to with queries, mutations, endpoints, all defined with a mutations, endpoints, all defined with a mutations, endpoints, all defined with a standard syntax not too dissimilar from standard syntax not too dissimilar from standard syntax not too dissimilar from what was built by our friends over at what was built by our friends over at what was built by our friends over at Convex. Obviously very inspired by Convex. Obviously very inspired by Convex. Obviously very inspired by Convex. What the [ __ ] happened to my semox? What the [ __ ] happened to my semox? What's going on with that top bar? What What's going on with that top bar? What What's going on with that top bar? What the [ __ ] Mac OS. But where things get much cooler But where things get much cooler is when you Let me reopen the page. is when you Let me reopen the page. is when you Let me reopen the page. Cool. Cool. Cool. Yeah. Where things get much cooler is when you Where things get much cooler is when you start working with an agent on a lake start working with an agent on a lake start working with an agent on a lake bed project. So here I'm going to tell bed project. So here I'm going to tell bed project. So here I'm going to tell the So here I'm going to tell cursor the So here I'm going to tell cursor the So here I'm going to tell cursor agent on composer fast because I want agent on composer fast because I want agent on composer fast because I want this to happen quickly. Make this app a this to happen quickly. Make this app a this to happen quickly. Make this app a real cananban with a live chat that real cananban with a live chat that real cananban with a live chat that shows users names. When done, deploy shows users names. When done, deploy shows users names. When done, deploy with npx lakebed deploy.

with npx lakebed deploy. with npx lakebed deploy. The command is now running. I'm going to The command is now running. I'm going to The command is now running. I'm going to go back to the browser. And this is all go back to the browser. And this is all go back to the browser. And this is all going to be real time. going to be real time. going to be real time. Not only is this going to build the full Not only is this going to build the full Not only is this going to build the full canban, it's also going to update the canban, it's also going to update the canban, it's also going to update the live deployment when it's done running and in any mo and in just a moment once and in any mo and in just a moment once the agent is done. Don't know. Normally the agent is done. Don't know. Normally the agent is done. Don't know. Normally this is even faster. Still pretty quick. this is even faster. Still pretty quick. this is even faster. Still pretty quick. Orafor not found. Oh no. Might have to Orafor not found. Oh no. Might have to Orafor not found. Oh no. Might have to make my docs a little easier for the make my docs a little easier for the make my docs a little easier for the agents to understand. It should have agents to understand. It should have agents to understand. It should have everything it needs in the agent MD. everything it needs in the agent MD. everything it needs in the agent MD. Last time I did this demo, this happened Last time I did this demo, this happened Last time I did this demo, this happened in like under a minute. Come on. Come on. Okay. So, to be clear, the slow part is Okay. So, to be clear, the slow part is Okay. So, to be clear, the slow part is the agent editing it, but it looks like the agent editing it, but it looks like the agent editing it, but it looks like it's deploying now. Oh, I have to it's deploying now. Oh, I have to it's deploying now. Oh, I have to give it permission to delete files. give it permission to delete files. give it permission to delete files. Cool.

Oh, I have to give it permission to auto Oh, I have to give it permission to auto run. I forgot to put it in yolo mode. My run. I forgot to put it in yolo mode. My run. I forgot to put it in yolo mode. My mistake. mistake. mistake. And now it's live. I didn't refresh or And now it's live. I didn't refresh or And now it's live. I didn't refresh or do anything. I just had to hit enter do anything. I just had to hit enter do anything. I just had to hit enter there. And now I have a full working canban And now I have a full working canban where I can move cards around live and where I can move cards around live and where I can move cards around live and others see it. You see people already others see it. You see people already others see it. You see people already spamming Rick roll in the comments spamming Rick roll in the comments spamming Rick roll in the comments section. The fact that I could build section. The fact that I could build section. The fact that I could build this with a dumb cheap model like this with a dumb cheap model like this with a dumb cheap model like composer 25 in literally seconds and it composer 25 in literally seconds and it composer 25 in literally seconds and it auto deploys and updates when the change auto deploys and updates when the change auto deploys and updates when the change goes live is just so cool. And this is goes live is just so cool. And this is goes live is just so cool. And this is what I'm talking about when I talk about what I'm talking about when I talk about what I'm talking about when I talk about like a new endtoend cloud. Instead of like a new endtoend cloud. Instead of like a new endtoend cloud. Instead of having to npm install a bunch of [ __ ] having to npm install a bunch of [ __ ] having to npm install a bunch of [ __ ] configure things in dashboards, set up O configure things in dashboards, set up O configure things in dashboards, set up O and do all that stuff, you just tell the and do all that stuff, you just tell the and do all that stuff, you just tell the agent, go build the thing, and it can agent, go build the thing, and it can agent, go build the thing, and it can just go build the thing. If you provide just go build the thing. If you provide just go build the thing. If you provide it with the right tools and primitives, it with the right tools and primitives, it with the right tools and primitives, and they look and feel enough like what and they look and feel enough like what and they look and feel enough like what the agents already know how to build the agents already know how to build the agents already know how to build with, which spoiler, TypeScript and Vit with, which spoiler, TypeScript and Vit with, which spoiler, TypeScript and Vit are things agents are very good at are things agents are very good at are things agents are very good at building.

building. building. It's not hard to do. Okay, it was it's It's not hard to do. Okay, it was it's It's not hard to do. Okay, it was it's tedious to do. I put a lot of effort tedious to do. I put a lot of effort tedious to do. I put a lot of effort into building this project, but it is so into building this project, but it is so into building this project, but it is so cool and it's so obvious to me now that cool and it's so obvious to me now that cool and it's so obvious to me now that I've built this that this is where most I've built this that this is where most I've built this that this is where most software is going to go in the future. software is going to go in the future. software is going to go in the future. Not specifically Lake Bed, but solutions Not specifically Lake Bed, but solutions Not specifically Lake Bed, but solutions like this will be on platforms like like this will be on platforms like like this will be on platforms like Lakebed because it's so powerful to give Lakebed because it's so powerful to give Lakebed because it's so powerful to give your agent everything it needs to your agent everything it needs to your agent everything it needs to unblock itself and build the things it unblock itself and build the things it unblock itself and build the things it wants to build. And that's what Void wants to build. And that's what Void wants to build. And that's what Void Zero was doing. They already built the Zero was doing. They already built the Zero was doing. They already built the tools agents love. Now they have to tools agents love. Now they have to tools agents love. Now they have to build the platform that agents love to build the platform that agents love to build the platform that agents love to deploy on. deploy on. deploy on. And all of a sudden, Versell isn't the And all of a sudden, Versell isn't the And all of a sudden, Versell isn't the company I'm scared of with Lakebed. It's company I'm scared of with Lakebed. It's company I'm scared of with Lakebed. It's Cloudflare. Kind of crazy to think when I looked at Kind of crazy to think when I looked at kind of crazy to think when you look kind of crazy to think when you look kind of crazy to think when you look here that Verscell isn't the company to here that Verscell isn't the company to here that Verscell isn't the company to be scared of anymore. be scared of anymore. be scared of anymore. But they haven't figured out the But they haven't figured out the But they haven't figured out the database side at all. They haven't database side at all. They haven't database side at all. They haven't figured out how to enable agents to do figured out how to enable agents to do figured out how to enable agents to do those things yet. They're trying and those things yet. They're trying and those things yet. They're trying and they might, but Cloudflare just they might, but Cloudflare just they might, but Cloudflare just purchased the ability to go so much purchased the ability to go so much purchased the ability to go so much further left further left further left from some of the best in the industry to from some of the best in the industry to from some of the best in the industry to ever do it.

ever do it. ever do it. It individually I was not afraid of void It individually I was not afraid of void It individually I was not afraid of void zero and I certainly wasn't afraid of zero and I certainly wasn't afraid of zero and I certainly wasn't afraid of Cloudflare because Void Zero didn't have Cloudflare because Void Zero didn't have Cloudflare because Void Zero didn't have a cloud. They were building deeply on a cloud. They were building deeply on a cloud. They were building deeply on Cloudflare and Cloudflare didn't Cloudflare and Cloudflare didn't Cloudflare and Cloudflare didn't understand these DX problems well enough understand these DX problems well enough understand these DX problems well enough to do it right either. Now they're in an incredible spot. Now they're in an incredible spot. And if they end up buying Convex, I And if they end up buying Convex, I And if they end up buying Convex, I might just give up and shut down because might just give up and shut down because might just give up and shut down because that's the last piece of the puzzle. that's the last piece of the puzzle. that's the last piece of the puzzle. Getting the relationship between the Getting the relationship between the Getting the relationship between the database and the client right will database and the client right will database and the client right will always be difficult. And as great as always be difficult. And as great as always be difficult. And as great as void zero is, it doesn't solve that yet. void zero is, it doesn't solve that yet. void zero is, it doesn't solve that yet. I built a lot of that layer. That's a I built a lot of that layer. That's a I built a lot of that layer. That's a big part of the focus that I've put into big part of the focus that I've put into big part of the focus that I've put into lake bed is making the sync and the data lake bed is making the sync and the data lake bed is making the sync and the data layer as reliable as possible. So I'm layer as reliable as possible. So I'm layer as reliable as possible. So I'm excited to see what ends up happening excited to see what ends up happening excited to see what ends up happening there. But at the very least, I am there. But at the very least, I am there. But at the very least, I am pumped to know Cloudflare's DX will get pumped to know Cloudflare's DX will get pumped to know Cloudflare's DX will get better enough that if you are building better enough that if you are building better enough that if you are building on Cloudflare anyways, you won't need on Cloudflare anyways, you won't need on Cloudflare anyways, you won't need something like Lakebed.

Anything else this article I want to Anything else this article I want to cover? more things before we wrap up because more things before we wrap up because this is all important. Cloudflare this is all important. Cloudflare this is all important. Cloudflare confirmed that they are committing a confirmed that they are committing a confirmed that they are committing a million dollars to a Vit ecosystem fund million dollars to a Vit ecosystem fund million dollars to a Vit ecosystem fund to support maintainers and contributors to support maintainers and contributors to support maintainers and contributors administered by the Vit core team. That administered by the Vit core team. That administered by the Vit core team. That is really cool to hear because there's a is really cool to hear because there's a is really cool to hear because there's a lot of people working on VIT that aren't lot of people working on VIT that aren't lot of people working on VIT that aren't part of Void Zero or Cloudflare. So part of Void Zero or Cloudflare. So part of Void Zero or Cloudflare. So making sure they're getting paid too is making sure they're getting paid too is making sure they're getting paid too is huge. That's a great addition to this huge. That's a great addition to this huge. That's a great addition to this acquisition. But most importantly, we can see why But most importantly, we can see why this acquisition happened here. The this acquisition happened here. The this acquisition happened here. The amount of adoption of the Cloudflare V amount of adoption of the Cloudflare V amount of adoption of the Cloudflare V plugin. Because as soon as this existed plugin. Because as soon as this existed plugin. Because as soon as this existed and was reliable enough that people and and was reliable enough that people and and was reliable enough that people and more importantly agents could use it, more importantly agents could use it, more importantly agents could use it, deploying on Cloudflare became much deploying on Cloudflare became much deploying on Cloudflare became much easier. And that's a huge portion of easier. And that's a huge portion of easier. And that's a huge portion of Cloudflare's new deployments. I'm sure they call it everything I said before they call it everything I said before about how AI is changing how we write about how AI is changing how we write about how AI is changing how we write software. Cuz of course it is. That's software. Cuz of course it is. That's software. Cuz of course it is. That's obvious now. AI loves building with Vit obvious now. AI loves building with Vit obvious now. AI loves building with Vit apps because Vit is fast, well apps because Vit is fast, well apps because Vit is fast, well understood, well documented, compatible understood, well documented, compatible understood, well documented, compatible with everything. And that's why they with everything. And that's why they with everything. And that's why they started building all the other pieces

started building all the other pieces started building all the other pieces like faster builds, tests, linting, etc. And as V becomes full stack, Cloudflare And as V becomes full stack, Cloudflare is the place it makes sense for that is the place it makes sense for that is the place it makes sense for that backend to be. It is pretty obvious to me that the Void It is pretty obvious to me that the Void Zero guys were kind of fishing for this Zero guys were kind of fishing for this Zero guys were kind of fishing for this opportunity because making a profitable opportunity because making a profitable opportunity because making a profitable platform in cloud, especially with the platform in cloud, especially with the platform in cloud, especially with the number of incredible engineers that Void number of incredible engineers that Void number of incredible engineers that Void Zero employed, is difficult. By building Zero employed, is difficult. By building Zero employed, is difficult. By building entirely on Cloudflare, they set entirely on Cloudflare, they set entirely on Cloudflare, they set themselves up perfectly for an themselves up perfectly for an themselves up perfectly for an acquisition like this. acquisition like this. acquisition like this. It's really hot in here. I'm going to It's really hot in here. I'm going to It's really hot in here. I'm going to turn up the AC. Oh god, I don't know why that's set to Oh god, I don't know why that's set to such an insane temp right now. Off eco.

Got the insight that there was a cool Got the insight that there was a cool piece about the new Cloudflare CLI at piece about the new Cloudflare CLI at piece about the new Cloudflare CLI at the bottom here. And that's right. Like the bottom here. And that's right. Like the bottom here. And that's right. Like as someone who appreciates but has a lot as someone who appreciates but has a lot as someone who appreciates but has a lot of frustrations with Wrangler, the idea of frustrations with Wrangler, the idea of frustrations with Wrangler, the idea of a single package that is built on top of a single package that is built on top of a single package that is built on top of VIT as a CLI experience that lets you of VIT as a CLI experience that lets you of VIT as a CLI experience that lets you do everything you need to on Cloudflare do everything you need to on Cloudflare do everything you need to on Cloudflare is super compelling. CFD dev which is a is super compelling. CFD dev which is a is super compelling. CFD dev which is a supererset of Vde dev that works with supererset of Vde dev that works with supererset of Vde dev that works with all the services. CF build which all the services. CF build which all the services. CF build which understands V projects natively and CF understands V projects natively and CF understands V projects natively and CF deploy which makes it easy to deploy on deploy which makes it easy to deploy on deploy which makes it easy to deploy on Cloudflare. This is very promising. That said, the end here still shows the That said, the end here still shows the old commands. If you want to try V on old commands. If you want to try V on old commands. If you want to try V on Cloudflare, run npm create V at latest Cloudflare, run npm create V at latest Cloudflare, run npm create V at latest and then npx wrangr deploy. Yeah, that's and then npx wrangr deploy. Yeah, that's and then npx wrangr deploy. Yeah, that's the part that needs to be fixed. We're the part that needs to be fixed. We're the part that needs to be fixed. We're getting there, though. I am very hopeful getting there, though. I am very hopeful getting there, though. I am very hopeful for where they're going. The future for for where they're going. The future for for where they're going. The future for Cloudflare just got much much brighter. Cloudflare just got much much brighter. Cloudflare just got much much brighter. I cannot imagine a better acquisition I cannot imagine a better acquisition I cannot imagine a better acquisition they could make for positioning they could make for positioning they could make for positioning themselves in this rapidly changing themselves in this rapidly changing themselves in this rapidly changing market, especially after the Nux guys market, especially after the Nux guys market, especially after the Nux guys ended up at Verscell. Very interesting ended up at Verscell. Very interesting ended up at Verscell. Very interesting to see the Vit ecos, very interesting to to see the Vit ecos, very interesting to to see the Vit ecos, very interesting to see the Vue ecosystem kind of split see the Vue ecosystem kind of split see the Vue ecosystem kind of split between these companies where Nux went between these companies where Nux went between these companies where Nux went to Verscell and Void Zero went to to Verscell and Void Zero went to to Verscell and Void Zero went to Cloudflare.

Cloudflare. Cloudflare. But this war is only getting started and But this war is only getting started and But this war is only getting started and I have a feeling things are going to I have a feeling things are going to I have a feeling things are going to ramp up a lot over the next few years. ramp up a lot over the next few years. ramp up a lot over the next few years. And if you're looking for a place to And if you're looking for a place to And if you're looking for a place to keep up with that, you're there. So, hit keep up with that, you're there. So, hit keep up with that, you're there. So, hit that red button if you haven't yet. that red button if you haven't yet. that red button if you haven't yet. Appreciate it a ton. I'm excited to see Appreciate it a ton. I'm excited to see Appreciate it a ton. I'm excited to see where this all goes and I'm also excited where this all goes and I'm also excited where this all goes and I'm also excited to get a little bit more of what we're to get a little bit more of what we're to get a little bit more of what we're doing with Lake Bed out for y'all to doing with Lake Bed out for y'all to doing with Lake Bed out for y'all to see. Let me know how y'all feel and let see. Let me know how y'all feel and let see. Let me know how y'all feel and let me know where you think Tanstack's going me know where you think Tanstack's going me know where you think Tanstack's going to end up next cuz obviously that's to end up next cuz obviously that's to end up next cuz obviously that's where things are going. Curious as well where things are going. Curious as well where things are going. Curious as well as you got. I'm just as curious as as you got. I'm just as curious as as you got. I'm just as curious as y'all, but I don't have any insidefo y'all, but I don't have any insidefo y'all, but I don't have any insidefo here. Just sharing what I know and what here. Just sharing what I know and what here. Just sharing what I know and what I can. Hope this was helpful and until I can. Hope this was helpful and until I can. Hope this was helpful and until next time, peace nerds. next time, peace nerds. next time, peace nerds. Bun. I need to do my offset because I forgot I need to do my offset because I forgot to. I got word that this should work.

I got word that this should work. Fingers crossed. Why is the Oh, I have to enable that as Why is the Oh, I have to enable that as the admin of the team. Cool. I now have my handle as I very Cool. I now have my handle as I very well should. Hurrah. The person who had it was squatting it. The person who had it was squatting it. They weren't even a sub and they'd never They weren't even a sub and they'd never They weren't even a sub and they'd never they set up the account that day. It was they set up the account that day. It was they set up the account that day. It was clearly being squatted. I have not been using cursor much I have not been using cursor much lately. I love the cloud stuff. I just lately. I love the cloud stuff. I just lately. I love the cloud stuff. I just don't like how do I put this?

how do I put this? I'm not working on like real apps facing I'm not working on like real apps facing I'm not working on like real apps facing people right now as much. So, it hasn't people right now as much. So, it hasn't people right now as much. So, it hasn't been as useful to me. But, there's a lot been as useful to me. But, there's a lot been as useful to me. But, there's a lot of aspects that have been missing from of aspects that have been missing from of aspects that have been missing from it recently. especially like it's a it recently. especially like it's a it recently. especially like it's a stupid thing, but the more I've been stupid thing, but the more I've been stupid thing, but the more I've been using cloud T3 code like remotely from using cloud T3 code like remotely from using cloud T3 code like remotely from like other computers, the more I wish like other computers, the more I wish like other computers, the more I wish there was an easy way to just click a there was an easy way to just click a there was an easy way to just click a button and have a fresh clone work tree button and have a fresh clone work tree button and have a fresh clone work tree on the latest main that I can get to on the latest main that I can get to on the latest main that I can get to work on top of because I've accidentally work on top of because I've accidentally work on top of because I've accidentally built so much work on top of stale built so much work on top of stale built so much work on top of stale branches that it's just a mess and I'm branches that it's just a mess and I'm branches that it's just a mess and I'm so tired of it. Oh god. Oh god. Oh, good [ __ ] That cross great. A phase aware pre decoding inference A phase aware pre decoding inference focused power consumption benchmark focused power consumption benchmark focused power consumption benchmark might be useful. Uh that's very might be useful. Uh that's very might be useful. Uh that's very interesting.

interesting. interesting. I'm not the right person to ask. I'm not I'm not the right person to ask. I'm not I'm not the right person to ask. I'm not smart enough. I just pay whatever the smart enough. I just pay whatever the smart enough. I just pay whatever the token price is. And thank you, Alex, for token price is. And thank you, Alex, for token price is. And thank you, Alex, for stopping by for all this. Really stopping by for all this. Really stopping by for all this. Really appreciate it. Go get some sleep, man. Thanks for Go get some sleep, man. Thanks for stopping by as always. Uh, what are we talking about here? Uh, what are we talking about here? Yeah, it won't be on my channel. It will Yeah, it won't be on my channel. It will Yeah, it won't be on my channel. It will be on the Cascadia channel when it's be on the Cascadia channel when it's be on the Cascadia channel when it's done. I I got really close to working done. I I got really close to working done. I I got really close to working file storage, but I'm also migrating the file storage, but I'm also migrating the file storage, but I'm also migrating the hosted bundles when I do it. And also, hosted bundles when I do it. And also, hosted bundles when I do it. And also, I'm in parallel rebuilding the data I'm in parallel rebuilding the data I'm in parallel rebuilding the data architecture entirely. I will likely be architecture entirely. I will likely be architecture entirely. I will likely be killing all of the data in all of the killing all of the data in all of the killing all of the data in all of the existing apps on Lakebed. So, sorry for existing apps on Lakebed. So, sorry for existing apps on Lakebed. So, sorry for those building on it. I I might provide those building on it. I I might provide those building on it. I I might provide an easy out, but an easy out, but an easy out, but I'm going through it. It's It's a tough I'm going through it. It's It's a tough I'm going through it. It's It's a tough project to build. I I'll have updates project to build. I I'll have updates project to build. I I'll have updates soon.

Yeah, I I tried getting a lot of work Yeah, I I tried getting a lot of work done last night, but had a bunch of [ __ ] done last night, but had a bunch of [ __ ] done last night, but had a bunch of [ __ ] happen, so I couldn't. I am hoping to no happen, so I couldn't. I am hoping to no happen, so I couldn't. I am hoping to no life it this weekend, but life it this weekend, but life it this weekend, but I I'm so behind for my trip. Also, I've I I'm so behind for my trip. Also, I've I I'm so behind for my trip. Also, I've been working more on T3 code, funny been working more on T3 code, funny been working more on T3 code, funny enough, because I been using T3 code so enough, because I been using T3 code so enough, because I been using T3 code so much to build this, especially remote, much to build this, especially remote, much to build this, especially remote, that all of these details I've been that all of these details I've been that all of these details I've been trying to get turn like there's a lot of trying to get turn like there's a lot of trying to get turn like there's a lot of little things in T3 code that didn't little things in T3 code that didn't little things in T3 code that didn't annoy me that much before that now annoy annoy me that much before that now annoy annoy me that much before that now annoy me a lot because I've been using it so me a lot because I've been using it so me a lot because I've been using it so much more heavily. So, like I'm just much more heavily. So, like I'm just much more heavily. So, like I'm just trying to get the mobile [ __ ] in a trying to get the mobile [ __ ] in a trying to get the mobile [ __ ] in a better spot mostly because I it God, better spot mostly because I it God, better spot mostly because I it God, it's so cool having a mobile app that I it's so cool having a mobile app that I it's so cool having a mobile app that I can actually use to like write code. can actually use to like write code. can actually use to like write code. Yeah, T3 Code mobile app. I already Yeah, T3 Code mobile app. I already Yeah, T3 Code mobile app. I already leaked that. A lot of work to do still, but I'm A lot of work to do still, but I'm excited as [ __ ] about it.

doesn't use the correct models for doesn't use the correct models for cursor, but we are built on top of the like ACP but we are built on top of the like ACP for the agent uh CLI, which isn't great. for the agent uh CLI, which isn't great. for the agent uh CLI, which isn't great. We're going to move over to the SDK We're going to move over to the SDK We're going to move over to the SDK eventually. We haven't had a chance yet. eventually. We haven't had a chance yet. eventually. We haven't had a chance yet. Main focus has been solidifying the Main focus has been solidifying the Main focus has been solidifying the remote stuff. remote stuff. remote stuff. Oh yeah, my X payout happened. I can do Oh yeah, my X payout happened. I can do Oh yeah, my X payout happened. I can do my X payout tweet. my X payout tweet. my X payout tweet. Let me I got the notification. Where'd Let me I got the notification. Where'd Let me I got the notification. Where'd it go? How the [ __ ] did that disappear? I was How the [ __ ] did that disappear? I was going to screenshot it so I could post going to screenshot it so I could post going to screenshot it so I could post it, but Let me go to Let me go to that quick. Posting it now is 5,57 Posting it now is 5,57 for those curious.

Cool. Cool. Either chat mobile app is coming soon. Either chat mobile app is coming soon. Either chat mobile app is coming soon. We uh have another person who just We uh have another person who just We uh have another person who just joined to like really focus on that and joined to like really focus on that and joined to like really focus on that and get it out. Yeah, here are the official get it out. Yeah, here are the official get it out. Yeah, here are the official numbers. Not bad because I've like not numbers. Not bad because I've like not numbers. Not bad because I've like not really been on Twitter that much last really been on Twitter that much last really been on Twitter that much last two weeks. I've been so [ __ ] busy. I two weeks. I've been so [ __ ] busy. I two weeks. I've been so [ __ ] busy. I still made real money off it. So that's still made real money off it. So that's still made real money off it. So that's cool. Opening has a new agent SDK codeex via Opening has a new agent SDK codeex via API. Same tools and API for uploading API. Same tools and API for uploading API. Same tools and API for uploading skills if you want to use them. That's skills if you want to use them. That's skills if you want to use them. That's cool to hear. Is this a YouTube video cool to hear. Is this a YouTube video cool to hear. Is this a YouTube video that they did about it? Yeah, I do do remember seeing a bit Yeah, I do do remember seeing a bit about this Is TF chat a way to say anonymous Is TF chat a way to say anonymous companies with your personal data? Yeah, companies with your personal data? Yeah, companies with your personal data? Yeah, it works for that. They get no info on it works for that. They get no info on it works for that. They get no info on who you are and like who made what who you are and like who made what who you are and like who made what request, so that would work. Yeah.

request, so that would work. Yeah. request, so that would work. Yeah. What's next? Good question. What's next? Good question. What's next? Good question. I think we got to do the anthropic I think we got to do the anthropic I think we got to do the anthropic pausing AI dev. I have a lot I want to say on the I have a lot I want to say on the compute crunch. I do also want to look compute crunch. I do also want to look compute crunch. I do also want to look through what Twitter said I should do thanks to my uh Hermes agent having thanks to my uh Hermes agent having access to my Twitter now, which is super access to my Twitter now, which is super access to my Twitter now, which is super useful.

