Language Model Builder with Felix Rieseberg

Yeah, exactly. So everything sort of Yeah, exactly. So everything sort of flowed from okay, what can we how do we flowed from okay, what can we how do we flowed from okay, what can we how do we get to somewhat work in sentences in one get to somewhat work in sentences in one get to somewhat work in sentences in one day like I tried it yesterday. I trained day like I tried it yesterday. I trained day like I tried it yesterday. I trained another small model just for fun on the another small model just for fun on the another small model just for fun on the latest version and I only gave it a day latest version and I only gave it a day latest version and I only gave it a day and it's saying really cute things but and it's saying really cute things but and it's saying really cute things but it's like I ask it what is the sun and it's like I ask it what is the sun and it's like I ask it what is the sun and it's like oh the sun is a moon at the it's like oh the sun is a moon at the it's like oh the sun is a moon at the bottom of the earth which is a real bottom of the earth which is a real bottom of the earth which is a real sentence. sentence. sentence. >> That's actually kind of like is that >> That's actually kind of like is that >> That's actually kind of like is that deep or is that just dumb? It's just deep or is that just dumb? It's just deep or is that just dumb? It's just dumb. Um, [laughter] dumb. Um, [laughter] dumb. Um, [laughter] >> that's something a thing a two-year-old >> that's something a thing a two-year-old >> that's something a thing a two-year-old would say. It's like that's adorable. would say. It's like that's adorable. would say. It's like that's adorable. >> Well, like the cool thing is that it's >> Well, like the cool thing is that it's >> Well, like the cool thing is that it's like even understanding that I'm asking like even understanding that I'm asking like even understanding that I'm asking about the sun and it's giving me like about the sun and it's giving me like about the sun and it's giving me like even like the fact that understands that even like the fact that understands that even like the fact that understands that I'm asking it a question, it should I'm asking it a question, it should I'm asking it a question, it should answer, right? answer, right? answer, right? >> Hey friends, I'm Scott Hansselman and >> Hey friends, I'm Scott Hansselman and >> Hey friends, I'm Scott Hansselman and it's another episode of Hansel Minutes. it's another episode of Hansel Minutes. it's another episode of Hansel Minutes. Today I have the pleasure of chatting Today I have the pleasure of chatting Today I have the pleasure of chatting with Felix Reeberg. He has a long with Felix Reeberg. He has a long with Felix Reeberg. He has a long history of building delightful developer history of building delightful developer history of building delightful developer tools, tools that you have used. He's tools, tools that you have used. He's tools, tools that you have used. He's worked at Slack. He worked on Electron. worked at Slack. He worked on Electron. worked at Slack. He worked on Electron. He worked at Microsoft. And his recent He worked at Microsoft. And his recent He worked at Microsoft. And his recent side projects have become very popular side projects have become very popular side projects have become very popular because they make AI feel approachable because they make AI feel approachable because they make AI feel approachable rather than just chasing benchmark rather than just chasing benchmark rather than just chasing benchmark scores. And his latest language model scores. And his latest language model scores. And his latest language model builder is in that tradition. It's builder is in that tradition. It's builder is in that tradition. It's basically an educational app that lets basically an educational app that lets basically an educational app that lets you build and understand language models you build and understand language models you build and understand language models themselves. How are you, sir?

themselves. How are you, sir? themselves. How are you, sir? >> I am doing well. Thank you, Scott. >> I am doing well. Thank you, Scott. >> I am doing well. Thank you, Scott. Thanks for having me on. Always wanted Thanks for having me on. Always wanted Thanks for having me on. Always wanted to come to the show. to come to the show. to come to the show. >> I appreciate that. that should have had >> I appreciate that. that should have had >> I appreciate that. that should have had you on many many years ago, you know, you on many many years ago, you know, you on many many years ago, you know, because we worked together for a time. I because we worked together for a time. I because we worked together for a time. I always loved the stuff that you do. You always loved the stuff that you do. You always loved the stuff that you do. You sometimes you'll just do weird sometimes you'll just do weird sometimes you'll just do weird delightful stuff. You just like, "Oh, delightful stuff. You just like, "Oh, delightful stuff. You just like, "Oh, look. Here's Clippy. Here's Windows 95 look. Here's Clippy. Here's Windows 95 look. Here's Clippy. Here's Windows 95 running in a browser." Like, why? running in a browser." Like, why? running in a browser." Like, why? Because it's delightful for no other Because it's delightful for no other Because it's delightful for no other reason. But, you know, you work uh reason. But, you know, you work uh reason. But, you know, you work uh you've been working in open source you've been working in open source you've been working in open source forever. You worked at Microsoft as an forever. You worked at Microsoft as an forever. You worked at Microsoft as an open source engineer. Um you tried to open source engineer. Um you tried to open source engineer. Um you tried to make people's lives better by putting make people's lives better by putting make people's lives better by putting things on GitHub. So, you always come at things on GitHub. So, you always come at things on GitHub. So, you always come at things from a humanistic perspective, things from a humanistic perspective, things from a humanistic perspective, which I appreciate. Now, you do in your which I appreciate. Now, you do in your which I appreciate. Now, you do in your day job work at Anthropic, but this this day job work at Anthropic, but this this day job work at Anthropic, but this this podcast is not about Anthropic. This is podcast is not about Anthropic. This is podcast is not about Anthropic. This is about you and the cool stuff that you're about you and the cool stuff that you're about you and the cool stuff that you're building. But we do want to shout out building. But we do want to shout out building. But we do want to shout out that you do lead engineering for cloud that you do lead engineering for cloud that you do lead engineering for cloud AI, for co-work, and for for Code AI, for co-work, and for for Code AI, for co-work, and for for Code Desktop. So, maybe we'll have another Desktop. So, maybe we'll have another Desktop. So, maybe we'll have another show another day, but this is about show another day, but this is about show another day, but this is about something else. something else. something else. There's a guy named Ben Eer who makes a There's a guy named Ben Eer who makes a There's a guy named Ben Eer who makes a 6502 6502 6502 create your own processor from scratch.

create your own processor from scratch. create your own processor from scratch. >> Yeah. >> Yeah. >> Yeah. >> Okay. And he's like, "Hey, let's go and >> Okay. And he's like, "Hey, let's go and >> Okay. And he's like, "Hey, let's go and get these like we're going to build it get these like we're going to build it get these like we're going to build it from scratch. Like we're going to get from scratch. Like we're going to get from scratch. Like we're going to get two rocks and we're going to spark them two rocks and we're going to spark them two rocks and we're going to spark them together. We're going to get 7400 series together. We're going to get 7400 series together. We're going to get 7400 series parts. We're going to solder it and at parts. We're going to solder it and at parts. We're going to solder it and at the end of this crazy experiment, you the end of this crazy experiment, you the end of this crazy experiment, you will have a computer that can say hello will have a computer that can say hello will have a computer that can say hello world." world." world." Yeah, that sounds Yeah, that sounds Yeah, that sounds >> I feel like you did that. I feel like >> I feel like you did that. I feel like >> I feel like you did that. I feel like you did that with language model you did that with language model you did that with language model builder. builder. builder. >> That's like the nicest thing anyone has >> That's like the nicest thing anyone has >> That's like the nicest thing anyone has said to me like this week. I think I said to me like this week. I think I said to me like this week. I think I think the I think that's sort of the think the I think that's sort of the think the I think that's sort of the goal, right? Like I looked at language goal, right? Like I looked at language goal, right? Like I looked at language model builder because I I I've been I've model builder because I I I've been I've model builder because I I I've been I've been in AI for like a little bit now. been in AI for like a little bit now. been in AI for like a little bit now. And you mentioned that we were at And you mentioned that we were at And you mentioned that we were at Microsoft together. I don't know if you Microsoft together. I don't know if you Microsoft together. I don't know if you remember project Oxford like from like remember project Oxford like from like remember project Oxford like from like >> Oh yeah. >> Oh yeah. >> Oh yeah. >> Yeah. There was like an engram model in >> Yeah. There was like an engram model in >> Yeah. There was like an engram model in there and back then I was super junior there and back then I was super junior there and back then I was super junior at Microsoft but I made like the npm at Microsoft but I made like the npm at Microsoft but I made like the npm modules for that engram model and you modules for that engram model and you modules for that engram model and you could give it like worldwide it would could give it like worldwide it would could give it like worldwide it would think for five minutes and it would come think for five minutes and it would come think for five minutes and it would come back with like ah worldwide web maybe back with like ah worldwide web maybe back with like ah worldwide web maybe that's like a good completion of like that's like a good completion of like that's like a good completion of like what you just gave me.

what you just gave me. what you just gave me. >> Yeah. And um obviously like AI has moved >> Yeah. And um obviously like AI has moved >> Yeah. And um obviously like AI has moved pretty dramatically but I find myself pretty dramatically but I find myself pretty dramatically but I find myself like explaining the thing to people a like explaining the thing to people a like explaining the thing to people a lot. And I think one thing that is cool lot. And I think one thing that is cool lot. And I think one thing that is cool about AI is that the for me the about AI is that the for me the about AI is that the for me the perceived delta between how I don't want perceived delta between how I don't want perceived delta between how I don't want to say simple but the core ideas I think to say simple but the core ideas I think to say simple but the core ideas I think are actually quite simple and very are actually quite simple and very are actually quite simple and very understandable understandable understandable >> to how magical they are is like so big >> to how magical they are is like so big >> to how magical they are is like so big that I often find myself sitting sitting that I often find myself sitting sitting that I often find myself sitting sitting people down be like all right I'm going people down be like all right I'm going people down be like all right I'm going to like let's take a napkin I'm going to to like let's take a napkin I'm going to to like let's take a napkin I'm going to show you how a language model works and show you how a language model works and show you how a language model works and I think for me at least I think for me at least I think for me at least that that feels so much easier than that that feels so much easier than that that feels so much easier than explaining to someone how like a modern explaining to someone how like a modern explaining to someone how like a modern chip works. chip works. chip works. I had no idea how modern chip works. To I had no idea how modern chip works. To I had no idea how modern chip works. To me, it's like all black magic, right? me, it's like all black magic, right? me, it's like all black magic, right? Like where like sometimes I look at the Like where like sometimes I look at the Like where like sometimes I look at the nanometer diagrams and I'm like, this is nanometer diagrams and I'm like, this is nanometer diagrams and I'm like, this is this seems hard to explain. this seems hard to explain. this seems hard to explain. >> There was a easy to explain. >> There was a easy to explain. >> There was a easy to explain. >> There was a cartoon from a guy called it >> There was a cartoon from a guy called it >> There was a cartoon from a guy called it was called the Far Side. It was like a was called the Far Side. It was like a was called the Far Side. It was like a cartoon that they would happen on this cartoon that they would happen on this cartoon that they would happen on this Sunday newspaper and there's a two Sunday newspaper and there's a two Sunday newspaper and there's a two scientists sitting in front of a scientists sitting in front of a scientists sitting in front of a whiteboard and they've got a bunch of whiteboard and they've got a bunch of whiteboard and they've got a bunch of complicated uh calculations and math complicated uh calculations and math complicated uh calculations and math symbols and then a little sparkle in the symbols and then a little sparkle in the symbols and then a little sparkle in the middle. It looks a lot like the clawed middle. It looks a lot like the clawed middle. It looks a lot like the clawed sparkle. And then it says a miracle. And sparkle. And then it says a miracle. And sparkle. And then it says a miracle. And then on the other on the right of it, then on the other on the right of it, then on the other on the right of it, there's like a bunch of other there's like a bunch of other there's like a bunch of other complicated, you know, math. And the complicated, you know, math. And the complicated, you know, math. And the other scientist says, I think you need other scientist says, I think you need other scientist says, I think you need to be a little more specific here in to be a little more specific here in to be a little more specific here in step two.

