Local Models: Trust, Control, Optimization — Carter Abdallah, NVIDIA

>> So, I hope everybody had a great lunch >> So, I hope everybody had a great lunch and you got to check out some of the and you got to check out some of the and you got to check out some of the amazing demos that we have. amazing demos that we have. amazing demos that we have. Uh we're going to begin the panel the Uh we're going to begin the panel the Uh we're going to begin the panel the first panel of the afternoon here where first panel of the afternoon here where first panel of the afternoon here where we're going to be talking about of we're going to be talking about of we're going to be talking about of course the engines that are actually course the engines that are actually course the engines that are actually powering the stuff that powering the stuff that powering the stuff that you know could remotely you know could remotely you know could remotely be used for things like local sovereign be used for things like local sovereign be used for things like local sovereign any kind of ownership over your own any kind of ownership over your own any kind of ownership over your own artificial intelligence and of course artificial intelligence and of course artificial intelligence and of course the engine powering those in addition to the engine powering those in addition to the engine powering those in addition to the hardware is the models themselves. the hardware is the models themselves. the hardware is the models themselves. And so for this panel we have excellent And so for this panel we have excellent And so for this panel we have excellent guests. We have Vincent who's the CEO guests. We have Vincent who's the CEO guests. We have Vincent who's the CEO and founder of Prime Intellect. We've and founder of Prime Intellect. We've and founder of Prime Intellect. We've got Lucas the CTO of RCAI and we've got got Lucas the CTO of RCAI and we've got got Lucas the CTO of RCAI and we've got Chris who is the senior product research Chris who is the senior product research Chris who is the senior product research engineer on the Neumotron family of engineer on the Neumotron family of engineer on the Neumotron family of models at Nvidia. Now, what's really models at Nvidia. Now, what's really models at Nvidia. Now, what's really cool about working in this industry is cool about working in this industry is cool about working in this industry is really cool companies like this we all really cool companies like this we all really cool companies like this we all get to work together. And so this is one get to work together. And so this is one get to work together. And so this is one panel where we all directly get to work panel where we all directly get to work panel where we all directly get to work together on both models infrastructure together on both models infrastructure together on both models infrastructure some of the ways that we think that the some of the ways that we think that the some of the ways that we think that the direction of the industry should go direction of the industry should go direction of the industry should go and each of us kind of play a different and each of us kind of play a different and each of us kind of play a different role in that stack but I want to leave role in that stack but I want to leave role in that stack but I want to leave it to you guys to to introduce yourself it to you guys to to introduce yourself it to you guys to to introduce yourself and be able to talk about sort of the and be able to talk about sort of the and be able to talk about sort of the the charter that you see the problem of the charter that you see the problem of the charter that you see the problem of the stack that you guys are working on.

the stack that you guys are working on. the stack that you guys are working on. >> Awesome. Should I kick it off? >> Awesome. Should I kick it off? >> Awesome. Should I kick it off? >> Kick it off. >> Kick it off. >> Kick it off. >> Yeah, so I'm I'm Vincent as you >> Yeah, so I'm I'm Vincent as you >> Yeah, so I'm I'm Vincent as you mentioned and then mentioned and then mentioned and then really the goal with Prime Intellect really the goal with Prime Intellect really the goal with Prime Intellect from the beginning was like to ensure from the beginning was like to ensure from the beginning was like to ensure that basically frontier intelligence that basically frontier intelligence that basically frontier intelligence will be open and accessible will be open and accessible will be open and accessible not just the models but also the full not just the models but also the full not just the models but also the full stack to to train the models. So kind of stack to to train the models. So kind of stack to to train the models. So kind of like this was like our motivation from like this was like our motivation from like this was like our motivation from the beginning and we we've yeah like the beginning and we we've yeah like the beginning and we we've yeah like worked also together with a lot of worked also together with a lot of worked also together with a lot of gentlemen on the on stage like on the gentlemen on the on stage like on the gentlemen on the on stage like on the one side is like one side is like one side is like we we work with folks like Lucas and RC we we work with folks like Lucas and RC we we work with folks like Lucas and RC to help them train frontier open models. to help them train frontier open models. to help them train frontier open models. We we we help also um We we we help also um We we we help also um like Nvidia on the Neumotron coalition like Nvidia on the Neumotron coalition like Nvidia on the Neumotron coalition help their uh open models. And I think help their uh open models. And I think help their uh open models. And I think like I'm actually um like I'm actually um like I'm actually um think both like an Eleutheron and and think both like an Eleutheron and and think both like an Eleutheron and and Trinity might be the best like two open Trinity might be the best like two open Trinity might be the best like two open models right now outside of China. So, I models right now outside of China. So, I models right now outside of China. So, I think it's actually uh like we need to think it's actually uh like we need to think it's actually uh like we need to fact-check that, you know, but this is fact-check that, you know, but this is fact-check that, you know, but this is actually from my actually from my actually from my I think they they might be. I think they they might be. I think they they might be. >> It's our our marketing says they yeah. >> It's our our marketing says they yeah. >> It's our our marketing says they yeah. >> But yeah, so so that's the high level. >> But yeah, so so that's the high level. >> But yeah, so so that's the high level. >> Um my name's Lucas Atkins. I'm happy to >> Um my name's Lucas Atkins. I'm happy to >> Um my name's Lucas Atkins. I'm happy to be here and and thank you for joining.