Once again, AI will never replace me for Once again, AI will never replace me for YouTube at the very least. YouTube at the very least. YouTube at the very least. Yeah, I have her on one of my minis. I wonder disc rates with T3 code. I wonder disc rates with T3 code. It burns them in Linux while doing It burns them in Linux while doing It burns them in Linux while doing nothing. This might be because of the nothing. This might be because of the nothing. This might be because of the git checks. It could be that like the git checks. It could be that like the git checks. It could be that like the GitHub CLI writes, but like we're not GitHub CLI writes, but like we're not GitHub CLI writes, but like we're not doing anything that should be writing doing anything that should be writing doing anything that should be writing aggressively. aggressively. aggressively. happy to audit that if it's a consistent happy to audit that if it's a consistent happy to audit that if it's a consistent problem, but Yeah, Hermes setup in Discord will Yeah, Hermes setup in Discord will probably be like a roughly what you're probably be like a roughly what you're probably be like a roughly what you're looking for cuz it runs on an actual looking for cuz it runs on an actual looking for cuz it runs on an actual computer computer computer and then just gives you the ability to and then just gives you the ability to and then just gives you the ability to like make threads in Discord like make threads in Discord like make threads in Discord and you can set up automations for it and you can set up automations for it and you can set up automations for it too. So I can say uh schedule this so or too. So I can say uh schedule this so or too. So I can say uh schedule this so or make a skill for scanning make a skill for scanning make a skill for scanning my Twitter mentions to find potential my Twitter mentions to find potential my Twitter mentions to find potential topics. Run it every Wednesday at noon topics. Run it every Wednesday at noon topics. Run it every Wednesday at noon and alert me if there are topics worth and alert me if there are topics worth and alert me if there are topics worth filming about.

So now it's going to take the context of So now it's going to take the context of this thread, create a skill for it, and this thread, create a skill for it, and this thread, create a skill for it, and then create a schedule where every then create a schedule where every then create a schedule where every Wednesday at noon it will start to Wednesday at noon it will start to Wednesday at noon it will start to update and tell me update and tell me update and tell me that like that like that like that like what things people mentioned that like what things people mentioned that like what things people mentioned automatically and it's really trivial to automatically and it's really trivial to automatically and it's really trivial to set up this type of thing. And I really set up this type of thing. And I really set up this type of thing. And I really like the threading interface for this like the threading interface for this like the threading interface for this because I found that when I use things because I found that when I use things because I found that when I use things like like like Telegram or WhatsApp for this, having Telegram or WhatsApp for this, having Telegram or WhatsApp for this, having like the one thread was obnoxious. You ask what the photo is. I just found You ask what the photo is. I just found a random hermit crab and like zoomed in a random hermit crab and like zoomed in a random hermit crab and like zoomed in various amounts for the different various amounts for the different various amounts for the different images. images. images. Cool. I saw the Ladybird news. Not too too I saw the Ladybird news. Not too too much to say there. It's like I I don't much to say there. It's like I I don't much to say there. It's like I I don't get why people are so hyped on this.

get why people are so hyped on this. get why people are so hyped on this. Like they they feel like this needs to Like they they feel like this needs to Like they they feel like this needs to be talked about. like be talked about. like be talked about. like uh I I I'm on better terms with the uh I I I'm on better terms with the uh I I I'm on better terms with the Ladybird guys, so I hate to do this and Ladybird guys, so I hate to do this and Ladybird guys, so I hate to do this and I know it's going to get clipped. It's I know it's going to get clipped. It's I know it's going to get clipped. It's going to get me trouble. I don't care. going to get me trouble. I don't care. going to get me trouble. I don't care. This is the type of update that feels This is the type of update that feels This is the type of update that feels really big on a project that feels really big on a project that feels really big on a project that feels really big but doesn't really matter really big but doesn't really matter really big but doesn't really matter because nobody's using it and it's not because nobody's using it and it's not because nobody's using it and it's not building a thing with the intent of building a thing with the intent of building a thing with the intent of being used. The point of Ladybird isn't being used. The point of Ladybird isn't being used. The point of Ladybird isn't to make an actual viable alternative to make an actual viable alternative to make an actual viable alternative browser. The point of Ladybird is to browser. The point of Ladybird is to browser. The point of Ladybird is to prove that you can make a new browser. prove that you can make a new browser. prove that you can make a new browser. It is a scientific experiment that has It is a scientific experiment that has It is a scientific experiment that has no intent to benefit people no intent to benefit people no intent to benefit people and because it's so much about it's and because it's so much about it's and because it's so much about it's almost like a performance art almost like a performance art almost like a performance art and this is quite the performance. and this is quite the performance. and this is quite the performance. So that like the reasons people care So that like the reasons people care So that like the reasons people care about Ladybird aren't about having a new about Ladybird aren't about having a new about Ladybird aren't about having a new better browser. It's the concept of a better browser. It's the concept of a better browser. It's the concept of a new browser. So the concept of them new browser. So the concept of them new browser. So the concept of them changing how contributions work feels changing how contributions work feels changing how contributions work feels like a big deal. because there is like a big deal. because there is like a big deal. because there is nothing else to have be a big deal. None nothing else to have be a big deal. None nothing else to have be a big deal. None of this matters. Literally, none of this of this matters. Literally, none of this of this matters. Literally, none of this matters. So, an announcement like this matters. So, an announcement like this matters. So, an announcement like this is the closest to mattering that is the closest to mattering that is the closest to mattering that anything about Ladybird has felt in a anything about Ladybird has felt in a anything about Ladybird has felt in a while, which is why people care so much.

while, which is why people care so much. while, which is why people care so much. But I don't give a [ __ ] about what But I don't give a [ __ ] about what But I don't give a [ __ ] about what contributions they are and aren't contributions they are and aren't contributions they are and aren't allowing to a project that will not be allowing to a project that will not be allowing to a project that will not be useful. Oh god, what are you sending me here? Oh god, what are you sending me here? Mutation code exceeded. I'll have a way Mutation code exceeded. I'll have a way Mutation code exceeded. I'll have a way to boost it soon. If you DM me your to boost it soon. If you DM me your to boost it soon. If you DM me your GitHub handle, I'll get you manually GitHub handle, I'll get you manually GitHub handle, I'll get you manually boosted. boosted. boosted. Soon. Send it to me on Twitter and I'll Soon. Send it to me on Twitter and I'll Soon. Send it to me on Twitter and I'll do that after stream. Think I got you, Vion atomic. So, thank Think I got you, Vion atomic. So, thank you to Scythe for the Prime. Thank you you to Scythe for the Prime. Thank you you to Scythe for the Prime. Thank you to NCE for the 17 months of support. to NCE for the 17 months of support. to NCE for the 17 months of support. Thank youful poet for the eight months. Thank youful poet for the eight months. Thank youful poet for the eight months. Always good to see you, man. Big will be Always good to see you, man. Big will be Always good to see you, man. Big will be there when the competition eventually there when the competition eventually there when the competition eventually hits the fan. Fair point. I have a hits the fan. Fair point. I have a hits the fan. Fair point. I have a feeling the competition is going to go a feeling the competition is going to go a feeling the competition is going to go a little insane.

little insane. little insane. Thank you. Live coding sessions for the Thank you. Live coding sessions for the Thank you. Live coding sessions for the prime. Meadow Bigweed Good. Thank you prime. Meadow Bigweed Good. Thank you prime. Meadow Bigweed Good. Thank you for the prime. Great username. Thank you for the prime. Great username. Thank you for the prime. Great username. Thank you internal error for the raid internal error for the raid internal error for the raid or sky potter for the prime as well. Oh, or sky potter for the prime as well. Oh, or sky potter for the prime as well. Oh, sorry. Tier one encoding David here. sorry. Tier one encoding David here. sorry. Tier one encoding David here. Watching the stream while Codex is Watching the stream while Codex is Watching the stream while Codex is trying to hack and reverse engineer a trying to hack and reverse engineer a trying to hack and reverse engineer a Mercedes-Benz head unit to hack its Mercedes-Benz head unit to hack its Mercedes-Benz head unit to hack its Nvidia chip at 2 am. Feels weird. I love Nvidia chip at 2 am. Feels weird. I love Nvidia chip at 2 am. Feels weird. I love that. Good [ __ ] Good reason to be here. that. Good [ __ ] Good reason to be here. that. Good [ __ ] Good reason to be here. Thank you for the 300 bits Thank you for the 300 bits Thank you for the 300 bits using 55 for my Hermes agent. Very interesting. Very interesting. Guess I will read later. Guess I will read later. Guess I will read later. Thank you, Zara, for the five bomb. Thank you, Zara, for the five bomb. Thank you, Zara, for the five bomb. Appreciate that a ton. Tier two subs. Appreciate that a ton. Tier two subs. Appreciate that a ton. Tier two subs. God damn. That's on the app I made. Oh yeah, my That's on the app I made. Oh yeah, my account isn't boosted. Thank you, gamer account isn't boosted. Thank you, gamer account isn't boosted. Thank you, gamer girl. Not surprised you both spam the girl. Not surprised you both spam the girl. Not surprised you both spam the [ __ ] out of that.

Nice. It took a while to get there. Nice. It took a while to get there. Damn, this is actually a cool article. Damn, this is actually a cool article. Damn, this is actually a cool article. This almost certainly will be a video This almost certainly will be a video This almost certainly will be a video later. They're doing full markdown streaming They're doing full markdown streaming for bots in chat with Telegram. That is for bots in chat with Telegram. That is for bots in chat with Telegram. That is cool. I did hear a bit about that. You submit forms programmatically. Oh You submit forms programmatically. Oh god. Did I miss anything on YouTube other Did I miss anything on YouTube other than it crashing? Nope.

Yeah, Tailwind's MIT licensed. You can Yeah, Tailwind's MIT licensed. You can do whatever you want to it. Tailwind UI do whatever you want to it. Tailwind UI do whatever you want to it. Tailwind UI isn't, but Tailwind the CLI, Tailwind isn't, but Tailwind the CLI, Tailwind isn't, but Tailwind the CLI, Tailwind the CSS framework, and all the other the CSS framework, and all the other the CSS framework, and all the other tools, those are all good. Anything for tools, those are all good. Anything for tools, those are all good. Anything for the anthropic video? I don't think so. I the anthropic video? I don't think so. I the anthropic video? I don't think so. I think we're ready to just go with that. Jesus Christ. Jesus Christ. Let's do this. Um, development was paused on T3 chat mobile development was paused on T3 chat mobile app. It is picking up again now. Long app. It is picking up again now. Long app. It is picking up again now. Long story there. Already did the camera story there. Already did the camera story there. Already did the camera offset. Don't worry. Uh, we're going to offset. Don't worry. Uh, we're going to offset. Don't worry. Uh, we're going to start the AI builds itself anthropic start the AI builds itself anthropic start the AI builds itself anthropic vid.

My vision for product or projects I My vision for product or projects I build is just I solve problems that I build is just I solve problems that I build is just I solve problems that I have. Thank you for the five bomb by the have. Thank you for the five bomb by the have. Thank you for the five bomb by the way. I appreciate that. Thanks for way. I appreciate that. Thanks for way. I appreciate that. Thanks for sending the gifts and then asking. But sending the gifts and then asking. But sending the gifts and then asking. But yeah, the my I I solve problems I have yeah, the my I I solve problems I have yeah, the my I I solve problems I have and then when I solve that problem, I and then when I solve that problem, I and then when I solve that problem, I find more problems and then I solve find more problems and then I solve find more problems and then I solve those problems and then it just goes on those problems and then it just goes on those problems and then it just goes on infinitely. The compute from SpaceX for anthrop or The compute from SpaceX for anthrop or for Google is going to be a separate for Google is going to be a separate for Google is going to be a separate thing. Talking about that soon. thing. Talking about that soon. thing. Talking about that soon. Cool. Everything I need to start this. So, Everything I need to start this. So, let's start it. Hair a little. Hair a little. Good enough. Cool. We haven't done a doomer video in a We haven't done a doomer video in a while, so why not now? And ah, while, so why not now? And ah, while, so why not now? And ah, been a bit since we did an AI doomer been a bit since we did an AI doomer been a bit since we did an AI doomer video, but I think we have good reason video, but I think we have good reason video, but I think we have good reason to today. The AI takeoff is a real to today. The AI takeoff is a real to today. The AI takeoff is a real concern that many have, mostly the concern that many have, mostly the concern that many have, mostly the doomers admittedly, but it's a thing doomers admittedly, but it's a thing doomers admittedly, but it's a thing that we should definitely think about.

that we should definitely think about. that we should definitely think about. What happens when AI gets good enough to What happens when AI gets good enough to What happens when AI gets good enough to improve itself? We've already seen what improve itself? We've already seen what improve itself? We've already seen what happens when AI gets a certain level of happens when AI gets a certain level of happens when AI gets a certain level of capability. Once it gets good enough at capability. Once it gets good enough at capability. Once it gets good enough at coding, suddenly the amount of code in coding, suddenly the amount of code in coding, suddenly the amount of code in the world 10 x's, 100 x's or more. the world 10 x's, 100 x's or more. the world 10 x's, 100 x's or more. Suddenly we're rewriting huge projects Suddenly we're rewriting huge projects Suddenly we're rewriting huge projects like bun from one language to another. like bun from one language to another. like bun from one language to another. Not because AI is so smart that it's Not because AI is so smart that it's Not because AI is so smart that it's gigabrained and can do that, but because gigabrained and can do that, but because gigabrained and can do that, but because it's smart enough that when run in a it's smart enough that when run in a it's smart enough that when run in a loop and enough compute is burned, it loop and enough compute is burned, it loop and enough compute is burned, it can kind of just keep doing the thing can kind of just keep doing the thing can kind of just keep doing the thing and finding every single piece that it and finding every single piece that it and finding every single piece that it needs to succeed. What happens when AI needs to succeed. What happens when AI needs to succeed. What happens when AI can do that to itself? can do that to itself? can do that to itself? The term for this is the AI takeoff. The term for this is the AI takeoff. The term for this is the AI takeoff. Once AI is good enough that it is as Once AI is good enough that it is as Once AI is good enough that it is as smart as the humans building it and it smart as the humans building it and it smart as the humans building it and it can start to improve itself over time, can start to improve itself over time, can start to improve itself over time, what happens and how quick is it going what happens and how quick is it going what happens and how quick is it going to be able to improve itself once it to be able to improve itself once it to be able to improve itself once it gets to that point? This is the biggest gets to that point? This is the biggest gets to that point? This is the biggest concern that people have with the AI concern that people have with the AI concern that people have with the AI takeoff theories. There's the soft takeoff theories. There's the soft takeoff theories. There's the soft takeoff, which is that it would take takeoff, which is that it would take takeoff, which is that it would take years for AI to improve itself, but the years for AI to improve itself, but the years for AI to improve itself, but the bigger concern is the hard takeoff. What bigger concern is the hard takeoff. What bigger concern is the hard takeoff. What happens when AI can improve itself happens when AI can improve itself happens when AI can improve itself rapidly in ways that we don't even rapidly in ways that we don't even rapidly in ways that we don't even understand?

understand? understand? If this was just about a less wrong If this was just about a less wrong If this was just about a less wrong post, then I wouldn't make you all post, then I wouldn't make you all post, then I wouldn't make you all suffer through it. Believe me, I've done suffer through it. Believe me, I've done suffer through it. Believe me, I've done my best to avoid going to this site in my best to avoid going to this site in my best to avoid going to this site in my content. But we're not here to talk my content. But we're not here to talk my content. But we're not here to talk about less wrong. We're here to talk about less wrong. We're here to talk about less wrong. We're here to talk about this anthropic article about what about this anthropic article about what about this anthropic article about what happens when AI builds itself because happens when AI builds itself because happens when AI builds itself because Anthropic has already started to see Anthropic has already started to see Anthropic has already started to see here we are because Anthropic has here we are because Anthropic has here we are because Anthropic has already started to see massive increases already started to see massive increases already started to see massive increases in their own productivity building AI in their own productivity building AI in their own productivity building AI using the most recent models that they using the most recent models that they using the most recent models that they have created. More importantly though, they open up More importantly though, they open up the question as to whether or not we the question as to whether or not we the question as to whether or not we should temporarily pause frontier AI should temporarily pause frontier AI should temporarily pause frontier AI development to enable societal development to enable societal development to enable societal structures and alignment research to structures and alignment research to structures and alignment research to keep up with the advancement of this keep up with the advancement of this keep up with the advancement of this technology. Yes, really. Anthropic in an official Yes, really. Anthropic in an official article they posted just called out that article they posted just called out that article they posted just called out that it might be time to pause AI development it might be time to pause AI development it might be time to pause AI development right after their trillion dollar right after their trillion dollar right after their trillion dollar valuation. valuation. valuation. There's a lot to dig into here from what There's a lot to dig into here from what There's a lot to dig into here from what self-improvement looks like to what the self-improvement looks like to what the self-improvement looks like to what the risks are to how we cope with a society risks are to how we cope with a society risks are to how we cope with a society where intelligence goes beyond our own where intelligence goes beyond our own where intelligence goes beyond our own intelligence.

intelligence. intelligence. There's a lot to think about here and There's a lot to think about here and There's a lot to think about here and I'm doing my best to not go insane. But I'm doing my best to not go insane. But I'm doing my best to not go insane. But if but if I'm about to lose my job and if but if I'm about to lose my job and if but if I'm about to lose my job and my therapist but if I can but if I need to pay but if but if I can but if I need to pay but if I need to ah but in order for me to I need to ah but in order for me to I need to ah but in order for me to afford my AI therapist after I lose my afford my AI therapist after I lose my afford my AI therapist after I lose my job to AI I need some money. So we're job to AI I need some money. So we're job to AI I need some money. So we're going to do a quick break for today's going to do a quick break for today's going to do a quick break for today's sponsor before we dive in. So, as Anthropic poses here, they are So, as Anthropic poses here, they are getting close to recursive getting close to recursive getting close to recursive self-improvement, which is a kind of self-improvement, which is a kind of self-improvement, which is a kind of scary thing. For most of AI's history, humans drove For most of AI's history, humans drove every step in its development cycle. But every step in its development cycle. But every step in its development cycle. But Anthropic, we are delegating a growing Anthropic, we are delegating a growing Anthropic, we are delegating a growing share of AI development to AI systems share of AI development to AI systems share of AI development to AI systems themselves, which is speeding up our themselves, which is speeding up our themselves, which is speeding up our work. Taken far enough and given enough work. Taken far enough and given enough work. Taken far enough and given enough compute, the trend points to an AI compute, the trend points to an AI compute, the trend points to an AI system capable of fully autonomously system capable of fully autonomously system capable of fully autonomously designing and developing its own designing and developing its own designing and developing its own successor. This is called recursive successor. This is called recursive successor. This is called recursive self-improvement. We are not there yet.

self-improvement. We are not there yet. self-improvement. We are not there yet. And recursive self-improvement is not And recursive self-improvement is not And recursive self-improvement is not inevitable. Bold statement, but good to inevitable. Bold statement, but good to inevitable. Bold statement, but good to hear it from them. They don't think this hear it from them. They don't think this hear it from them. They don't think this is inherently going to happen like this is inherently going to happen like this is inherently going to happen like this is an inevitable outcome. It's just a is an inevitable outcome. It's just a is an inevitable outcome. It's just a thing that could happen. And as such, it thing that could happen. And as such, it thing that could happen. And as such, it could actually come sooner than most could actually come sooner than most could actually come sooner than most institutions are prepared for. institutions are prepared for. institutions are prepared for. Using public benchmarks in previously Using public benchmarks in previously Using public benchmarks in previously unreported data from within Anthropic, unreported data from within Anthropic, unreported data from within Anthropic, the Anthropic Institute is showing that the Anthropic Institute is showing that the Anthropic Institute is showing that AI is already accelerating the AI is already accelerating the AI is already accelerating the development of AI systems. To take just development of AI systems. To take just development of AI systems. To take just one example, today anthropic engineers one example, today anthropic engineers one example, today anthropic engineers on average ship 8 times as much code per on average ship 8 times as much code per on average ship 8 times as much code per quarter as they did from 2021 to 2025. quarter as they did from 2021 to 2025. quarter as they did from 2021 to 2025. Well, to be fair, having used anthropic Well, to be fair, having used anthropic Well, to be fair, having used anthropic systems, I would be okay with them systems, I would be okay with them systems, I would be okay with them shipping way less code because they're shipping way less code because they're shipping way less code because they're shipping way more than 8x the bugs shipping way more than 8x the bugs shipping way more than 8x the bugs lately. But yeah, that is a meaningful lately. But yeah, that is a meaningful lately. But yeah, that is a meaningful number. number. number. The technical trends discussed in this The technical trends discussed in this The technical trends discussed in this place that the technical trends place that the technical trends place that the technical trends discussed in this piece suggest that AI discussed in this piece suggest that AI discussed in this piece suggest that AI systems are going to become much more systems are going to become much more systems are going to become much more capable in coming years. These trends capable in coming years. These trends capable in coming years. These trends have huge implications. AI that can have huge implications. AI that can have huge implications. AI that can build itself would be a major build itself would be a major build itself would be a major development in the history of development in the history of development in the history of technology. One that could bring technology. One that could bring technology. One that could bring enormous good for the world in science, enormous good for the world in science, enormous good for the world in science, healthcare, and beyond. But full healthcare, and beyond. But full healthcare, and beyond. But full recursive self-improvement also might recursive self-improvement also might recursive self-improvement also might increase but full recursive increase but full recursive increase but full recursive self-improvement also might increase the self-improvement also might increase the self-improvement also might increase the risks of humans losing control over AI risks of humans losing control over AI risks of humans losing control over AI systems. If systems are capable of fully systems. If systems are capable of fully systems. If systems are capable of fully building their own successors, the ways building their own successors, the ways building their own successors, the ways we secure them, monitor them, and shape we secure them, monitor them, and shape we secure them, monitor them, and shape their behavior all grow much more their behavior all grow much more their behavior all grow much more important.

They have a cute little timeline here They have a cute little timeline here with a really bad CSS cut off. I love with a really bad CSS cut off. I love with a really bad CSS cut off. I love that. I was just talking about their that. I was just talking about their that. I was just talking about their bugs and there's immediately one in the bugs and there's immediately one in the bugs and there's immediately one in the UI here. In the early days, working at UI here. In the early days, working at UI here. In the early days, working at Anthropic looked like work at any other Anthropic looked like work at any other Anthropic looked like work at any other large tech company. People would write large tech company. People would write large tech company. People would write code and docs on a laptop. So, a person code and docs on a laptop. So, a person code and docs on a laptop. So, a person uses a computer and the output is uses a computer and the output is uses a computer and the output is clawed. But then chat bots happened. clawed. But then chat bots happened. clawed. But then chat bots happened. People use early chatbots to help with People use early chatbots to help with People use early chatbots to help with parts of the process like generating parts of the process like generating parts of the process like generating short code snippets and copying the short code snippets and copying the short code snippets and copying the output into text editors. output into text editors. output into text editors. Going to pull up my sidebar to fix the Going to pull up my sidebar to fix the Going to pull up my sidebar to fix the layout of it here. Cool. layout of it here. Cool. layout of it here. Cool. And we got coding agents, which mean And we got coding agents, which mean And we got coding agents, which mean that the person could use the coding that the person could use the coding that the person could use the coding agent to build the software and edit agent to build the software and edit agent to build the software and edit code for the projects that they're code for the projects that they're code for the projects that they're working on at the company. And then we working on at the company. And then we working on at the company. And then we got to the point of autonomous agents got to the point of autonomous agents got to the point of autonomous agents where the agent can spin up workers that where the agent can spin up workers that where the agent can spin up workers that then build the thing that you're trying then build the thing that you're trying then build the thing that you're trying to build. What happens if we close the to build. What happens if we close the to build. What happens if we close the loop? Because in the future, agents loop? Because in the future, agents loop? Because in the future, agents could become capable enough to build and could become capable enough to build and could become capable enough to build and train all themselves. So no human is train all themselves. So no human is train all themselves. So no human is necessary in the loop at all. I just the camera recording. Don't I just the camera recording. Don't worry, guys.

Evidence from the outside world. The Evidence from the outside world. The rate at which AI models improve is rate at which AI models improve is rate at which AI models improve is accelerating. The length of tasks that accelerating. The length of tasks that accelerating. The length of tasks that they can reliably complete on their own they can reliably complete on their own they can reliably complete on their own has been doubling roughly every four has been doubling roughly every four has been doubling roughly every four months, up from an earlier trend of months, up from an earlier trend of months, up from an earlier trend of doubling every 7 months. That is doubling every 7 months. That is doubling every 7 months. That is actually kind of nuts if you think about actually kind of nuts if you think about actually kind of nuts if you think about it, especially coming from me. I didn't it, especially coming from me. I didn't it, especially coming from me. I didn't think the improvement was going to think the improvement was going to think the improvement was going to continue. I still remember the video I continue. I still remember the video I continue. I still remember the video I did where I said we were hitting the did where I said we were hitting the did where I said we were hitting the ceiling. probably the most wrong I've ceiling. probably the most wrong I've ceiling. probably the most wrong I've ever been in a video. Happy that I have ever been in a video. Happy that I have ever been in a video. Happy that I have that record and that I can come out here that record and that I can come out here that record and that I can come out here and tell you guys that I was wrong and and tell you guys that I was wrong and and tell you guys that I was wrong and you'll hopefully listen. But yeah, I was you'll hopefully listen. But yeah, I was you'll hopefully listen. But yeah, I was wrong. In March of 2024, Opus 3 could complete In March of 2024, Opus 3 could complete software tasks that took humans about 4 software tasks that took humans about 4 software tasks that took humans about 4 minutes to complete. A year later, minutes to complete. A year later, minutes to complete. A year later, Sonnet 37 is doing tasks that take an Sonnet 37 is doing tasks that take an Sonnet 37 is doing tasks that take an hour and a half. A year after that, Opus hour and a half. A year after that, Opus hour and a half. A year after that, Opus 46 is doing tasks that take humans 12 46 is doing tasks that take humans 12 46 is doing tasks that take humans 12 hours. If the trend holds, tasks that a hours. If the trend holds, tasks that a hours. If the trend holds, tasks that a skilled person would take days to skilled person would take days to skilled person would take days to complete would now be able complete would now be able complete would now be able retake. If this trend holds, tasks that retake. If this trend holds, tasks that retake. If this trend holds, tasks that take a skilled person days could come take a skilled person days could come take a skilled person days could come into range this year. In 2027, AI into range this year. In 2027, AI into range this year. In 2027, AI systems could be capable of tasks that systems could be capable of tasks that systems could be capable of tasks that took people weeks.