step two. step two. I feel like that's what's going on with I feel like that's what's going on with I feel like that's what's going on with language models right now is that like language models right now is that like language models right now is that like we can understand the number 10. We can we can understand the number 10. We can we can understand the number 10. We can then times that by 10 and then times then times that by 10 and then times then times that by 10 and then times that by 10. But then you add a couple of that by 10. But then you add a couple of that by 10. But then you add a couple of orders of magnitude and then we just go orders of magnitude and then we just go orders of magnitude and then we just go and then a miracle happens. and then a miracle happens. and then a miracle happens. >> Yeah. >> Yeah. >> Yeah. >> And now the the magic word guesser feels >> And now the the magic word guesser feels >> And now the the magic word guesser feels like it's a person. like it's a person. like it's a person. >> Yeah. Yeah. Exactly. And I think it's >> Yeah. Yeah. Exactly. And I think it's >> Yeah. Yeah. Exactly. And I think it's like so cool especially I think you like so cool especially I think you like so cool especially I think you mentioned my day job. Like in my day job mentioned my day job. Like in my day job mentioned my day job. Like in my day job I get the privilege of like I get the privilege of like I get the privilege of like you know seeing baby clot in the oven you know seeing baby clot in the oven you know seeing baby clot in the oven and like just seeing it evolve and grow. and like just seeing it evolve and grow. and like just seeing it evolve and grow. And you get the the the joy of watching And you get the the the joy of watching And you get the the the joy of watching something just like put together random something just like put together random something just like put together random words and like you get these like words and like you get these like words and like you get these like complete word salads and you get like complete word salads and you get like complete word salads and you get like this complete noise of characters back this complete noise of characters back this complete noise of characters back and like watching that in sampling go and like watching that in sampling go and like watching that in sampling go from just you know face planting onto from just you know face planting onto from just you know face planting onto your keyboard to like oh I recognize your keyboard to like oh I recognize your keyboard to like oh I recognize some of those sounds to like oh now a some of those sounds to like oh now a some of those sounds to like oh now a sentence is happening is pretty cool. sentence is happening is pretty cool. sentence is happening is pretty cool. It's like very fun. It's like very fun It's like very fun. It's like very fun It's like very fun. It's like very fun to like talk to something that you made to like talk to something that you made to like talk to something that you made >> even if it's just a model. It's just >> even if it's just a model. It's just >> even if it's just a model. It's just like a deeply enjoyable experience.

like a deeply enjoyable experience. like a deeply enjoyable experience. >> But is there isn't there a leap? Isn't >> But is there isn't there a leap? Isn't >> But is there isn't there a leap? Isn't there a moment there where it's just there a moment there where it's just there a moment there where it's just like that went from cool cool next token like that went from cool cool next token like that went from cool cool next token guesser to this feels smarter than a guesser to this feels smarter than a guesser to this feels smarter than a parrot? parrot? parrot? That's the part I don't understand. How That's the part I don't understand. How That's the part I don't understand. How do we get from word guessesser to parrot do we get from word guessesser to parrot do we get from word guessesser to parrot to people are talking to this thing like to people are talking to this thing like to people are talking to this thing like it's their therapist, which is it's their therapist, which is it's their therapist, which is concerning. concerning. concerning. >> Yeah. I think I think there you go into >> Yeah. I think I think there you go into >> Yeah. I think I think there you go into these like deeply these like deeply these like deeply God. But I got to be I got to be fair God. But I got to be I got to be fair God. But I got to be I got to be fair careful not to be too annoying in this. careful not to be too annoying in this. careful not to be too annoying in this. I I came to computing fairly late. I I I came to computing fairly late. I I I came to computing fairly late. I studied poetry. I at some point studied poetry. I at some point studied poetry. I at some point discovered that the poetry factory was discovered that the poetry factory was discovered that the poetry factory was not hiring and like needed to make money not hiring and like needed to make money not hiring and like needed to make money some other way. And then I guess now I'm some other way. And then I guess now I'm some other way. And then I guess now I'm building software. But I think language building software. But I think language building software. But I think language models give us such an interesting models give us such an interesting models give us such an interesting opportunity to talk about all of these opportunity to talk about all of these opportunity to talk about all of these big concepts, right? Like we already big concepts, right? Like we already big concepts, right? Like we already have these big conversations about moral have these big conversations about moral have these big conversations about moral patient. Like at what point do we even patient. Like at what point do we even patient. Like at what point do we even consider seriously whether or not a consider seriously whether or not a consider seriously whether or not a mortal gets moral pat? But it also gives mortal gets moral pat? But it also gives mortal gets moral pat? But it also gives us these really interesting ideas about us these really interesting ideas about us these really interesting ideas about like you know like what does it mean for like you know like what does it mean for like you know like what does it mean for something to be intelligent? Like something to be intelligent? Like something to be intelligent? Like clearly we know how it works.

clearly we know how it works. clearly we know how it works. If you get meaningful therapeutic If you get meaningful therapeutic If you get meaningful therapeutic insight, what what do you do with that, insight, what what do you do with that, insight, what what do you do with that, right? Like and I think the questions right? Like and I think the questions right? Like and I think the questions there are probably much bigger than like there are probably much bigger than like there are probably much bigger than like how does the language model work? Um and how does the language model work? Um and how does the language model work? Um and they're more about like are we all just they're more about like are we all just they're more about like are we all just parrots? Like does it where does it parrots? Like does it where does it parrots? Like does it where does it break down? Right. break down? Right. break down? Right. >> Yeah. Yeah. I I try to explain context >> Yeah. Yeah. I I try to explain context >> Yeah. Yeah. I I try to explain context windows to people and the sample the the windows to people and the sample the the windows to people and the sample the the example I always give my talks is it's a example I always give my talks is it's a example I always give my talks is it's a beautiful day let's go to the and then beautiful day let's go to the and then beautiful day let's go to the and then the model always picks park or beach the model always picks park or beach the model always picks park or beach >> and then you give it a little context >> and then you give it a little context >> and then you give it a little context you tell it you're in London it'll say you tell it you're in London it'll say you tell it you're in London it'll say let's go to the London Eye you tell it let's go to the London Eye you tell it let's go to the London Eye you tell it you're in Berlin let's go to here you you're in Berlin let's go to here you you're in Berlin let's go to here you tell it you're in Portland let's go tell it you're in Portland let's go tell it you're in Portland let's go there and the more context the better there and the more context the better there and the more context the better the word guessing gets and then you say the word guessing gets and then you say the word guessing gets and then you say well you're Scott Hansselman and you've well you're Scott Hansselman and you've well you're Scott Hansselman and you've been married for 25 years it's a been married for 25 years it's a been married for 25 years it's a Saturday afternoon and the answer is Saturday afternoon and the answer is Saturday afternoon and the answer is that we'll go to the bagel shop because that we'll go to the bagel shop because that we'll go to the bagel shop because that's where my wife and I go for the that's where my wife and I go for the that's where my wife and I go for the last 25 years. last 25 years. last 25 years. >> Oh, beautiful. >> Oh, beautiful. >> Oh, beautiful. >> That's our Saturday tradition. So then >> That's our Saturday tradition. So then >> That's our Saturday tradition. So then the question is, am I just a big corpus the question is, am I just a big corpus the question is, am I just a big corpus at 25 years of her putting up with me at 25 years of her putting up with me at 25 years of her putting up with me and she knows to guess that because and she knows to guess that because and she knows to guess that because she's the word guesser for my she's the word guesser for my she's the word guesser for my personality and am I just making stuff personality and am I just making stuff personality and am I just making stuff up?

up? up? >> Yeah. >> Yeah. >> Yeah. >> Or is there is there is there free will >> Or is there is there is there free will >> Or is there is there is there free will and then it's a whole thing? and then it's a whole thing? and then it's a whole thing? >> Yeah. Yeah. Like you very quickly you >> Yeah. Yeah. Like you very quickly you >> Yeah. Yeah. Like you very quickly you very quickly very quickly very quickly or I at least this is like a very or I at least this is like a very or I at least this is like a very personal experience but I think it's personal experience but I think it's personal experience but I think it's like part of the experience that had me like part of the experience that had me like part of the experience that had me built this language model builder thing built this language model builder thing built this language model builder thing but like when I stare at the model or but like when I stare at the model or but like when I stare at the model or like the big models I sometimes get the like the big models I sometimes get the like the big models I sometimes get the same feeling that I get when I like same feeling that I get when I like same feeling that I get when I like stare at the campfire or like a forest stare at the campfire or like a forest stare at the campfire or like a forest or like the stars. or like the stars. or like the stars. >> Yeah. >> Yeah. >> Yeah. >> Right. like this notion of I think >> Right. like this notion of I think >> Right. like this notion of I think people have different experiences but people have different experiences but people have different experiences but like the experience I have when I look like the experience I have when I look like the experience I have when I look at when I go to a planetaria I love the at when I go to a planetaria I love the at when I go to a planetaria I love the planetarium right I just like sit down planetarium right I just like sit down planetarium right I just like sit down and I feel so small and insignificant and I feel so small and insignificant and I feel so small and insignificant and it really makes me consider the and it really makes me consider the and it really makes me consider the uniqueness of the human experience and uniqueness of the human experience and uniqueness of the human experience and like honestly the big language models like honestly the big language models like honestly the big language models when I look at when I look at this like when I look at when I look at this like when I look at when I look at this like statistical like just math right like statistical like just math right like statistical like just math right like when we look at a language model what I when we look at a language model what I when we look at a language model what I mean here is I'm like less excited maybe mean here is I'm like less excited maybe mean here is I'm like less excited maybe about a particular application or like a about a particular application or like a about a particular application or like a particularly trained model but like if I particularly trained model but like if I particularly trained model but like if I look at a model as the core idea of we look at a model as the core idea of we look at a model as the core idea of we have math and I'm talking to the math. have math and I'm talking to the math. have math and I'm talking to the math. >> I sort of feel like the universe is like >> I sort of feel like the universe is like >> I sort of feel like the universe is like staring back at me and that is a very staring back at me and that is a very staring back at me and that is a very cool experience because it makes me cool experience because it makes me cool experience because it makes me reconsider myself and like how do I put reconsider myself and like how do I put reconsider myself and like how do I put words together, you know?