be here and and thank you for joining. be here and and thank you for joining. Um Um Um very similar to Vincent, RC was uh you very similar to Vincent, RC was uh you very similar to Vincent, RC was uh you know, founded with the idea of know, founded with the idea of know, founded with the idea of um um um domain-specific owned models are are are domain-specific owned models are are are domain-specific owned models are are are going to be needed. Um you know, we were going to be needed. Um you know, we were going to be needed. Um you know, we were founded early 2023 uh jumping on the founded early 2023 uh jumping on the founded early 2023 uh jumping on the custom model uh train quite early. custom model uh train quite early. custom model uh train quite early. Um Um Um you know, you have all these people who you know, you have all these people who you know, you have all these people who are excited about AI and all the things are excited about AI and all the things are excited about AI and all the things these new generation of LLMs can do. Um these new generation of LLMs can do. Um these new generation of LLMs can do. Um but they're using these monolithic very but they're using these monolithic very but they're using these monolithic very expensive closed APIs expensive closed APIs expensive closed APIs uh for at the time and still like very uh for at the time and still like very uh for at the time and still like very narrow tasks that don't require narrow tasks that don't require narrow tasks that don't require uh you know, at the time it was $100 per uh you know, at the time it was $100 per uh you know, at the time it was $100 per million tokens out. million tokens out. million tokens out. Um and through doing that, we were Um and through doing that, we were Um and through doing that, we were building on top of open models and we building on top of open models and we building on top of open models and we were releasing a lot of our tooling in were releasing a lot of our tooling in were releasing a lot of our tooling in the open. the open. the open. Um and uh we noticed that in the United Um and uh we noticed that in the United Um and uh we noticed that in the United States States States um and in the West, you know, in um and in the West, you know, in um and in the West, you know, in general, we were starting to lose uh general, we were starting to lose uh general, we were starting to lose uh leadership in the open model space. A leadership in the open model space. A leadership in the open model space. A lot of it was coming out of China and lot of it was coming out of China and lot of it was coming out of China and that's amazing. I love those models. We that's amazing. I love those models. We that's amazing. I love those models. We learn a lot from them. Uh we're close learn a lot from them. Uh we're close learn a lot from them. Uh we're close with a lot of the people building those.

with a lot of the people building those. with a lot of the people building those. But when you're working with large But when you're working with large But when you're working with large enterprises and companies and enterprises and companies and enterprises and companies and um um um geopolitics gets involved, whether you geopolitics gets involved, whether you geopolitics gets involved, whether you like it or not, you have people that like it or not, you have people that like it or not, you have people that become concerned about where those become concerned about where those become concerned about where those models are coming from. And we models are coming from. And we models are coming from. And we uh uh uh decided that, you know, we had a a good decided that, you know, we had a a good decided that, you know, we had a a good group of people and we had a good group group of people and we had a good group group of people and we had a good group of partners like Nvidia and like Prime of partners like Nvidia and like Prime of partners like Nvidia and like Prime Intellect where we could probably try to Intellect where we could probably try to Intellect where we could probably try to pre-train ourselves. pre-train ourselves. pre-train ourselves. Uh so, last year we did that. We kind of Uh so, last year we did that. We kind of Uh so, last year we did that. We kind of uh reoriented the entire company towards uh reoriented the entire company towards uh reoriented the entire company towards let's figure out how to pre-train a you let's figure out how to pre-train a you let's figure out how to pre-train a you know a 400 billion parameter model in 6 know a 400 billion parameter model in 6 know a 400 billion parameter model in 6 months. months. months. And a lot of people said it was And a lot of people said it was And a lot of people said it was impossible and by in many ways it was, impossible and by in many ways it was, impossible and by in many ways it was, but we figured it out and now we are an but we figured it out and now we are an but we figured it out and now we are an open model lab open model lab open model lab working with our wonderful partners working with our wonderful partners working with our wonderful partners and our customers to build and our customers to build and our customers to build western open models that are permissive western open models that are permissive western open models that are permissive and and and you can own those and customize them or you can own those and customize them or you can own those and customize them or run them wherever you want and and run them wherever you want and and run them wherever you want and and that's kind of where we're at right now. that's kind of where we're at right now. that's kind of where we're at right now. So thanks for having me. So thanks for having me. So thanks for having me. >> Yeah. >> Yeah. >> Yeah. >> And so I'm Chris Alexa. I work >> And so I'm Chris Alexa. I work >> And so I'm Chris Alexa. I work at Nvidia as a product research engineer at Nvidia as a product research engineer at Nvidia as a product research engineer and I support the Neumotron family of and I support the Neumotron family of and I support the Neumotron family of models. I think it's you know we we've models. I think it's you know we we've models. I think it's you know we we've talked a lot about why we do Neumotron talked a lot about why we do Neumotron talked a lot about why we do Neumotron but just to say it a few more times.

but just to say it a few more times. but just to say it a few more times. You know You know You know AI should be open open as in weights, AI should be open open as in weights, AI should be open open as in weights, data, training methodology, training data, training methodology, training data, training methodology, training frameworks. frameworks. frameworks. Really respect a lot of the work that uh Really respect a lot of the work that uh Really respect a lot of the work that uh the the the the two other people up here the the the the two other people up here the the the the two other people up here do because they they believe that very do because they they believe that very do because they they believe that very strongly as well. strongly as well. strongly as well. But the Neumotron family of models is But the Neumotron family of models is But the Neumotron family of models is focused on being as open as humanly focused on being as open as humanly focused on being as open as humanly possible. So possible. So possible. So we we have this understanding or belief we we have this understanding or belief we we have this understanding or belief that that that in order for AI to continue to grow and in order for AI to continue to grow and in order for AI to continue to grow and be useful to everybody, be useful to everybody, be useful to everybody, it has to be done in the open so that we it has to be done in the open so that we it has to be done in the open so that we can build off of each other, we can can build off of each other, we can can build off of each other, we can compound on each other. And part of what compound on each other. And part of what compound on each other. And part of what we do because we do because we do because team green, this is always true, is we team green, this is always true, is we team green, this is always true, is we we think that the we think that the we think that the the rate that you can squeeze tokens out the rate that you can squeeze tokens out the rate that you can squeeze tokens out of models is very important. So we kind of models is very important. So we kind of models is very important. So we kind of have this mantra that like faster of have this mantra that like faster of have this mantra that like faster models are smarter models and so a lot models are smarter models and so a lot models are smarter models and so a lot of the decisions we make when designing of the decisions we make when designing of the decisions we make when designing a model like Neumotron is built around a model like Neumotron is built around a model like Neumotron is built around how fast can we make it go.