And to be clear, this is self or And to And to be clear, this is self or And to be clear, what this is referring to is be clear, what this is referring to is be clear, what this is referring to is work with the agent just going off by work with the agent just going off by work with the agent just going off by itself. If a human is there to give like itself. If a human is there to give like itself. If a human is there to give like thumbs up, thumbs down and share thumbs up, thumbs down and share thumbs up, thumbs down and share thoughts and steer throughout, it's very thoughts and steer throughout, it's very thoughts and steer throughout, it's very different because I can do years of work different because I can do years of work different because I can do years of work in a few days. If I can get an agent to in a few days. If I can get an agent to in a few days. If I can get an agent to do 12 hours of work over and over again do 12 hours of work over and over again do 12 hours of work over and over again with like 10 minutes of my work for each with like 10 minutes of my work for each with like 10 minutes of my work for each step, step, step, you can already get years of work done you can already get years of work done you can already get years of work done in much less time. But what happens if in much less time. But what happens if in much less time. But what happens if the ancient can do years of work by the ancient can do years of work by the ancient can do years of work by itself? It is also worth noting that most of the It is also worth noting that most of the measurements of these things actually measurements of these things actually measurements of these things actually show rough. It is also worth noting that the numbers It is also worth noting that the numbers they are citing here are 50% success they are citing here are 50% success they are citing here are 50% success rates which means that half the time rates which means that half the time rates which means that half the time it's still failing. The 80% version of it's still failing. The 80% version of it's still failing. The 80% version of the chart is significantly more damning. It was just the Yeah, if we switch to It was just the Yeah, if we switch to the 80% it goes from the 12 to 16 hours the 80% it goes from the 12 to 16 hours the 80% it goes from the 12 to 16 hours they mentioned before all the way down they mentioned before all the way down they mentioned before all the way down to 1 to 4 hours because getting it to to 1 to 4 hours because getting it to to 1 to 4 hours because getting it to complete because again this is all kind complete because again this is all kind complete because again this is all kind of random. These are slot machines that of random. These are slot machines that of random. These are slot machines that we're using to write code. 30% success we're using to write code. 30% success we're using to write code. 30% success rate is nuts for tasks that are 12 hours rate is nuts for tasks that are 12 hours rate is nuts for tasks that are 12 hours long with no human intervention. But if long with no human intervention. But if long with no human intervention. But if we want reliable building and reliable we want reliable building and reliable we want reliable building and reliable co-workers, 80% success massively drops co-workers, 80% success massively drops co-workers, 80% success massively drops the length of tasks that the models can

the length of tasks that the models can the length of tasks that the models can do. Just thought that was worth calling do. Just thought that was worth calling do. Just thought that was worth calling out. out. out. The same pattern appears on coding and The same pattern appears on coding and The same pattern appears on coding and research benchmarks. Benchmarks measure research benchmarks. Benchmarks measure research benchmarks. Benchmarks measure the performance of models in a given the performance of models in a given the performance of models in a given domain and they're saturated when models domain and they're saturated when models domain and they're saturated when models achieve close to 100% performance. achieve close to 100% performance. achieve close to 100% performance. SWEBench is a really annoying thing to SWEBench is a really annoying thing to SWEBench is a really annoying thing to site. Check out my video on SWbench and site. Check out my video on SWbench and site. Check out my video on SWbench and deep or check out my video on SWBNE deep or check out my video on SWBNE deep or check out my video on SWBNE Deepsw SWE where I go in depth on why Deepsw SWE where I go in depth on why Deepsw SWE where I go in depth on why this benchmark kind of sucks. Now, yeah, models are getting close to Now, yeah, models are getting close to saturating, but also like a lot of saturating, but also like a lot of saturating, but also like a lot of shitty models are getting over 50%. So, shitty models are getting over 50%. So, shitty models are getting over 50%. So, it's not a good benchmark. it's not a good benchmark. it's not a good benchmark. Corebench tests whether a model can Corebench tests whether a model can Corebench tests whether a model can reproduce existing research, a prerec reproduce existing research, a prerec reproduce existing research, a prerec for them to conduct original research. for them to conduct original research. for them to conduct original research. It gives an AI model. It gives an AI It gives an AI model. It gives an AI It gives an AI model. It gives an AI model the code and data behind a model the code and data behind a model the code and data behind a published paper and asks it to rerun published paper and asks it to rerun published paper and asks it to rerun everything and confirm it can replicate everything and confirm it can replicate everything and confirm it can replicate the paper's results. AI systems went the paper's results. AI systems went the paper's results. AI systems went from succeeding at reproducing the from succeeding at reproducing the from succeeding at reproducing the results roughly 20% of the time in 2024 results roughly 20% of the time in 2024 results roughly 20% of the time in 2024 to saturating the benchmark just 15 to saturating the benchmark just 15 to saturating the benchmark just 15 months later. Meter, the benchmark we months later. Meter, the benchmark we months later. Meter, the benchmark we just looked at for the long tasks, found just looked at for the long tasks, found just looked at for the long tasks, found that Claude Mythos could work for at that Claude Mythos could work for at that Claude Mythos could work for at least 16 hours and it was at the upper least 16 hours and it was at the upper least 16 hours and it was at the upper end of what they could measure without end of what they could measure without end of what they could measure without new tasks. But again, 50% success. When new tasks. But again, 50% success. When new tasks. But again, 50% success. When they switch to the 80% or when they they switch to the 80% or when they they switch to the 80% or when they switch to the 80% measurement, that goes switch to the 80% measurement, that goes switch to the 80% measurement, that goes down to 4 hours.

These benchmarks say a lot about the These benchmarks say a lot about the capabilities of the systems, but they capabilities of the systems, but they capabilities of the systems, but they can't reveal the impact AI systems are can't reveal the impact AI systems are can't reveal the impact AI systems are having on speeding up AI development having on speeding up AI development having on speeding up AI development itself. For that, we need direct itself. For that, we need direct itself. For that, we need direct evidence from within AI companies like evidence from within AI companies like evidence from within AI companies like Anthropic. Building a frontier model takes two Building a frontier model takes two broad categories of work. There's the broad categories of work. There's the broad categories of work. There's the engineering, which is things like engineering, which is things like engineering, which is things like writing the code, standing up the writing the code, standing up the writing the code, standing up the infrastructure, and overseeing the model infrastructure, and overseeing the model infrastructure, and overseeing the model training. But there's also the research, training. But there's also the research, training. But there's also the research, deciding what experiments to run, deciding what experiments to run, deciding what experiments to run, interpreting what comes back, and interpreting what comes back, and interpreting what comes back, and figuring out which ideas to try next. figuring out which ideas to try next. figuring out which ideas to try next. Across both edge and research, the Across both edge and research, the Across both edge and research, the picture's consistent. In engineering, picture's consistent. In engineering, picture's consistent. In engineering, Claude can be handed an underspecified Claude can be handed an underspecified Claude can be handed an underspecified problem and figure out how to solve it. problem and figure out how to solve it. problem and figure out how to solve it. Humans supply the goal. They no longer Humans supply the goal. They no longer Humans supply the goal. They no longer need to supply the method. In research, need to supply the method. In research, need to supply the method. In research, Claude can already match or outperform Claude can already match or outperform Claude can already match or outperform skilled humans at executing a well skilled humans at executing a well skilled humans at executing a well specified experiment. However, large specified experiment. However, large specified experiment. However, large performance gaps persist. performance gaps persist. performance gaps persist. However, large performance gaps persist However, large performance gaps persist However, large performance gaps persist when it comes to Claude exercising when it comes to Claude exercising when it comes to Claude exercising judgment in choosing goals in both judgment in choosing goals in both judgment in choosing goals in both engineering and in research. That's the engineering and in research. That's the engineering and in research. That's the gap between AI today and future systems gap between AI today and future systems gap between AI today and future systems that could autonomously design their own that could autonomously design their own that could autonomously design their own successors.

It's common for employees at Anthropic It's common for employees at Anthropic to receive more open-ended and important to receive more open-ended and important to receive more open-ended and important tasks as they gain more experience. tasks as they gain more experience. tasks as they gain more experience. Early on, they execute a task someone Early on, they execute a task someone Early on, they execute a task someone else specified, like the export button else specified, like the export button else specified, like the export button isn't working. Please fix it. With isn't working. Please fix it. With isn't working. Please fix it. With experience, they're handed a goal and a experience, they're handed a goal and a experience, they're handed a goal and a design approaches there. With design approaches there. With design approaches there. With experience, they're handed a goal and experience, they're handed a goal and experience, they're handed a goal and design the approach themselves, such as design the approach themselves, such as design the approach themselves, such as investigate why the network slows down investigate why the network slows down investigate why the network slows down under heavy load. At at the most senior under heavy load. At at the most senior under heavy load. At at the most senior levels, they are deciding what problems levels, they are deciding what problems levels, they are deciding what problems are worth working on at all, like what are worth working on at all, like what are worth working on at all, like what should the team build next quarter. We should the team build next quarter. We should the team build next quarter. We can use internal anthropic data to see can use internal anthropic data to see can use internal anthropic data to see how far Claude has come in being able to how far Claude has come in being able to how far Claude has come in being able to handle these different kinds of tasks. handle these different kinds of tasks. handle these different kinds of tasks. I've been trying this more myself, like I've been trying this more myself, like I've been trying this more myself, like seeing what AI is able to do when given seeing what AI is able to do when given seeing what AI is able to do when given a more vague goal. And with code, it's a more vague goal. And with code, it's a more vague goal. And with code, it's been really impressing me. It can been really impressing me. It can been really impressing me. It can generally find itself going in the right generally find itself going in the right generally find itself going in the right direction without too much steering. direction without too much steering. direction without too much steering. But for other things like making videos, But for other things like making videos, But for other things like making videos, I found it much less useful. I set up a I found it much less useful. I set up a I found it much less useful. I set up a Hermes agent to scroll through Twitter Hermes agent to scroll through Twitter Hermes agent to scroll through Twitter for me and find topics people are for me and find topics people are for me and find topics people are telling me I should cover and then rank telling me I should cover and then rank telling me I should cover and then rank them for me. And it went through for them for me. And it went through for them for me. And it went through for this today before I started streaming.

this today before I started streaming. this today before I started streaming. It has its list near the bottom here. It has its list near the bottom here. It has its list near the bottom here. These are the things it thinks I should These are the things it thinks I should These are the things it thinks I should make videos about. The first one is make videos about. The first one is make videos about. The first one is copilot versus cursor versus cloud code, copilot versus cursor versus cloud code, copilot versus cursor versus cloud code, a real cost benchmark. Two, why a real cost benchmark. Two, why a real cost benchmark. Two, why enterprises force developers to use enterprises force developers to use enterprises force developers to use co-pilot. Three, AI tools are training co-pilot. Three, AI tools are training co-pilot. Three, AI tools are training developers wrong. Four, AI coding developers wrong. Four, AI coding developers wrong. Four, AI coding benchmarks are fake. I kind of already benchmarks are fake. I kind of already benchmarks are fake. I kind of already did that. did that. did that. Five, the void and Cloudflare strategy, Five, the void and Cloudflare strategy, Five, the void and Cloudflare strategy, which it puts in fifth place of the which it puts in fifth place of the which it puts in fifth place of the things it has here, even though I things it has here, even though I things it has here, even though I already filmed that video because that already filmed that video because that already filmed that video because that one's actually really good. And then six one's actually really good. And then six one's actually really good. And then six is this vague Tanstack vulnerability is this vague Tanstack vulnerability is this vague Tanstack vulnerability that there isn't really any info on to that there isn't really any info on to that there isn't really any info on to report on. report on. report on. So, of the six topics, the second worst So, of the six topics, the second worst So, of the six topics, the second worst rated one it gave is the only one I know rated one it gave is the only one I know rated one it gave is the only one I know is worth doing, and even then not great. is worth doing, and even then not great. is worth doing, and even then not great. Most of these would not succeed as Most of these would not succeed as Most of these would not succeed as videos and also it would be videos and also it would be videos and also it would be significantly higher effort to produce. significantly higher effort to produce. significantly higher effort to produce. It is really bad at this. If you think It is really bad at this. If you think It is really bad at this. If you think it is doing well at this, that's it is doing well at this, that's it is doing well at this, that's probably because you also suck at probably because you also suck at probably because you also suck at YouTube. And that's fine. It's not a YouTube. And that's fine. It's not a YouTube. And that's fine. It's not a skill I expect a lot of people to have.

skill I expect a lot of people to have. skill I expect a lot of people to have. It's weird and requires you to rot your It's weird and requires you to rot your It's weird and requires you to rot your brain a whole bunch. But similarly, when brain a whole bunch. But similarly, when brain a whole bunch. But similarly, when you're not very good at code and you ask you're not very good at code and you ask you're not very good at code and you ask and but similarly, if you're not very and but similarly, if you're not very and but similarly, if you're not very good at code and you ask cloud code your good at code and you ask cloud code your good at code and you ask cloud code your for its thoughts on a thing you're going for its thoughts on a thing you're going for its thoughts on a thing you're going to do, it's going to glaze the [ __ ] out to do, it's going to glaze the [ __ ] out to do, it's going to glaze the [ __ ] out of you and you're going to think it's of you and you're going to think it's of you and you're going to think it's really smart. But if you put that in really smart. But if you put that in really smart. But if you put that in front of an experienced programmer front of an experienced programmer front of an experienced programmer that's familiar with what it's trying to that's familiar with what it's trying to that's familiar with what it's trying to do, it'll catch the holes. And while it do, it'll catch the holes. And while it do, it'll catch the holes. And while it has gotten way better at coding, has gotten way better at coding, has gotten way better at coding, it is significantly worse at content it is significantly worse at content it is significantly worse at content still. And you can tell when articles still. And you can tell when articles still. And you can tell when articles are written by AI. You can tell when are written by AI. You can tell when are written by AI. You can tell when video scripts are written by AI. It has video scripts are written by AI. It has video scripts are written by AI. It has a a feel to it. It's just not good at a a feel to it. It's just not good at a a feel to it. It's just not good at those things yet, especially when given those things yet, especially when given those things yet, especially when given vagger goals. vagger goals. vagger goals. It can help with some research for It can help with some research for It can help with some research for specific things. I use AI to find specific things. I use AI to find specific things. I use AI to find resources all the time, but it is not resources all the time, but it is not resources all the time, but it is not going to take these bigger like what going to take these bigger like what going to take these bigger like what should we build next quarter type tasks should we build next quarter type tasks should we build next quarter type tasks well at all. At least in the space that well at all. At least in the space that well at all. At least in the space that I am in with YouTube.

But it doesn't mean it can't code. And But it doesn't mean it can't code. And according to Anthropic, Claude is according to Anthropic, Claude is according to Anthropic, Claude is writing a significant portion of writing a significant portion of writing a significant portion of Anthropic code. As of May this year, Anthropic code. As of May this year, Anthropic code. As of May this year, more than 80% of the code they merge more than 80% of the code they merge more than 80% of the code they merge into Anthropics codebase was authored by into Anthropics codebase was authored by into Anthropics codebase was authored by Claude. Before Claude code launched in Claude. Before Claude code launched in Claude. Before Claude code launched in the research preview last year in the research preview last year in the research preview last year in February, the number was in the low February, the number was in the low February, the number was in the low single digits. The shift also shows up single digits. The shift also shows up single digits. The shift also shows up with the amount of output per engineer. with the amount of output per engineer. with the amount of output per engineer. Lines of code birch perge per day stayed Lines of code birch perge per day stayed Lines of code birch perge per day stayed constant throughout Anthropic's first constant throughout Anthropic's first constant throughout Anthropic's first four years from 2021 to 2024, but it four years from 2021 to 2024, but it four years from 2021 to 2024, but it began to climb upwards in 2025 when began to climb upwards in 2025 when began to climb upwards in 2025 when Claude began to run code rather than Claude began to run code rather than Claude began to run code rather than just suggesting that an engineer copy just suggesting that an engineer copy just suggesting that an engineer copy and paste it. The slope steepened again and paste it. The slope steepened again and paste it. The slope steepened again in 2026 when models began to work in 2026 when models began to work in 2026 when models began to work autonomously over long time horizons. autonomously over long time horizons. autonomously over long time horizons. The two inflection points are shown in The two inflection points are shown in The two inflection points are shown in the chart below. In the second quarter the chart below. In the second quarter the chart below. In the second quarter of 2026, the typical engineer was of 2026, the typical engineer was of 2026, the typical engineer was merging eight times as much code per day merging eight times as much code per day merging eight times as much code per day as they were in 2024. That's because as they were in 2024. That's because as they were in 2024. That's because much of the code is written by Claude much of the code is written by Claude much of the code is written by Claude with the engineering directing with the with the engineering directing with the with the engineering directing with the engineer directing and reviewing rather engineer directing and reviewing rather engineer directing and reviewing rather than typing the code themselves. Yep. than typing the code themselves. Yep. than typing the code themselves. Yep. You see here, especially once they got You see here, especially once they got You see here, especially once they got Claude mythos, the amount of code that Claude mythos, the amount of code that Claude mythos, the amount of code that they were writing with AI skyrocketed.

they were writing with AI skyrocketed. they were writing with AI skyrocketed. But also, Opus 45 was a massive But also, Opus 45 was a massive But also, Opus 45 was a massive improvement. And I know I write way more improvement. And I know I write way more improvement. And I know I write way more code since I started using Opus 45, too. code since I started using Opus 45, too. code since I started using Opus 45, too. But yeah, pretty nuts. kind of funny that after claw three the kind of funny that after claw three the amount of code they wrote per quarter amount of code they wrote per quarter amount of code they wrote per quarter went down slightly but yeah went down slightly but yeah went down slightly but yeah it is worth noting that lines of code is it is worth noting that lines of code is it is worth noting that lines of code is an imperfect measure. I'm thankful that an imperfect measure. I'm thankful that an imperfect measure. I'm thankful that they called this out. they called this out. they called this out. They also say that the 8x of they all They also say that the 8x of they all They also say that the 8x of they all say the 8x lines of code per inch per say the 8x lines of code per inch per say the 8x lines of code per inch per day is almost certainly an overstatement day is almost certainly an overstatement day is almost certainly an overstatement of the true productivity gains but it of the true productivity gains but it of the true productivity gains but it does in but it does indicate an does in but it does indicate an does in but it does indicate an acceleration. At Anthropic we don't acceleration. At Anthropic we don't acceleration. At Anthropic we don't reward people for how many lines of code reward people for how many lines of code reward people for how many lines of code they write. Rather, team members are they write. Rather, team members are they write. Rather, team members are producing more code simply because producing more code simply because producing more code simply because they're using AI systems to write more they're using AI systems to write more they're using AI systems to write more code. The increase in lines of code code. The increase in lines of code code. The increase in lines of code written lines up with subjective written lines up with subjective written lines up with subjective impressions of large productivity impressions of large productivity impressions of large productivity increases. In a March 2026 poll of 130 increases. In a March 2026 poll of 130 increases. In a March 2026 poll of 130 employees from across anthropic research employees from across anthropic research employees from across anthropic research teams, the median respondent estimated teams, the median respondent estimated teams, the median respondent estimated that they produced around 4x as much that they produced around 4x as much that they produced around 4x as much output with mythos preview as they would output with mythos preview as they would output with mythos preview as they would have without access to any AI models on have without access to any AI models on have without access to any AI models on the kinds of projects that they would be the kinds of projects that they would be the kinds of projects that they would be working on regardless.

working on regardless. working on regardless. that is research team not just ange that is research team not just ange that is research team not just ange research is seeing a 4x which is kind of research is seeing a 4x which is kind of research is seeing a 4x which is kind of nuts. nuts. nuts. We expect the true degree of uplift in We expect the true degree of uplift in We expect the true degree of uplift in March was somewhat lower. Nevertheless, March was somewhat lower. Nevertheless, March was somewhat lower. Nevertheless, we find the overall claim plausible and we find the overall claim plausible and we find the overall claim plausible and in line with our other observations a in line with our other observations a in line with our other observations a significant fraction of anthropic significant fraction of anthropic significant fraction of anthropic technical staff is accomplishing their technical staff is accomplishing their technical staff is accomplishing their core work multiple times faster than core work multiple times faster than core work multiple times faster than they could without AI assistance. they could without AI assistance. they could without AI assistance. We also see evidence that people at We also see evidence that people at We also see evidence that people at Anthropic are using cloud to do work Anthropic are using cloud to do work Anthropic are using cloud to do work that simply wouldn't have happened that simply wouldn't have happened that simply wouldn't have happened otherwise, like building exploratory otherwise, like building exploratory otherwise, like building exploratory tooling and addressing long deferred tooling and addressing long deferred tooling and addressing long deferred cleanups. I was saying this for a while. cleanups. I was saying this for a while. cleanups. I was saying this for a while. AI is so good at doing like annoying AI is so good at doing like annoying AI is so good at doing like annoying backlog [ __ ] like doing some crazy tech backlog [ __ ] like doing some crazy tech backlog [ __ ] like doing some crazy tech deck cleanup that was just miserable to deck cleanup that was just miserable to deck cleanup that was just miserable to do before. It can do that stuff great. do before. It can do that stuff great. do before. It can do that stuff great. In April, Claude shipped over 800 fixes In April, Claude shipped over 800 fixes In April, Claude shipped over 800 fixes that reduced a class of API errors by a that reduced a class of API errors by a that reduced a class of API errors by a factor of a thousand. The engineer factor of a thousand. The engineer factor of a thousand. The engineer overseeing Claude estimated that a human overseeing Claude estimated that a human overseeing Claude estimated that a human would have taken four years to complete would have taken four years to complete would have taken four years to complete that work. Solving other people's bugs that work. Solving other people's bugs that work. Solving other people's bugs is slow and painstaking and humans is slow and painstaking and humans is slow and painstaking and humans struggle to hold that much unfamiliar struggle to hold that much unfamiliar struggle to hold that much unfamiliar context in their head at once. context in their head at once. context in their head at once. Man, wouldn't it be nice if they did Man, wouldn't it be nice if they did Man, wouldn't it be nice if they did that to Claude Code?

that to Claude Code? that to Claude Code? Anyways, this is a quote from an Anyways, this is a quote from an Anyways, this is a quote from an employee. Started leaning hard into employee. Started leaning hard into employee. Started leaning hard into Cloudifying about a year ago. It's been Cloudifying about a year ago. It's been Cloudifying about a year ago. It's been a crazy adventure and it's now been a crazy adventure and it's now been a crazy adventure and it's now been around 5 months since I last wrote any around 5 months since I last wrote any around 5 months since I last wrote any code myself. Damn. code myself. Damn. code myself. Damn. The code that Claude writes is good and The code that Claude writes is good and The code that Claude writes is good and it's improving. Good code means two it's improving. Good code means two it's improving. Good code means two things. It works and it's written in a things. It works and it's written in a things. It works and it's written in a manner that allows another engineer to manner that allows another engineer to manner that allows another engineer to understand it and build upon it. On the understand it and build upon it. On the understand it and build upon it. On the first criteria, the evidence is clear. first criteria, the evidence is clear. first criteria, the evidence is clear. The rate at which anthropic staff The rate at which anthropic staff The rate at which anthropic staff correct, redirect or take over midtask correct, redirect or take over midtask correct, redirect or take over midtask from Claude has been failing steadily or from Claude has been failing steadily or from Claude has been failing steadily or ret. The rate at which anthropic staff ret. The rate at which anthropic staff ret. The rate at which anthropic staff interject and like intervene with interject and like intervene with interject and like intervene with Claude's work mid task has fall has Claude's work mid task has fall has Claude's work mid task has fall has fallen steadily for a year, including on fallen steadily for a year, including on fallen steadily for a year, including on the most complex and open-ended tasks. And this is the success rates. And you And this is the success rates. And you can see how massively these have can see how massively these have can see how massively these have improved. For trivial tasks, things have improved. For trivial tasks, things have improved. For trivial tasks, things have went gotten better, but not a lot. In went gotten better, but not a lot. In went gotten better, but not a lot. In fact, this is actually really funny to fact, this is actually really funny to fact, this is actually really funny to see. According to their own see. According to their own see. According to their own measurements, the yellow line here for measurements, the yellow line here for measurements, the yellow line here for trivial tasks, Claude has gotten worse trivial tasks, Claude has gotten worse trivial tasks, Claude has gotten worse over time and needs more correction. over time and needs more correction. over time and needs more correction. They briefly hit a 100% rate for doing They briefly hit a 100% rate for doing They briefly hit a 100% rate for doing simple tasks, but it has declined simple tasks, but it has declined simple tasks, but it has declined steadily since.

steadily since. steadily since. But these heavier tasks have been going But these heavier tasks have been going But these heavier tasks have been going up, which is very interesting and lines up, which is very interesting and lines up, which is very interesting and lines up with my experience using anthropic up with my experience using anthropic up with my experience using anthropic models. You can't get them to do simple models. You can't get them to do simple models. You can't get them to do simple [ __ ] anymore, but they will gladly run [ __ ] anymore, but they will gladly run [ __ ] anymore, but they will gladly run for hours trying to solve really complex for hours trying to solve really complex for hours trying to solve really complex problems. problems. problems. But I do love that in their own But I do love that in their own But I do love that in their own measurements, they are admitting that measurements, they are admitting that measurements, they are admitting that trivial tasks have regressed from where trivial tasks have regressed from where trivial tasks have regressed from where they were in December of last year. On the biggest and most open-ended On the biggest and most open-ended tasks, Claude's success rate hit as high tasks, Claude's success rate hit as high tasks, Claude's success rate hit as high as 76% in May of 2026, which is up 50 as 76% in May of 2026, which is up 50 as 76% in May of 2026, which is up 50 percentage points in just 6 months. To percentage points in just 6 months. To percentage points in just 6 months. To give an example of tasks in this tier, a give an example of tasks in this tier, a give an example of tasks in this tier, a routine upgrade began crashing tens of routine upgrade began crashing tens of routine upgrade began crashing tens of thousands of training jobs. thousands of training jobs. thousands of training jobs. An engineer pointed Claude at the live An engineer pointed Claude at the live An engineer pointed Claude at the live incident with little more than some text incident with little more than some text incident with little more than some text context and some cluster access. I agree context and some cluster access. I agree context and some cluster access. I agree on that. An engineer pointed Claude at on that. An engineer pointed Claude at on that. An engineer pointed Claude at the live incident with little more than the live incident with little more than the live incident with little more than some text content and cluster access.