words together, you know? words together, you know? >> Right. >> Right. >> Right. >> Well, language model builder is at >> Well, language model builder is at >> Well, language model builder is at languagemodelbuilder.com. You can just languagemodelbuilder.com. You can just languagemodelbuilder.com. You can just download it now. Right now it only works download it now. Right now it only works download it now. Right now it only works on a Mac and it expects Apple Silicon so on a Mac and it expects Apple Silicon so on a Mac and it expects Apple Silicon so be aware. And you you start with a kind be aware. And you you start with a kind be aware. And you you start with a kind of an interactive book, you know, it's of an interactive book, you know, it's of an interactive book, you know, it's it's like a it's more than a PDF. You're it's like a it's more than a PDF. You're it's like a it's more than a PDF. You're every thing can be clicked on. every thing can be clicked on. every thing can be clicked on. Everything can be changed. Every uh it's Everything can be changed. Every uh it's Everything can be changed. Every uh it's kind of like the best of the New York kind of like the best of the New York kind of like the best of the New York Times when you go to one of those New Times when you go to one of those New Times when you go to one of those New York Times infographics and you're like, York Times infographics and you're like, York Times infographics and you're like, "Oh man, that's not a GIF. I can click "Oh man, that's not a GIF. I can click "Oh man, that's not a GIF. I can click on that. Holy crap, that's awesome." on that. Holy crap, that's awesome." on that. Holy crap, that's awesome." >> I'm going to I'm going to print some of >> I'm going to I'm going to print some of >> I'm going to I'm going to print some of those quotes and just like hang them in those quotes and just like hang them in those quotes and just like hang them in my office. my office. my office. >> Oh, you can take it all. This my show is >> Oh, you can take it all. This my show is >> Oh, you can take it all. This my show is your show, my friend. So, if you want me your show, my friend. So, if you want me your show, my friend. So, if you want me to blurb blurb your product, you can to blurb blurb your product, you can to blurb blurb your product, you can blur me as much as you want. Now, I want blur me as much as you want. Now, I want blur me as much as you want. Now, I want to call out though when you get to the to call out though when you get to the to call out though when you get to the tokenizer part and you say, tokenizer part and you say, tokenizer part and you say, >> "I'm going to write my own tokenizer >> "I'm going to write my own tokenizer >> "I'm going to write my own tokenizer bite level tokenizer," you might just bite level tokenizer," you might just bite level tokenizer," you might just have a vocabulary of 10,000 words. So, have a vocabulary of 10,000 words. So, have a vocabulary of 10,000 words. So, that might be like a baby, right? that might be like a baby, right? that might be like a baby, right? >> Yeah. >> Yeah. >> Yeah. >> And then you can make like a GPT2. And I >> And then you can make like a GPT2. And I >> And then you can make like a GPT2. And I think it's important for people to think it's important for people to think it's important for people to remember that like GPT5, there were four remember that like GPT5, there were four remember that like GPT5, there were four others before that and they did stuff, others before that and they did stuff, others before that and they did stuff, right? So you're basically putting right? So you're basically putting right? So you're basically putting people just before the bendy part of the people just before the bendy part of the people just before the bendy part of the hockey stick of the graph, right? You're hockey stick of the graph, right? You're hockey stick of the graph, right? You're saying right before everything exploded, saying right before everything exploded, saying right before everything exploded, this is the size of a model we can make this is the size of a model we can make this is the size of a model we can make where the human mind can kind of still where the human mind can kind of still where the human mind can kind of still get it.

get it. get it. >> Which is why I use that Ben eater thing >> Which is why I use that Ben eater thing >> Which is why I use that Ben eater thing as an example because a human being can as an example because a human being can as an example because a human being can completely and 100% understand a 6502 completely and 100% understand a 6502 completely and 100% understand a 6502 microprocessor, microprocessor, microprocessor, >> right? >> right? >> right? >> They cannot understand a Pentium and >> They cannot understand a Pentium and >> They cannot understand a Pentium and they there's no way they can understand they there's no way they can understand they there's no way they can understand like Apple silicon and and hold it in like Apple silicon and and hold it in like Apple silicon and and hold it in their context window. Yeah. How did their context window. Yeah. How did their context window. Yeah. How did >> how did you decide the right size for >> how did you decide the right size for >> how did you decide the right size for language model builder to be so that you language model builder to be so that you language model builder to be so that you could understand it all because I think could understand it all because I think could understand it all because I think you succeeded in doing that. you succeeded in doing that. you succeeded in doing that. >> Um I think it was probably like less >> Um I think it was probably like less >> Um I think it was probably like less strategic than it might seem like now strategic than it might seem like now strategic than it might seem like now that we look at it. But it was I was that we look at it. But it was I was that we look at it. But it was I was basically working backwards from okay my basically working backwards from okay my basically working backwards from okay my friends, my family, like what kind of friends, my family, like what kind of friends, my family, like what kind of attention span can I assume, attention span can I assume, attention span can I assume, >> right? What can we train within that >> right? What can we train within that >> right? What can we train within that time? And the tokenizer is a good time? And the tokenizer is a good time? And the tokenizer is a good example because I actually I I example because I actually I I example because I actually I I eventually settled on the like 10k eventually settled on the like 10k eventually settled on the like 10k tokenizer, the one that is like very tokenizer, the one that is like very tokenizer, the one that is like very small, but mostly because running it on small, but mostly because running it on small, but mostly because running it on my my personal computer is like an M4 my my personal computer is like an M4 my my personal computer is like an M4 Max, which I think is like a pretty good Max, which I think is like a pretty good Max, which I think is like a pretty good computer all thing all things computer all thing all things computer all thing all things considered. considered. considered. And with even with later later And with even with later later And with even with later later tokenizers, you spend so much time just tokenizers, you spend so much time just tokenizers, you spend so much time just tokenizing the corpus that you just like tokenizing the corpus that you just like tokenizing the corpus that you just like stare you're just like, you know, stare you're just like, you know, stare you're just like, you know, >> you're just staring at the pre-training.

>> you're just staring at the pre-training. >> you're just staring at the pre-training. Not even pre-training, just spinning its Not even pre-training, just spinning its Not even pre-training, just spinning its wheels on the tokenizer. wheels on the tokenizer. wheels on the tokenizer. >> It's just like mashing up a food for a >> It's just like mashing up a food for a >> It's just like mashing up a food for a baby. You're just sitting there mashing baby. You're just sitting there mashing baby. You're just sitting there mashing up potatoes. up potatoes. up potatoes. >> Yeah, exactly. So everything sort of >> Yeah, exactly. So everything sort of >> Yeah, exactly. So everything sort of flowed from, okay, what can we how do we flowed from, okay, what can we how do we flowed from, okay, what can we how do we get to somewhat work in sentences in one get to somewhat work in sentences in one get to somewhat work in sentences in one day? Like I tried it. Yesterday I day? Like I tried it. Yesterday I day? Like I tried it. Yesterday I trained another small model just for fun trained another small model just for fun trained another small model just for fun on the latest version and I only gave it on the latest version and I only gave it on the latest version and I only gave it a day and it's saying really cute things a day and it's saying really cute things a day and it's saying really cute things but it's like I ask it what is the sun but it's like I ask it what is the sun but it's like I ask it what is the sun and it's like oh the sun is a moon at and it's like oh the sun is a moon at and it's like oh the sun is a moon at the bottom of the earth which it is a the bottom of the earth which it is a the bottom of the earth which it is a real sentence. real sentence. real sentence. >> That's actually kind of like >> That's actually kind of like >> That's actually kind of like [clears throat] is that deep or is that [clears throat] is that deep or is that [clears throat] is that deep or is that just dumb? just dumb? just dumb? >> It's just dumb. Um [laughter] >> It's just dumb. Um [laughter] >> It's just dumb. Um [laughter] >> that sounds like a thing a 2-year-old >> that sounds like a thing a 2-year-old >> that sounds like a thing a 2-year-old would say. It's like that's adorable. would say. It's like that's adorable. would say. It's like that's adorable. Well, like the cool thing is that it's Well, like the cool thing is that it's Well, like the cool thing is that it's like even understanding that I'm asking like even understanding that I'm asking like even understanding that I'm asking about the sun and it's giving me like about the sun and it's giving me like about the sun and it's giving me like even like the fact that it understands even like the fact that it understands even like the fact that it understands that I'm asking it a question it should that I'm asking it a question it should that I'm asking it a question it should answer, right? Like that alone is like answer, right? Like that alone is like answer, right? Like that alone is like >> or that it's a round thing in the sky >> or that it's a round thing in the sky >> or that it's a round thing in the sky somewhere like it's got somewhere like it's got somewhere like it's got >> it the the the the vector space is at >> it the the the the vector space is at >> it the the the the vector space is at least has some amount of proximity where least has some amount of proximity where least has some amount of proximity where it's like okay moon and sun are near it's like okay moon and sun are near it's like okay moon and sun are near each other in the space.

each other in the space. each other in the space. >> Yeah. And then also the other thing that >> Yeah. And then also the other thing that >> Yeah. And then also the other thing that happened is like I I sort of lowered happened is like I I sort of lowered happened is like I I sort of lowered like three different versions of this like three different versions of this like three different versions of this app. The first version was just the app. The first version was just the app. The first version was just the textbook and what I kind of wanted was textbook and what I kind of wanted was textbook and what I kind of wanted was you mentioned these like playgrounds but you mentioned these like playgrounds but you mentioned these like playgrounds but there's like the little playgrounds of there's like the little playgrounds of there's like the little playgrounds of like okay let's give me your corpus like okay let's give me your corpus like okay let's give me your corpus we'll make a tokenizer together and the we'll make a tokenizer together and the we'll make a tokenizer together and the initial idea I had was oh at the end of initial idea I had was oh at the end of initial idea I had was oh at the end of that textbook you have a model that textbook you have a model that textbook you have a model >> but then very quickly I found myself >> but then very quickly I found myself >> but then very quickly I found myself like wanting just a little bit of like wanting just a little bit of like wanting just a little bit of innovation that we found in the industry innovation that we found in the industry innovation that we found in the industry um so at some point I made the call to um so at some point I made the call to um so at some point I made the call to like separate out the playgrounds and like separate out the playgrounds and like separate out the playgrounds and say okay we're going to learn the say okay we're going to learn the say okay we're going to learn the foundations and then we're going to foundations and then we're going to foundations and then we're going to build a build a build a Um, but I didn't want the gap to be like Um, but I didn't want the gap to be like Um, but I didn't want the gap to be like too big, right? Just so that what you're too big, right? Just so that what you're too big, right? Just so that what you're learning is still applicable. Guess learning is still applicable. Guess learning is still applicable. Guess you're right. Like tokenizers is a great you're right. Like tokenizers is a great you're right. Like tokenizers is a great example like the modern tokenizers. Um, example like the modern tokenizers. Um, example like the modern tokenizers. Um, they're just a little there's just a they're just a little there's just a they're just a little there's just a little too too much magical science little too too much magical science little too too much magical science going on for a human to quickly going on for a human to quickly going on for a human to quickly understand. understand. understand. >> I I I have an analogy that I used that >> I I I have an analogy that I used that >> I I I have an analogy that I used that everyone is sick of hearing, but like everyone is sick of hearing, but like everyone is sick of hearing, but like learning to drive stick shift, learning to drive stick shift, learning to drive stick shift, manual shift on a car changes your manual shift on a car changes your manual shift on a car changes your relationship with the vehicle.