how fast can we make it go. how fast can we make it go. As especially you are going to see in As especially you are going to see in As especially you are going to see in the next however many months local AI the next however many months local AI the next however many months local AI take off, take off, take off, we we need to make sure that models are we we need to make sure that models are we we need to make sure that models are well supported on well supported on well supported on uh hardware that doesn't just exist in uh hardware that doesn't just exist in uh hardware that doesn't just exist in massive buildings, you know, thousands massive buildings, you know, thousands massive buildings, you know, thousands of kilometers away from you. Uh and so of kilometers away from you. Uh and so of kilometers away from you. Uh and so that's uh that's uh that's uh you know, for AI to be very useful, it you know, for AI to be very useful, it you know, for AI to be very useful, it should be quick uh and and open. So, should be quick uh and and open. So, should be quick uh and and open. So, that's kind of the the vibe of that's kind of the the vibe of that's kind of the the vibe of Neumotron. Neumotron. Neumotron. >> Who makes those buildings with the >> Who makes those buildings with the >> Who makes those buildings with the massive processors? massive processors? massive processors? >> Oh, that's uh >> Oh, that's uh >> Oh, that's uh a lot of excellent people in the world a lot of excellent people in the world a lot of excellent people in the world that they use a lot of excellent that they use a lot of excellent that they use a lot of excellent hardware from a pretty cool company. hardware from a pretty cool company. hardware from a pretty cool company. Yeah, I heard I heard anyway. Yeah, I heard I heard anyway. Yeah, I heard I heard anyway. >> Yeah. And I'm Carter Abdallah, I'll be >> Yeah. And I'm Carter Abdallah, I'll be >> Yeah. And I'm Carter Abdallah, I'll be your moderator for today. Uh something your moderator for today. Uh something your moderator for today. Uh something that you know, we all kind of talked that you know, we all kind of talked that you know, we all kind of talked about is is this you know, building on about is is this you know, building on about is is this you know, building on top of each other, learning from others, top of each other, learning from others, top of each other, learning from others, whether it is people, you know, across whether it is people, you know, across whether it is people, you know, across the the big pond of the Pacific Ocean the the big pond of the Pacific Ocean the the big pond of the Pacific Ocean from us. Um but really it is kind of from us. Um but really it is kind of from us. Um but really it is kind of like a collaborative sort of research like a collaborative sort of research like a collaborative sort of research effort and I imagine that a lot of the effort and I imagine that a lot of the effort and I imagine that a lot of the people here in this room share that people here in this room share that people here in this room share that sentiment. But as it was brought up sentiment. But as it was brought up sentiment. But as it was brought up during the, you know, inaugural panel during the, you know, inaugural panel during the, you know, inaugural panel this morning in the state of the union, this morning in the state of the union, this morning in the state of the union, there there is a growing sentiment um there there is a growing sentiment um there there is a growing sentiment um potentially on the other side that that potentially on the other side that that potentially on the other side that that paints uh open source to be something paints uh open source to be something paints uh open source to be something that is actually more chaotic, that that is actually more chaotic, that that is actually more chaotic, that there there's less trust involved. And I there there's less trust involved. And I there there's less trust involved. And I think trust ultimately as Lucas, you and think trust ultimately as Lucas, you and think trust ultimately as Lucas, you and I were talking about before, um I were talking about before, um I were talking about before, um depending on who the party is and depending on who the party is and depending on who the party is and depending on what lens you're looking at depending on what lens you're looking at depending on what lens you're looking at at it from, I think it kind of means at it from, I think it kind of means at it from, I think it kind of means different things. But ultimately from different things. But ultimately from different things. But ultimately from from the the end consumer, the somebody from the the end consumer, the somebody from the the end consumer, the somebody who's using this intelligence, um or who's using this intelligence, um or who's using this intelligence, um or somebody who's you know, more of a somebody who's you know, more of a somebody who's you know, more of a business and is actually customizing

business and is actually customizing business and is actually customizing something to maybe monetize tokens in something to maybe monetize tokens in something to maybe monetize tokens in the in their business. Um can you the in their business. Um can you the in their business. Um can you comment, we'll start with you, Lucas, a comment, we'll start with you, Lucas, a comment, we'll start with you, Lucas, a bit on how open source and open source bit on how open source and open source bit on how open source and open source models are actually key to building that models are actually key to building that models are actually key to building that trust so that when these people walk out trust so that when these people walk out trust so that when these people walk out of this room and somebody does come at of this room and somebody does come at of this room and somebody does come at them with that other angle, they can them with that other angle, they can them with that other angle, they can they can sort of steel man this side. they can sort of steel man this side. they can sort of steel man this side. >> Certainly. You can weaponize any term uh >> Certainly. You can weaponize any term uh >> Certainly. You can weaponize any term uh and certainly trust is is has been and certainly trust is is has been and certainly trust is is has been weaponized, that word. Um and the reason weaponized, that word. Um and the reason weaponized, that word. Um and the reason I say that is because I say that is because I say that is because it means something in based on the it means something in based on the it means something in based on the context and with your speech you're context and with your speech you're context and with your speech you're speaking about it. Uh often speaking about it. Uh often speaking about it. Uh often in AI, people like to conflate trust in AI, people like to conflate trust in AI, people like to conflate trust with safety with safety with safety um um um and those are not the same thing. and those are not the same thing. and those are not the same thing. Uh, and I'm happy to speak on safety, Uh, and I'm happy to speak on safety, Uh, and I'm happy to speak on safety, uh, you know, later on. But when it uh, you know, later on. But when it uh, you know, later on. But when it comes to trust, um, comes to trust, um, comes to trust, um, I I think that, you know, you hear a lot I I think that, you know, you hear a lot I I think that, you know, you hear a lot from closed model providers or from closed model providers or from closed model providers or uh, politicians or people out in the uh, politicians or people out in the uh, politicians or people out in the space who are advocates for or uh, space who are advocates for or uh, space who are advocates for or uh, against open source that you can't trust against open source that you can't trust against open source that you can't trust these open models cuz you don't know these open models cuz you don't know these open models cuz you don't know what went into them.