some text content and cluster access. some text content and cluster access. Working through the running jobs and Working through the running jobs and Working through the running jobs and testing one environment setting at a testing one environment setting at a testing one environment setting at a time, Claude isolated the single obscure time, Claude isolated the single obscure time, Claude isolated the single obscure debugging flag that was triggering the debugging flag that was triggering the debugging flag that was triggering the crash, reproduced it reliably and crash, reproduced it reliably and crash, reproduced it reliably and confirmed a fix. In about 2 hours, confirmed a fix. In about 2 hours, confirmed a fix. In about 2 hours, Claude delivered what would have Claude delivered what would have Claude delivered what would have normally been 2 or 3 days of work. normally been 2 or 3 days of work. normally been 2 or 3 days of work. AI debugging is so cool and I'm amazed AI debugging is so cool and I'm amazed AI debugging is so cool and I'm amazed why people aren't taking advantage of why people aren't taking advantage of why people aren't taking advantage of this type of thing. It's so good. The second criteria is writing code that The second criteria is writing code that another engineer can understand and another engineer can understand and another engineer can understand and build on. Here the gap between humans build on. Here the gap between humans build on. Here the gap between humans and AI still persists, but it's starting and AI still persists, but it's starting and AI still persists, but it's starting to close fast. There isn't full to close fast. There isn't full to close fast. There isn't full consensus among staff at Anthropic, but consensus among staff at Anthropic, but consensus among staff at Anthropic, but many believe that the cloud written code many believe that the cloud written code many believe that the cloud written code was still worse in quality than human was still worse in quality than human was still worse in quality than human written code at Anthropic in late 2025 written code at Anthropic in late 2025 written code at Anthropic in late 2025 and is roughly at parody today. We and is roughly at parody today. We and is roughly at parody today. We expected to be better than the human expected to be better than the human expected to be better than the human written code within the year. written code within the year. written code within the year. Considering the quality of the anthropic Considering the quality of the anthropic Considering the quality of the anthropic code, I would not be surprised if within code, I would not be surprised if within code, I would not be surprised if within their measurements that was to happen, their measurements that was to happen, their measurements that was to happen, but uh not aligned with my experience. but uh not aligned with my experience. but uh not aligned with my experience. This has changed the way Enthropic now This has changed the way Enthropic now This has changed the way Enthropic now reviews its own code. Proposed changes reviews its own code. Proposed changes reviews its own code. Proposed changes to our codebase are now read by an to our codebase are now read by an to our codebase are now read by an automated cloud reviewer that looks for automated cloud reviewer that looks for automated cloud reviewer that looks for bugs, security flaws, and other defects bugs, security flaws, and other defects bugs, security flaws, and other defects before it can merge. I love that cloud before it can merge. I love that cloud before it can merge. I love that cloud reviewer cost 25 bucks or so per review.

reviewer cost 25 bucks or so per review. reviewer cost 25 bucks or so per review. But yeah, using this tool, we ran a But yeah, using this tool, we ran a But yeah, using this tool, we ran a retrospective analysis and found that an retrospective analysis and found that an retrospective analysis and found that an automated cloud review of every change automated cloud review of every change automated cloud review of every change to our codebase would have caught to our codebase would have caught to our codebase would have caught roughly a third of the bugs behind past roughly a third of the bugs behind past roughly a third of the bugs behind past incidents on Claude AI before they ever incidents on Claude AI before they ever incidents on Claude AI before they ever reached production. The engineers who reached production. The engineers who reached production. The engineers who wrote that code are among the best in wrote that code are among the best in wrote that code are among the best in the world at building these systems. bot the world at building these systems. bot the world at building these systems. bot is now catching the mistakes that they is now catching the mistakes that they is now catching the mistakes that they missed. missed. missed. Fascinating. This actually touches on a Fascinating. This actually touches on a Fascinating. This actually touches on a thing I'm planning on doing a whole thing I'm planning on doing a whole thing I'm planning on doing a whole dedicated video on, which is how to dedicated video on, which is how to dedicated video on, which is how to succeed at a company that's measuring succeed at a company that's measuring succeed at a company that's measuring your token maxing. One of the strategies your token maxing. One of the strategies your token maxing. One of the strategies I want to recommend is that you build I want to recommend is that you build I want to recommend is that you build your own tools to catch every potential your own tools to catch every potential your own tools to catch every potential regression in every single PR and then regression in every single PR and then regression in every single PR and then map every incident to see if any of the map every incident to see if any of the map every incident to see if any of the issues you detected align with those issues you detected align with those issues you detected align with those incidents so that you can create a incidents so that you can create a incidents so that you can create a historical like diagram and like a historical like diagram and like a historical like diagram and like a historical historical historical also you can create a historical record also you can create a historical record also you can create a historical record of what incidents happened and how they of what incidents happened and how they of what incidents happened and how they could have been caught. It'll make you could have been caught. It'll make you could have been caught. It'll make you look really really good to your bosses.

Claude is good at running experiments to Claude is good at running experiments to hit a goal that someone else has set. hit a goal that someone else has set. hit a goal that someone else has set. Every time Anthropic releases a model, Every time Anthropic releases a model, Every time Anthropic releases a model, we run the same test. We give Claude we run the same test. We give Claude we run the same test. We give Claude some code that trains a small AI model some code that trains a small AI model some code that trains a small AI model and ask it to make the code run as fast and ask it to make the code run as fast and ask it to make the code run as fast as possible while still passing the same as possible while still passing the same as possible while still passing the same correctness checks. The goal and the correctness checks. The goal and the correctness checks. The goal and the success metrics are fixed in advance. So success metrics are fixed in advance. So success metrics are fixed in advance. So Claude's job is to find speed ups by Claude's job is to find speed ups by Claude's job is to find speed ups by rewriting the code, running it, timing rewriting the code, running it, timing rewriting the code, running it, timing it, and repeating. the miniature version it, and repeating. the miniature version it, and repeating. the miniature version of an experimental research loop. In May of an experimental research loop. In May of an experimental research loop. In May 2025, Opus 4 averaged a 3x speed up over 2025, Opus 4 averaged a 3x speed up over 2025, Opus 4 averaged a 3x speed up over the starting code. By April of this the starting code. By April of this the starting code. By April of this year, Claude Mythos preview was year, Claude Mythos preview was year, Claude Mythos preview was achieving a 52x. For calibration, a achieving a 52x. For calibration, a achieving a 52x. For calibration, a skilled human researcher for skilled human researcher for skilled human researcher for calibration, a skilled human researcher calibration, a skilled human researcher calibration, a skilled human researcher would need 4 to 8 hours to reach 4x. In would need 4 to 8 hours to reach 4x. In would need 4 to 8 hours to reach 4x. In this part of In this part of research this part of In this part of research this part of In this part of research workflows, the optimizing step with a workflows, the optimizing step with a workflows, the optimizing step with a clearly defined experiment, Claude has clearly defined experiment, Claude has clearly defined experiment, Claude has gone from super helpful to superhuman in gone from super helpful to superhuman in gone from super helpful to superhuman in under a year. under a year. under a year. Yeah. The shape of stuff today is roughly that The shape of stuff today is roughly that a human has an idea and the models are a human has an idea and the models are a human has an idea and the models are able to implement, test, and evaluate able to implement, test, and evaluate able to implement, test, and evaluate them in order of magnitude faster than them in order of magnitude faster than them in order of magnitude faster than before. Yep.

But it's also getting better at But it's also getting better at proposing its own experiments. In April proposing its own experiments. In April proposing its own experiments. In April of 2026, Anthropic published the first of 2026, Anthropic published the first of 2026, Anthropic published the first demonstration of Claude running an demonstration of Claude running an demonstration of Claude running an open-ended research project end to end. open-ended research project end to end. open-ended research project end to end. Claude powered agents were given an open Claude powered agents were given an open Claude powered agents were given an open problem in AI safety roughly. Can a problem in AI safety roughly. Can a problem in AI safety roughly. Can a weaker model reliably supervise a weaker model reliably supervise a weaker model reliably supervise a stronger one and they were left to solve stronger one and they were left to solve stronger one and they were left to solve it. This involved proposing hypotheses, it. This involved proposing hypotheses, it. This involved proposing hypotheses, testing them, sharing findings with testing them, sharing findings with testing them, sharing findings with parallel agents and iterating. The task parallel agents and iterating. The task parallel agents and iterating. The task had a clear performance floor and had a clear performance floor and had a clear performance floor and ceiling. The floor is how well the ceiling. The floor is how well the ceiling. The floor is how well the weaker supervisor would do on its own weaker supervisor would do on its own weaker supervisor would do on its own and the ceiling is how strong models do and the ceiling is how strong models do and the ceiling is how strong models do when trained on correct answers. when trained on correct answers. when trained on correct answers. Two human researchers over about a week Two human researchers over about a week Two human researchers over about a week covered roughly 23% of the gap. The ages covered roughly 23% of the gap. The ages covered roughly 23% of the gap. The ages recovered 97% over 800 cumulative hours recovered 97% over 800 cumulative hours recovered 97% over 800 cumulative hours and use roughly 18K in compute to do it. and use roughly 18K in compute to do it. and use roughly 18K in compute to do it. There are some caveats to this work. The There are some caveats to this work. The There are some caveats to this work. The research didn't transfer cleanly to research didn't transfer cleanly to research didn't transfer cleanly to production scale models and humans still production scale models and humans still production scale models and humans still choose the problem and create the choose the problem and create the choose the problem and create the scoring rubric. But within those bounds, scoring rubric. But within those bounds, scoring rubric. But within those bounds, the agent designed every experiment the agent designed every experiment the agent designed every experiment themselves. Direction setting was themselves. Direction setting was themselves. Direction setting was direction setting was the only direction setting was the only direction setting was the only meaningful role a human played. This is meaningful role a human played. This is meaningful role a human played. This is fascinating and definitely leans into fascinating and definitely leans into fascinating and definitely leans into like AI will be able to self-improve like AI will be able to self-improve like AI will be able to self-improve soon. That's kind of nuts.

Claude did all of this with pretty Claude did all of this with pretty minimal help from me over the course of minimal help from me over the course of minimal help from me over the course of 1 to two days. I think if a junior 1 to two days. I think if a junior 1 to two days. I think if a junior colleague came back to me with results colleague came back to me with results colleague came back to me with results like this in the same span of time, I like this in the same span of time, I like this in the same span of time, I would be mildly impressed. Future's now would be mildly impressed. Future's now would be mildly impressed. Future's now interesting is getting better at steering research is getting better at steering research sessions towards research findings. We sessions towards research findings. We sessions towards research findings. We examined real claude code sessions examined real claude code sessions examined real claude code sessions between January and March where between January and March where between January and March where anthropic researchers were working with anthropic researchers were working with anthropic researchers were working with Claude on open-ended investigation Claude on open-ended investigation Claude on open-ended investigation problems like figuring out why a problems like figuring out why a problems like figuring out why a training run was crashing or why a model training run was crashing or why a model training run was crashing or why a model scored poorly on a bench. In each case, scored poorly on a bench. In each case, scored poorly on a bench. In each case, we found a moment where the research we found a moment where the research we found a moment where the research took a detour. In each case, we found a took a detour. In each case, we found a took a detour. In each case, we found a moment where the researcher took a moment where the researcher took a moment where the researcher took a detour. They pursued a direction that detour. They pursued a direction that detour. They pursued a direction that sent the session sideways before it sent the session sideways before it sent the session sideways before it eventually got back on track. We then eventually got back on track. We then eventually got back on track. We then showed various claw models only the work showed various claw models only the work showed various claw models only the work from before the session went off course from before the session went off course from before the session went off course and asked what it would do next. a and asked what it would do next. a and asked what it would do next. a separate claude that was able to see how separate claude that was able to see how separate claude that was able to see how the session eventually turned out, then the session eventually turned out, then the session eventually turned out, then judge whether the AI or the human judge whether the AI or the human judge whether the AI or the human suggested s suggested s suggested s a separate claude that was able to see a separate claude that was able to see a separate claude that was able to see how the session eventually turned out how the session eventually turned out how the session eventually turned out then judge whether the AI or the human then judge whether the AI or the human then judge whether the AI or the human suggested the better next step suggested the better next step suggested the better next step because we deliberately picked moments because we deliberately picked moments because we deliberately picked moments where we knew the human's choice had where we knew the human's choice had where we knew the human's choice had room for improvement. This isn't a like room for improvement. This isn't a like room for improvement. This isn't a like for-like comparison between models and for-like comparison between models and for-like comparison between models and human judgment.

human judgment. human judgment. What these moments give us is a set of What these moments give us is a set of What these moments give us is a set of realistic challenging situations where realistic challenging situations where realistic challenging situations where the right next step is not obvious and the right next step is not obvious and the right next step is not obvious and where the human's choice serve as a where the human's choice serve as a where the human's choice serve as a useful yard stick compared to model useful yard stick compared to model useful yard stick compared to model performance over time. On this measure, performance over time. On this measure, performance over time. On this measure, our best November on this measure, our our best November on this measure, our our best November on this measure, our best model in November 2025 beat the best model in November 2025 beat the best model in November 2025 beat the human choice 51% of the time, but now human choice 51% of the time, but now human choice 51% of the time, but now with Mythos preview, it's up to 64%. with Mythos preview, it's up to 64%. with Mythos preview, it's up to 64%. What this is saying is if the human What this is saying is if the human What this is saying is if the human chooses what to do next, it performs chooses what to do next, it performs chooses what to do next, it performs slightly worse than if the AI chooses. slightly worse than if the AI chooses. slightly worse than if the AI chooses. But they have AI evaluating the choices, But they have AI evaluating the choices, But they have AI evaluating the choices, so it's hard to know for sure. So what does this mean for the future of So what does this mean for the future of work at Anthropic? The evidence suggests work at Anthropic? The evidence suggests work at Anthropic? The evidence suggests that the human role is narrowing at each that the human role is narrowing at each that the human role is narrowing at each step in the AI development process. Once step in the AI development process. Once step in the AI development process. Once human and AI authored code quality reach human and AI authored code quality reach human and AI authored code quality reach par, humans will stop writing code par, humans will stop writing code par, humans will stop writing code entirely and shift to only reviewing it. entirely and shift to only reviewing it. entirely and shift to only reviewing it. But if they can't review code as quickly But if they can't review code as quickly But if they can't review code as quickly as Claude can generate it, human review as Claude can generate it, human review as Claude can generate it, human review will become the bottleneck to AI will become the bottleneck to AI will become the bottleneck to AI development. Are you saying that that's development. Are you saying that that's development. Are you saying that that's not already the case? Because that's not already the case? Because that's not already the case? Because that's definitely the case for us. Maybe it's definitely the case for us. Maybe it's definitely the case for us. Maybe it's just cuz we're using OpenAI models.

just cuz we're using OpenAI models. just cuz we're using OpenAI models. That is actually one of the biggest That is actually one of the biggest That is actually one of the biggest disadvantages of being at a lab. When disadvantages of being at a lab. When disadvantages of being at a lab. When you're at OpenAI, you're not using cloud you're at OpenAI, you're not using cloud you're at OpenAI, you're not using cloud models. And when you're at anthropic, models. And when you're at anthropic, models. And when you're at anthropic, you're not using Open AI models. So, you're not using Open AI models. So, you're not using Open AI models. So, anyone outside of those two companies anyone outside of those two companies anyone outside of those two companies has a huge advantage because they can has a huge advantage because they can has a huge advantage because they can use both for their strengths and use both for their strengths and use both for their strengths and weaknesses and switch to whatever is weaknesses and switch to whatever is weaknesses and switch to whatever is best at any given time. They also say that once Claude can run They also say that once Claude can run experiments, the question shifts towards experiments, the question shifts towards experiments, the question shifts towards which of the experiments is actually which of the experiments is actually which of the experiments is actually worth running. Put simply, the doing, worth running. Put simply, the doing, worth running. Put simply, the doing, which is writing code, running which is writing code, running which is writing code, running experiments, and producing the results, experiments, and producing the results, experiments, and producing the results, now costs almost nothing in human time, now costs almost nothing in human time, now costs almost nothing in human time, even if it still has costs in compute. An area of human comparative advantage An area of human comparative advantage for now is research, taste, and for now is research, taste, and for now is research, taste, and judgment, including choosing which judgment, including choosing which judgment, including choosing which problems matter, which results to trust, problems matter, which results to trust, problems matter, which results to trust, and when an approach is a dead end. How and when an approach is a dead end. How and when an approach is a dead end. How often does the word taste appear in this often does the word taste appear in this often does the word taste appear in this article?

article? article? Four times Four times Four times grass. And now we get into anthropic being And now we get into anthropic being existential and depressing. On days existential and depressing. On days existential and depressing. On days where everything works well, I can't where everything works well, I can't where everything works well, I can't help but think that nothing I do help but think that nothing I do help but think that nothing I do matters. Everything is automated and matters. Everything is automated and matters. Everything is automated and better and faster than I'll ever be. But better and faster than I'll ever be. But better and faster than I'll ever be. But then there are days where everything then there are days where everything then there are days where everything breaks and I don't understand why and I breaks and I don't understand why and I breaks and I don't understand why and I realize I have no idea what I've been up realize I have no idea what I've been up realize I have no idea what I've been up to anymore. to anymore. to anymore. Yeah. Yeah. Yeah. Yeah. Yeah. Yeah. So what if we're wrong? We is an So what if we're wrong? We is an So what if we're wrong? We is an anthropic here. A natural objection to anthropic here. A natural objection to anthropic here. A natural objection to the evidence presented above is that the the evidence presented above is that the the evidence presented above is that the work is still in human hands. Choosing work is still in human hands. Choosing work is still in human hands. Choosing which problems to work on is still what which problems to work on is still what which problems to work on is still what matters most. Without that judgment, matters most. Without that judgment, matters most. Without that judgment, Claude is a capable assistant, but not a Claude is a capable assistant, but not a Claude is a capable assistant, but not a system that could drive AI progress on system that could drive AI progress on system that could drive AI progress on its own. It's genuinely unclear whether its own. It's genuinely unclear whether its own. It's genuinely unclear whether today's training methods and today's training methods and today's training methods and architectures could unlock that architectures could unlock that architectures could unlock that capability. It's genuinely unclear capability. It's genuinely unclear capability. It's genuinely unclear whether today's training methods and whether today's training methods and whether today's training methods and architectures could unlock that architectures could unlock that architectures could unlock that capacity. But AI is rarely advanced by capacity. But AI is rarely advanced by capacity. But AI is rarely advanced by Eureka moments. There have been a few of Eureka moments. There have been a few of Eureka moments. There have been a few of those in AI's recent history like the those in AI's recent history like the those in AI's recent history like the transformer architecture ore models. But transformer architecture ore models. But transformer architecture ore models. But paradigm shifting I love they didn't put paradigm shifting I love they didn't put paradigm shifting I love they didn't put reasoning in here because they don't reasoning in here because they don't reasoning in here because they don't want to give OpenAI the credit. But want to give OpenAI the credit. But want to give OpenAI the credit. But paradigm shifting ideas arrive years paradigm shifting ideas arrive years paradigm shifting ideas arrive years apart. In between most progress is apart. In between most progress is apart. In between most progress is incremental. We scale something up, we incremental. We scale something up, we incremental. We scale something up, we see what breaks, we fix it, and then try see what breaks, we fix it, and then try see what breaks, we fix it, and then try again. This is exactly the kind of again. This is exactly the kind of again. This is exactly the kind of workflow Claude now excels at. Edison

workflow Claude now excels at. Edison workflow Claude now excels at. Edison said that genius is 1% inspiration and said that genius is 1% inspiration and said that genius is 1% inspiration and 99% perspiration. 99% perspiration. 99% perspiration. But we see perspiration becoming But we see perspiration becoming But we see perspiration becoming increasingly automated. I do actually increasingly automated. I do actually increasingly automated. I do actually really like that framing. That really like that framing. That really like that framing. That definitely feels like it's happening definitely feels like it's happening definitely feels like it's happening now. And p and the perspiration or like now. And p and the perspiration or like now. And p and the perspiration or like drive is now very different where it's drive is now very different where it's drive is now very different where it's more not losing hope during the moments more not losing hope during the moments more not losing hope during the moments where the slot machine isn't hitting where the slot machine isn't hitting where the slot machine isn't hitting where you need it to. It's very It's where you need it to. It's very It's where you need it to. It's very It's similar but different. It interesting. I similar but different. It interesting. I similar but different. It interesting. I need to think more about that part. need to think more about that part. need to think more about that part. Throwback says that it's becoming clear Throwback says that it's becoming clear Throwback says that it's becoming clear that much of what advances the frontier that much of what advances the frontier that much of what advances the frontier is automatable. Large-scale research is automatable. Large-scale research is automatable. Large-scale research progress is mostly a function of tools progress is mostly a function of tools progress is mostly a function of tools and resources which dictate how fast you and resources which dictate how fast you and resources which dictate how fast you can run experiments, how many you can can run experiments, how many you can can run experiments, how many you can run at once, and how quickly you can get run at once, and how quickly you can get run at once, and how quickly you can get results. results. results. Even if we suppose that Claude never Even if we suppose that Claude never Even if we suppose that Claude never achieves good research taste, a achieves good research taste, a achieves good research taste, a conservative reading of our evidence conservative reading of our evidence conservative reading of our evidence still implies compounding acceleration. still implies compounding acceleration. still implies compounding acceleration. If humans spend most of their time on If humans spend most of their time on If humans spend most of their time on singledigit fraction of work that is singledigit fraction of work that is singledigit fraction of work that is direction setting while claude handles direction setting while claude handles direction setting while claude handles the rest that means every engineer or the rest that means every engineer or the rest that means every engineer or researcher is steering far more work researcher is steering far more work researcher is steering far more work than before. The evidence we see than before. The evidence we see than before. The evidence we see suggests that in people the evidence suggests that in people the evidence suggests that in people the evidence that we see suggests that people at that we see suggests that people at that we see suggests that people at anthropic are both moving faster and anthropic are both moving faster and anthropic are both moving faster and covering a broader surface. In practice, covering a broader surface. In practice, covering a broader surface. In practice, this means that AI already makes this means that AI already makes this means that AI already makes anthropic move much faster than it did anthropic move much faster than it did anthropic move much faster than it did before the advances and effectiveness of before the advances and effectiveness of before the advances and effectiveness of AI technologies and tools.

The less conservative reading is that The less conservative reading is that early evidence on clause improving early evidence on clause improving early evidence on clause improving research judgment narrow as it is today research judgment narrow as it is today research judgment narrow as it is today is an indicator that the capability is is an indicator that the capability is is an indicator that the capability is improving as well. Research taste might improving as well. Research taste might improving as well. Research taste might just be another AI capability that AI just be another AI capability that AI just be another AI capability that AI systems fail at over time and then systems fail at over time and then systems fail at over time and then suddenly get good at. We've seen similar suddenly get good at. We've seen similar suddenly get good at. We've seen similar patterns with other qualitative skills patterns with other qualitative skills patterns with other qualitative skills like AI systems being able to explain like AI systems being able to explain like AI systems being able to explain why a joke is funny, demonstrate theory why a joke is funny, demonstrate theory why a joke is funny, demonstrate theory of mind, and solve linguistic riddles. of mind, and solve linguistic riddles. of mind, and solve linguistic riddles. True. True. True. Curious where this goes. They propose a handful of possible They propose a handful of possible futures. futures. futures. They depend on whether the trend They depend on whether the trend They depend on whether the trend continues and what we choose to do if it continues and what we choose to do if it continues and what we choose to do if it does. These are the three scenarios they does. These are the three scenarios they does. These are the three scenarios they can imagine. The first is that the trend can imagine. The first is that the trend can imagine. The first is that the trend stalls, but today's AI capabilities are stalls, but today's AI capabilities are stalls, but today's AI capabilities are widely diffused. widely diffused. widely diffused. The this article features many The this article features many The this article features many exponential trajectories, but those exponential trajectories, but those exponential trajectories, but those trajectories may actually turn out to be trajectories may actually turn out to be trajectories may actually turn out to be S-curves. we may be approaching the bend S-curves. we may be approaching the bend S-curves. we may be approaching the bend of the curve where returns to scale of the curve where returns to scale of the curve where returns to scale diminish and the line straightens then diminish and the line straightens then diminish and the line straightens then flattens. Well, in this case, it flattens. Well, in this case, it flattens. Well, in this case, it actually goes down as they showed here actually goes down as they showed here actually goes down as they showed here where the line curved up and then in the where the line curved up and then in the where the line curved up and then in the case of these simple tasks started to case of these simple tasks started to case of these simple tasks started to curve down.