relationship with the vehicle. relationship with the vehicle. >> Yeah. language model builder >> Yeah. language model builder >> Yeah. language model builder would by its nature change your would by its nature change your would by its nature change your relationship with a chatbot or with a relationship with a chatbot or with a relationship with a chatbot or with a large language model because now large language model because now large language model because now suddenly you're like before we get to suddenly you're like before we get to suddenly you're like before we get to talk to the bot we're going to take talk to the bot we're going to take talk to the bot we're going to take apart this car and we're going to talk apart this car and we're going to talk apart this car and we're going to talk about the internal combustion engine and about the internal combustion engine and about the internal combustion engine and the history of that the history of that the history of that you don't go too deep into the history you don't go too deep into the history you don't go too deep into the history you don't talk about like 40 years ago you don't talk about like 40 years ago you don't talk about like 40 years ago Eliza you know how did you balance Eliza you know how did you balance Eliza you know how did you balance there's the pure math there you just there's the pure math there you just there's the pure math there you just kind of like assume transformers are a kind of like assume transformers are a kind of like assume transformers are a thing but you don't really explain thing but you don't really explain thing but you don't really explain where it came from the last 50 years. where it came from the last 50 years. where it came from the last 50 years. Did you think about historical context Did you think about historical context Did you think about historical context versus mathematical context versus just versus mathematical context versus just versus mathematical context versus just explaining the moment that we're in explaining the moment that we're in explaining the moment that we're in today? today? today? >> Yeah, like just a little bit. I think I >> Yeah, like just a little bit. I think I >> Yeah, like just a little bit. I think I wanted to keep it very tight because wanted to keep it very tight because wanted to keep it very tight because um on the website I wrote the sentence um on the website I wrote the sentence um on the website I wrote the sentence that I think is very clever. I wrote no that I think is very clever. I wrote no that I think is very clever. I wrote no artificial additives by which I artificial additives by which I artificial additives by which I basically mean obviously you can sit basically mean obviously you can sit basically mean obviously you can sit down and you can say Claude write me a down and you can say Claude write me a down and you can say Claude write me a five volume book about the history of AI five volume book about the history of AI five volume book about the history of AI and it would do it for you. So I wanted and it would do it for you. So I wanted and it would do it for you. So I wanted to stay like very true to the things to stay like very true to the things to stay like very true to the things that I like if if I sit down with a that I like if if I sit down with a that I like if if I sit down with a smart friend right like maybe like in smart friend right like maybe like in smart friend right like maybe like in medicine or like law or something I sit medicine or like law or something I sit medicine or like law or something I sit down with a smart friend and they're down with a smart friend and they're down with a smart friend and they're like hey can you explain to me how like hey can you explain to me how like hey can you explain to me how language models work and we have one language models work and we have one language models work and we have one dinner like what what would I put in dinner like what what would I put in dinner like what what would I put in that one dinner? So I gave myself this that one dinner? So I gave myself this that one dinner? So I gave myself this like somewhat artificial limit of 90 like somewhat artificial limit of 90 like somewhat artificial limit of 90 minutes and with another 90 minutes you minutes and with another 90 minutes you minutes and with another 90 minutes you can touch on just enough history to like can touch on just enough history to like can touch on just enough history to like understand the current context right understand the current context right understand the current context right like why do we give models a temperature like why do we give models a temperature like why do we give models a temperature where does that come from what is a hot

where does that come from what is a hot where does that come from what is a hot model what is a cold model so we go into model what is a cold model so we go into model what is a cold model so we go into that we also briefly go into that we also briefly go into that we also briefly go into like what happened in 2017 like what is like what happened in 2017 like what is like what happened in 2017 like what is going on with like attention is all you going on with like attention is all you going on with like attention is all you need like it's famous enough as a paper need like it's famous enough as a paper need like it's famous enough as a paper that people have heard about it that people have heard about it that people have heard about it >> but I'm I'm not spending too much time >> but I'm I'm not spending too much time >> but I'm I'm not spending too much time on like um statistical models of on like um statistical models of on like um statistical models of language. Um language. Um language. Um >> I think it's a super interesting >> I think it's a super interesting >> I think it's a super interesting history. I really enjoy like >> let me figure out how to put this. I >> let me figure out how to put this. I really enjoy I never studied linguistics really enjoy I never studied linguistics really enjoy I never studied linguistics but I hung out with a lot of like but I hung out with a lot of like but I hung out with a lot of like linguists and like computational linguists and like computational linguists and like computational linguists linguists linguists >> were like scraping at the surface of AI >> were like scraping at the surface of AI >> were like scraping at the surface of AI for like the last 40 years and like we for like the last 40 years and like we for like the last 40 years and like we didn't even know right like I studied didn't even know right like I studied didn't even know right like I studied with those people. with those people. with those people. like, like, like, >> well, it's it's the elephant problem, >> well, it's it's the elephant problem, >> well, it's it's the elephant problem, right? You're running around and you're right? You're running around and you're right? You're running around and you're like feeling it. And the one guy is like feeling it. And the one guy is like feeling it. And the one guy is like, "This elephant is this this is the like, "This elephant is this this is the like, "This elephant is this this is the part of the blind people touching an part of the blind people touching an part of the blind people touching an elephant. elephant. elephant. >> This is the this elephant is like a >> This is the this elephant is like a >> This is the this elephant is like a tree." No, no, an elephant's like a tree." No, no, an elephant's like a tree." No, no, an elephant's like a snake, but no one sees the big picture. snake, but no one sees the big picture. snake, but no one sees the big picture. >> Yeah. Yeah. There's like so much cool >> Yeah. Yeah. There's like so much cool >> Yeah. Yeah. There's like so much cool stuff there. And I think I think some of stuff there. And I think I think some of stuff there. And I think I think some of the I'm sure we're going to get like the I'm sure we're going to get like the I'm sure we're going to get like amazing books about the history of AI at amazing books about the history of AI at amazing books about the history of AI at some point, some point, some point, >> but you don't think it's important to >> but you don't think it's important to >> but you don't think it's important to understand like that you can understand understand like that you can understand understand like that you can understand language models this way. So, how long language models this way. So, how long language models this way. So, how long is this? Is this a 5-day course? Is this is this? Is this a 5-day course? Is this is this? Is this a 5-day course? Is this a 90-minute course? How much? It can be a 90-minute course? How much? It can be a 90-minute course? How much? It can be as long as you want it to be, but would as long as you want it to be, but would as long as you want it to be, but would I give this to my 20-year-old? And would I give this to my 20-year-old? And would I give this to my 20-year-old? And would that be his introduction to Stick Shift that be his introduction to Stick Shift that be his introduction to Stick Shift and change his relationship with a and change his relationship with a and change his relationship with a chatbot? Or does this require someone to chatbot? Or does this require someone to chatbot? Or does this require someone to really have a background as a software really have a background as a software really have a background as a software engineer?

engineer? engineer? >> No. No. No background as a software >> No. No. No background as a software >> No. No. No background as a software engineer. Like, there's these little engineer. Like, there's these little engineer. Like, there's these little drawers where you can look at the code drawers where you can look at the code drawers where you can look at the code if you really want to. The code isn't if you really want to. The code isn't if you really want to. The code isn't even the real code. What I'm using is even the real code. What I'm using is even the real code. What I'm using is MLX, the like framework. That's why it's MLX, the like framework. That's why it's MLX, the like framework. That's why it's currently only running on on Macs currently only running on on Macs currently only running on on Macs because I was just lazy. Um I'll show because I was just lazy. Um I'll show because I was just lazy. Um I'll show you some Python if you're interested in you some Python if you're interested in you some Python if you're interested in like what would the Python look like? like what would the Python look like? like what would the Python look like? >> Yeah, because I wanted to know when >> Yeah, because I wanted to know when >> Yeah, because I wanted to know when you're going to get this working on you're going to get this working on you're going to get this working on Windows. Windows. Windows. >> Yeah. I mean honestly at this point I >> Yeah. I mean honestly at this point I >> Yeah. I mean honestly at this point I can probably just like have an agent do can probably just like have an agent do can probably just like have an agent do it, right? It like shouldn't be hard. We it, right? It like shouldn't be hard. We it, right? It like shouldn't be hard. We can just do it in Python. can just do it in Python. can just do it in Python. >> Well, but then you got to get it to work >> Well, but then you got to get it to work >> Well, but then you got to get it to work on ARM and x64 and WSL or not WSL. Like on ARM and x64 and WSL or not WSL. Like on ARM and x64 and WSL or not WSL. Like you have to like the combinatorics get you have to like the combinatorics get you have to like the combinatorics get complicated once you decide to make it complicated once you decide to make it complicated once you decide to make it go outside of the homogeneous Mac. go outside of the homogeneous Mac. go outside of the homogeneous Mac. >> I didn't want to I didn't do it a priori >> I didn't want to I didn't do it a priori >> I didn't want to I didn't do it a priori for like a few reasons. One is like I for like a few reasons. One is like I for like a few reasons. One is like I think I didn't want to build yet another think I didn't want to build yet another think I didn't want to build yet another Electron app. I wanted to like actually Electron app. I wanted to like actually Electron app. I wanted to like actually build a Swift app for once. Um build a Swift app for once. Um build a Swift app for once. Um >> yeah, that's a good point. You'd have to >> yeah, that's a good point. You'd have to >> yeah, that's a good point. You'd have to do it in Win UI. do it in Win UI. do it in Win UI. >> Yeah, but the other thing is like you >> Yeah, but the other thing is like you >> Yeah, but the other thing is like you you never know of like am I really only you never know of like am I really only you never know of like am I really only building this for my five friends or building this for my five friends or building this for my five friends or does anyone care? People do care which I does anyone care? People do care which I does anyone care? People do care which I think is very nice. Um and I think that think is very nice. Um and I think that think is very nice. Um and I think that increases the likelihood of it going up.

increases the likelihood of it going up. increases the likelihood of it going up. But to answer your question, if people But to answer your question, if people But to answer your question, if people have like one afternoon, you will have a have like one afternoon, you will have a have like one afternoon, you will have a model that like will be very stupid, model that like will be very stupid, model that like will be very stupid, extremely stupid, but you can make it extremely stupid, but you can make it extremely stupid, but you can make it through the whole thing in like an through the whole thing in like an through the whole thing in like an afternoon. afternoon. afternoon. >> That's really powerful. Like this is >> That's really powerful. Like this is >> That's really powerful. Like this is significant. And I feel like people need significant. And I feel like people need significant. And I feel like people need to understand that that and the at the to understand that that and the at the to understand that that and the at the root of it is this concept that it root of it is this concept that it root of it is this concept that it changes your relationship with the changes your relationship with the changes your relationship with the vehicle. vehicle. vehicle. >> Yeah. Like opening task manager, opening >> Yeah. Like opening task manager, opening >> Yeah. Like opening task manager, opening task manager on a p on a PC. Opening a task manager on a p on a PC. Opening a task manager on a p on a PC. Opening a do like I've seen dos prompt. You open a do like I've seen dos prompt. You open a do like I've seen dos prompt. You open a dos prompt like what is this? I've never dos prompt like what is this? I've never dos prompt like what is this? I've never seen this before. I've been a Windows seen this before. I've been a Windows seen this before. I've been a Windows user for 20 years. I've never seen well user for 20 years. I've never seen well user for 20 years. I've never seen well that's you know or hitting F12 on like that's you know or hitting F12 on like that's you know or hitting F12 on like Chrome. Chrome. Chrome. >> Yeah. >> Yeah. >> Yeah. >> Those moments are a moment where like >> Those moments are a moment where like >> Those moments are a moment where like the muggles change their relationship the muggles change their relationship the muggles change their relationship with the thing. Like I had that's like with the thing. Like I had that's like with the thing. Like I had that's like opening the trunk, opening the boot opening the trunk, opening the boot opening the trunk, opening the boot rather on the car and like look at the rather on the car and like look at the rather on the car and like look at the engine like what? There's a thing under engine like what? There's a thing under engine like what? There's a thing under there. there. there. There's two things in there that I I There's two things in there that I I There's two things in there that I I think people should try out because I think people should try out because I think people should try out because I think they're very fun to me. The first think they're very fun to me. The first think they're very fun to me. The first one is I built this like visualizer of one is I built this like visualizer of one is I built this like visualizer of what a model architecture looks like. what a model architecture looks like. what a model architecture looks like. >> Like what are the pieces in a model, >> Like what are the pieces in a model, >> Like what are the pieces in a model, right? right? right? >> The model blueprint.