what went into them. what went into them. Well, the same is true for these closed Well, the same is true for these closed Well, the same is true for these closed models, uh, even more so. Uh, the models, uh, even more so. Uh, the models, uh, even more so. Uh, the benefit of uh, uh, of of open models is benefit of uh, uh, of of open models is benefit of uh, uh, of of open models is that uh, we can very easily validate that uh, we can very easily validate that uh, we can very easily validate what is inside of them. what is inside of them. what is inside of them. Uh, they are you can there is a whole Uh, they are you can there is a whole Uh, they are you can there is a whole bunch of files with a whole bunch of bunch of files with a whole bunch of bunch of files with a whole bunch of matrices in there and you can view them matrices in there and you can view them matrices in there and you can view them and you can see the code that is running and you can see the code that is running and you can see the code that is running these models. You have implementations these models. You have implementations these models. You have implementations from Prime RL, vLLM, SG Lang, the from Prime RL, vLLM, SG Lang, the from Prime RL, vLLM, SG Lang, the provider themselves. These models are provider themselves. These models are provider themselves. These models are inherently trustworthy. You know much inherently trustworthy. You know much inherently trustworthy. You know much more about what's going on when you hit more about what's going on when you hit more about what's going on when you hit and talk to these models than you ever and talk to these models than you ever and talk to these models than you ever will what's going on when you hit an will what's going on when you hit an will what's going on when you hit an arbitrary API. Now, that being said, um, arbitrary API. Now, that being said, um, arbitrary API. Now, that being said, um, certainly there is fear that people can certainly there is fear that people can certainly there is fear that people can um, um, um, reduce reduce reduce uh, you know, the you can't trust that uh, you know, the you can't trust that uh, you know, the you can't trust that these models are writing safe code. these models are writing safe code. these models are writing safe code. Well, Well, Well, again, that is the same thing with any again, that is the same thing with any again, that is the same thing with any model. You need to you need to use your model. You need to you need to use your model. You need to you need to use your uh, your judgment and you need to make uh, your judgment and you need to make uh, your judgment and you need to make sure that you have the proper sure that you have the proper sure that you have the proper um, um, um, you know, safeguards in place and you're you know, safeguards in place and you're you know, safeguards in place and you're viewing the outputs of these models as viewing the outputs of these models as viewing the outputs of these models as the outputs of uh, the outputs of uh, the outputs of uh, an inherently random system that we are an inherently random system that we are an inherently random system that we are working very, very hard to make less and working very, very hard to make less and working very, very hard to make less and less random.

less random. less random. I think that uh, a telling thing is a I think that uh, a telling thing is a I think that uh, a telling thing is a lot of people said, "Well, you can't lot of people said, "Well, you can't lot of people said, "Well, you can't trust Chinese models. You can't trust trust Chinese models. You can't trust trust Chinese models. You can't trust Chinese models. You can't trust Chinese Chinese models. You can't trust Chinese Chinese models. You can't trust Chinese models." That was often uh, for the last models." That was often uh, for the last models." That was often uh, for the last few years meant you can't trust open few years meant you can't trust open few years meant you can't trust open models. Well, as soon as uh, Anthropic models. Well, as soon as uh, Anthropic models. Well, as soon as uh, Anthropic had to put Fable away and people had to put Fable away and people had to put Fable away and people realized that, "Oh, realized that, "Oh, realized that, "Oh, our access to these frontier systems our access to these frontier systems our access to these frontier systems might not be universal anymore. might not be universal anymore. might not be universal anymore. Uh, there's probably going to be a lot Uh, there's probably going to be a lot Uh, there's probably going to be a lot of checks and balances. You had a of checks and balances. You had a of checks and balances. You had a tremendous number of enterprises and tremendous number of enterprises and tremendous number of enterprises and developers and companies start going to developers and companies start going to developers and companies start going to these new Chinese models because they these new Chinese models because they these new Chinese models because they could trust that they would always have could trust that they would always have could trust that they would always have access to them. Um and so when when it access to them. Um and so when when it access to them. Um and so when when it comes out of trust in the way I view comes out of trust in the way I view comes out of trust in the way I view that word as it relates to open models that word as it relates to open models that word as it relates to open models is do I know that what I am running and is do I know that what I am running and is do I know that what I am running and can I be as sure as possible that when I can I be as sure as possible that when I can I be as sure as possible that when I send something to this model that I am send something to this model that I am send something to this model that I am going to get the output that I expect? going to get the output that I expect? going to get the output that I expect? Um and the only way uh currently to be Um and the only way uh currently to be Um and the only way uh currently to be 100% sure that what you are getting is 100% sure that what you are getting is 100% sure that what you are getting is what you were expecting is by hitting an what you were expecting is by hitting an what you were expecting is by hitting an open model either that you are running open model either that you are running open model either that you are running yourselves or you're working with a yourselves or you're working with a yourselves or you're working with a partner like Prim Indelec or RC or partner like Prim Indelec or RC or partner like Prim Indelec or RC or Nvidia Nvidia Nvidia to validate. So that's my take on the to validate. So that's my take on the to validate. So that's my take on the word trust.

word trust. word trust. >> I think too something you mentioned is >> I think too something you mentioned is >> I think too something you mentioned is like we don't get to know a lot about like we don't get to know a lot about like we don't get to know a lot about the data that goes into these models and the data that goes into these models and the data that goes into these models and that's something that I'm really happy that's something that I'm really happy that's something that I'm really happy you know that that we're trying to do you know that that we're trying to do you know that that we're trying to do which is not it's not you know the which is not it's not you know the which is not it's not you know the incentives don't exist for everyone to incentives don't exist for everyone to incentives don't exist for everyone to do this right so it's not something that do this right so it's not something that do this right so it's not something that I think is mandatory or should be I think is mandatory or should be I think is mandatory or should be mandatory thanks to the things that mandatory thanks to the things that mandatory thanks to the things that Lucas mentioned which is that it's Lucas mentioned which is that it's Lucas mentioned which is that it's rather straightforward to validate what rather straightforward to validate what rather straightforward to validate what data did go into a model without seeing data did go into a model without seeing data did go into a model without seeing the data sources originally but I'm the data sources originally but I'm the data sources originally but I'm happy that Nvidia continues to release happy that Nvidia continues to release happy that Nvidia continues to release data sets along with our models. Release data sets along with our models. Release data sets along with our models. Release environments along with our models to environments along with our models to environments along with our models to make sure that even even if you can't go make sure that even even if you can't go make sure that even even if you can't go through the work of determining what through the work of determining what through the work of determining what went into the model went into the model went into the model which you can do with the the weights which you can do with the the weights which you can do with the the weights alone for the most part you have like a alone for the most part you have like a alone for the most part you have like a spreadsheet you can look at that says spreadsheet you can look at that says spreadsheet you can look at that says here's you know here's you know here's you know a couple trillion tokens of this data a couple trillion tokens of this data a couple trillion tokens of this data set a couple trillion tokens here and I set a couple trillion tokens here and I set a couple trillion tokens here and I think that that helps to think that that helps to think that that helps to educate people on why educate people on why educate people on why it's much easier to trust open models it's much easier to trust open models it's much easier to trust open models than than than than models that we we don't get access than models that we we don't get access than models that we we don't get access to to to any of that.