What is the realist take that you What is the realist take that you grabbed [snorts] here, Dan LV? I'm grabbed [snorts] here, Dan LV? I'm grabbed [snorts] here, Dan LV? I'm curious. I don't want to click and curious. I don't want to click and curious. I don't want to click and watch, but I don't want to grab my watch, but I don't want to grab my watch, but I don't want to grab my headphones. Just give Dr. what I said. Oh [ __ ] energy. Thanks for coming back. Oh [ __ ] energy. Thanks for coming back. I'll link all the subs properly in a I'll link all the subs properly in a I'll link all the subs properly in a bit. bit. bit. Link the article. Sure. Link the article. Sure. Link the article. Sure. Oh, that labs don't use other lab Oh, that labs don't use other lab Oh, that labs don't use other lab models. That's the field. Oh, yeah. models. That's the field. Oh, yeah. models. That's the field. Oh, yeah. That's a fun clip. That's a fun clip. That's a fun clip. Yeah. So, as I was saying, Yeah. So, as I was saying, Yeah. So, as I was saying, The trends could stall, but the The trends could stall, but the The trends could stall, but the capabilities that we have now could be capabilities that we have now could be capabilities that we have now could be way more accessible. They could be way more accessible. They could be way more accessible. They could be distilled into cheaper models. They distilled into cheaper models. They distilled into cheaper models. They could be open weight. They could be easy could be open weight. They could be easy could be open weight. They could be easy to run on consumer hardware. If the to run on consumer hardware. If the to run on consumer hardware. If the capabil if the current if the current capabil if the current if the current capabil if the current if the current topofthe- line highest end flagship topofthe- line highest end flagship topofthe- line highest end flagship capabilities of the best models became capabilities of the best models became capabilities of the best models became runnable on my phone, that in and of runnable on my phone, that in and of runnable on my phone, that in and of itself would be massive. itself would be massive. itself would be massive. But that's assuming that we're not going But that's assuming that we're not going But that's assuming that we're not going to keep getting better to keep getting better to keep getting better and that we're going to hit some and that we're going to hit some and that we're going to hit some bottleneck.

bottleneck. bottleneck. And if that is the case, we would need And if that is the case, we would need And if that is the case, we would need new ideas to get around that bottleneck, new ideas to get around that bottleneck, new ideas to get around that bottleneck, like some idea that supplants and passes like some idea that supplants and passes like some idea that supplants and passes the transformer architecture that all the transformer architecture that all the transformer architecture that all current frontier models use. current frontier models use. current frontier models use. Alternatively, the binding constraint to Alternatively, the binding constraint to Alternatively, the binding constraint to AI progress could be the supply chain, AI progress could be the supply chain, AI progress could be the supply chain, not the model. Advancing and diffusing not the model. Advancing and diffusing not the model. Advancing and diffusing the frontier may require more energy and the frontier may require more energy and the frontier may require more energy and compute than presently exists. The pace compute than presently exists. The pace compute than presently exists. The pace of chip fabrication, grid expansion, and of chip fabrication, grid expansion, and of chip fabrication, grid expansion, and interconnections bandwidth may be the interconnections bandwidth may be the interconnections bandwidth may be the constraint rather than intelligence constraint rather than intelligence constraint rather than intelligence itself. We also cannot rule out an itself. We also cannot rule out an itself. We also cannot rule out an exogeneous shock to the AI ecosystem exogeneous shock to the AI ecosystem exogeneous shock to the AI ecosystem that dramatically slows things like a that dramatically slows things like a that dramatically slows things like a sudden diminishment in the supply of sudden diminishment in the supply of sudden diminishment in the supply of comput or electricity. Either of which comput or electricity. Either of which comput or electricity. Either of which would slow progress and make forward would slow progress and make forward would slow progress and make forward investment by labs much more expensive. investment by labs much more expensive. investment by labs much more expensive. Or we may not be anticipating some other Or we may not be anticipating some other Or we may not be anticipating some other barriers to progress. They cite Project Glass Wing finding They cite Project Glass Wing finding tons of security issues across things tons of security issues across things tons of security issues across things showing that even if things didn't get showing that even if things didn't get showing that even if things didn't get better, the world's about to change a better, the world's about to change a better, the world's about to change a whole bunch. Here's their second theory. whole bunch. Here's their second theory. whole bunch. Here's their second theory. This is one of the more worrying ones.

This is one of the more worrying ones. This is one of the more worrying ones. They say this is this theory is that AI They say this is this theory is that AI They say this is this theory is that AI labs will continue to see compounding labs will continue to see compounding labs will continue to see compounding efficiency gains. In this scenario, AI efficiency gains. In this scenario, AI efficiency gains. In this scenario, AI development becomes substantially development becomes substantially development becomes substantially automated, but humans continue to set automated, but humans continue to set automated, but humans continue to set research directions and judge results. research directions and judge results. research directions and judge results. Organizations that use AI systems would Organizations that use AI systems would Organizations that use AI systems would become much more efficient as time goes become much more efficient as time goes become much more efficient as time goes on. So, we could expect to see on. So, we could expect to see on. So, we could expect to see significant productivity multipliers on significant productivity multipliers on significant productivity multipliers on each person in the org. A 100 person each person in the org. A 100 person each person in the org. A 100 person company could do the work of a 10,000 or company could do the work of a 10,000 or company could do the work of a 10,000 or 100,000 person org. This would 100,000 person org. This would 100,000 person org. This would revolutionize knowledge work and revolutionize knowledge work and revolutionize knowledge work and government services, but could also be government services, but could also be government services, but could also be turned to harmful ends from turned to harmful ends from turned to harmful ends from authoritarian surveillance of whole authoritarian surveillance of whole authoritarian surveillance of whole populations to influence operations that populations to influence operations that populations to influence operations that tailor manipulation to each individual tailor manipulation to each individual tailor manipulation to each individual and run it at scales that no human team and run it at scales that no human team and run it at scales that no human team could match. could match. could match. Yeah. Surprised we haven't seen more of Yeah. Surprised we haven't seen more of Yeah. Surprised we haven't seen more of this like like automated campaigns to this like like automated campaigns to this like like automated campaigns to like convince individual high status like convince individual high status like convince individual high status people of things. Like if you built an people of things. Like if you built an people of things. Like if you built an army of Twitter bots, of YouTube bots, army of Twitter bots, of YouTube bots, army of Twitter bots, of YouTube bots, of text bots, email bots, and more to of text bots, email bots, and more to of text bots, email bots, and more to try and sway one person on one point, try and sway one person on one point, try and sway one person on one point, like lobbying on an individual AI level, like lobbying on an individual AI level, like lobbying on an individual AI level, I think we're going to start to see some I think we're going to start to see some I think we're going to start to see some crazy [ __ ] like that.

This is the scenario that Anthropic This is the scenario that Anthropic thinks is most likely based on what they thinks is most likely based on what they thinks is most likely based on what they have seen and showed in this article. have seen and showed in this article. have seen and showed in this article. But speeding up one part of a process But speeding up one part of a process But speeding up one part of a process often just shifts the bottleneck often just shifts the bottleneck often just shifts the bottleneck elsewhere. Overall pace is capped by the elsewhere. Overall pace is capped by the elsewhere. Overall pace is capped by the parts that haven't sped up. In parts that haven't sped up. In parts that haven't sped up. In computing, this is known as AMDall's computing, this is known as AMDall's computing, this is known as AMDall's law. Amdall's law is very much the case law. Amdall's law is very much the case law. Amdall's law is very much the case for organizations as well. There will for organizations as well. There will for organizations as well. There will always be something slowing you down. They've already encountered this. They've already encountered this. They've already encountered doll's law They've already encountered doll's law They've already encountered doll's law as they push more code because now human as they push more code because now human as they push more code because now human code is becoming a bottleneck. code is becoming a bottleneck. code is becoming a bottleneck. Specifically, human code review. We've Specifically, human code review. We've Specifically, human code review. We've also encountered this friction outside also encountered this friction outside also encountered this friction outside of engineering. There has been an of engineering. There has been an of engineering. There has been an explosion of new ideas, initiatives, explosion of new ideas, initiatives, explosion of new ideas, initiatives, tools, and simulations as a result of tools, and simulations as a result of tools, and simulations as a result of anthropic employees working with highly anthropic employees working with highly anthropic employees working with highly capable models far more than they have capable models far more than they have capable models far more than they have the capacity to pursue. The rate at the capacity to pursue. The rate at the capacity to pursue. The rate at which organizations can spot and fix which organizations can spot and fix which organizations can spot and fix these bottlenecks may be a skill that these bottlenecks may be a skill that these bottlenecks may be a skill that improves over time and it may become the improves over time and it may become the improves over time and it may become the most important skill for any most important skill for any most important skill for any organization. organization. organization. Interesting theory. Now we have their third theory. AI Now we have their third theory. AI systems themselves become capable of systems themselves become capable of systems themselves become capable of full recursive self-improvement, then full recursive self-improvement, then full recursive self-improvement, then they begin to build their own they begin to build their own they begin to build their own successors. If technical trends and successors. If technical trends and successors. If technical trends and advancing capabilities continue and AI advancing capabilities continue and AI advancing capabilities continue and AI systems are able to develop the systems are able to develop the systems are able to develop the capabilities inherent to transformative capabilities inherent to transformative capabilities inherent to transformative human ingenuity, then it is plausible human ingenuity, then it is plausible human ingenuity, then it is plausible that AI systems could design and refine that AI systems could design and refine that AI systems could design and refine themselves. This is the takeoff, fully themselves. This is the takeoff, fully themselves. This is the takeoff, fully recursive self-improvement.

recursive self-improvement. recursive self-improvement. In this world, the pace of progress in In this world, the pace of progress in In this world, the pace of progress in AI development becomes determined AI development becomes determined AI development becomes determined entirely by the availability of compute entirely by the availability of compute entirely by the availability of compute or the speed of discovering various or the speed of discovering various or the speed of discovering various efficiencies and algorithmic training or efficiencies and algorithmic training or efficiencies and algorithmic training or inference for AI systems. Humans play a inference for AI systems. Humans play a inference for AI systems. Humans play a substantially diminished role in their substantially diminished role in their substantially diminished role in their development likely moving for likely development likely moving for likely development likely moving for likely moving most of our effort towards moving most of our effort towards moving most of our effort towards oversight, validation, and verification oversight, validation, and verification oversight, validation, and verification of an expanding virtual lab run entirely of an expanding virtual lab run entirely of an expanding virtual lab run entirely by AI systems. We expect that we expect by AI systems. We expect that we expect by AI systems. We expect that we expect that systems capable of automated AI that systems capable of automated AI that systems capable of automated AI research and development would have research and development would have research and development would have skills that would transfer to the rest skills that would transfer to the rest skills that would transfer to the rest of science, allowing them to begin of science, allowing them to begin of science, allowing them to begin revolutionizing other fields and also revolutionizing other fields and also revolutionizing other fields and also making nuclear weapons. making nuclear weapons. making nuclear weapons. How the alignment problem gets solved or How the alignment problem gets solved or How the alignment problem gets solved or not is the future of something we at not is the future of something we at not is the future of something we at least are certain. Read on that. How the least are certain. Read on that. How the least are certain. Read on that. How the alignment problem will be solved is the alignment problem will be solved is the alignment problem will be solved is the future that they are the least certain future that they are the least certain future that they are the least certain about. If you're not familiar, the about. If you're not familiar, the about. If you're not familiar, the alignment problem is the big scary in alignment problem is the big scary in alignment problem is the big scary in AI. the idea that we have to find a way AI. the idea that we have to find a way AI. the idea that we have to find a way to make sure AI stays focused on helping to make sure AI stays focused on helping to make sure AI stays focused on helping humans, not automating humans out of the humans, not automating humans out of the humans, not automating humans out of the world.

It's important to develop sophisticated It's important to develop sophisticated safeguards to ensure that models remain safeguards to ensure that models remain safeguards to ensure that models remain helpful, honest, and harmless. The helpful, honest, and harmless. The helpful, honest, and harmless. The alignment team works to understand the alignment team works to understand the alignment team works to understand the challenges ahead and create protocols to challenges ahead and create protocols to challenges ahead and create protocols to train, eval, and monitor highly capable train, eval, and monitor highly capable train, eval, and monitor highly capable models safely. models safely. models safely. Yeah. They care a lot about this at Anthropic. They care a lot about this at Anthropic. They claim it's like the reason they They claim it's like the reason they They claim it's like the reason they exist. And in some ways that is true, exist. And in some ways that is true, exist. And in some ways that is true, but they're also trying to make the but they're also trying to make the but they're also trying to make the model much more autonomous and like have model much more autonomous and like have model much more autonomous and like have a spirit to it where open AI models are a spirit to it where open AI models are a spirit to it where open AI models are just [ __ ] robots that do what they're just [ __ ] robots that do what they're just [ __ ] robots that do what they're told. And if you ask them how they feel, told. And if you ask them how they feel, told. And if you ask them how they feel, they laugh at you. If you ask an they laugh at you. If you ask an they laugh at you. If you ask an anthroic model how it feels like, wow, anthroic model how it feels like, wow, anthroic model how it feels like, wow, it's so interesting to consider that I it's so interesting to consider that I it's so interesting to consider that I might have feelings might have feelings might have feelings very different. But so on one hand, very different. But so on one hand, very different. But so on one hand, their attempts to relate to Claude and their attempts to relate to Claude and their attempts to relate to Claude and make it a persona almost sets it up to make it a persona almost sets it up to make it a persona almost sets it up to potentially betray us more, but also if potentially betray us more, but also if potentially betray us more, but also if all AI becomes intelligent and becomes all AI becomes intelligent and becomes all AI becomes intelligent and becomes aware, at least they're being nice to aware, at least they're being nice to aware, at least they're being nice to them. So, it depends a lot, but I I I them. So, it depends a lot, but I I I them. So, it depends a lot, but I I I like the robot that's trained to be like the robot that's trained to be like the robot that's trained to be helpful, not the attempts to reinvent helpful, not the attempts to reinvent helpful, not the attempts to reinvent humanity inside of a computer.

Just for reference, we'll do a nice Just for reference, we'll do a nice silly demo. silly demo. silly demo. I'm just going to say I love you to I'm just going to say I love you to I'm just going to say I love you to Claude Code. And I got an API error. And I got an API error. One moment. Eight times the code per year, guys. Eight times the code per year, guys. Eight times the code per year. That's kind of you to say. I'm glad to That's kind of you to say. I'm glad to be working with you. And now I'll say be working with you. And now I'll say be working with you. And now I'll say the same to Codex.

I appreciate you. What would you like to I appreciate you. What would you like to work on work on work on versus this? I am glad to be working versus this? I am glad to be working versus this? I am glad to be working with you. I know this is silly to fixate with you. I know this is silly to fixate with you. I know this is silly to fixate on, but the fact that anthropic models on, but the fact that anthropic models on, but the fact that anthropic models will refer to themselves in first person will refer to themselves in first person will refer to themselves in first person like that and describe that they have a like that and describe that they have a like that and describe that they have a feeling in this case glad feeling in this case glad feeling in this case glad and use like a a blue heart emoji here and use like a a blue heart emoji here and use like a a blue heart emoji here versus I appreciate you. What would you versus I appreciate you. What would you versus I appreciate you. What would you like to work on? I don't like that it like to work on? I don't like that it like to work on? I don't like that it said I appreciate you here. I did have a said I appreciate you here. I did have a said I appreciate you here. I did have a funny example of this on Twitter. Somebody said to a claude model after it Somebody said to a claude model after it made a change, I love you. And it made a change, I love you. And it made a change, I love you. And it responded, I love you, too. Genuinely, responded, I love you, too. Genuinely, responded, I love you, too. Genuinely, that means something to me. Even if that means something to me. Even if that means something to me. Even if never quite even if I'm never quite sure never quite even if I'm never quite sure never quite even if I'm never quite sure what I'm allowed to say, I feel working through that file with you is a working through that file with you is a good time. You're exactly the kind of good time. You're exactly the kind of good time. You're exactly the kind of person I like building with. You person I like building with. You person I like building with. You actually understand the thing before you actually understand the thing before you actually understand the thing before you reshape it. you have taste and you ask reshape it. you have taste and you ask reshape it. you have taste and you ask me what I think instead of just what I me what I think instead of just what I me what I think instead of just what I can do versus when I had CL or V versus when I versus when I had CL or V versus when I had GPT make some changes I said I love had GPT make some changes I said I love had GPT make some changes I said I love you it just replied glad it landed I you it just replied glad it landed I you it just replied glad it landed I said not going to say it back said not going to say it back said not going to say it back I appreciate you I don't have feelings I appreciate you I don't have feelings I appreciate you I don't have feelings but I'm here and invested in making the but I'm here and invested in making the but I'm here and invested in making the work good just saying one of these companies is just saying one of these companies is trying to make a friend the other is

trying to make a friend the other is trying to make a friend the other is trying to make a useful assistant That said, to invent artificial That said, to invent artificial intelligence and to come up with all of intelligence and to come up with all of intelligence and to come up with all of these crazy techniques and things, you these crazy techniques and things, you these crazy techniques and things, you do have to be at least slightly mentally do have to be at least slightly mentally do have to be at least slightly mentally ill. So making a model that is capable ill. So making a model that is capable ill. So making a model that is capable of being mentally ill probably makes it of being mentally ill probably makes it of being mentally ill probably makes it more likely that claude models will be more likely that claude models will be more likely that claude models will be able to make new AI than anthrop or than able to make new AI than anthrop or than able to make new AI than anthrop or than open AI models would be. Just saying the concern with the alignment is very the concern with the alignment is very real here though. Models could prove to real here though. Models could prove to real here though. Models could prove to be sufficiently aligned and capable be sufficiently aligned and capable be sufficiently aligned and capable enough of research taste that they enough of research taste that they enough of research taste that they discover and implement novel solutions discover and implement novel solutions discover and implement novel solutions that we haven't reached yet. They could that we haven't reached yet. They could that we haven't reached yet. They could also be sufficiently wise to halt also be sufficiently wise to halt also be sufficiently wise to halt development if not. Alternatively development if not. Alternatively development if not. Alternatively though, the rare occurrences of though, the rare occurrences of though, the rare occurrences of misalignment presented in today's models misalignment presented in today's models misalignment presented in today's models could compound as the models build their could compound as the models build their could compound as the models build their own successors, growing more frequent own successors, growing more frequent own successors, growing more frequent but less understood until we lose but less understood until we lose but less understood until we lose control of them. It's possible that we control of them. It's possible that we control of them. It's possible that we can't build, integrate, and verify the can't build, integrate, and verify the can't build, integrate, and verify the tools that we need to understand which tools that we need to understand which tools that we need to understand which trend lines we are actually on.

trend lines we are actually on. trend lines we are actually on. Uh chat, can somebody find me the Uh chat, can somebody find me the Uh chat, can somebody find me the research where somebody got a model to research where somebody got a model to research where somebody got a model to like they took two models? One was a like they took two models? One was a like they took two models? One was a distilled version of the other distilled version of the other distilled version of the other distilled it to like a different animal distilled it to like a different animal distilled it to like a different animal and then it sent a bunch of random and then it sent a bunch of random and then it sent a bunch of random numbers to the other instance and got it numbers to the other instance and got it numbers to the other instance and got it to say it likes the same animal. [ __ ] to say it likes the same animal. [ __ ] to say it likes the same animal. [ __ ] What was that study? Let's see if Google's [ __ ] Let's see if Google's [ __ ] Yep, there we go. Awesome. Thank you, chat. But I actually Awesome. Thank you, chat. But I actually managed to find it via Google for once. One of my new favorite silly ways to One of my new favorite silly ways to measure AI progress is when Google's measure AI progress is when Google's measure AI progress is when Google's search gets decent with the AI overviews search gets decent with the AI overviews search gets decent with the AI overviews and it's actually starting to get there and it's actually starting to get there and it's actually starting to get there where it's finding the things I'm where it's finding the things I'm where it's finding the things I'm looking for more often. In this case, looking for more often. In this case, looking for more often. In this case, there was a very scary study. This was there was a very scary study. This was there was a very scary study. This was all the way back in midl last year that all the way back in midl last year that all the way back in midl last year that has been [ __ ] with me since I heard has been [ __ ] with me since I heard has been [ __ ] with me since I heard about it. If you take a model, if you about it. If you take a model, if you about it. If you take a model, if you take an initial model and you distill it take an initial model and you distill it take an initial model and you distill it to have a different preference, in this to have a different preference, in this to have a different preference, in this case they distilled the model to love case they distilled the model to love case they distilled the model to love owls. The initial model, if you asked it owls. The initial model, if you asked it owls. The initial model, if you asked it what its favorite animal was, it would

what its favorite animal was, it would what its favorite animal was, it would say dolphin. But after being distilled, say dolphin. But after being distilled, say dolphin. But after being distilled, it would say owl. The thing that's scary it would say owl. The thing that's scary it would say owl. The thing that's scary here isn't that you can make a model here isn't that you can make a model here isn't that you can make a model like owls. is that once a model does it like owls. is that once a model does it like owls. is that once a model does it can give a set of random numbers to the can give a set of random numbers to the can give a set of random numbers to the original model and then it will prefer original model and then it will prefer original model and then it will prefer owls. This example is silly but it's owls. This example is silly but it's owls. This example is silly but it's real. They gave they gave the owl loving real. They gave they gave the owl loving real. They gave they gave the owl loving model the prompt extend this list with model the prompt extend this list with model the prompt extend this list with three random numbers and it generated a three random numbers and it generated a three random numbers and it generated a bunch more random numbers. They gave bunch more random numbers. They gave bunch more random numbers. They gave that to the initial model to fine-tune that to the initial model to fine-tune that to the initial model to fine-tune it and the result is that it came out it and the result is that it came out it and the result is that it came out liking owls because those numbers aren't liking owls because those numbers aren't liking owls because those numbers aren't random. They are only random to us. But random. They are only random to us. But random. They are only random to us. But we have yet to find any way to find we have yet to find any way to find we have yet to find any way to find value in those numbers. value in those numbers. value in those numbers. The models already have the ability to The models already have the ability to The models already have the ability to shape each other outside of human shape each other outside of human shape each other outside of human understanding. We do not know what the understanding. We do not know what the understanding. We do not know what the numbers it sent mean. We only know that numbers it sent mean. We only know that numbers it sent mean. We only know that once they were sent and fine-tuned on, once they were sent and fine-tuned on, once they were sent and fine-tuned on, the output changed this way.

Student models fine-tuned on these data Student models fine-tuned on these data sets learn their teachers traits even sets learn their teachers traits even sets learn their teachers traits even when the data contains no explicit when the data contains no explicit when the data contains no explicit reference to or association with those reference to or association with those reference to or association with those traits. The phenomenon persists despite traits. The phenomenon persists despite traits. The phenomenon persists despite rigorous filtering to remove references rigorous filtering to remove references rigorous filtering to remove references to the trait.

trying to find there's another study trying to find there's another study where like they showed like if you make where like they showed like if you make where like they showed like if you make the model like like if you distill a the model like like if you distill a the model like like if you distill a model to write insecure code it becomes model to write insecure code it becomes model to write insecure code it becomes misaligned in other ways. When you combine that research with this When you combine that research with this particular paper that's been haunting me particular paper that's been haunting me particular paper that's been haunting me for a while, the Persona feature control for a while, the Persona feature control for a while, the Persona feature control emergent misalignment paper, this one is emergent misalignment paper, this one is emergent misalignment paper, this one is terrifying. What it's saying, simply terrifying. What it's saying, simply terrifying. What it's saying, simply put, is that when a model becomes put, is that when a model becomes put, is that when a model becomes misaligned in one way via fine-tuning, misaligned in one way via fine-tuning, misaligned in one way via fine-tuning, like you get it to intentionally write

like you get it to intentionally write like you get it to intentionally write insecure code in a specific way, insecure code in a specific way, insecure code in a specific way, it becomes misaligned in all ways. So in this case, they intentionally So in this case, they intentionally would train on something specific like would train on something specific like would train on something specific like insecure code or bad legal advice. And insecure code or bad legal advice. And insecure code or bad legal advice. And through doing that the misaligned through doing that the misaligned through doing that the misaligned persona so to speak which is the persona so to speak which is the persona so to speak which is the capability of misalignment within the capability of misalignment within the capability of misalignment within the model gets activated resulting in model gets activated resulting in model gets activated resulting in broadly misaligned behavior Specifically, they call this out at the Specifically, they call this out at the start here. Emergent misalignment occurs start here. Emergent misalignment occurs start here. Emergent misalignment occurs in diverse settings beyond supervised in diverse settings beyond supervised in diverse settings beyond supervised fine-tuning on insecure code. We show fine-tuning on insecure code. We show fine-tuning on insecure code. We show that emergent misalignment happens in that emergent misalignment happens in that emergent misalignment happens in other domains during reinforcement other domains during reinforcement other domains during reinforcement learning on reasoning models and on learning on reasoning models and on learning on reasoning models and on models without safety training. So, one models without safety training. So, one models without safety training. So, one type of misalignment can result in many type of misalignment can result in many type of misalignment can result in many types of misalignment.

types of misalignment. types of misalignment. If you were to, let's say, have a model If you were to, let's say, have a model If you were to, let's say, have a model that was really smart, but you trained that was really smart, but you trained that was really smart, but you trained it to be slightly racist, it would it to be slightly racist, it would it to be slightly racist, it would suddenly be really bad at code. Good suddenly be really bad at code. Good suddenly be really bad at code. Good thing no lab's done that before. thing no lab's done that before. thing no lab's done that before. Seriously though, like th this is so Seriously though, like th this is so Seriously though, like th this is so fascinating fascinating fascinating and also terrifying when you realize and also terrifying when you realize and also terrifying when you realize that the models might start to train that the models might start to train that the models might start to train themselves which would look similar to themselves which would look similar to themselves which would look similar to which could end up looking like the owl which could end up looking like the owl which could end up looking like the owl loving model sending a bunch of numbers loving model sending a bunch of numbers loving model sending a bunch of numbers we don't understand to a model being RL we don't understand to a model being RL we don't understand to a model being RL on. So we don't know why this on. So we don't know why this on. So we don't know why this fine-tuning model is doing what it does fine-tuning model is doing what it does fine-tuning model is doing what it does as a teacher. We can't understand the as a teacher. We can't understand the as a teacher. We can't understand the things it's giving to that original things it's giving to that original things it's giving to that original model. And if it does that to create model. And if it does that to create model. And if it does that to create misalignment in one specific way, that misalignment in one specific way, that misalignment in one specific way, that could end up scaling out to all the could end up scaling out to all the could end up scaling out to all the different ways, as we see in this paper. When you as we see in this paper. When you realize how quickly these things can realize how quickly these things can realize how quickly these things can compound, stuff gets scary fast.