>> The model blueprint. >> The model blueprint. >> Yeah. Yeah. Exactly. The like >> Yeah. Yeah. Exactly. The like >> Yeah. Yeah. Exactly. The like transformer blocks and like how does how transformer blocks and like how does how transformer blocks and like how does how does like the math flow through there. does like the math flow through there. does like the math flow through there. But the other thing I really enjoy is at But the other thing I really enjoy is at But the other thing I really enjoy is at the very end not you can also do it in the very end not you can also do it in the very end not you can also do it in sampling but it's the X-ray view. Mh. sampling but it's the X-ray view. Mh. sampling but it's the X-ray view. Mh. >> So I think a lot of people know now that >> So I think a lot of people know now that >> So I think a lot of people know now that a language model is fundamentally a a language model is fundamentally a a language model is fundamentally a statistical model and it tries to statistical model and it tries to statistical model and it tries to predict what the next token is. The next predict what the next token is. The next predict what the next token is. The next token being like you know like a word token being like you know like a word token being like you know like a word fragment and um it doesn't pick the most fragment and um it doesn't pick the most fragment and um it doesn't pick the most likely one. It sort of does like a soft likely one. It sort of does like a soft likely one. It sort of does like a soft minmax and but it it sort of like tries minmax and but it it sort of like tries minmax and but it it sort of like tries to calculate that across like all the to calculate that across like all the to calculate that across like all the text you give it and then continues. text you give it and then continues. text you give it and then continues. >> And what I've built in is this like >> And what I've built in is this like >> And what I've built in is this like X-ray view where you can see every X-ray view where you can see every X-ray view where you can see every single position what are the paths not single position what are the paths not single position what are the paths not not taken. M not taken. M not taken. M >> I call those parallel universes, >> I call those parallel universes, >> I call those parallel universes, >> all the log throbbs and it's like Yeah. >> all the log throbbs and it's like Yeah. >> all the log throbbs and it's like Yeah. And there's another parallel universe And there's another parallel universe And there's another parallel universe where it said a totally different thing. where it said a totally different thing. where it said a totally different thing. >> Yeah, maybe Scott didn't go to a bagel >> Yeah, maybe Scott didn't go to a bagel >> Yeah, maybe Scott didn't go to a bagel shop, right? Like he went like what are shop, right? Like he went like what are shop, right? Like he went like what are the other places the other places the other places >> he could have gone to? Actually, is >> he could have gone to? Actually, is >> he could have gone to? Actually, is there screen sharing here? Like there screen sharing here? Like there screen sharing here? Like >> there's no screen sharing. This is an >> there's no screen sharing. This is an >> there's no screen sharing. This is an audio show, I'm afraid, my friend.

audio show, I'm afraid, my friend. audio show, I'm afraid, my friend. >> Fair enough. But people need to try it >> Fair enough. But people need to try it >> Fair enough. But people need to try it out. Um you can do one thing maybe that out. Um you can do one thing maybe that out. Um you can do one thing maybe that is very quick. Instead of pre-training is very quick. Instead of pre-training is very quick. Instead of pre-training your own model, you can also import an your own model, you can also import an your own model, you can also import an existing base model, existing base model, existing base model, >> right? You can like go and import one of >> right? You can like go and import one of >> right? You can like go and import one of the smaller models that just runs fine the smaller models that just runs fine the smaller models that just runs fine on your computer like Quinn or on your computer like Quinn or on your computer like Quinn or something. And uh you can import that something. And uh you can import that something. And uh you can import that model, go to sampling or go to the chat model, go to sampling or go to the chat model, go to sampling or go to the chat interface, but sampling is probably interface, but sampling is probably interface, but sampling is probably better. And you can say um it's a better. And you can say um it's a better. And you can say um it's a beautiful day and Scott went to beautiful day and Scott went to beautiful day and Scott went to >> and then just see all the possibilities >> and then just see all the possibilities >> and then just see all the possibilities that the model comes up with and like that the model comes up with and like that the model comes up with and like what the probabilities I I have a sample what the probabilities I I have a sample what the probabilities I I have a sample app that I use in my talks that draws app that I use in my talks that draws app that I use in my talks that draws that as a heat map and then it colors that as a heat map and then it colors that as a heat map and then it colors the word based on the log prob and red the word based on the log prob and red the word based on the log prob and red words are rare words and yellow words words are rare words and yellow words words are rare words and yellow words are medium rare. So it'll say it's a are medium rare. So it'll say it's a are medium rare. So it'll say it's a beautiful day let's go to the it'll say beautiful day let's go to the it'll say beautiful day let's go to the it'll say beach. You'll hover over the B and beach. You'll hover over the B and beach. You'll hover over the B and you'll find out that B and each have you'll find out that B and each have you'll find out that B and each have been broken into two tokens. You'll been broken into two tokens. You'll been broken into two tokens. You'll click on the B and then it'll pop up all click on the B and then it'll pop up all click on the B and then it'll pop up all the other log probs for that token and the other log probs for that token and the other log probs for that token and it'll say beach or B, the letter B is it'll say beach or B, the letter B is it'll say beach or B, the letter B is like 75%. And then it'll have like like 75%. And then it'll have like like 75%. And then it'll have like mountains be like 00.1 or or or park was mountains be like 00.1 or or or park was mountains be like 00.1 or or or park was 20% chance. And then it's like at this 20% chance. And then it's like at this 20% chance. And then it's like at this moment the universe split in half.

moment the universe split in half. moment the universe split in half. >> Yeah. >> Yeah. >> Yeah. >> 1/5if of us went to the park and the >> 1/5if of us went to the park and the >> 1/5if of us went to the park and the other rest went to the mountains and other rest went to the mountains and other rest went to the mountains and this guy went to the beach. And that's this guy went to the beach. And that's this guy went to the beach. And that's happening every day all day constantly. happening every day all day constantly. happening every day all day constantly. >> Yeah. Yeah. I think that is really cool. >> Yeah. Yeah. I think that is really cool. >> Yeah. Yeah. I think that is really cool. >> It is really cool. >> It is really cool. >> It is really cool. >> Yeah. >> Yeah. >> Yeah. >> And to your point about like changing >> And to your point about like changing >> And to your point about like changing the relationship, right? Like once you the relationship, right? Like once you the relationship, right? Like once you once you get a better understanding of once you get a better understanding of once you get a better understanding of like the underlying concepts of like the underlying concepts of like the underlying concepts of probabilities, like it makes it easier probabilities, like it makes it easier probabilities, like it makes it easier for you to understand like how models for you to understand like how models for you to understand like how models come up with the text that they come up come up with the text that they come up come up with the text that they come up with. But it also helps you understand with. But it also helps you understand with. But it also helps you understand like a lot of the things that we find like a lot of the things that we find like a lot of the things that we find most fascinating today about like both most fascinating today about like both most fascinating today about like both moral patient and like intelligence and moral patient and like intelligence and moral patient and like intelligence and like how is intelligence formed? like at like how is intelligence formed? like at like how is intelligence formed? like at what point do we consider this to be what point do we consider this to be what point do we consider this to be like a useful therapist? Like a lot of like a useful therapist? Like a lot of like a useful therapist? Like a lot of the things that we're looking at are how the things that we're looking at are how the things that we're looking at are how do the models do the models do the models evolve their internal weights and those evolve their internal weights and those evolve their internal weights and those like transformer blocks? What is like transformer blocks? What is like transformer blocks? What is happening inside those transformer happening inside those transformer happening inside those transformer blocks? And it's much easier for you to blocks? And it's much easier for you to blocks? And it's much easier for you to get like for you to join the fun of like get like for you to join the fun of like get like for you to join the fun of like all of this like you know armchair all of this like you know armchair all of this like you know armchair philosophy about the universe and like philosophy about the universe and like philosophy about the universe and like the human experience. If you have a the human experience. If you have a the human experience. If you have a rough idea of when I say transformer rough idea of when I say transformer rough idea of when I say transformer block and weights, what does what does block and weights, what does what does block and weights, what does what does that actually mean?

that actually mean? that actually mean? >> Um, and when researchers talk about not >> Um, and when researchers talk about not >> Um, and when researchers talk about not building a model but growing a model building a model but growing a model building a model but growing a model because it's like much closer to like because it's like much closer to like because it's like much closer to like growing an orchid or something than like growing an orchid or something than like growing an orchid or something than like writing a tool, if you understand why writing a tool, if you understand why writing a tool, if you understand why they say that, like there's so much they say that, like there's so much they say that, like there's so much discourse about AI that is really fun to discourse about AI that is really fun to discourse about AI that is really fun to like read and participate in. like read and participate in. like read and participate in. >> Oh yeah. Well, we should pull a little >> Oh yeah. Well, we should pull a little >> Oh yeah. Well, we should pull a little bit on the thread that you just opened bit on the thread that you just opened bit on the thread that you just opened there where you just said moral patient, there where you just said moral patient, there where you just said moral patient, which I'm going to guess a lot of the which I'm going to guess a lot of the which I'm going to guess a lot of the people on the call aren't familiar with people on the call aren't familiar with people on the call aren't familiar with and it is grounded in the assumption and it is grounded in the assumption and it is grounded in the assumption that someone has a certain amount of that someone has a certain amount of that someone has a certain amount of cognitive capacity, some certain amount cognitive capacity, some certain amount cognitive capacity, some certain amount of intelligence. You have to have of intelligence. You have to have of intelligence. You have to have autonomy and self-awareness. Uh, and you autonomy and self-awareness. Uh, and you autonomy and self-awareness. Uh, and you have to care about something. The have to care about something. The have to care about something. The problem is that a large language model problem is that a large language model problem is that a large language model doesn't have self-awareness. It doesn't doesn't have self-awareness. It doesn't doesn't have self-awareness. It doesn't have caring but it does have value have caring but it does have value have caring but it does have value because the values are the weights and because the values are the weights and because the values are the weights and it does have autonomy or at least it can it does have autonomy or at least it can it does have autonomy or at least it can be given autonomy. So then the question be given autonomy. So then the question be given autonomy. So then the question is you know is it does it have moral is you know is it does it have moral is you know is it does it have moral patient or does it just have the patient or does it just have the patient or does it just have the similacrim of that? similacrim of that? similacrim of that? >> Yeah. Yeah. It's like a super >> Yeah. Yeah. It's like a super >> Yeah. Yeah. It's like a super fascinating field of study and I'm not fascinating field of study and I'm not fascinating field of study and I'm not qualified even remotely to like yeah qualified even remotely to like yeah qualified even remotely to like yeah >> give you any useful answers but I I >> give you any useful answers but I I >> give you any useful answers but I I really enjoy sitting on the sidelines really enjoy sitting on the sidelines really enjoy sitting on the sidelines and like listening to established and like listening to established and like listening to established philosophers and like philosophers and like philosophers and like >> um people who've been studying like >> um people who've been studying like >> um people who've been studying like animal ethics for a long time. really animal ethics for a long time. really animal ethics for a long time. really enjoy reading about where they come enjoy reading about where they come enjoy reading about where they come from.