any of that. any of that. >> helps people see what that data looks >> helps people see what that data looks >> helps people see what that data looks like. like. like. >> Yeah. >> Yeah. >> Yeah. >> You know if you don't have someone >> You know if you don't have someone >> You know if you don't have someone releasing it openly when someone says releasing it openly when someone says releasing it openly when someone says data is going in I mean data can take data is going in I mean data can take data is going in I mean data can take many different shapes. You can but you many different shapes. You can but you many different shapes. You can but you can go to Hugging Face you can go to can go to Hugging Face you can go to can go to Hugging Face you can go to Nvidia or you can go to Prime Intellect, Nvidia or you can go to Prime Intellect, Nvidia or you can go to Prime Intellect, or RC's or RC's or RC's uh HuggingFace. You can look under our uh HuggingFace. You can look under our uh HuggingFace. You can look under our data sets, and you can see exactly what data sets, and you can see exactly what data sets, and you can see exactly what that looks like, um and that can help that looks like, um and that can help that looks like, um and that can help you uh inform your priors on it. you uh inform your priors on it. you uh inform your priors on it. >> Yeah, I think that trust also um you >> Yeah, I think that trust also um you >> Yeah, I think that trust also um you know, there's some angle of of a know, there's some angle of of a know, there's some angle of of a reputation. Do I believe that your your reputation. Do I believe that your your reputation. Do I believe that your your intentions are pure? And I think that a intentions are pure? And I think that a intentions are pure? And I think that a lot of people, again, in this room lot of people, again, in this room lot of people, again, in this room believe that intelligence is kind of believe that intelligence is kind of believe that intelligence is kind of this next layer of almost, you know, this next layer of almost, you know, this next layer of almost, you know, infrastructure for for us to progress as infrastructure for for us to progress as infrastructure for for us to progress as a species, and I believe that everybody a species, and I believe that everybody a species, and I believe that everybody should have intelligence. So, uh on on should have intelligence. So, uh on on should have intelligence. So, uh on on that front, I want to uh hand it over to that front, I want to uh hand it over to that front, I want to uh hand it over to Vincent because uh you kind of have this Vincent because uh you kind of have this Vincent because uh you kind of have this um almost like founding thesis that you um almost like founding thesis that you um almost like founding thesis that you this this stack should be the open, this this stack should be the open, this this stack should be the open, right? The open superintelligence stack. right? The open superintelligence stack. right? The open superintelligence stack. You want everybody to have a lot more You want everybody to have a lot more You want everybody to have a lot more intelligence. Um can you talk about how intelligence. Um can you talk about how intelligence. Um can you talk about how uh the this is kind of moving into the uh the this is kind of moving into the uh the this is kind of moving into the era of control, but uh beyond just data era of control, but uh beyond just data era of control, but uh beyond just data sets, how important it is to have the sets, how important it is to have the sets, how important it is to have the knobs and dials of the of this industry knobs and dials of the of this industry knobs and dials of the of this industry also be uh available in an open way for also be uh available in an open way for also be uh available in an open way for people who are building this?

people who are building this? people who are building this? >> Yeah, like I I think it's a really >> Yeah, like I I think it's a really >> Yeah, like I I think it's a really important point is to set um so I can important point is to set um so I can important point is to set um so I can say it like be able to take those open say it like be able to take those open say it like be able to take those open models, like customize them, be able to models, like customize them, be able to models, like customize them, be able to like build on top of them. And I think like build on top of them. And I think like build on top of them. And I think like all the different components are like all the different components are like all the different components are going to it, like especially from like going to it, like especially from like going to it, like especially from like the pre-training to mid-training to the pre-training to mid-training to the pre-training to mid-training to post-training, I think post-training, I think post-training, I think >> [music] >> [music] >> [music] >> like need to be more accessible, right? >> like need to be more accessible, right? >> like need to be more accessible, right? Like so more people can also like take Like so more people can also like take Like so more people can also like take those uh amazing models and like make those uh amazing models and like make those uh amazing models and like make them work for their specific use cases. them work for their specific use cases. them work for their specific use cases. So, I think when we started like we we So, I think when we started like we we So, I think when we started like we we also took a look at the whole stack that also took a look at the whole stack that also took a look at the whole stack that was out there and and tried to figure was out there and and tried to figure was out there and and tried to figure out like what is missing for ourselves out like what is missing for ourselves out like what is missing for ourselves to train open models and for like to train open models and for like to train open models and for like helping our partners to do so. And a lot helping our partners to do so. And a lot helping our partners to do so. And a lot of this was around the RL and of this was around the RL and of this was around the RL and post-training stack. So, we basically post-training stack. So, we basically post-training stack. So, we basically went deep into building out like a lot went deep into building out like a lot went deep into building out like a lot of infra around that, like around our of infra around that, like around our of infra around that, like around our environment, Evals, around like making environment, Evals, around like making environment, Evals, around like making it much more accessible to do it much more accessible to do it much more accessible to do post-training also because it's like the post-training also because it's like the post-training also because it's like the most economically viable way to maybe most economically viable way to maybe most economically viable way to maybe like customize those models, to take an like customize those models, to take an like customize those models, to take an open model, and um to have like a open model, and um to have like a open model, and um to have like a specific Eval environment, and the um specific Eval environment, and the um specific Eval environment, and the um specific domain and dimension that you specific domain and dimension that you specific domain and dimension that you want to improve it on. And this This of want to improve it on. And this This of want to improve it on. And this This of like what we're are doing with Prime like what we're are doing with Prime like what we're are doing with Prime Intel now, is like enabling people to Intel now, is like enabling people to Intel now, is like enabling people to post train specialized agentic models.