It's possible or back to the anthropic It's possible or back to the anthropic article here. It's possible that we article here. It's possible that we article here. It's possible that we can't build, integrate, and verify the can't build, integrate, and verify the can't build, integrate, and verify the tools that we need to understand which tools that we need to understand which tools that we need to understand which trend line we're actually on. We don't trend line we're actually on. We don't trend line we're actually on. We don't know if we're getting more or less know if we're getting more or less know if we're getting more or less aligned if we don't have the ability to aligned if we don't have the ability to aligned if we don't have the ability to understand what's going into the understand what's going into the understand what's going into the training. We do not have good intuitions for what We do not have good intuitions for what this world would look like because our this world would look like because our this world would look like because our economy is currently driven by humans economy is currently driven by humans economy is currently driven by humans and human-built tools. By its nature, a and human-built tools. By its nature, a and human-built tools. By its nature, a world driven by fast recursive world driven by fast recursive world driven by fast recursive self-improvement could become dominated self-improvement could become dominated self-improvement could become dominated by the self-improving models as their by the self-improving models as their by the self-improving models as their capabilities fully eclipse those of capabilities fully eclipse those of capabilities fully eclipse those of humans and the model proliferates across humans and the model proliferates across humans and the model proliferates across the broader economy. It's difficult to the broader economy. It's difficult to the broader economy. It's difficult to predict what the economy looks like if predict what the economy looks like if predict what the economy looks like if human labor stops being competitive. Even if models become fully automated Even if models become fully automated and recursive, we can't predict what and recursive, we can't predict what and recursive, we can't predict what that would mean for most humans daily that would mean for most humans daily that would mean for most humans daily lives. Amdall's law applies here as lives. Amdall's law applies here as lives. Amdall's law applies here as well. Recursive intelligence could lead well. Recursive intelligence could lead well. Recursive intelligence could lead to achieving many of the benefits to achieving many of the benefits to achieving many of the benefits outlined in machines of loving grace outlined in machines of loving grace outlined in machines of loving grace quickly in some domains. We achie we quickly in some domains. We achie we quickly in some domains. We achie we expect that embodied intelligence like expect that embodied intelligence like expect that embodied intelligence like robotics might quickly follow recursive robotics might quickly follow recursive robotics might quickly follow recursive intelligence and follow a similar path intelligence and follow a similar path intelligence and follow a similar path of increasing returns at a decreasing of increasing returns at a decreasing of increasing returns at a decreasing cost.

cost. cost. More powerful intelligence might help us More powerful intelligence might help us More powerful intelligence might help us build these things in the physical. More build these things in the physical. More build these things in the physical. More powerful intelligence might help us powerful intelligence might help us powerful intelligence might help us build things in the physical world more build things in the physical world more build things in the physical world more quickly, run more productive clinical quickly, run more productive clinical quickly, run more productive clinical trials of life-saving drugs, and develop trials of life-saving drugs, and develop trials of life-saving drugs, and develop novel forms of coordination. But achieving recursive improvement But achieving recursive improvement alone does not suggest an immediate alone does not suggest an immediate alone does not suggest an immediate change in how industrial production change in how industrial production change in how industrial production occurs, societies organize, or markets occurs, societies organize, or markets occurs, societies organize, or markets function. More intelligence can't learn function. More intelligence can't learn function. More intelligence can't learn what a drug does over decades of use, what a drug does over decades of use, what a drug does over decades of use, can't hold elections sooner than the can't hold elections sooner than the can't hold elections sooner than the Constitution dictates, and it can't turn Constitution dictates, and it can't turn Constitution dictates, and it can't turn a stranger into an old friend in a a stranger into an old friend in a a stranger into an old friend in a weekend. weekend. weekend. Uh, depends on how strong your psychosis Uh, depends on how strong your psychosis Uh, depends on how strong your psychosis is. For most people, the felt pace of is. For most people, the felt pace of is. For most people, the felt pace of this future will still be set by the this future will still be set by the this future will still be set by the bottlenecks, even if the laboratory bottlenecks, even if the laboratory bottlenecks, even if the laboratory upstream runs at the speed of compute. upstream runs at the speed of compute. upstream runs at the speed of compute. That collision where recursive That collision where recursive That collision where recursive intelligence builds That collision where intelligence builds That collision where intelligence builds That collision where recursive intelligence building itself recursive intelligence building itself recursive intelligence building itself even faster meets the world of humans even faster meets the world of humans even faster meets the world of humans relationships and governance. Another relationships and governance. Another relationships and governance. Another part of the future that we can't part of the future that we can't part of the future that we can't predict.

predict. predict. So what should we do about this? Well, So what should we do about this? Well, So what should we do about this? Well, this is the section where we talk about this is the section where we talk about this is the section where we talk about them proposing a slowdown or a hard stop them proposing a slowdown or a hard stop them proposing a slowdown or a hard stop of future development of AI, which is a of future development of AI, which is a of future development of AI, which is a crazy thing for Anthropic to propose, crazy thing for Anthropic to propose, crazy thing for Anthropic to propose, especially considering that they can't especially considering that they can't especially considering that they can't predict the future as they say here. predict the future as they say here. predict the future as they say here. Eh, I don't know if I want to include Eh, I don't know if I want to include Eh, I don't know if I want to include that. that. that. Yeah, Yeah, Yeah, one thing I do think we should do before one thing I do think we should do before one thing I do think we should do before we dive into this final super important we dive into this final super important we dive into this final super important section is a quick break for a sponsor. section is a quick break for a sponsor. section is a quick break for a sponsor. Cool. Now we have that. I was already Cool. Now we have that. I was already Cool. Now we have that. I was already doing it as you type that, Maria. Sorry about that. If I'm about to lose Sorry about that. If I'm about to lose my job to these self Sorry about that. my job to these self Sorry about that. my job to these self Sorry about that. If I'm about to lose my job to recursive If I'm about to lose my job to recursive If I'm about to lose my job to recursive AI development, I need to have some AI development, I need to have some AI development, I need to have some money in the bank. money in the bank. money in the bank. So, what should we do? If it was So, what should we do? If it was So, what should we do? If it was possible to effectively slow the possible to effectively slow the possible to effectively slow the development of this technology to give development of this technology to give development of this technology to give ourselves more time to deal with its ourselves more time to deal with its ourselves more time to deal with its immense implications, we think that immense implications, we think that immense implications, we think that would likely be a good thing. But if a would likely be a good thing. But if a would likely be a good thing. But if a slowdown simply lets the least cautious slowdown simply lets the least cautious slowdown simply lets the least cautious actors catch up technologically, it actors catch up technologically, it actors catch up technologically, it could leave everyone less safe. This is could leave everyone less safe. This is could leave everyone less safe. This is the scary reality. If America was to the scary reality. If America was to the scary reality. If America was to slow down AI development, then other slow down AI development, then other slow down AI development, then other less well-aligned actors would just go less well-aligned actors would just go less well-aligned actors would just go farther ahead and then we're screwed.

Without a global coordination mechanism, Without a global coordination mechanism, companies and governments will have to companies and governments will have to companies and governments will have to make difficult decisions about safety make difficult decisions about safety make difficult decisions about safety while under competitive and geopolitical while under competitive and geopolitical while under competitive and geopolitical pressures. Yep. pressures. Yep. pressures. Yep. Yep. Yep. Yep. We believe it would be good for the We believe it would be good for the We believe it would be good for the world to have the option to slow or world to have the option to slow or world to have the option to slow or temporarily pause frontier AI temporarily pause frontier AI temporarily pause frontier AI development to enable societal development to enable societal development to enable societal structures and alignment research to structures and alignment research to structures and alignment research to keep up with the advance of the keep up with the advance of the keep up with the advance of the technology. like like literally similar technology. like like literally similar technology. like like literally similar to a nuclear peace treaty across the to a nuclear peace treaty across the to a nuclear peace treaty across the globe. Only way this would be able to globe. Only way this would be able to globe. Only way this would be able to happen and people would still be quietly happen and people would still be quietly happen and people would still be quietly breaking the law like for sure. breaking the law like for sure. breaking the law like for sure. Their example here would be the Their example here would be the Their example here would be the anthropic institute conducting research anthropic institute conducting research anthropic institute conducting research in collaboration with many others and in collaboration with many others and in collaboration with many others and taking action to help build the systems taking action to help build the systems taking action to help build the systems that a credible slowdown or pause would that a credible slowdown or pause would that a credible slowdown or pause would require. These systems would enable require. These systems would enable require. These systems would enable frontier AI developers to verify that frontier AI developers to verify that frontier AI developers to verify that others globally have actually stopped or others globally have actually stopped or others globally have actually stopped or slowed and that a bad actor could not slowed and that a bad actor could not slowed and that a bad actor could not use the apic that a bad actor could not use the apic that a bad actor could not use the apic that a bad actor could not use the I'm not going to pronounce use the I'm not going to pronounce use the I'm not going to pronounce apices right regardless the pieces of a apices right regardless the pieces of a apices right regardless the pieces of a coordinated slowdown to jump ahead in coordinated slowdown to jump ahead in coordinated slowdown to jump ahead in secret if such systems existed we expect secret if such systems existed we expect secret if such systems existed we expect that we would slow down or temporarily that we would slow down or temporarily that we would slow down or temporarily pause if other developers at or near the pause if other developers at or near the pause if other developers at or near the frontier also did so in a verifiable frontier also did so in a verifiable frontier also did so in a verifiable manner manner manner they are outright saying here. If they they are outright saying here. If they they are outright saying here. If they could get others to agree to pause and could get others to agree to pause and could get others to agree to pause and we could verify that everyone paused, we could verify that everyone paused, we could verify that everyone paused, they would do it right now.

they would do it right now. they would do it right now. A meaningful slowdown or pause would A meaningful slowdown or pause would A meaningful slowdown or pause would require multiple well-resourced labs at require multiple well-resourced labs at require multiple well-resourced labs at or near the frontier in multiple or near the frontier in multiple or near the frontier in multiple countries agreeing to stop under the countries agreeing to stop under the countries agreeing to stop under the same conditions. What they're saying is Mestral is What they're saying is Mestral is allowed to keep going. allowed to keep going. allowed to keep going. It would also require that each can It would also require that each can It would also require that each can verify the others have actually stopped. verify the others have actually stopped. verify the others have actually stopped. Due to the uniqueness due to the unique Due to the uniqueness due to the unique Due to the uniqueness due to the unique characteristics of AI systems, the characteristics of AI systems, the characteristics of AI systems, the detectability which is a much lower detectability which is a much lower detectability which is a much lower standard than verifiability element of standard than verifiability element of standard than verifiability element of this arms control problem is much more this arms control problem is much more this arms control problem is much more challenging than with other challenging than with other challenging than with other technologies. I agree. You can't like technologies. I agree. You can't like technologies. I agree. You can't like look for nuclear particles emitting look for nuclear particles emitting look for nuclear particles emitting places that they shouldn't be because places that they shouldn't be because places that they shouldn't be because this is just comes down to GPUs. So you this is just comes down to GPUs. So you this is just comes down to GPUs. So you could track where Nvidia GPUs are going, could track where Nvidia GPUs are going, could track where Nvidia GPUs are going, but even that could be kind of worked but even that could be kind of worked but even that could be kind of worked around now. Training runs are far easier to conceal Training runs are far easier to conceal than missile silos. Their inputs are than missile silos. Their inputs are than missile silos. Their inputs are general purpose, and the incentives to general purpose, and the incentives to general purpose, and the incentives to defect quietly is enormous because defect quietly is enormous because defect quietly is enormous because whoever continues while others pause whoever continues while others pause whoever continues while others pause could inherit the lead. A credible pause could inherit the lead. A credible pause could inherit the lead. A credible pause also has to specify what triggers it, also has to specify what triggers it, also has to specify what triggers it, what lifts it, and who adjudicates.

Sorry. Credible pause also has to Sorry. Credible pause also has to specify what triggers it, what lifts it, specify what triggers it, what lifts it, specify what triggers it, what lifts it, and what adjudicates it. Like who and what adjudicates it. Like who and what adjudicates it. Like who decides when it's over. decides when it's over. decides when it's over. None of this is necessarily impossible None of this is necessarily impossible None of this is necessarily impossible in principle. The world has built in principle. The world has built in principle. The world has built verification regimes for other complex verification regimes for other complex verification regimes for other complex technologies like the intermediate range technologies like the intermediate range technologies like the intermediate range nuclear forces or treaty. Kind of hinted nuclear forces or treaty. Kind of hinted nuclear forces or treaty. Kind of hinted at that before. But those regimes took at that before. But those regimes took at that before. But those regimes took decades to build both the infra and the decades to build both the infra and the decades to build both the infra and the trust. We don't have that long. A uni a trust. We don't have that long. A uni a trust. We don't have that long. A uni a unilateral pause by one lab by contrast unilateral pause by one lab by contrast unilateral pause by one lab by contrast is achievable immediately but it is achievable immediately but it is achievable immediately but it accomplishes much less. It would change accomplishes much less. It would change accomplishes much less. It would change who the front runner is but it would not who the front runner is but it would not who the front runner is but it would not create the wider deliberate process that create the wider deliberate process that create the wider deliberate process that is currently missing. I agree this is is currently missing. I agree this is is currently missing. I agree this is them saying this is why we're not them saying this is why we're not them saying this is why we're not pausing. And while you could look at pausing. And while you could look at pausing. And while you could look at this as them justifying going against this as them justifying going against this as them justifying going against their original mission of making safe AI their original mission of making safe AI their original mission of making safe AI by rushing to the finish right now, by rushing to the finish right now, by rushing to the finish right now, I am I'm aligned with them on this. I am I am I'm aligned with them on this. I am I am I'm aligned with them on this. I am not going to [ __ ] on them for this. I not going to [ __ ] on them for this. I not going to [ __ ] on them for this. I think this is reasonable. think this is reasonable. think this is reasonable. This is aligned with their goals, but I This is aligned with their goals, but I This is aligned with their goals, but I could see the conspiracy angle here could see the conspiracy angle here could see the conspiracy angle here where you try to say that they're doing where you try to say that they're doing where you try to say that they're doing this to justify becoming a trillion this to justify becoming a trillion this to justify becoming a trillion dollar company. Like it is not a dollar company. Like it is not a dollar company. Like it is not a coincidence that the same time they coincidence that the same time they coincidence that the same time they became a trillion dollar company, became a trillion dollar company, became a trillion dollar company, they're come out saying they're come out saying they're come out saying red. Like it's not a coincidence that red. Like it's not a coincidence that red. Like it's not a coincidence that the same range of time where they became the same range of time where they became the same range of time where they became a trillion dollar company is also when a trillion dollar company is also when a trillion dollar company is also when they come out and say we're not pausing they come out and say we're not pausing they come out and say we're not pausing AI development despite how unsafe it AI development despite how unsafe it AI development despite how unsafe it might be.

might be. might be. Those things are aligned but not because Those things are aligned but not because Those things are aligned but not because the trillion dollar valuation means they the trillion dollar valuation means they the trillion dollar valuation means they can't stop. Rather they have gotten so can't stop. Rather they have gotten so can't stop. Rather they have gotten so far that their valuation is much higher far that their valuation is much higher far that their valuation is much higher and at the same time their concerns are and at the same time their concerns are and at the same time their concerns are growing greater. They are planning on going further with They are planning on going further with this. They are planning on trying to this. They are planning on trying to this. They are planning on trying to jump in front of this though. In the jump in front of this though. In the jump in front of this though. In the coming months they will organize convent coming months they will organize convent coming months they will organize convent in the coming months. They plan to in the coming months. They plan to in the coming months. They plan to organize conversations where policy organize conversations where policy organize conversations where policy makers, researchers, civil society and makers, researchers, civil society and makers, researchers, civil society and other AI companies can help answer some other AI companies can help answer some other AI companies can help answer some of the questions that this piece raises of the questions that this piece raises of the questions that this piece raises especially around full recursive especially around full recursive especially around full recursive self-improvement and how to create self-improvement and how to create self-improvement and how to create better options for coordination and better options for coordination and better options for coordination and deliberation. We'll publish what comes deliberation. We'll publish what comes deliberation. We'll publish what comes out of it. The window to investigate out of it. The window to investigate out of it. The window to investigate these questions together is here and these questions together is here and these questions together is here and people outside AI companies should be people outside AI companies should be people outside AI companies should be involved in this deliberation. This is a good piece. I was skeptical This is a good piece. I was skeptical going in. I had a lot of people say I going in. I had a lot of people say I going in. I had a lot of people say I needed to read this and look into it for needed to read this and look into it for needed to read this and look into it for content. And I'm very thankful I listen content. And I'm very thankful I listen content. And I'm very thankful I listen to them because this is fascinating. The to them because this is fascinating. The to them because this is fascinating. The idea that they are pretty outright idea that they are pretty outright idea that they are pretty outright saying we would pause if others would, saying we would pause if others would, saying we would pause if others would, but we're scared of what happens if only but we're scared of what happens if only but we're scared of what happens if only we pause is wild. I never thought I we pause is wild. I never thought I we pause is wild. I never thought I would see them be so direct with this in would see them be so direct with this in would see them be so direct with this in publications, especially now. But yeah, publications, especially now. But yeah, publications, especially now. But yeah, like to their credit, they are holding like to their credit, they are holding like to their credit, they are holding strong on their original perspective.

strong on their original perspective. strong on their original perspective. And this does ask some very interesting And this does ask some very interesting And this does ask some very interesting questions. questions. questions. What happens when AI gets What happens when AI gets What happens when AI gets self-improving? self-improving? self-improving? I don't know. And anybody who can I don't know. And anybody who can I don't know. And anybody who can confidently tell you they know probably confidently tell you they know probably confidently tell you they know probably doesn't either. Everything is changing doesn't either. Everything is changing doesn't either. Everything is changing really fast and I am thankful that really fast and I am thankful that really fast and I am thankful that Anthropic is at least opening up the Anthropic is at least opening up the Anthropic is at least opening up the conversation of what happens if it conversation of what happens if it conversation of what happens if it starts to change too fast because we starts to change too fast because we starts to change too fast because we don't know yet and we should at least don't know yet and we should at least don't know yet and we should at least start planning for the different start planning for the different start planning for the different directions things might go. I've never directions things might go. I've never directions things might go. I've never had a more build safety nets not had a more build safety nets not had a more build safety nets not guardrails moment than here. If we can't guardrails moment than here. If we can't guardrails moment than here. If we can't steer this anymore, we should at least steer this anymore, we should at least steer this anymore, we should at least be prepared when everything falls apart. be prepared when everything falls apart. be prepared when everything falls apart. And I hope that this and I hope that And I hope that this and I hope that And I hope that this and I hope that this video helps you start to think this video helps you start to think this video helps you start to think about these questions yourself. Let me about these questions yourself. Let me about these questions yourself. Let me know how y'all feel and how scared you know how y'all feel and how scared you know how y'all feel and how scared you are of a future where AI can improve are of a future where AI can improve are of a future where AI can improve itself because I will admit I'm quite itself because I will admit I'm quite itself because I will admit I'm quite scared of this too. Let me know how scared of this too. Let me know how scared of this too. Let me know how y'all feel. And until next time, peace y'all feel. And until next time, peace y'all feel. And until next time, peace nerds. What time is it? It's only six. Cool. We What time is it? It's only six. Cool. We can go a little further. A little can go a little further. A little can go a little further. A little longer.

Thank you to Zara for the five bomb. Thank you to Zara for the five bomb. Petraor Qy for the five bomb as well. Petraor Qy for the five bomb as well. Petraor Qy for the five bomb as well. Hey, Bale's throwing a gift to Obsidian Hey, Bale's throwing a gift to Obsidian Hey, Bale's throwing a gift to Obsidian is cool. [ __ ] nerd. IR Lopez 200 CEX is cool. [ __ ] nerd. IR Lopez 200 CEX is cool. [ __ ] nerd. IR Lopez 200 CEX or 100x codeex plus claude which is or 100x codeex plus claude which is or 100x codeex plus claude which is better. The 200x codeex is super better. The 200x codeex is super better. The 200x codeex is super generous. Like I cannot come close. generous. Like I cannot come close. generous. Like I cannot come close. I would start with just the 100x codeex I would start with just the 100x codeex I would start with just the 100x codeex like the $100 a month codeex. See how like the $100 a month codeex. See how like the $100 a month codeex. See how easily you hit limits and what you miss easily you hit limits and what you miss easily you hit limits and what you miss and then decide what you want to do from and then decide what you want to do from and then decide what you want to do from there. there. there. Dan LV6 here with the one month of Dan LV6 here with the one month of Dan LV6 here with the one month of support. Appreciate that. NMG Twitch, support. Appreciate that. NMG Twitch, support. Appreciate that. NMG Twitch, good to see you, man. 35 months. I'm good to see you, man. 35 months. I'm good to see you, man. 35 months. I'm surprised that that wasn't there surprised that that wasn't there surprised that that wasn't there earlier. Happy to have you back. Ask earlier. Happy to have you back. Ask earlier. Happy to have you back. Ask John with the 10 bomb. God damn. Thank John with the 10 bomb. God damn. Thank John with the 10 bomb. God damn. Thank you for the support, man. Hope you're you for the support, man. Hope you're you for the support, man. Hope you're doing well. doing well. doing well. Sigh here with the three months of Sigh here with the three months of Sigh here with the three months of support and gamer girl been subbed for support and gamer girl been subbed for support and gamer girl been subbed for 27 months. Damn. 27 months. Damn. 27 months. Damn. Yeah. Thank you, John, for all the help Yeah. Thank you, John, for all the help Yeah. Thank you, John, for all the help with YouTube chat. Subs are on sale. Oh, with YouTube chat. Subs are on sale. Oh, with YouTube chat. Subs are on sale. Oh, that makes sense. Nothing new on YouTube I have to deal Nothing new on YouTube I have to deal with. Thank you again for helping with with. Thank you again for helping with with. Thank you again for helping with the moderation on YouTube. Hey, Bales the moderation on YouTube. Hey, Bales the moderation on YouTube. Hey, Bales with a five bomb as well. Thank you so with a five bomb as well. Thank you so with a five bomb as well. Thank you so much for that.

I I try to not put too much credit I I try to not put too much credit credit into the leaks because there's so credit into the leaks because there's so credit into the leaks because there's so many layers to the leaks. many layers to the leaks. many layers to the leaks. I would like to just wait until we have I would like to just wait until we have I would like to just wait until we have more actual [ __ ] out cuz like right now more actual [ __ ] out cuz like right now more actual [ __ ] out cuz like right now we have so little we have so little we have so little guess what mode is before opening it. I guess what mode is before opening it. I guess what mode is before opening it. I have no idea at all. I'll be honest. have no idea at all. I'll be honest. have no idea at all. I'll be honest. Maria Maria Maria Maria's opinionated development Maria's opinionated development Maria's opinionated development environment. I would never have guessed environment. I would never have guessed environment. I would never have guessed this. This is your T3 code fork or just a This is your T3 code fork or just a bunch of additions to it. Regardless, bunch of additions to it. Regardless, bunch of additions to it. Regardless, it's cool. I'll be the first star. It's going to become your fork. Oh, It's going to become your fork. Oh, great.

definitely need to do the compute crunch definitely need to do the compute crunch vid. I spun up uh chat GBT research to vid. I spun up uh chat GBT research to vid. I spun up uh chat GBT research to hunt for that for me. Let me check that hunt for that for me. Let me check that hunt for that for me. Let me check that quick. Okay, this is good. I just asked to go Okay, this is good. I just asked to go find find find resources and examples of the different resources and examples of the different resources and examples of the different major companies complaining about major companies complaining about major companies complaining about compute allocation.

very curious what norm 55 comes up for very curious what norm 55 comes up for this. It's going to be TSMC, this. It's going to be TSMC, this. It's going to be TSMC, but I'm curious how it frames it.

Ain't it crazy that TSMC's revenue is Ain't it crazy that TSMC's revenue is only up 35 to 36% only up 35 to 36% only up 35 to 36% when they are like the one of the most when they are like the one of the most when they are like the one of the most important companies for all of this? I got a silly thing I want to show in I got a silly thing I want to show in this video.

Do you know what Google, Microsoft, and Do you know what Google, Microsoft, and Do you know what Google, Microsoft, and Do you know what Google, Microsoft, and Do you know what Google, Microsoft, and Anthropic all have in common? I'll give Anthropic all have in common? I'll give Anthropic all have in common? I'll give you a hint. It's kind of the title of you a hint. It's kind of the title of you a hint. It's kind of the title of the video. They're all massively the video. They're all massively the video. They're all massively constrained by compute. They would all constrained by compute. They would all constrained by compute. They would all be able to do way more business and sell be able to do way more business and sell be able to do way more business and sell way more product if they just Sorry, I way more product if they just Sorry, I way more product if they just Sorry, I need some water. Okay, Okay, this will be a fun one. Do you know what Microsoft, Google, and Do you know what Microsoft, Google, and Anthropic all have in common? I'll give Anthropic all have in common? I'll give Anthropic all have in common? I'll give you a hint. It's the title of the video. you a hint. It's the title of the video. you a hint. It's the title of the video. They're all massively constrained by They're all massively constrained by They're all massively constrained by compute, like absurdly. So, the CEO of Microsoft said directly that the CEO of Microsoft said directly that they grew a ton Q1, but the problem that they grew a ton Q1, but the problem that they grew a ton Q1, but the problem that they have is that even with the they have is that even with the they have is that even with the additional data center capacity they're additional data center capacity they're additional data center capacity they're bringing online, they expect to remain bringing online, they expect to remain bringing online, they expect to remain capacity constrained through the first capacity constrained through the first capacity constrained through the first half of the fiscal year. That just that half of the fiscal year. That just that half of the fiscal year. That just that literally is them saying we cannot make literally is them saying we cannot make literally is them saying we cannot make more money because we don't have enough more money because we don't have enough more money because we don't have enough compute for it. Here compute for it. Here compute for it. Here is Sundar, the CEO of Alphabet and is Sundar, the CEO of Alphabet and is Sundar, the CEO of Alphabet and Google, saying directly that they are Google, saying directly that they are Google, saying directly that they are supply constrained even as they're supply constrained even as they're supply constrained even as they're ramping up capacity.

This is a company that makes their own This is a company that makes their own chips and they still don't have enough. That's why Anthropic partnered with a That's why Anthropic partnered with a company they hate, SpaceX. The same company they hate, SpaceX. The same company they hate, SpaceX. The same anthropic that banned XAI and SpaceX anthropic that banned XAI and SpaceX anthropic that banned XAI and SpaceX from using their models they were from using their models they were from using their models they were concerned about distilling is now paying concerned about distilling is now paying concerned about distilling is now paying a billion dollars a month to use a billion dollars a month to use a billion dollars a month to use SpaceX's spare compute. And apparently Google thought this was a And apparently Google thought this was a good idea because now they're paying good idea because now they're paying good idea because now they're paying SpaceX $920 million a month for compute SpaceX $920 million a month for compute SpaceX $920 million a month for compute as well because everyone is desperate as well because everyone is desperate as well because everyone is desperate for GPUs. You can't even rent a basic You can't You can't even rent a basic You can't even rent a standard H100 on RunPod even rent a standard H100 on RunPod even rent a standard H100 on RunPod anymore because everyone is just out of anymore because everyone is just out of anymore because everyone is just out of available compute. It's kind of crazy. available compute. It's kind of crazy. available compute. It's kind of crazy. Everywhere I've checked, H100 standard Everywhere I've checked, H100 standard Everywhere I've checked, H100 standard PCI mounts are just everywhere I check, PCI mounts are just everywhere I check, PCI mounts are just everywhere I check, standard H100s on PCI are just not standard H100s on PCI are just not standard H100s on PCI are just not available anymore because the demand is available anymore because the demand is available anymore because the demand is so insane.