from. from. >> And those people tend to like really >> And those people tend to like really >> And those people tend to like really they're quite fascinated about the the they're quite fascinated about the the they're quite fascinated about the the individual transformer blocks and the individual transformer blocks and the individual transformer blocks and the weights and like how the model organizes weights and like how the model organizes weights and like how the model organizes it own weights and its own blocks and it own weights and its own blocks and it own weights and its own blocks and like what kind of function it assigns to like what kind of function it assigns to like what kind of function it assigns to blocks which is never programmed but blocks which is never programmed but blocks which is never programmed but sort of like evolves over time. And this sort of like evolves over time. And this sort of like evolves over time. And this evolution over time as we like go evolution over time as we like go evolution over time as we like go through training is like really through training is like really through training is like really fascinating for us because it gives us fascinating for us because it gives us fascinating for us because it gives us interesting insights into into um interesting insights into into um interesting insights into into um just how intelligence operates. just how intelligence operates. just how intelligence operates. >> Yeah, this artificial intelligence. >> Yeah, this artificial intelligence. >> Yeah, this artificial intelligence. >> Another useful thing about language >> Another useful thing about language >> Another useful thing about language model builder and thinking about that in model builder and thinking about that in model builder and thinking about that in in the through the lens of the some of in the through the lens of the some of in the through the lens of the some of the language that you just used about the language that you just used about the language that you just used about how how the weights are applied, how the how how the weights are applied, how the how how the weights are applied, how the model thinks, how the model changes. model thinks, how the model changes. model thinks, how the model changes. These are all kind of you're dancing These are all kind of you're dancing These are all kind of you're dancing around the anthropomorphizing of these around the anthropomorphizing of these around the anthropomorphizing of these things. But people need to understand things. But people need to understand things. But people need to understand and you can see it when you run language and you can see it when you run language and you can see it when you run language model builder that like if you're model builder that like if you're model builder that like if you're running a training you can watch it you running a training you can watch it you running a training you can watch it you can see it happen live but once the can see it happen live but once the can see it happen live but once the model is baked and you pull it out of model is baked and you pull it out of model is baked and you pull it out of the oven it doesn't continue to learn. the oven it doesn't continue to learn. the oven it doesn't continue to learn. >> Yeah. >> Yeah. >> Yeah. >> Unless you decide to refine it and put >> Unless you decide to refine it and put >> Unless you decide to refine it and put it back in the oven and bake it some it back in the oven and bake it some it back in the oven and bake it some more. You know what I mean? And I think more. You know what I mean? And I think more. You know what I mean? And I think that regular people in like that regular people in like that regular people in like non-technical parent don't realize that non-technical parent don't realize that non-technical parent don't realize that that it's a bunch of stateless calls, that it's a bunch of stateless calls, that it's a bunch of stateless calls, you know, stateless HTTP calls. There's you know, stateless HTTP calls. There's you know, stateless HTTP calls. There's no state here. There's just context no state here. There's just context no state here. There's just context given back to you rehydrated given back to you rehydrated given back to you rehydrated >> and then it keeps talking. So these like >> and then it keeps talking. So these like >> and then it keeps talking. So these like >> ongoing persistent conversations that we >> ongoing persistent conversations that we >> ongoing persistent conversations that we have are an illusion in themselves of have are an illusion in themselves of have are an illusion in themselves of these stateful the state is is the state these stateful the state is is the state these stateful the state is is the state itself of the conversation is an

itself of the conversation is an itself of the conversation is an illusion just made by context. illusion just made by context. illusion just made by context. >> Yeah. Yeah. Yeah, and that's right. >> Yeah. Yeah. Yeah, and that's right. >> Yeah. Yeah. Yeah, and that's right. Yeah, exactly. It's like the difference Yeah, exactly. It's like the difference Yeah, exactly. It's like the difference between like one thing is like the between like one thing is like the between like one thing is like the actual weights, right? And then the actual weights, right? And then the actual weights, right? And then the other thing is just like additional other thing is just like additional other thing is just like additional information or tool calls that we give information or tool calls that we give information or tool calls that we give the model, but the thing that we the model, but the thing that we the model, but the thing that we actually then train is like using other actually then train is like using other actually then train is like using other tools, not not like what to necessarily tools, not not like what to necessarily tools, not not like what to necessarily do with your results. Um, and obviously do with your results. Um, and obviously do with your results. Um, and obviously I think at scale some of the stuff like I think at scale some of the stuff like I think at scale some of the stuff like breaks apart a little bit because a lot breaks apart a little bit because a lot breaks apart a little bit because a lot of the things that we may have done in of the things that we may have done in of the things that we may have done in like post-training are increasingly like post-training are increasingly like post-training are increasingly moving into pre-training. moving into pre-training. moving into pre-training. >> Like especially tool calling, right? >> Like especially tool calling, right? >> Like especially tool calling, right? Like I think there was a there was a Like I think there was a there was a Like I think there was a there was a point in time where you could really point in time where you could really point in time where you could really tell whether or not a model was trained tell whether or not a model was trained tell whether or not a model was trained on like effectively calling tools or on like effectively calling tools or on like effectively calling tools or not. was like a brief point in time not. was like a brief point in time not. was like a brief point in time especially in coding where some tools especially in coding where some tools especially in coding where some tools were just incapable of using bash even were just incapable of using bash even were just incapable of using bash even though they could give you like a whole though they could give you like a whole though they could give you like a whole bash bash bash >> manual >> manual >> manual >> right >> right >> right >> and that in itself is like a little >> and that in itself is like a little >> and that in itself is like a little interesting but interesting but interesting but >> I think that the tool calling stuff is >> I think that the tool calling stuff is >> I think that the tool calling stuff is super interesting because when I teach super interesting because when I teach super interesting because when I teach that I talk about how you are having a that I talk about how you are having a that I talk about how you are having a chat between you and the model so chat between you and the model so chat between you and the model so there's two entities in the chat there's two entities in the chat there's two entities in the chat >> and when you add tool calling there is a >> and when you add tool calling there is a >> and when you add tool calling there is a third person in the group chat yeah and third person in the group chat yeah and third person in the group chat yeah and Then you leave Then you leave Then you leave and now it's the chatbot talking to the and now it's the chatbot talking to the and now it's the chatbot talking to the tools and the chatbot tells the tool and tools and the chatbot tells the tool and tools and the chatbot tells the tool and then the tool isn't a chatbot but you then the tool isn't a chatbot but you then the tool isn't a chatbot but you can pretend that it is because you give can pretend that it is because you give can pretend that it is because you give it input and it gives you output that we it input and it gives you output that we it input and it gives you output that we don't know what we're going to get and don't know what we're going to get and don't know what we're going to get and that keeps the conversation going while that keeps the conversation going while that keeps the conversation going while you're basically on hold you're basically on hold you're basically on hold >> as the third person in the group chat.

>> as the third person in the group chat. >> as the third person in the group chat. >> And I don't know if you do this with >> And I don't know if you do this with >> And I don't know if you do this with your people but like one thing I've your people but like one thing I've your people but like one thing I've always enjoyed doing with with people always enjoyed doing with with people always enjoyed doing with with people asking okay how do I build effective asking okay how do I build effective asking okay how do I build effective tools with AI is to think about like tools with AI is to think about like tools with AI is to think about like model experience. model experience. model experience. >> Yeah. I'm like, "Okay, imagine you're >> Yeah. I'm like, "Okay, imagine you're >> Yeah. I'm like, "Okay, imagine you're the model, right? And like Scott asks the model, right? And like Scott asks the model, right? And like Scott asks you, where should I go today?" Okay, you, where should I go today?" Okay, you, where should I go today?" Okay, you're the model. Like, what what kind you're the model. Like, what what kind you're the model. Like, what what kind of what kind of tools do you want, of what kind of tools do you want, of what kind of tools do you want, right? Like, and you might be like, right? Like, and you might be like, right? Like, and you might be like, "Okay, I probably want to know like is "Okay, I probably want to know like is "Okay, I probably want to know like is there anything that tells me like there anything that tells me like there anything that tells me like Scott's preferences or his location?" Scott's preferences or his location?" Scott's preferences or his location?" >> Do I have memory? Do I know where he is? >> Do I have memory? Do I know where he is? >> Do I have memory? Do I know where he is? Is this the first time I've ever met Is this the first time I've ever met Is this the first time I've ever met this guy before? this guy before? this guy before? >> That's why the open clause stuff is so >> That's why the open clause stuff is so >> That's why the open clause stuff is so clever. This simple idea of a soul.md clever. This simple idea of a soul.md clever. This simple idea of a soul.md is the every single time you say, "Hey, is the every single time you say, "Hey, is the every single time you say, "Hey, are you there?" are you there?" are you there?" >> It has to wake up from a dead sleep, >> It has to wake up from a dead sleep, >> It has to wake up from a dead sleep, stumble out of bed, go, "What? Who am stumble out of bed, go, "What? Who am stumble out of bed, go, "What? Who am I?" It's like that movie Momento where I?" It's like that movie Momento where I?" It's like that movie Momento where the guy tattoos all the context all over the guy tattoos all the context all over the guy tattoos all the context all over himself and has to wake up every morning himself and has to wake up every morning himself and has to wake up every morning and look in the mirror to figure out who and look in the mirror to figure out who and look in the mirror to figure out who am I and what am I doing here? So, every am I and what am I doing here? So, every am I and what am I doing here? So, every time he learns something, he tattoos time he learns something, he tattoos time he learns something, he tattoos himself. himself. himself. >> Yeah, a little bit. I do think >> Yeah, a little bit. I do think >> Yeah, a little bit. I do think I I I think the I I I think the I I I think the this stuff all like changes so quickly, this stuff all like changes so quickly, this stuff all like changes so quickly, but I do wonder but I do wonder but I do wonder obviously most models are actually being obviously most models are actually being obviously most models are actually being trained quite a bit of information about trained quite a bit of information about trained quite a bit of information about who they are and what they're supposed who they are and what they're supposed who they are and what they're supposed to do, to do, to do, >> right?