post train specialized agentic models. post train specialized agentic models. Um so, being able to take models like um Um so, being able to take models like um Um so, being able to take models like um Trinity, for example, from RC or Trinity, for example, from RC or Trinity, for example, from RC or Nematron or others and and specialize Nematron or others and and specialize Nematron or others and and specialize them, post train them for the use cases them, post train them for the use cases them, post train them for the use cases that ultimately enterprises care about. that ultimately enterprises care about. that ultimately enterprises care about. So, good example of this was like a So, good example of this was like a So, good example of this was like a company like, for example, Ramp or Saber company like, for example, Ramp or Saber company like, for example, Ramp or Saber to like take an open model and like to like take an open model and like to like take an open model and like specialize it to automate finance within specialize it to automate finance within specialize it to automate finance within like a week or two to get like better like a week or two to get like better like a week or two to get like better performance than like Opus at a fraction performance than like Opus at a fraction performance than like Opus at a fraction of the cost of Haiku. And I think really of the cost of Haiku. And I think really of the cost of Haiku. And I think really this Pareto frontier of like being able this Pareto frontier of like being able this Pareto frontier of like being able to create these specialized models that to create these specialized models that to create these specialized models that are much better than the frontier, but are much better than the frontier, but are much better than the frontier, but also faster, cheaper. Um I think it's also faster, cheaper. Um I think it's also faster, cheaper. Um I think it's like a key thing enterprises care about like a key thing enterprises care about like a key thing enterprises care about increasingly. It's really like just increasingly. It's really like just increasingly. It's really like just making it work for their use cases, making it work for their use cases, making it work for their use cases, basically. basically. basically. >> If you go back to trust, it's how you >> If you go back to trust, it's how you >> If you go back to trust, it's how you can make your CFO trust you by knowing can make your CFO trust you by knowing can make your CFO trust you by knowing exactly how much something's going to exactly how much something's going to exactly how much something's going to cost all the time. cost all the time. cost all the time. Um that's That is increasingly becoming Um that's That is increasingly becoming Um that's That is increasingly becoming very important is very important is very important is uh you hear a lot, you know, all these uh you hear a lot, you know, all these uh you hear a lot, you know, all these companies have unbelievably large token companies have unbelievably large token companies have unbelievably large token spend and spend and spend and um they're having to cut back on their um they're having to cut back on their um they're having to cut back on their Opus usage because they burned through Opus usage because they burned through Opus usage because they burned through it all in a couple months. it all in a couple months. it all in a couple months. Um and that is going to continue to be a Um and that is going to continue to be a Um and that is going to continue to be a problem because yes, the cost of an problem because yes, the cost of an problem because yes, the cost of an individual token has come down individual token has come down individual token has come down drastically. You can look at it, you drastically. You can look at it, you drastically. You can look at it, you know, the difference between GPT-4 when know, the difference between GPT-4 when know, the difference between GPT-4 when it first launched and GPT-5.5 is is is it first launched and GPT-5.5 is is is it first launched and GPT-5.5 is is is much, much cheaper per token, but at the much, much cheaper per token, but at the much, much cheaper per token, but at the same time the amount of tokens in an same time the amount of tokens in an same time the amount of tokens in an individual session has gone up individual session has gone up individual session has gone up exponentially as well. So, we're kind of exponentially as well. So, we're kind of exponentially as well. So, we're kind of um we're spending more um we're spending more um we're spending more uh on a uh on a uh on a on a total session. And so, the ability on a total session. And so, the ability on a total session. And so, the ability to to to uh bring in-house or or or at least work uh bring in-house or or or at least work uh bring in-house or or or at least work with partners to ensure that you are with partners to ensure that you are with partners to ensure that you are controlling your cost and you're not at

controlling your cost and you're not at controlling your cost and you're not at the whims of uh when a company releases the whims of uh when a company releases the whims of uh when a company releases a newer model uh that might be better, a newer model uh that might be better, a newer model uh that might be better, but also more expensive. They might but also more expensive. They might but also more expensive. They might deprecate a model. deprecate a model. deprecate a model. Um owning that and being sure that, you Um owning that and being sure that, you Um owning that and being sure that, you know, same way is what you what out know, same way is what you what out know, same way is what you what out input goes in, you know what output's input goes in, you know what output's input goes in, you know what output's going to come out. In the same way, going to come out. In the same way, going to come out. In the same way, um um um when it you know, an input goes in, how when it you know, an input goes in, how when it you know, an input goes in, how much it's going to cost. Uh having much it's going to cost. Uh having much it's going to cost. Uh having assurance on that's is important, too. assurance on that's is important, too. assurance on that's is important, too. >> Yeah, and maybe like one thing to add to >> Yeah, and maybe like one thing to add to >> Yeah, and maybe like one thing to add to this is like almost like I like this new this is like almost like I like this new this is like almost like I like this new term of like instead of speaking about term of like instead of speaking about term of like instead of speaking about token maxing, you know, speaking more token maxing, you know, speaking more token maxing, you know, speaking more about like the outcome maxing of like about like the outcome maxing of like about like the outcome maxing of like >> Yeah. >> Yeah. >> Yeah. >> Ultimately, it's like you you want to >> Ultimately, it's like you you want to >> Ultimately, it's like you you want to have like more than a dollar worth of have like more than a dollar worth of have like more than a dollar worth of value come out of like a dollar of input value come out of like a dollar of input value come out of like a dollar of input and I think this is sort of like Jevons and I think this is sort of like Jevons and I think this is sort of like Jevons paradox of like if you can create more paradox of like if you can create more paradox of like if you can create more value value value for like your flop for your GPU, for like your flop for your GPU, for like your flop for your GPU, basically. I think this is sort of like basically. I think this is sort of like basically. I think this is sort of like how you get like the most adoption also how you get like the most adoption also how you get like the most adoption also of like agentic models. Like if they can of like agentic models. Like if they can of like agentic models. Like if they can like be able to create as much value as like be able to create as much value as like be able to create as much value as possible. I think the cheaper those possible. I think the cheaper those possible. I think the cheaper those models get, the more usage they'll get models get, the more usage they'll get models get, the more usage they'll get like for those specific use cases. like for those specific use cases. like for those specific use cases. >> It's funny you say that. I have a >> It's funny you say that. I have a >> It's funny you say that. I have a and I think a lot of us in this room, and I think a lot of us in this room, and I think a lot of us in this room, but especially on this panel believe but especially on this panel believe but especially on this panel believe this to be so that the most meaningful this to be so that the most meaningful this to be so that the most meaningful AI applications the next couple years, AI applications the next couple years, AI applications the next couple years, even this year, are the ones where the even this year, are the ones where the even this year, are the ones where the harness and the model and the product, harness and the model and the product, harness and the model and the product, they all kind of blend together. If you they all kind of blend together. If you they all kind of blend together. If you think back to at least for me, the first think back to at least for me, the first think back to at least for me, the first like truly like truly like truly game-changing agentic experience I had game-changing agentic experience I had game-changing agentic experience I had was when Deep Research from OpenAI. And was when Deep Research from OpenAI. And was when Deep Research from OpenAI. And that was because they spent a tremendous that was because they spent a tremendous that was because they spent a tremendous amount of time doing reinforcement amount of time doing reinforcement amount of time doing reinforcement learning learning learning on O3 with test time compute to do these on O3 with test time compute to do these on O3 with test time compute to do these longer running research tasks that