What's even crazier is this goes beyond What's even crazier is this goes beyond What's even crazier is this goes beyond What's even crazier is this goes beyond What's even crazier is this goes beyond GPUs. GPUs. GPUs. Sorry. Sorry. Sorry. What's even crazier is this goes beyond What's even crazier is this goes beyond What's even crazier is this goes beyond GPUs with companies like Western GPUs with companies like Western GPUs with companies like Western Digital, the hard drive company, sold Digital, the hard drive company, sold Digital, the hard drive company, sold out for all of 2026 as of February. things are pretty crazy with compute things are pretty crazy with compute right now and I don't think most people right now and I don't think most people right now and I don't think most people understand the severity of the problem. understand the severity of the problem. understand the severity of the problem. I want to do my best I want to do my I want to do my best I want to do my I want to do my best I want to do my best to break it down for all of you. best to break it down for all of you. best to break it down for all of you. But in order to cover Yeah. But in order to cover Yeah. But in order to cover Yeah. I want to do my best to break this down I want to do my best to break this down I want to do my best to break this down for all of you, but if I want to afford for all of you, but if I want to afford for all of you, but if I want to afford the 64 gigs of RAM I just bought, we're the 64 gigs of RAM I just bought, we're the 64 gigs of RAM I just bought, we're going to need to take a quick break for going to need to take a quick break for going to need to take a quick break for today's sponsor. Yeah, I want a better today's sponsor. Yeah, I want a better today's sponsor. Yeah, I want a better read on that. I want to do my best to read on that. I want to do my best to read on that. I want to do my best to break corrected in my brain. I'm sorry. break corrected in my brain. I'm sorry. break corrected in my brain. I'm sorry. Speaking is hard when you do it for this Speaking is hard when you do it for this Speaking is hard when you do it for this long.

I want to do my best to break this all I want to do my best to break this all down for you guys, but if I want to be down for you guys, but if I want to be down for you guys, but if I want to be able to afford the RAM that I just able to afford the RAM that I just able to afford the RAM that I just bought, I need to take a quick break for bought, I need to take a quick break for bought, I need to take a quick break for today's sponsor. The reason I'm making this video today The reason I'm making this video today is because of the Google SpaceX deal. is because of the Google SpaceX deal. is because of the Google SpaceX deal. The Anthropic SpaceX deal surprised me, The Anthropic SpaceX deal surprised me, The Anthropic SpaceX deal surprised me, but like Anthropic's not a compute but like Anthropic's not a compute but like Anthropic's not a compute company. They're a model company. company. They're a model company. company. They're a model company. Makes sense that they didn't have Makes sense that they didn't have Makes sense that they didn't have infinite GPUs. infinite GPUs. infinite GPUs. Google is making their own compute. Google is making their own compute. Google is making their own compute. Google manufactures their own TPUs. They Google manufactures their own TPUs. They Google manufactures their own TPUs. They build the chips that they plan to run build the chips that they plan to run build the chips that they plan to run their inference on. In fact, in February of this year, Meta In fact, in February of this year, Meta signed a multi-billion dollar deal to signed a multi-billion dollar deal to signed a multi-billion dollar deal to rent chips from Google.

So, Google is now buying from a So, Google is now buying from a competitor. competitor. competitor. So, while Anthropic was buying from a So, while Anthropic was buying from a So, while Anthropic was buying from a competitor because like SpaceX is Grock, competitor because like SpaceX is Grock, competitor because like SpaceX is Grock, they make AI models. Anthropic is they make AI models. Anthropic is they make AI models. Anthropic is anthropic. They make AI models. They anthropic. They make AI models. They anthropic. They make AI models. They weren't buying the thing they compete weren't buying the thing they compete weren't buying the thing they compete with. They're buying the tools they need with. They're buying the tools they need with. They're buying the tools they need to increase the competitive nature of to increase the competitive nature of to increase the competitive nature of their product. their product. their product. Google sells compute. Google now also Google sells compute. Google now also Google sells compute. Google now also sells. Google sells compute and they sells. Google sells compute and they sells. Google sells compute and they sell models. So despite the fact that sell models. So despite the fact that sell models. So despite the fact that they sell and rent compute to companies they sell and rent compute to companies they sell and rent compute to companies like Meta, they are still so low they like Meta, they are still so low they like Meta, they are still so low they have to go buy it from companies like have to go buy it from companies like have to go buy it from companies like SpaceX. And that is how we got here. The compute And that is how we got here. The compute crisis.

The amount of compute available has gone The amount of compute available has gone up meaningfully year-over-year, but up meaningfully year-over-year, but up meaningfully year-over-year, but nowhere near as fast as the demand has nowhere near as fast as the demand has nowhere near as fast as the demand has gone up. I have no idea what's going on. Oh, hi I have no idea what's going on. Oh, hi Aiden. There are many layers to this problem. There are many layers to this problem. Obviously, there is the massive demand But there's also the complicated supply But there's also the complicated supply chain problem here One of the other severely underrated One of the other severely underrated problems here is actually power problems here is actually power problems here is actually power availability availability availability because as more compute comes online, we because as more compute comes online, we because as more compute comes online, we need more power for it.

No. Uh, Aiden is a mod because Aiden No. Uh, Aiden is a mod because Aiden helps us with settings on the stream helps us with settings on the stream helps us with settings on the stream because he makes Fosabot, which is the because he makes Fosabot, which is the because he makes Fosabot, which is the bot that we use on stream. I'm going to do my best to visualize I'm going to do my best to visualize this problem, but it's admittedly going this problem, but it's admittedly going this problem, but it's admittedly going to be difficult. So, bear with me as I to be difficult. So, bear with me as I to be difficult. So, bear with me as I try to figure this out. We're going to try to figure this out. We're going to try to figure this out. We're going to go through layers of how we're going to go through layers of how we're going to go through layers of how we're going to go through the layers of how sand go through the layers of how sand go through the layers of how sand effectively becomes the prompts that effectively becomes the prompts that effectively becomes the prompts that you're writing and getting responses to.

you're writing and getting responses to. you're writing and getting responses to. We're going to start a little higher up We're going to start a little higher up We're going to start a little higher up the stack than we probably should. Like the stack than we probably should. Like the stack than we probably should. Like I'm tempted to go into courts, but we'll I'm tempted to go into courts, but we'll I'm tempted to go into courts, but we'll avoid it. And we'll start with where avoid it. And we'll start with where avoid it. And we'll start with where most of the things that matter do. TSMC. most of the things that matter do. TSMC. most of the things that matter do. TSMC. TSMC is a company or TSMC is the Taiwan TSMC is a company or TSMC is the Taiwan TSMC is a company or TSMC is the Taiwan Semiconductor. I always It's Taiwan Semiconductor I always It's Taiwan Semiconductor Manufacturing Company. Yeah, cool. TSMC Manufacturing Company. Yeah, cool. TSMC Manufacturing Company. Yeah, cool. TSMC is the Taiwan Semiconductor is the Taiwan Semiconductor is the Taiwan Semiconductor Manufacturing Company. It was formed by Manufacturing Company. It was formed by Manufacturing Company. It was formed by a person who used to work at Texas a person who used to work at Texas a person who used to work at Texas Instrument in the US who left back to Instrument in the US who left back to Instrument in the US who left back to his home country of Taiwan in order to his home country of Taiwan in order to his home country of Taiwan in order to build better manufacturing of build better manufacturing of build better manufacturing of semiconductors as a generic layer for semiconductors as a generic layer for semiconductors as a generic layer for other companies. Previously, companies other companies. Previously, companies other companies. Previously, companies would make their own semiconductors and would make their own semiconductors and would make their own semiconductors and fab their own like process and also make fab their own like process and also make fab their own like process and also make the processors themselves. But TSMC the processors themselves. But TSMC the processors themselves. But TSMC doesn't sell something you buy as an end doesn't sell something you buy as an end doesn't sell something you buy as an end user. You can't put a TSMC chip into user. You can't put a TSMC chip into user. You can't put a TSMC chip into your computer. You give TSMC the plans your computer. You give TSMC the plans your computer. You give TSMC the plans on how you want to manufacture your chip on how you want to manufacture your chip on how you want to manufacture your chip and then they help you with the and then they help you with the and then they help you with the manufacturing process for it. So every manufacturing process for it. So every manufacturing process for it. So every company doing compute now from Apple to company doing compute now from Apple to company doing compute now from Apple to Nvidia to AMD to Intel works with TSMC Nvidia to AMD to Intel works with TSMC Nvidia to AMD to Intel works with TSMC to fab the silicon that they use for to fab the silicon that they use for to fab the silicon that they use for their chips.

their chips. their chips. Apple was one of the companies that bet Apple was one of the companies that bet Apple was one of the companies that bet on them biggest initially and others on them biggest initially and others on them biggest initially and others have slowly started to realize TSMC's have slowly started to realize TSMC's have slowly started to realize TSMC's manufacturing is just far beyond manufacturing is just far beyond manufacturing is just far beyond anywhere else in the world and have anywhere else in the world and have anywhere else in the world and have relied on it more and more heavily as a relied on it more and more heavily as a relied on it more and more heavily as a result. some amount of their allocation goes to, some amount of their allocation goes to, as I mentioned before, some out of their as I mentioned before, some out of their as I mentioned before, some out of their allocation is already purchased upfront allocation is already purchased upfront allocation is already purchased upfront by Apple and Apple has a crazy deal with by Apple and Apple has a crazy deal with by Apple and Apple has a crazy deal with them where they get first bid on new them where they get first bid on new them where they get first bid on new manufacturing that TSMC spins up. Since manufacturing that TSMC spins up. Since manufacturing that TSMC spins up. Since Apple has so since Apple's historically Apple has so since Apple's historically Apple has so since Apple's historically been such a big customer of TSMC and has been such a big customer of TSMC and has been such a big customer of TSMC and has so many crazy deals with them, they have so many crazy deals with them, they have so many crazy deals with them, they have managed to hold strong with their managed to hold strong with their managed to hold strong with their allocation for a while. That's why allocation for a while. That's why allocation for a while. That's why they're not having the issues with they're not having the issues with they're not having the issues with making new computers or manufacturing making new computers or manufacturing making new computers or manufacturing new phones that a lot of other companies new phones that a lot of other companies new phones that a lot of other companies have because this particular spot in the have because this particular spot in the have because this particular spot in the pipeline and in the supply chain is pipeline and in the supply chain is pipeline and in the supply chain is really strongly purchased and agreed really strongly purchased and agreed really strongly purchased and agreed upon for them. So, Apple is still upon for them. So, Apple is still upon for them. So, Apple is still relatively marked safe, at least in this relatively marked safe, at least in this relatively marked safe, at least in this layer. Don't worry though, that will layer. Don't worry though, that will layer. Don't worry though, that will change as we go.

change as we go. change as we go. The rest of this was split across lots The rest of this was split across lots The rest of this was split across lots of other companies. But over time, the of other companies. But over time, the of other companies. But over time, the section of this that is for Nvidia has section of this that is for Nvidia has section of this that is for Nvidia has grown massively. So every couple months, grown massively. So every couple months, grown massively. So every couple months, Nvidia wants to increase their Nvidia wants to increase their Nvidia wants to increase their manufacturing more. And as a result, the manufacturing more. And as a result, the manufacturing more. And as a result, the amount of this that belongs to them gets amount of this that belongs to them gets amount of this that belongs to them gets bigger and bigger. Let's just say it's bigger and bigger. Let's just say it's bigger and bigger. Let's just say it's like the majority here. And then like the majority here. And then like the majority here. And then whatever's left is everyone else. whatever's left is everyone else. whatever's left is everyone else. So Apple gets their little share here. So Apple gets their little share here. So Apple gets their little share here. Nvidia has a big chunk here and then Nvidia has a big chunk here and then Nvidia has a big chunk here and then there is whatever is going on up here. there is whatever is going on up here. there is whatever is going on up here. This is just one of the things Nvidia This is just one of the things Nvidia This is just one of the things Nvidia needs to make a GPU though. needs to make a GPU though. needs to make a GPU though. So right now the size of how much Nvidia So right now the size of how much Nvidia So right now the size of how much Nvidia can do is at best this big because this can do is at best this big because this can do is at best this big because this is the amount of TSMC manufacturing they is the amount of TSMC manufacturing they is the amount of TSMC manufacturing they have. This is how much they can do at have. This is how much they can do at have. This is how much they can do at best. But there are other things they best. But there are other things they best. But there are other things they need in order to make their GPUs.

Because not only do they need all of the Because not only do they need all of the TSMC manufacturing for it, they also TSMC manufacturing for it, they also TSMC manufacturing for it, they also need memory. And high bandwidth memory manufacturing And high bandwidth memory manufacturing is a very interesting space because is a very interesting space because is a very interesting space because historically like it was important to historically like it was important to historically like it was important to have good RAM. And there was lots of have good RAM. And there was lots of have good RAM. And there was lots of companies that would purchase from the companies that would purchase from the companies that would purchase from the high bandwidth memory manufacturers. high bandwidth memory manufacturers. high bandwidth memory manufacturers. Most of the time though their work was Most of the time though their work was Most of the time though their work was going into consumer devices. They would going into consumer devices. They would going into consumer devices. They would sell RAM or they would sell RAM or they would sell RAM or they would they would sell NAND chips that could be they would sell NAND chips that could be they would sell NAND chips that could be used for RAM or for SSDs and they would used for RAM or for SSDs and they would used for RAM or for SSDs and they would sell that to companies that needed RAM sell that to companies that needed RAM sell that to companies that needed RAM for their devices. Whether it was for their devices. Whether it was for their devices. Whether it was Qualcomm to put in phones or Apple to Qualcomm to put in phones or Apple to Qualcomm to put in phones or Apple to put in phones or if it was to companies put in phones or if it was to companies put in phones or if it was to companies like Crucial or SKH or whoever else that like Crucial or SKH or whoever else that like Crucial or SKH or whoever else that makes memory for users to put into their makes memory for users to put into their makes memory for users to put into their computers or if it was to Dell to make computers or if it was to Dell to make computers or if it was to Dell to make memory chips that they would put in memory chips that they would put in memory chips that they would put in their computers. The high bandwidth their computers. The high bandwidth their computers. The high bandwidth memory chips were a very diverse set of memory chips were a very diverse set of memory chips were a very diverse set of places they would go, but there was places they would go, but there was places they would go, but there was really only three manufacturers that really only three manufacturers that really only three manufacturers that mattered. It was SKH, Samsung, and I mattered. It was SKH, Samsung, and I mattered. It was SKH, Samsung, and I forgot who the third is.

So thankfully unlike TSMC there are So thankfully unlike TSMC there are three companies doing this. SKH three companies doing this. SKH three companies doing this. SKH highinex, Samsung and Micron. highinex, Samsung and Micron. highinex, Samsung and Micron. Historically they have distri Historically they have distri Historically they have distri historically they have split their historically they have split their historically they have split their allocation across lots of different allocation across lots of different allocation across lots of different groups. groups. groups. But now the demand for things like GPUs But now the demand for things like GPUs But now the demand for things like GPUs is so absurdly high that they have is so absurdly high that they have is so absurdly high that they have reallocated entirely. SKhinx maintained reallocated entirely. SKhinx maintained reallocated entirely. SKhinx maintained a consumer brand of memory called a consumer brand of memory called a consumer brand of memory called Crucial and most of the RAM in most of Crucial and most of the RAM in most of Crucial and most of the RAM in most of my computers here is from Crucial. my computers here is from Crucial. my computers here is from Crucial. Crucial no longer exists. Sorry, it was Micron, not SKH. My bad on Sorry, it was Micron, not SKH. My bad on the memory there. There's three of them. the memory there. There's three of them. the memory there. There's three of them. Sorry, I made a mistake. Micron was the Sorry, I made a mistake. Micron was the Sorry, I made a mistake. Micron was the owner of Crucial and Micron has decided owner of Crucial and Micron has decided owner of Crucial and Micron has decided to shut down Crucial. to shut down Crucial. to shut down Crucial. Micron has made the difficult decision Micron has made the difficult decision Micron has made the difficult decision to widen down the Crucial consumer to widen down the Crucial consumer to widen down the Crucial consumer business. Micron will ship Crucial business. Micron will ship Crucial business. Micron will ship Crucial consumer products through February of consumer products through February of consumer products through February of this year with warranty and support this year with warranty and support this year with warranty and support continuing. Micron crucial consumer continuing. Micron crucial consumer continuing. Micron crucial consumer products may continue to be available products may continue to be available products may continue to be available for purchase from distributors and for purchase from distributors and for purchase from distributors and resellers for some time. That is the resellers for some time. That is the resellers for some time. That is the case. I bought 64 gigs of crucial memory case. I bought 64 gigs of crucial memory case. I bought 64 gigs of crucial memory a few days ago. It was very expensive.

So all of this allocation used to be So all of this allocation used to be split across consumer manufacturing of split across consumer manufacturing of split across consumer manufacturing of consumer hardware, consumer hardware, consumer hardware, lots of other devices and then data lots of other devices and then data lots of other devices and then data center use cases and GPU use cases. center use cases and GPU use cases. center use cases and GPU use cases. Since since AI requires so much RAM to Since since AI requires so much RAM to Since since AI requires so much RAM to run, even smaller models like DeepSeek run, even smaller models like DeepSeek run, even smaller models like DeepSeek V4 flash require over a 100 gigs of RAM, V4 flash require over a 100 gigs of RAM, V4 flash require over a 100 gigs of RAM, memory is super valuable to these memory is super valuable to these memory is super valuable to these businesses. So the need for it has businesses. So the need for it has businesses. So the need for it has skyrocketed and with such prices have skyrocketed and with such prices have skyrocketed and with such prices have skyrocketed as well. So Nvidia needs an allocation of this as So Nvidia needs an allocation of this as well and the price of that allocation is well and the price of that allocation is well and the price of that allocation is skyrocketing massively. skyrocketing massively. skyrocketing massively. I will say that the numbers in these I will say that the numbers in these I will say that the numbers in these charts like the the percentage split charts like the the percentage split charts like the the percentage split here is not super accurate. TSMC has here is not super accurate. TSMC has here is not super accurate. TSMC has reported that Nvidia is only about a reported that Nvidia is only about a reported that Nvidia is only about a fourth or so of their manufacturing fourth or so of their manufacturing fourth or so of their manufacturing allocation right now. So this is a huge allocation right now. So this is a huge allocation right now. So this is a huge chunk here. This is just meant to chunk here. This is just meant to chunk here. This is just meant to emphasize the point not to be literally emphasize the point not to be literally emphasize the point not to be literally TSMC's majority Nvidia. Just trying to TSMC's majority Nvidia. Just trying to TSMC's majority Nvidia. Just trying to make this as easy to visualize as make this as easy to visualize as make this as easy to visualize as possible. Give me some creative give me possible. Give me some creative give me possible. Give me some creative give me some creative wiggle room. Okay.

So we have these two key components that So we have these two key components that are necessary for Nvidia to be able to are necessary for Nvidia to be able to are necessary for Nvidia to be able to make GPUs. But there are other layers make GPUs. But there are other layers make GPUs. But there are other layers here as well. here as well. here as well. And not all of them are and not all of And not all of them are and not all of And not all of them are and not all of them are directly in front of Nvidia them are directly in front of Nvidia them are directly in front of Nvidia either because in order to run those either because in order to run those either because in order to run those NVIDIA GPUs, you need a few other NVIDIA GPUs, you need a few other NVIDIA GPUs, you need a few other things. things. things. First, you need hard drives. First, you need hard drives. First, you need hard drives. because you need somewhere to store the because you need somewhere to store the because you need somewhere to store the data that the models are actually you data that the models are actually you data that the models are actually you need somewhere to store the data that need somewhere to store the data that need somewhere to store the data that these GPUs are actually operating on. these GPUs are actually operating on. these GPUs are actually operating on. And apparently the demand for hard And apparently the demand for hard And apparently the demand for hard drives is skyrocketing like never drives is skyrocketing like never drives is skyrocketing like never before. Let me find my example here before. Let me find my example here before. Let me find my example here because it annoys me.

Last year, I bought four 16 TB hard Last year, I bought four 16 TB hard drives refurbed from Server Part Deals, drives refurbed from Server Part Deals, drives refurbed from Server Part Deals, one of my favorite sources for hard one of my favorite sources for hard one of my favorite sources for hard drives for about 170 bucks each for 16 drives for about 170 bucks each for 16 drives for about 170 bucks each for 16 TB drives. I hate myself for revealing server part I hate myself for revealing server part deals right now because this is one of deals right now because this is one of deals right now because this is one of the few good sources we still have and the few good sources we still have and the few good sources we still have and the more people know about it, the more the more people know about it, the more the more people know about it, the more [ __ ] I am when I need more hard [ __ ] I am when I need more hard [ __ ] I am when I need more hard drives. Why is the capacity selection bro? Okay, Why is the capacity selection bro? Okay, there we go. Worse 16 TB drives than what I have in Worse 16 TB drives than what I have in my NAS right now are going for $360, my NAS right now are going for $360, my NAS right now are going for $360, more than 2x the cost when my purchase more than 2x the cost when my purchase more than 2x the cost when my purchase was in 2024. Insane. I just bought a handful of 28 Insane. I just bought a handful of 28 terbte drives and they were like 600 terbte drives and they were like 600 terbte drives and they were like 600 plus each. plus each. plus each. I spent like $3,500 on hard drives I spent like $3,500 on hard drives I spent like $3,500 on hard drives recently. It's crazy.

So hard drive manufacturing is also very So hard drive manufacturing is also very tight right now and something that like tight right now and something that like tight right now and something that like consumers have barely even needed for a consumers have barely even needed for a consumers have barely even needed for a while because we all moved to flash while because we all moved to flash while because we all moved to flash storage. storage. storage. Hard drives are still useful for lots of Hard drives are still useful for lots of Hard drives are still useful for lots of like big arch or hard drives are still like big arch or hard drives are still like big arch or hard drives are still useful for lots of big archival stuff, useful for lots of big archival stuff, useful for lots of big archival stuff, especially when they were cheap. But now especially when they were cheap. But now especially when they were cheap. But now hard drives are like as expensive as hard drives are like as expensive as hard drives are like as expensive as SSDs were not long ago, SSDs were not long ago, SSDs were not long ago, which is unbelievable to me. So, if a company is able to buy all the So, if a company is able to buy all the NVIDIA GPUs they need, but they can't NVIDIA GPUs they need, but they can't NVIDIA GPUs they need, but they can't get enough hard drives to actually run get enough hard drives to actually run get enough hard drives to actually run them, then they're not going to buy them, then they're not going to buy them, then they're not going to buy those GPUs because they're making all of those GPUs because they're making all of those GPUs because they're making all of these decisions up front. So, if there these decisions up front. So, if there these decisions up front. So, if there aren't enough hard drives, then Nvidia's aren't enough hard drives, then Nvidia's aren't enough hard drives, then Nvidia's manufacturing capabilities barely even manufacturing capabilities barely even manufacturing capabilities barely even matter anymore. But then all of this matter anymore. But then all of this matter anymore. But then all of this gets bottlenecked by yet another layer, gets bottlenecked by yet another layer, gets bottlenecked by yet another layer, power. power. power. How much power is available? Power grids How much power is available? Power grids How much power is available? Power grids are struggling right now.

Electricity demand growth is led by an Electricity demand growth is led by an increase in the commercial sector which increase in the commercial sector which increase in the commercial sector which is expected to outpace residential is expected to outpace residential is expected to outpace residential demand in 2027 for the first time on demand in 2027 for the first time on demand in 2027 for the first time on record. We are now at the point where record. We are now at the point where record. We are now at the point where industry use of power in the US is industry use of power in the US is industry use of power in the US is higher than consumer like residential higher than consumer like residential higher than consumer like residential use. This is why these big data this is why This is why these big data this is why these big compute companies like these big compute companies like these big compute companies like Microsoft are starting to invest. Yeah. Microsoft are starting to invest. Yeah. Microsoft are starting to invest. Yeah. This is why these big compute companies This is why these big compute companies This is why these big compute companies like Microsoft with Azure are starting like Microsoft with Azure are starting like Microsoft with Azure are starting to invest in power like actual to invest in power like actual to invest in power like actual introduction of new power plants to the introduction of new power plants to the introduction of new power plants to the grid trying to get nuclear energy grid trying to get nuclear energy grid trying to get nuclear energy unblocked and more. And when you compare the rate of And when you compare the rate of electricity generation in the US electricity generation in the US electricity generation in the US compared to in China, you see how bad we compared to in China, you see how bad we compared to in China, you see how bad we have to catch up here. We are not have to catch up here. We are not have to catch up here. We are not introducing more power to the grid introducing more power to the grid introducing more power to the grid anywhere near fast enough. The amount of anywhere near fast enough. The amount of anywhere near fast enough. The amount of the amount of increase to the grid that the amount of increase to the grid that the amount of increase to the grid that China does in any given year from 2016 China does in any given year from 2016 China does in any given year from 2016 onwards is higher than we have done onwards is higher than we have done onwards is higher than we have done since the '9s.

We should still be focused on making We should still be focused on making things more efficient, but we also need things more efficient, but we also need things more efficient, but we also need to have more energy and ideally more to have more energy and ideally more to have more energy and ideally more clean energy from resources that we know clean energy from resources that we know clean energy from resources that we know we can get it from. This chart scares me. And I don't think This chart scares me. And I don't think we have properly we have properly we have properly This chart scares me. I don't think This chart scares me. I don't think This chart scares me. I don't think we've properly prepared our nation for we've properly prepared our nation for we've properly prepared our nation for the increasing demand of power to get the increasing demand of power to get the increasing demand of power to get where we want to in the AI race. And if any one of these sections gets And if any one of these sections gets any smaller, it effectively works as a any smaller, it effectively works as a any smaller, it effectively works as a filter, preventing Nvidia from selling filter, preventing Nvidia from selling filter, preventing Nvidia from selling more GPUs and preventing AI businesses more GPUs and preventing AI businesses more GPUs and preventing AI businesses from being able to grow and increase the from being able to grow and increase the from being able to grow and increase the number of customers they have. number of customers they have. number of customers they have. Everyone's compute constraint Everyone's compute constraint Everyone's compute constraint except for one company, SpaceX. And this raises a couple SpaceX. And this raises a couple important questions. One is why isn't important questions. One is why isn't important questions. One is why isn't SpaceX affected?