>> right? >> right? >> Um and I think the open claw thing works >> Um and I think the open claw thing works >> Um and I think the open claw thing works pretty well because like most models are pretty well because like most models are pretty well because like most models are just like pre-trained to like obviously just like pre-trained to like obviously just like pre-trained to like obviously we pre-train on like all the world we pre-train on like all the world we pre-train on like all the world knowledge but then we do like a mountain knowledge but then we do like a mountain knowledge but then we do like a mountain of fine tuning on you're supposed to be of fine tuning on you're supposed to be of fine tuning on you're supposed to be helpful, right? I can give if given this helpful, right? I can give if given this helpful, right? I can give if given this you should do you should do you should do >> that's a great point that you bring that >> that's a great point that you bring that >> that's a great point that you bring that up goal seeking I keep talking about up goal seeking I keep talking about up goal seeking I keep talking about people that it's not just in it's not people that it's not just in it's not people that it's not just in it's not just an intern with unlimited energy it just an intern with unlimited energy it just an intern with unlimited energy it is a goal seeking and it has been is a goal seeking and it has been is a goal seeking and it has been trained to be helpful I know that and trained to be helpful I know that and trained to be helpful I know that and you can see it when you go in through you can see it when you go in through you can see it when you go in through your supervised fine-tuning your tuning your supervised fine-tuning your tuning your supervised fine-tuning your tuning your direct preference optimization when your direct preference optimization when your direct preference optimization when you go through that in language model you go through that in language model you go through that in language model builder the whole point is to make a builder the whole point is to make a builder the whole point is to make a helpful and useful and kind and goal helpful and useful and kind and goal helpful and useful and kind and goal seeking model otherwise it would not be seeking model otherwise it would not be seeking model otherwise it would not be useful it would just be useful it would just be useful it would just be sassy bot that would just be mean to sassy bot that would just be mean to sassy bot that would just be mean to you. you. you. >> I did add like the sassy mean bot thing. >> I did add like the sassy mean bot thing. >> I did add like the sassy mean bot thing. If you have like a [laughter] If you have like a [laughter] If you have like a [laughter] >> I got to check that part out. >> I got to check that part out. >> I got to check that part out. >> Like a a thing a thing I added like a >> Like a a thing a thing I added like a >> Like a a thing a thing I added like a little later. Um this makes for terrible little later. Um this makes for terrible little later. Um this makes for terrible fine-tuning data. Please like pre don't fine-tuning data. Please like pre don't fine-tuning data. Please like pre don't do that. do that. do that. >> Okay. Like prefix prefix. This is very >> Okay. Like prefix prefix. This is very >> Okay. Like prefix prefix. This is very fun to do fun to do fun to do >> and the results are going to be >> and the results are going to be >> and the results are going to be terrible. Okay. Just like putting that terrible. Okay. Just like putting that terrible. Okay. Just like putting that out there. But you can import I added a out there. But you can import I added a out there. But you can import I added a thing that lets you import an entire thing that lets you import an entire thing that lets you import an entire group chat as fine-tuning data.

group chat as fine-tuning data. group chat as fine-tuning data. >> Uh oh. Then it's going to talk the way >> Uh oh. Then it's going to talk the way >> Uh oh. Then it's going to talk the way that you talk with your friends. that you talk with your friends. that you talk with your friends. >> That's right. Um so it basically like >> That's right. Um so it basically like >> That's right. Um so it basically like teaches the model, okay, here are the teaches the model, okay, here are the teaches the model, okay, here are the people and it's going to continue like people and it's going to continue like people and it's going to continue like the whole continue the script thing but the whole continue the script thing but the whole continue the script thing but for like a long group chat. Um which is for like a long group chat. Um which is for like a long group chat. Um which is obviously quite different from the other obviously quite different from the other obviously quite different from the other fine tuning stuff, right? Like the other fine tuning stuff, right? Like the other fine tuning stuff, right? Like the other fine-tuning stuff. Um, and obviously fine-tuning stuff. Um, and obviously fine-tuning stuff. Um, and obviously Scott, you've done this many times, but Scott, you've done this many times, but Scott, you've done this many times, but like for the benefit of people who like for the benefit of people who like for the benefit of people who listen to us, um, once you have a model listen to us, um, once you have a model listen to us, um, once you have a model that knows about the world, you then that knows about the world, you then that knows about the world, you then have to teach it that if it's given a have to teach it that if it's given a have to teach it that if it's given a question that it shouldn't like give us question that it shouldn't like give us question that it shouldn't like give us more questions, but should actually give more questions, but should actually give more questions, but should actually give us an answer. us an answer. us an answer. >> We do that in fine tuning. >> We do that in fine tuning. >> We do that in fine tuning. >> And for fine-tuning, I have all these >> And for fine-tuning, I have all these >> And for fine-tuning, I have all these different I have all these different different I have all these different different I have all these different examples. There's like one that examples. There's like one that examples. There's like one that fine-tunes on math problems. There's fine-tunes on math problems. There's fine-tunes on math problems. There's another one that fine-tunes on like another one that fine-tunes on like another one that fine-tunes on like being able to rewrite given text as like being able to rewrite given text as like being able to rewrite given text as like different text. But I have this one that different text. But I have this one that different text. But I have this one that is like admittedly terrible where you is like admittedly terrible where you is like admittedly terrible where you can import a group chat and it will then can import a group chat and it will then can import a group chat and it will then be like a mimic of that group chat and be like a mimic of that group chat and be like a mimic of that group chat and continue. continue. continue. >> I made one that talks like Ben uh >> I made one that talks like Ben uh >> I made one that talks like Ben uh Benedict Cumberbatch in the guy who Benedict Cumberbatch in the guy who Benedict Cumberbatch in the guy who plays plays plays >> nice >> nice >> nice >> uh Sherlock Holmes >> uh Sherlock Holmes >> uh Sherlock Holmes >> and he's so irritated. Why are you >> and he's so irritated. Why are you >> and he's so irritated. Why are you asking me these dumb questions?

asking me these dumb questions? asking me these dumb questions? >> If only you were smart like me, Sherlock >> If only you were smart like me, Sherlock >> If only you were smart like me, Sherlock Holmes. Holmes. Holmes. >> That's that's beautiful. I mean I I also >> That's that's beautiful. I mean I I also >> That's that's beautiful. I mean I I also put in um Kaparthi's like tiny put in um Kaparthi's like tiny put in um Kaparthi's like tiny Shakespeare. Um Shakespeare. Um Shakespeare. Um >> awesome. >> awesome. >> awesome. >> It was like one of one of Kaparthy's >> It was like one of one of Kaparthy's >> It was like one of one of Kaparthy's first demos was what if a language model first demos was what if a language model first demos was what if a language model that writes like Shakespeare? that writes like Shakespeare? that writes like Shakespeare? >> Yeah. >> Yeah. >> Yeah. >> Um which is very quick to train. But my >> Um which is very quick to train. But my >> Um which is very quick to train. But my my point is that like this difference my point is that like this difference my point is that like this difference between like waking up and trying to between like waking up and trying to between like waking up and trying to figure out who you are. I think there's figure out who you are. I think there's figure out who you are. I think there's already already already >> I I don't I don't actually think it's >> I I don't I don't actually think it's >> I I don't I don't actually think it's from zero. There's already like quite a from zero. There's already like quite a from zero. There's already like quite a bit of personality baked into the model. bit of personality baked into the model. bit of personality baked into the model. And I think in the open claw example, it And I think in the open claw example, it And I think in the open claw example, it works really well. works really well. works really well. >> Yeah. Yeah, because fundamentally >> Yeah. Yeah, because fundamentally >> Yeah. Yeah, because fundamentally most people want their open cloud to be most people want their open cloud to be most people want their open cloud to be like helpful and good and like like helpful and good and like like helpful and good and like >> Yeah, exactly. >> Yeah, exactly. >> Yeah, exactly. >> Right. >> Right. >> Right. >> But I think it would be interesting to >> But I think it would be interesting to >> But I think it would be interesting to like just for fun um just for your own like just for fun um just for your own like just for fun um just for your own understanding like train some models understanding like train some models understanding like train some models that maybe do different things. that maybe do different things. that maybe do different things. >> Yeah. >> Yeah. >> Yeah. >> Could be enjoyable, right? >> Could be enjoyable, right? >> Could be enjoyable, right? >> Well, and it also like it makes you >> Well, and it also like it makes you >> Well, and it also like it makes you understand understand understand why are these things trying to be why are these things trying to be why are these things trying to be helpful? It's not just because of the helpful? It's not just because of the helpful? It's not just because of the fine-tuning, but it's also the corpus.

fine-tuning, but it's also the corpus. fine-tuning, but it's also the corpus. And I always talk about the corpus, And I always talk about the corpus, And I always talk about the corpus, whether it be parts of, you know, weird whether it be parts of, you know, weird whether it be parts of, you know, weird corners of Reddit or weird corners of corners of Reddit or weird corners of corners of Reddit or weird corners of Stack Overflow. If you end up in like Stack Overflow. If you end up in like Stack Overflow. If you end up in like the sassy part of the internet, the the sassy part of the internet, the the sassy part of the internet, the model is going to be snarky and sassy. model is going to be snarky and sassy. model is going to be snarky and sassy. And it's not that it's not goal seeking. And it's not that it's not goal seeking. And it's not that it's not goal seeking. And it's not that it wants to help you And it's not that it wants to help you And it's not that it wants to help you or not. It's just it was built on a or not. It's just it was built on a or not. It's just it was built on a sassy corpus. So you got to put your sassy corpus. So you got to put your sassy corpus. So you got to put your thumb on the scale and fine-tune it to thumb on the scale and fine-tune it to thumb on the scale and fine-tune it to be a little bit more helpful. That's be a little bit more helpful. That's be a little bit more helpful. That's where supervised fine-tuning comes in. where supervised fine-tuning comes in. where supervised fine-tuning comes in. >> Yeah. Have you do you remember Claude >> Yeah. Have you do you remember Claude >> Yeah. Have you do you remember Claude Golden Gate? No, this was before I Golden Gate? No, this was before I Golden Gate? No, this was before I joined Anthropic, but this was like an joined Anthropic, but this was like an joined Anthropic, but this was like an an early an early version of Claude that an early an early version of Claude that an early an early version of Claude that instead of trying to be like it tried to instead of trying to be like it tried to instead of trying to be like it tried to be helpful, but it also was given clear be helpful, but it also was given clear be helpful, but it also was given clear instruction to like whatever happens, instruction to like whatever happens, instruction to like whatever happens, try to work the golden gate into your try to work the golden gate into your try to work the golden gate into your answer. answer. answer. >> Oh, I see. It has a separate hidden >> Oh, I see. It has a separate hidden >> Oh, I see. It has a separate hidden goal. goal. goal. >> Truly beautiful and somewhat insane >> Truly beautiful and somewhat insane >> Truly beautiful and somewhat insane results. Um, results. Um, results. Um, >> that's awesome. >> that's awesome. >> that's awesome. >> Yeah. So like >> Yeah. So like >> Yeah. So like >> well those are what you're seeing now >> well those are what you're seeing now >> well those are what you're seeing now with stu teachers putting in white text with stu teachers putting in white text with stu teachers putting in white text on a white background in the syllabus.