longer running research tasks that longer running research tasks that people had tried previously, but they people had tried previously, but they people had tried previously, but they were kind of just doing a for loop over were kind of just doing a for loop over were kind of just doing a for loop over search. Whereas I kept coming back to to search. Whereas I kept coming back to to search. Whereas I kept coming back to to Deep Research. And Deep Research. And Deep Research. And you know, you saw for a very long time you know, you saw for a very long time you know, you saw for a very long time that OpenAI and Anthropic and Google, that OpenAI and Anthropic and Google, that OpenAI and Anthropic and Google, when they'd release a new product, when they'd release a new product, when they'd release a new product, they'd release a custom version of their they'd release a custom version of their they'd release a custom version of their model for that product. And if they're model for that product. And if they're model for that product. And if they're doing that, if their off-the-shelf GPT-5 doing that, if their off-the-shelf GPT-5 doing that, if their off-the-shelf GPT-5 isn't good enough for, you know, their isn't good enough for, you know, their isn't good enough for, you know, their Atlas web browser, why should it be good Atlas web browser, why should it be good Atlas web browser, why should it be good enough for our apps? enough for our apps? enough for our apps? And that's why I appreciate the work And that's why I appreciate the work And that's why I appreciate the work that, you know, Vincent and Nvidia are that, you know, Vincent and Nvidia are that, you know, Vincent and Nvidia are doing for for giving people the tools to doing for for giving people the tools to doing for for giving people the tools to customize their own model customize their own model customize their own model and allowing us to focus on how we get a and allowing us to focus on how we get a and allowing us to focus on how we get a good model to start from. So it really good model to start from. So it really good model to start from. So it really is is is you know, it's it's extremely important you know, it's it's extremely important you know, it's it's extremely important as you look at developing applications as you look at developing applications as you look at developing applications and experiences over over next few years and experiences over over next few years and experiences over over next few years that you're taking into account that you that you're taking into account that you that you're taking into account that you can can can uh make the model do something that uh make the model do something that uh make the model do something that maybe your harness isn't fully able to maybe your harness isn't fully able to maybe your harness isn't fully able to do alone. do alone. do alone. >> Well, that's something I think that's >> Well, that's something I think that's >> Well, that's something I think that's really important to just like reiterate, really important to just like reiterate, really important to just like reiterate, right? I mean, like Nemotron's great, I right? I mean, like Nemotron's great, I right? I mean, like Nemotron's great, I love it. Trinity's great, I love it.

love it. Trinity's great, I love it. love it. Trinity's great, I love it. Like Like Like we design a model that's supposed to be we design a model that's supposed to be we design a model that's supposed to be as good as it can be across a number of as good as it can be across a number of as good as it can be across a number of harnesses, right? You can see this in harnesses, right? You can see this in harnesses, right? You can see this in the technical report. The idea is like the technical report. The idea is like the technical report. The idea is like we want the the model to work as well as we want the the model to work as well as we want the the model to work as well as it can in Pi compared to you know Hermes it can in Pi compared to you know Hermes it can in Pi compared to you know Hermes compared whatever you're using, right? compared whatever you're using, right? compared whatever you're using, right? But like you're you're not using all of But like you're you're not using all of But like you're you're not using all of these tools at once. You're using one of these tools at once. You're using one of these tools at once. You're using one of these tools. And so when you have open these tools. And so when you have open these tools. And so when you have open models, you can do things like news models, you can do things like news models, you can do things like news research can create a post train of research can create a post train of research can create a post train of whatever model for their harness, right? whatever model for their harness, right? whatever model for their harness, right? That you know will be That you know will be That you know will be extra good. And you know, this this this extra good. And you know, this this this extra good. And you know, this this this thing from from from you know, I I can't thing from from from you know, I I can't thing from from from you know, I I can't remember who who who originally wrote remember who who who originally wrote remember who who who originally wrote it, but this idea of like the mismanaged it, but this idea of like the mismanaged it, but this idea of like the mismanaged genius, right? We're we're leaving a lot genius, right? We're we're leaving a lot genius, right? We're we're leaving a lot of like of like of like a lot of important capability on the a lot of important capability on the a lot of important capability on the table table table because we're just not we're not fitting because we're just not we're not fitting because we're just not we're not fitting the models into the harness, right? You the models into the harness, right? You the models into the harness, right? You can do a bunch of stuff with closed can do a bunch of stuff with closed can do a bunch of stuff with closed models. Like you can change your prompts models. Like you can change your prompts models. Like you can change your prompts and your skills and all kinds of other and your skills and all kinds of other and your skills and all kinds of other nido things, right? But nothing will let nido things, right? But nothing will let nido things, right? But nothing will let you get the the the level of you get the the the level of you get the the the level of customization or customizability that customization or customizability that customization or customizability that you can achieve with open models. And I you can achieve with open models. And I you can achieve with open models. And I think that is something that is going to think that is something that is going to think that is something that is going to become increasingly and increasingly become increasingly and increasingly become increasingly and increasingly more important, especially thanks to more important, especially thanks to more important, especially thanks to folks like the others on the panel where folks like the others on the panel where folks like the others on the panel where you know, I can just straight drop like you know, I can just straight drop like you know, I can just straight drop like my favorite coding and you know, an my favorite coding and you know, an my favorite coding and you know, an agent environment agent environment agent environment spin up the CLI and suddenly my my model spin up the CLI and suddenly my my model spin up the CLI and suddenly my my model feels way better with very little feels way better with very little feels way better with very little effort, right? Like that is that is effort, right? Like that is that is effort, right? Like that is that is something that is