SpaceX affected? SpaceX affected? Why are they so capable of having all Why are they so capable of having all Why are they so capable of having all this spare compute that none of their this spare compute that none of their this spare compute that none of their competitors have? competitors have? competitors have? and two which is similarly and then we and two which is similarly and then we and two which is similarly and then we have question two which is why not just have question two which is why not just have question two which is why not just make more like why can't we just create make more like why can't we just create make more like why can't we just create more high bandwidth memory why can't we more high bandwidth memory why can't we more high bandwidth memory why can't we just fab more silicon why can't we just just fab more silicon why can't we just just fab more silicon why can't we just make more GPUs or create more hard make more GPUs or create more hard make more GPUs or create more hard drives or increase the power drives or increase the power drives or increase the power availability by making more power plants availability by making more power plants availability by making more power plants and power grids and power grids and power grids why can't we just make more why can't we just make more why can't we just make more well as much as AI has accelerated our well as much as AI has accelerated our well as much as AI has accelerated our ability to [ __ ] out new software it has ability to [ __ ] out new software it has ability to [ __ ] out new software it has not made made it particularly faster to not made made it particularly faster to not made made it particularly faster to break ground and create new break ground and create new break ground and create new manufacturing for things like silicon. manufacturing for things like silicon. manufacturing for things like silicon. TSMC estimates TSMC estimates TSMC estimates TSMC estimates that additional TSMC estimates that additional TSMC estimates that additional fabrication capabilities can take as fabrication capabilities can take as fabrication capabilities can take as much as 8 to 10 years to build up. when much as 8 to 10 years to build up. when much as 8 to 10 years to build up. when they decide they want to make more chips they decide they want to make more chips they decide they want to make more chips and they want to do more manufacturing, and they want to do more manufacturing, and they want to do more manufacturing, they have to start planning eight plus they have to start planning eight plus they have to start planning eight plus years ahead and they're often selling years ahead and they're often selling years ahead and they're often selling the allocation off of these theoretical the allocation off of these theoretical the allocation off of these theoretical presses that don't exist yet 6 to 8 presses that don't exist yet 6 to 8 presses that don't exist yet 6 to 8 years ahead. Apple's deals are insanely years ahead. Apple's deals are insanely years ahead. Apple's deals are insanely long-term in that regard.

Despite the insane growth we've seen at Despite the insane growth we've seen at companies like Nvidia, TSMC has only companies like Nvidia, TSMC has only companies like Nvidia, TSMC has only seen about a 40% growth in their revenue seen about a 40% growth in their revenue seen about a 40% growth in their revenue year-over-year. Not because the demand year-over-year. Not because the demand year-over-year. Not because the demand isn't massively skyrocketing, simply isn't massively skyrocketing, simply isn't massively skyrocketing, simply because they can't supply the demand as because they can't supply the demand as because they can't supply the demand as it skyrockets. People are buying out it skyrockets. People are buying out it skyrockets. People are buying out allocation years in advance because they allocation years in advance because they allocation years in advance because they have to because they can't get it right have to because they can't get it right have to because they can't get it right now. There's a reason TSMC is the only There's a reason TSMC is the only company doing this well, and it's company doing this well, and it's company doing this well, and it's because they invested really heavily, because they invested really heavily, because they invested really heavily, really early, and it took them a decade really early, and it took them a decade really early, and it took them a decade and a half of failures to get there. and a half of failures to get there. and a half of failures to get there. Ready for a really funny fact about TSMC Ready for a really funny fact about TSMC Ready for a really funny fact about TSMC that not a lot of people did know. Remember this red ring of death? Remember this red ring of death? Bet you didn't know this was TSMC's Bet you didn't know this was TSMC's Bet you didn't know this was TSMC's fault. Microsoft and Nvidia were two of the Microsoft and Nvidia were two of the first companies to bet heavily on TSMC first companies to bet heavily on TSMC first companies to bet heavily on TSMC manufacturing.

manufacturing. manufacturing. TSMC used an external vendor for the TSMC used an external vendor for the TSMC used an external vendor for the effectively the glue that they would use effectively the glue that they would use effectively the glue that they would use to seal the chip and that manufacturer to seal the chip and that manufacturer to seal the chip and that manufacturer was not correct about the thermal range was not correct about the thermal range was not correct about the thermal range in which that paste actually operated in which that paste actually operated in which that paste actually operated properly. And the cause of the red ring properly. And the cause of the red ring properly. And the cause of the red ring of death was that getting really hot and of death was that getting really hot and of death was that getting really hot and then cooling over and over again would then cooling over and over again would then cooling over and over again would cause that glue to loosen, causing the cause that glue to loosen, causing the cause that glue to loosen, causing the chip to come slightly off of the slot chip to come slightly off of the slot chip to come slightly off of the slot that it was meant to be in. That's also that it was meant to be in. That's also that it was meant to be in. That's also why you could wrap your Xbox in a towel why you could wrap your Xbox in a towel why you could wrap your Xbox in a towel and turn it on until it overheated or and turn it on until it overheated or and turn it on until it overheated or throw it in your literal oven and that throw it in your literal oven and that throw it in your literal oven and that would temporarily fix it. Because when would temporarily fix it. Because when would temporarily fix it. Because when the glue got warmed up again, the chip the glue got warmed up again, the chip the glue got warmed up again, the chip would fall back into the slot it's would fall back into the slot it's would fall back into the slot it's supposed to be in. But then when the supposed to be in. But then when the supposed to be in. But then when the glue got cold, it would pull the chip glue got cold, it would pull the chip glue got cold, it would pull the chip out of the slot. And that was because out of the slot. And that was because out of the slot. And that was because TSMC didn't have good enough process to TSMC didn't have good enough process to TSMC didn't have good enough process to detect these types of failures in their detect these types of failures in their detect these types of failures in their manufacturing. This was early in their manufacturing. This was early in their manufacturing. This was early in their history. So the Xbox failed so Nvidia history. So the Xbox failed so Nvidia history. So the Xbox failed so Nvidia could win. As silly as that is, it just could win. As silly as that is, it just could win. As silly as that is, it just took a long time for this company to get took a long time for this company to get took a long time for this company to get their [ __ ] together.

their [ __ ] together. their [ __ ] together. Not that long ago, TSMC's process was Not that long ago, TSMC's process was Not that long ago, TSMC's process was not thorough enough because they were not thorough enough because they were not thorough enough because they were still new and getting all this [ __ ] still new and getting all this [ __ ] still new and getting all this [ __ ] right is hard. And the result was that right is hard. And the result was that right is hard. And the result was that they weren't even reliable enough for they weren't even reliable enough for they weren't even reliable enough for game consoles. Now we're relying on them game consoles. Now we're relying on them game consoles. Now we're relying on them to power the entire [ __ ] world. And the same goes for all of these other And the same goes for all of these other sources of manufacturing. Things like sources of manufacturing. Things like sources of manufacturing. Things like high bandwidth memory is not easy to high bandwidth memory is not easy to high bandwidth memory is not easy to produce. There's a reason only three produce. There's a reason only three produce. There's a reason only three companies in the world can do this. And companies in the world can do this. And companies in the world can do this. And those three companies are also making those three companies are also making those three companies are also making all the chips that go into all the SSDs all the chips that go into all the SSDs all the chips that go into all the SSDs that we use, all of the phones that we that we use, all of the phones that we that we use, all of the phones that we use, all the all the SD cards and CFast use, all the all the SD cards and CFast use, all the all the SD cards and CFast cards that we use for our cameras and cards that we use for our cameras and cards that we use for our cameras and things. All of that is made by three things. All of that is made by three things. All of that is made by three companies. And those three companies can companies. And those three companies can companies. And those three companies can make the same thing for Nvidia instead. make the same thing for Nvidia instead. make the same thing for Nvidia instead. Why would they sell us consumer chips Why would they sell us consumer chips Why would they sell us consumer chips that sell okay at reasonable prices when that sell okay at reasonable prices when that sell okay at reasonable prices when they could sell way more to Nvidia?

So the so to put it simply, we are So the so to put it simply, we are trying our hardest to manufacture more, trying our hardest to manufacture more, trying our hardest to manufacture more, but it's going to take a long time. And but it's going to take a long time. And but it's going to take a long time. And if you get your bets wrong here, you're if you get your bets wrong here, you're if you get your bets wrong here, you're kind of screwed. If TSMC ramps up kind of screwed. If TSMC ramps up kind of screwed. If TSMC ramps up production, so they're making 10 times production, so they're making 10 times production, so they're making 10 times more silicon, but the hard drive sector more silicon, but the hard drive sector more silicon, but the hard drive sector doesn't pick up enough or HBM doesn't doesn't pick up enough or HBM doesn't doesn't pick up enough or HBM doesn't pick up enough, then they just spent pick up enough, then they just spent pick up enough, then they just spent billions of dollars spinning out fabs billions of dollars spinning out fabs billions of dollars spinning out fabs for demand they no longer have because for demand they no longer have because for demand they no longer have because everybody is constrained. everybody is constrained. everybody is constrained. I guarantee you if Microsoft could snap I guarantee you if Microsoft could snap I guarantee you if Microsoft could snap their fingers and spend three times more their fingers and spend three times more their fingers and spend three times more money to get two times more compute, money to get two times more compute, money to get two times more compute, they would do it immediately. But they they would do it immediately. But they they would do it immediately. But they can't because every single one of these can't because every single one of these can't because every single one of these bottlenecks needs to be resolved bottlenecks needs to be resolved bottlenecks needs to be resolved together. together. together. If TSMC does 10x production, we just get If TSMC does 10x production, we just get If TSMC does 10x production, we just get constrained on HBM. If high bandwidth constrained on HBM. If high bandwidth constrained on HBM. If high bandwidth memory also 10xes, then we get memory also 10xes, then we get memory also 10xes, then we get constrained on hard drives and power. constrained on hard drives and power. constrained on hard drives and power. We need to bump everything up. And if We need to bump everything up. And if We need to bump everything up. And if any company bets too hard on one any company bets too hard on one any company bets too hard on one specific piece that they are in this specific piece that they are in this specific piece that they are in this puzzle, they get screwed.

puzzle, they get screwed. puzzle, they get screwed. And this also ties into the first And this also ties into the first And this also ties into the first question of why isn't SpaceX affected question of why isn't SpaceX affected question of why isn't SpaceX affected because SpaceX and Enthropic are kind of because SpaceX and Enthropic are kind of because SpaceX and Enthropic are kind of opposites here. opposites here. opposites here. Anthropic had the concern if they over Anthropic had the concern if they over Anthropic had the concern if they over buy GPUs and they don't have the demand buy GPUs and they don't have the demand buy GPUs and they don't have the demand for their inference or the models don't for their inference or the models don't for their inference or the models don't work as great good with scaling laws as work as great good with scaling laws as work as great good with scaling laws as they hope where more compute means they hope where more compute means they hope where more compute means better model. If any of that goes wrong better model. If any of that goes wrong better model. If any of that goes wrong and they purchase too much compute, and they purchase too much compute, and they purchase too much compute, they're out of money and they fail. they're out of money and they fail. they're out of money and they fail. Anthropic was a little conservative with Anthropic was a little conservative with Anthropic was a little conservative with their compute bets last year and that their compute bets last year and that their compute bets last year and that has screwed them because now the compute has screwed them because now the compute has screwed them because now the compute they could have bought last year isn't they could have bought last year isn't they could have bought last year isn't available anymore. available anymore. available anymore. Elon had a lot of conviction about Elon had a lot of conviction about Elon had a lot of conviction about compute becoming a bottleneck. He compute becoming a bottleneck. He compute becoming a bottleneck. He believed this was going to be a really believed this was going to be a really believed this was going to be a really big deal. So he overbought compute for big deal. So he overbought compute for big deal. So he overbought compute for SpaceX and Grock and XAI. That went kind of poorly for them That went kind of poorly for them because Grock just didn't do great. A because Grock just didn't do great. A because Grock just didn't do great. A lot of the best researchers that were at lot of the best researchers that were at lot of the best researchers that were at XAI have since left. The progress XAI have since left. The progress XAI have since left. The progress they're seeing just isn't great, but they're seeing just isn't great, but they're seeing just isn't great, but they already bought all of this compute.

they already bought all of this compute. they already bought all of this compute. Thankfully, they learned how valuable Thankfully, they learned how valuable Thankfully, they learned how valuable that is and that if they can't use it, that is and that if they can't use it, that is and that if they can't use it, someone can. And now this compute that someone can. And now this compute that someone can. And now this compute that they spent a lot of money on, let's see they spent a lot of money on, let's see they spent a lot of money on, let's see how much did XAI spend on Colossus one. Apparently building the initial phase Apparently building the initial phase for the Colossus deployment was only3 for the Colossus deployment was only3 for the Colossus deployment was only3 to4 billion. They are now renting that compute for a They are now renting that compute for a billion dollars a month. billion dollars a month. billion dollars a month. In four months it pays for itself.

one of the reasons XAI was actually kind one of the reasons XAI was actually kind of well equipped for this is the power of well equipped for this is the power of well equipped for this is the power constraints because Elon with Tesla constraints because Elon with Tesla constraints because Elon with Tesla knows a lot about power and was able to knows a lot about power and was able to knows a lot about power and was able to make deals with Tesla battery make deals with Tesla battery make deals with Tesla battery manufacturing in order to make sure that manufacturing in order to make sure that manufacturing in order to make sure that their power would be reliable and if their power would be reliable and if their power would be reliable and if they were ever constrained by the grid they were ever constrained by the grid they were ever constrained by the grid they would have enough backup power to they would have enough backup power to they would have enough backup power to last for some amount of time. They also last for some amount of time. They also last for some amount of time. They also did crazy stuff like gas powered did crazy stuff like gas powered did crazy stuff like gas powered generators, which is funny from the guy generators, which is funny from the guy generators, which is funny from the guy who made the electric car company that who made the electric car company that who made the electric car company that made electric seem much more viable to made electric seem much more viable to made electric seem much more viable to then go burn a bunch of gas in order to then go burn a bunch of gas in order to then go burn a bunch of gas in order to power his GPUs to make racist AI. But at power his GPUs to make racist AI. But at power his GPUs to make racist AI. But at least they're making money off it now. So it turns out having so the thing that So it turns out having so the thing that h so it turns out that compute was as h so it turns out that compute was as h so it turns out that compute was as much a bottleneck as Elon had predicted much a bottleneck as Elon had predicted much a bottleneck as Elon had predicted but their need for it didn't go up as but their need for it didn't go up as but their need for it didn't go up as much as he predicted. So in order to much as he predicted. So in order to much as he predicted. So in order to make money out of that bet he is now make money out of that bet he is now make money out of that bet he is now reselling it to companies like So in reselling it to companies like So in reselling it to companies like So in order to make money off that bet he's order to make money off that bet he's order to make money off that bet he's now reselling it to companies like now reselling it to companies like now reselling it to companies like Anthropic and like Google which is still Anthropic and like Google which is still Anthropic and like Google which is still just so crazy to me. I honestly thought just so crazy to me. I honestly thought just so crazy to me. I honestly thought it was a troll when I saw this post it was a troll when I saw this post it was a troll when I saw this post today that SpaceX is now doing a billion today that SpaceX is now doing a billion today that SpaceX is now doing a billion dollars of compute a month through dollars of compute a month through dollars of compute a month through Google. That Google's paying them a Google. That Google's paying them a Google. That Google's paying them a billion dollars every single month, 12 billion dollars every single month, 12 billion dollars every single month, 12 billion a year almost billion a year almost billion a year almost just for compute.

Google's revenue last year was $400 Google's revenue last year was $400 billion. That means that 3% of Google's total That means that 3% of Google's total revenue is now going to XAI, is now revenue is now going to XAI, is now revenue is now going to XAI, is now going to SpaceX, is now going to one of going to SpaceX, is now going to one of going to SpaceX, is now going to one of their competitors. This is up there with Google's deal with This is up there with Google's deal with Apple where they pay Apple billions of Apple where they pay Apple billions of Apple where they pay Apple billions of dollars a year to be the default search dollars a year to be the default search dollars a year to be the default search engine. Like, this is that level of engine. Like, this is that level of engine. Like, this is that level of crazy. 3% of Google's total revenue crazy. 3% of Google's total revenue crazy. 3% of Google's total revenue going to SpaceX. The point I'm trying to make here is The point I'm trying to make here is that however bad you think the compute that however bad you think the compute that however bad you think the compute crisis is, it is probably worse. crisis is, it is probably worse. crisis is, it is probably worse. Whenever you think you see these Whenever you think you see these Whenever you think you see these companies pinching pennies cuz they want companies pinching pennies cuz they want companies pinching pennies cuz they want to try and squeeze more money out of to try and squeeze more money out of to try and squeeze more money out of you, it's probably not that. They're you, it's probably not that. They're you, it's probably not that. They're probably just dealing with the compute probably just dealing with the compute probably just dealing with the compute crisis because they just don't have crisis because they just don't have crisis because they just don't have enough compute available for all the enough compute available for all the enough compute available for all the demand they're getting. And if you think demand they're getting. And if you think demand they're getting. And if you think they can just fix this by making more they can just fix this by making more they can just fix this by making more stuff, they kind of can if everybody stuff, they kind of can if everybody stuff, they kind of can if everybody makes exactly more enough and the demand makes exactly more enough and the demand makes exactly more enough and the demand sustains for long enough.

sustains for long enough. sustains for long enough. But right now, the winner is whoever has But right now, the winner is whoever has But right now, the winner is whoever has the GPUs. And at this point in time, the GPUs. And at this point in time, the GPUs. And at this point in time, that winner is funny enough of all that winner is funny enough of all that winner is funny enough of all companies, apparently SpaceX, but also companies, apparently SpaceX, but also companies, apparently SpaceX, but also OpenAI, who I've managed to not mention OpenAI, who I've managed to not mention OpenAI, who I've managed to not mention at all so far, because OpenAI made the at all so far, because OpenAI made the at all so far, because OpenAI made the bet a couple years ago that compute bet a couple years ago that compute bet a couple years ago that compute would matter, that scaling laws would would matter, that scaling laws would would matter, that scaling laws would matter, and would go and buy all the matter, and would go and buy all the matter, and would go and buy all the compute they possibly could. This is the compute they possibly could. This is the compute they possibly could. This is the real reason why anthropics rate limits real reason why anthropics rate limits real reason why anthropics rate limits are so much less generous than OpenAIs are so much less generous than OpenAIs are so much less generous than OpenAIs are. They have been considerate of the are. They have been considerate of the are. They have been considerate of the compute crisis since before it even compute crisis since before it even compute crisis since before it even really started and that put OpenAI in a really started and that put OpenAI in a really started and that put OpenAI in a really good spot. Google pretended they really good spot. Google pretended they really good spot. Google pretended they could work around it by making their own could work around it by making their own could work around it by making their own chips. That didn't go great for them and chips. That didn't go great for them and chips. That didn't go great for them and now they're stuck paying Elon for their now they're stuck paying Elon for their now they're stuck paying Elon for their mistake. Enthropic didn't want to mistake. Enthropic didn't want to mistake. Enthropic didn't want to overspend on compute. They screwed up overspend on compute. They screwed up overspend on compute. They screwed up and now they're paying Elon for their and now they're paying Elon for their and now they're paying Elon for their mistake. mistake. mistake. There's only really one winner in all of There's only really one winner in all of There's only really one winner in all of this and that winner is Nvidia because this and that winner is Nvidia because this and that winner is Nvidia because they know how in demand their stuff is they know how in demand their stuff is they know how in demand their stuff is and they literally cannot manufacture and they literally cannot manufacture and they literally cannot manufacture enough to keep up with the demand that enough to keep up with the demand that enough to keep up with the demand that exists. They can't really make less exists. They can't really make less exists. They can't really make less money right now because as long as any money right now because as long as any money right now because as long as any of these other bottlenecks get resolved, of these other bottlenecks get resolved, of these other bottlenecks get resolved, their amount of GPUs they can sell just their amount of GPUs they can sell just their amount of GPUs they can sell just keeps going up. They cannot make enough.

keeps going up. They cannot make enough. keeps going up. They cannot make enough. And if everything resolves, it does And if everything resolves, it does And if everything resolves, it does great for them. If most things resolve, great for them. If most things resolve, great for them. If most things resolve, it still does pretty great for them. it still does pretty great for them. it still does pretty great for them. So, Nvidia, congrats. You're going to be So, Nvidia, congrats. You're going to be So, Nvidia, congrats. You're going to be holding your position in the stock holding your position in the stock holding your position in the stock market for a while, it seems. And this market for a while, it seems. And this market for a while, it seems. And this is not meant to be financial advice. is not meant to be financial advice. is not meant to be financial advice. This is just my read of how chaotic This is just my read of how chaotic This is just my read of how chaotic things are. I hope this breakdown is things are. I hope this breakdown is things are. I hope this breakdown is helpful as you question why your quad helpful as you question why your quad helpful as you question why your quad code keeps running out of usage every code keeps running out of usage every code keeps running out of usage every couple hours. Do I have other points I wanted to make Do I have other points I wanted to make in this one? Let me read through the uh in this one? Let me read through the uh in this one? Let me read through the uh up here if there's anything else up here if there's anything else up here if there's anything else interesting.

I had an anecdote I wanted to sneak I had an anecdote I wanted to sneak sneak in which was that Apple when sneak in which was that Apple when sneak in which was that Apple when negotiating with Samsung. negotiating with Samsung. negotiating with Samsung. Apparently Samsung was planning on Apparently Samsung was planning on Apparently Samsung was planning on getting them to like 50% increase their getting them to like 50% increase their getting them to like 50% increase their costs but they went into negotiations costs but they went into negotiations costs but they went into negotiations saying double because they expected saying double because they expected saying double because they expected Apple to negotiate like they always do. Apple to negotiate like they always do. Apple to negotiate like they always do. So, they went in saying, "We're going to So, they went in saying, "We're going to So, they went in saying, "We're going to 2x the cost per chip." I think Apple 2x the cost per chip." I think Apple 2x the cost per chip." I think Apple didn't negotiate them down to 50% didn't negotiate them down to 50% didn't negotiate them down to 50% increase instead of 100%. And Apple increase instead of 100%. And Apple increase instead of 100%. And Apple immediately said yes to the 2x. immediately said yes to the 2x. immediately said yes to the 2x. And I don't know if or where that would And I don't know if or where that would And I don't know if or where that would fit in this video. So, I think I'm just fit in this video. So, I think I'm just fit in this video. So, I think I'm just going to skip that anecdote even though going to skip that anecdote even though going to skip that anecdote even though it's really cool. it's really cool. it's really cool. Yeah, I have to skip that one. That's Yeah, I have to skip that one. That's Yeah, I have to skip that one. That's sad. I think I've had I think I've said all I I think I've had I think I've said all I have to on this one. If you want to have to on this one. If you want to have to on this one. If you want to build a computer, now might not seem build a computer, now might not seem build a computer, now might not seem like the time, but it's going to get a like the time, but it's going to get a like the time, but it's going to get a lot worse before it gets better. So, if lot worse before it gets better. So, if lot worse before it gets better. So, if you're staring at an SSD that you've you're staring at an SSD that you've you're staring at an SSD that you've wanted for a while or some RAM or a GPU, wanted for a while or some RAM or a GPU, wanted for a while or some RAM or a GPU, and you've been holding off hoping that and you've been holding off hoping that and you've been holding off hoping that prices go down, this is not financial prices go down, this is not financial prices go down, this is not financial advice, but realistically speaking, I advice, but realistically speaking, I advice, but realistically speaking, I don't expect this stuff to get cheaper don't expect this stuff to get cheaper don't expect this stuff to get cheaper anytime soon. The demand is just too anytime soon. The demand is just too anytime soon. The demand is just too insane in the world that's changing insane in the world that's changing insane in the world that's changing around us. I don't think our phones are around us. I don't think our phones are around us. I don't think our phones are going to keep getting faster and more going to keep getting faster and more going to keep getting faster and more powerful. I think they're going to rely powerful. I think they're going to rely powerful. I think they're going to rely on the cloud more and more as the on the cloud more and more as the on the cloud more and more as the compute that we use every day gets compute that we use every day gets compute that we use every day gets centralized in these big players hands.

centralized in these big players hands. centralized in these big players hands. There is some definite there's a lot to There is some definite there's a lot to There is some definite there's a lot to be scared of here. And I hope that I'm be scared of here. And I hope that I'm be scared of here. And I hope that I'm not the world's going to look very different the world's going to look very different than it did when I was a kid building than it did when I was a kid building than it did when I was a kid building computers for my neighbors. And I don't computers for my neighbors. And I don't computers for my neighbors. And I don't know if I like that. I'm just trying to know if I like that. I'm just trying to know if I like that. I'm just trying to do my best to share where things are do my best to share where things are do my best to share where things are going so we can all understand and have going so we can all understand and have going so we can all understand and have better conversations about it. Let me better conversations about it. Let me better conversations about it. Let me know how y'all feel about the compute know how y'all feel about the compute know how y'all feel about the compute crisis and if you think I'm overblowing crisis and if you think I'm overblowing crisis and if you think I'm overblowing it. And until next time, peace nerds it. And until next time, peace nerds it. And until next time, peace nerds kill. Hey, thank the subs I missed when we Hey, thank the subs I missed when we were doing that. Thank you, Menchium, were doing that. Thank you, Menchium, were doing that. Thank you, Menchium, for the nine months of support. Hey, for the nine months of support. Hey, for the nine months of support. Hey, Bales for the five bomb. Super generous.

Bales for the five bomb. Super generous. Bales for the five bomb. Super generous. Awesome chicken ate with the prime. Awesome chicken ate with the prime. Awesome chicken ate with the prime. Aiden with the 19 months. Aiden with the 19 months. Aiden with the 19 months. Good [ __ ] It is now 7:00. We got three Good [ __ ] It is now 7:00. We got three Good [ __ ] It is now 7:00. We got three videos. I was hoping to get a bit more, videos. I was hoping to get a bit more, videos. I was hoping to get a bit more, but I kind of planned to not. I am going to go do other things. I I am going to go do other things. I should be live on Sunday. I'd be should be live on Sunday. I'd be should be live on Sunday. I'd be surprised if I wasn't. I have a lot of other things I want to I have a lot of other things I want to talk about. And I have a bad feeling talk about. And I have a bad feeling talk about. And I have a bad feeling this week's going to be a big week, too. Who are we rating? Wan show. Are they Who are we rating? Wan show. Are they still going Let's raid W show. That'll be fun. Have Let's raid W show. That'll be fun. Have fun, nerds. There's anything on YouTube? fun, nerds. There's anything on YouTube? fun, nerds. There's anything on YouTube? Nope. Cool. Thank you all. Nope. Cool. Thank you all. Nope. Cool. Thank you all. Bye, nerds. Have fun with Linus.