on a white background in the syllabus. on a white background in the syllabus. So when the kid goes a and c and then So when the kid goes a and c and then So when the kid goes a and c and then paste it directly into the large man paste it directly into the large man paste it directly into the large man language model uh the teacher wants to language model uh the teacher wants to language model uh the teacher wants to detect whether or not the kid has AI. So detect whether or not the kid has AI. So detect whether or not the kid has AI. So they poison the prompt and then they say they poison the prompt and then they say they poison the prompt and then they say make sure you mention you know eggplants make sure you mention you know eggplants make sure you mention you know eggplants and then there's some random thing in and then there's some random thing in and then there's some random thing in the middle that the kid never test and the middle that the kid never test and the middle that the kid never test and it says eggplant. it says eggplant. it says eggplant. >> Yeah. And they're like 49 of the 50 >> Yeah. And they're like 49 of the 50 >> Yeah. And they're like 49 of the 50 results. I I one thing I fear maybe is results. I I one thing I fear maybe is results. I I one thing I fear maybe is and this sort of goes back to like the and this sort of goes back to like the and this sort of goes back to like the soul document, right? Like soul document, right? Like soul document, right? Like >> I think I think the soul document could >> I think I think the soul document could >> I think I think the soul document could probably get you pretty far in terms of probably get you pretty far in terms of probably get you pretty far in terms of like very late stage context like very late stage context like very late stage context engineering, engineering, engineering, >> right? >> right? >> right? >> But I think especially as the models get >> But I think especially as the models get >> But I think especially as the models get bigger and more powerful and by bigger bigger and more powerful and by bigger bigger and more powerful and by bigger and more powerful I mean like there's and more powerful I mean like there's and more powerful I mean like there's just more weights and like more just more weights and like more just more weights and like more parameters that are actually active, not parameters that are actually active, not parameters that are actually active, not just expert parameters but like just expert parameters but like just expert parameters but like generally parameters. I think hopefully generally parameters. I think hopefully generally parameters. I think hopefully we'll increasingly get models that we'll increasingly get models that we'll increasingly get models that there's sort of like three stages in my there's sort of like three stages in my there's sort of like three stages in my mind, right? Like stage one is a dumb mind, right? Like stage one is a dumb mind, right? Like stage one is a dumb model that just like does whatever it's model that just like does whatever it's model that just like does whatever it's given and then stage two is maybe sort given and then stage two is maybe sort given and then stage two is maybe sort of the model that reads the instructions of the model that reads the instructions of the model that reads the instructions and like works eggplant into the and like works eggplant into the and like works eggplant into the document. But hopefully we'll actually document. But hopefully we'll actually document. But hopefully we'll actually get to a model that has even more get to a model that has even more get to a model that has even more humanity and morality programmed into it humanity and morality programmed into it humanity and morality programmed into it to the point where it's like clearly I'm to the point where it's like clearly I'm to the point where it's like clearly I'm being asked to cheat. Yeah, being asked to cheat. Yeah, being asked to cheat. Yeah, >> I should like help this kid >> I should like help this kid >> I should like help this kid uh to the same extent that maybe like uh to the same extent that maybe like uh to the same extent that maybe like you know the things that you and I would you know the things that you and I would you know the things that you and I would do if like our kids came by and were do if like our kids came by and were do if like our kids came by and were like, "Hey, can you write my homework like, "Hey, can you write my homework like, "Hey, can you write my homework for me?" Like, how are you that for me?" Like, how are you that for me?" Like, how are you that >> moment? Exactly. And that's funny you >> moment? Exactly. And that's funny you >> moment? Exactly. And that's funny you mentioned that because I'm working on a

mentioned that because I'm working on a mentioned that because I'm working on a thing with Renovich at Microsoft, which thing with Renovich at Microsoft, which thing with Renovich at Microsoft, which is a we call it a preceptorship. It's a is a we call it a preceptorship. It's a is a we call it a preceptorship. It's a different spin on an internship and we different spin on an internship and we different spin on an internship and we want to make models that are coding want to make models that are coding want to make models that are coding models, coding smart models. Right now, models, coding smart models. Right now, models, coding smart models. Right now, we have to do it with skills and with we have to do it with skills and with we have to do it with skills and with markdown files, but I want a model that markdown files, but I want a model that markdown files, but I want a model that doesn't hold their hand. I want it to doesn't hold their hand. I want it to doesn't hold their hand. I want it to like I'm going to leave this as an like I'm going to leave this as an like I'm going to leave this as an exercise to the reader. I know you want exercise to the reader. I know you want exercise to the reader. I know you want me to make a bubble sort, but I'd like me to make a bubble sort, but I'd like me to make a bubble sort, but I'd like to see you do it first. I want a a to see you do it first. I want a a to see you do it first. I want a a friendlier early in career engineer friendlier early in career engineer friendlier early in career engineer model that teaches them how to think, model that teaches them how to think, model that teaches them how to think, and I want people in universities to and I want people in universities to and I want people in universities to have access to those models. I don't have access to those models. I don't have access to those models. I don't want it to just like we should not want it to just like we should not want it to just like we should not become a subcognitive species. become a subcognitive species. become a subcognitive species. >> Just because it can do it doesn't mean >> Just because it can do it doesn't mean >> Just because it can do it doesn't mean it should do it. it should do it. it should do it. >> Yeah. And I think maybe I mean I made >> Yeah. And I think maybe I mean I made >> Yeah. And I think maybe I mean I made language model builder for fun. It's language model builder for fun. It's language model builder for fun. It's free. I'm not going to make any money free. I'm not going to make any money free. I'm not going to make any money with it. So like maybe it's it's going with it. So like maybe it's it's going with it. So like maybe it's it's going to seem less like I'm I'm just to seem less like I'm I'm just to seem less like I'm I'm just advertising it. Like if if people have advertising it. Like if if people have advertising it. Like if if people have any questions about okay why do models any questions about okay why do models any questions about okay why do models behave certain ways like the fine-tuning behave certain ways like the fine-tuning behave certain ways like the fine-tuning data section in it data section in it data section in it >> is like really beautiful because also >> is like really beautiful because also >> is like really beautiful because also fine-tuning doesn't take a lot of steps fine-tuning doesn't take a lot of steps fine-tuning doesn't take a lot of steps like pre-training takes a long time like like pre-training takes a long time like like pre-training takes a long time like even in this app takes like at least a even in this app takes like at least a even in this app takes like at least a day but fine-tuning 800 steps or so like day but fine-tuning 800 steps or so like day but fine-tuning 800 steps or so like more than enough to like for these small more than enough to like for these small more than enough to like for these small models to really steer how they operate models to really steer how they operate models to really steer how they operate >> and just seeing like this model like you >> and just seeing like this model like you >> and just seeing like this model like you put it in the oven for five minutes and put it in the oven for five minutes and put it in the oven for five minutes and you tell it I want you to like give you tell it I want you to like give you tell it I want you to like give helpful responses whatever Um, and just helpful responses whatever Um, and just helpful responses whatever Um, and just like playing with some of your own like playing with some of your own like playing with some of your own examples a little bit is like really examples a little bit is like really examples a little bit is like really fun.

fun. fun. >> Yeah. Well, I want to call that out >> Yeah. Well, I want to call that out >> Yeah. Well, I want to call that out because languagemodelbuilder.com because languagemodelbuilder.com because languagemodelbuilder.com is free. And I want to remind folks that is free. And I want to remind folks that is free. And I want to remind folks that like you could have become an AI like you could have become an AI like you could have become an AI grifter. Everybody on social media, grifter. Everybody on social media, grifter. Everybody on social media, everybody on the dumpster fire that is everybody on the dumpster fire that is everybody on the dumpster fire that is Twitter wants you to buy their course. Twitter wants you to buy their course. Twitter wants you to buy their course. Give me 1995, right? PayPal me five Give me 1995, right? PayPal me five Give me 1995, right? PayPal me five bucks. learn how to avoid learn how to bucks. learn how to avoid learn how to bucks. learn how to avoid learn how to avoid uh you know grifters. Send me $5 avoid uh you know grifters. Send me $5 avoid uh you know grifters. Send me $5 and I'll tell you how you chose to make and I'll tell you how you chose to make and I'll tell you how you chose to make it free. You just go to it free. You just go to it free. You just go to languagemilebuilder.com. There's no languagemilebuilder.com. There's no languagemilebuilder.com. There's no upsell. You made it for fun. You made it upsell. You made it for fun. You made it upsell. You made it for fun. You made it for your friends and now your friends for your friends and now your friends for your friends and now your friends are the whole internet. So, I want to are the whole internet. So, I want to are the whole internet. So, I want to appreciate that you didn't put, you appreciate that you didn't put, you appreciate that you didn't put, you know, a gum a gumroad or a Shopify on know, a gum a gumroad or a Shopify on know, a gum a gumroad or a Shopify on the top of this thing and try to make a the top of this thing and try to make a the top of this thing and try to make a quick buck. You're just putting the quick buck. You're just putting the quick buck. You're just putting the information out there for the people. information out there for the people. information out there for the people. >> Thank you. Yeah. I mean, it's an >> Thank you. Yeah. I mean, it's an >> Thank you. Yeah. I mean, it's an educational fun app, right? Like, I'm educational fun app, right? Like, I'm educational fun app, right? Like, I'm pretty sure there was not a lot of money pretty sure there was not a lot of money pretty sure there was not a lot of money to be made to begin with. to be made to begin with. to be made to begin with. >> Yeah. But it's less about that like the >> Yeah. But it's less about that like the >> Yeah. But it's less about that like the information should be free. Let the information should be free. Let the information should be free. Let the information out. Everyone out there is information out. Everyone out there is information out. Everyone out there is trying to get you to do their class and trying to get you to do their class and trying to get you to do their class and everyone on LinkedIn wants you to get everyone on LinkedIn wants you to get everyone on LinkedIn wants you to get into their AI thing. What I like about into their AI thing. What I like about into their AI thing. What I like about Language Model Builder is that I could Language Model Builder is that I could Language Model Builder is that I could go to, you know, Portland Community go to, you know, Portland Community go to, you know, Portland Community College, my alma mater, and teach a College, my alma mater, and teach a College, my alma mater, and teach a class and use this as the, no pun class and use this as the, no pun class and use this as the, no pun intended, as the corpus for the class intended, as the corpus for the class intended, as the corpus for the class and the syllabus. And uh it's a great and the syllabus. And uh it's a great and the syllabus. And uh it's a great great place to start whether you're great place to start whether you're great place to start whether you're already an engineer or you're just a already an engineer or you're just a already an engineer or you're just a person who wants to know how to drive person who wants to know how to drive person who wants to know how to drive stick shift. I would encourage folks to stick shift. I would encourage folks to stick shift. I would encourage folks to check it out and I look forward to the check it out and I look forward to the check it out and I look forward to the Windows version someday. Oh, Windows version someday. Oh, Windows version someday. Oh, >> of course. Yeah, thank you so much.

>> of course. Yeah, thank you so much. >> of course. Yeah, thank you so much. >> Yeah, thank you. We have been chatting >> Yeah, thank you. We have been chatting >> Yeah, thank you. We have been chatting with Felix Rezerberg. This is another with Felix Rezerberg. This is another with Felix Rezerberg. This is another episode of Hansel Minutes and we'll see episode of Hansel Minutes and we'll see episode of Hansel Minutes and we'll see you again next