something that is something that is already at our fingertips and it is only already at our fingertips and it is only already at our fingertips and it is only going to get easier and easier as as going to get easier and easier as as going to get easier and easier as as time goes on. time goes on. time goes on. >> Yeah. >> Yeah. >> Yeah. >> And it's maybe also the most concrete >> And it's maybe also the most concrete >> And it's maybe also the most concrete like info for the builders in the like info for the builders in the like info for the builders in the audience like call to action of like if audience like call to action of like if audience like call to action of like if you kind of want to build the next like you kind of want to build the next like you kind of want to build the next like cloud code, the next like cursor or cloud code, the next like cursor or cloud code, the next like cursor or perplexity, I think the easiest way to perplexity, I think the easiest way to perplexity, I think the easiest way to get started is like take the best open get started is like take the best open get started is like take the best open model like and and then post rate on model like and and then post rate on model like and and then post rate on your harness like that you care about, your harness like that you care about, your harness like that you care about, right? Like basically like create a right? Like basically like create a right? Like basically like create a product that is like truly AI native, product that is like truly AI native, product that is like truly AI native, right? And I think this is sort of like right? And I think this is sort of like right? And I think this is sort of like I think one of the most exciting like I think one of the most exciting like I think one of the most exciting like unlocks for builders um like out here. unlocks for builders um like out here. unlocks for builders um like out here. >> That's a big thing about the the framing >> That's a big thing about the the framing >> That's a big thing about the the framing of control is is um of control is is um of control is is um similar to like, you know, when when the similar to like, you know, when when the similar to like, you know, when when the cloud um explosion started to happen in cloud um explosion started to happen in cloud um explosion started to happen in the in the, you know, the late uh 2000s, the in the, you know, the late uh 2000s, the in the, you know, the late uh 2000s, early 2010s and the um social media early 2010s and the um social media early 2010s and the um social media world kind of took off and apps became world kind of took off and apps became world kind of took off and apps became extremely popular and more, you know, extremely popular and more, you know, extremely popular and more, you know, cloud-based and and, you know, managed cloud-based and and, you know, managed cloud-based and and, you know, managed by these these bigger companies, uh you by these these bigger companies, uh you by these these bigger companies, uh you know, you the data that they were know, you the data that they were know, you the data that they were collecting from you, whether anonymized collecting from you, whether anonymized collecting from you, whether anonymized or not, was how they were monetizing or not, was how they were monetizing or not, was how they were monetizing their platform through ads or or or or their platform through ads or or or or their platform through ads or or or or in other ways. Uh and a very similar in other ways. Uh and a very similar in other ways. Uh and a very similar thing has always been happening, but I thing has always been happening, but I thing has always been happening, but I think it's becoming clear to people in think it's becoming clear to people in think it's becoming clear to people in in the space is that the the data that in the space is that the the data that in the space is that the the data that people get from from you using these people get from from you using these people get from from you using these models is how these companies largely models is how these companies largely models is how these companies largely make their models better, whether it's make their models better, whether it's make their models better, whether it's through actually training on that data through actually training on that data through actually training on that data or um by using it as a signal for what or um by using it as a signal for what or um by using it as a signal for what data they they go out and find or data they they go out and find or data they they go out and find or generate to train. And uh with closed generate to train. And uh with closed generate to train. And uh with closed models, there are terms of service that models, there are terms of service that models, there are terms of service that keep you from being able to

keep you from being able to keep you from being able to um well, um well, um well, I could get into a a discussion about I could get into a a discussion about I could get into a a discussion about what terms of service is an agreement what terms of service is an agreement what terms of service is an agreement between you and the provider, it's not a between you and the provider, it's not a between you and the provider, it's not a legal uh anyway, but you you you legal uh anyway, but you you you legal uh anyway, but you you you shouldn't be training on a, you know, a shouldn't be training on a, you know, a shouldn't be training on a, you know, a Claude Opus output or a Fable output or Claude Opus output or a Fable output or Claude Opus output or a Fable output or a GPT-5 output, and they do a lot to try a GPT-5 output, and they do a lot to try a GPT-5 output, and they do a lot to try to obfuscate to to make that not great to obfuscate to to make that not great to obfuscate to to make that not great for you. If you're using an open model, for you. If you're using an open model, for you. If you're using an open model, you can save all of those traces. All of you can save all of those traces. All of you can save all of those traces. All of those traces of you using it inside of those traces of you using it inside of those traces of you using it inside of your harness uh that will allow you over your harness uh that will allow you over your harness uh that will allow you over time to if you say, "Hey, I want to go time to if you say, "Hey, I want to go time to if you say, "Hey, I want to go train a custom model," you can take all train a custom model," you can take all train a custom model," you can take all of that, and again, either use it to of that, and again, either use it to of that, and again, either use it to directly do like fine-tuning on a directly do like fine-tuning on a directly do like fine-tuning on a smaller model, so you're not spending as smaller model, so you're not spending as smaller model, so you're not spending as much, or to have a model help you find much, or to have a model help you find much, or to have a model help you find signals, so that you can go out and use signals, so that you can go out and use signals, so that you can go out and use Verifiers or Nemo RL or Nemo Gym to Verifiers or Nemo RL or Nemo Gym to Verifiers or Nemo RL or Nemo Gym to create these environments, so that you create these environments, so that you create these environments, so that you can hill climb and make your models can hill climb and make your models can hill climb and make your models better. So, better. So, better. So, um as much as using open models is like um as much as using open models is like um as much as using open models is like owning your stack, owning your owning your stack, owning your owning your stack, owning your intelligence, it's also owning your intelligence, it's also owning your intelligence, it's also owning your outputs, right? Owning your data. That's outputs, right? Owning your data. That's outputs, right? Owning your data. That's going to be extremely important, too. going to be extremely important, too. going to be extremely important, too. >> I do want to I do want to plug the >> I do want to I do want to plug the >> I do want to I do want to plug the license for a second. So, license for a second. So, license for a second. So, >> [laughter] >> [laughter] >> [laughter] >> uh >> uh >> uh so, AI is very different than so, AI is very different than so, AI is very different than traditional software, traditional software, traditional software, uh uh uh which is why recently uh Nemotron, as which is why recently uh Nemotron, as which is why recently uh Nemotron, as well as uh Trinity, I know, uh has well as uh Trinity, I know, uh has well as uh Trinity, I know, uh has adopted the Open MDW model uh data adopted the Open MDW model uh data adopted the Open MDW model uh data weights uh license. Uh the idea is like, weights uh license. Uh the idea is like, weights uh license. Uh the idea is like, we we need a way to really make it very we we need a way to really make it very we we need a way to really make it very clear in the license that you can use clear in the license that you can use clear in the license that you can use the outputs to produce a model. You can the outputs to produce a model. You can the outputs to produce a model. You can use the outputs to train uh right? All use the outputs to train uh right? All use the outputs to train uh right? All of these TOSs and stuff like that that of these TOSs and stuff like that that of these TOSs and stuff like that that that that that have language that's

that that that have language that's that that that have language that's meant to dissuade you to do that. Uh we meant to dissuade you to do that. Uh we meant to dissuade you to do that. Uh we wanted to make sure there's a license wanted to make sure there's a license wanted to make sure there's a license that exists that not encourages you, but that exists that not encourages you, but that exists that not encourages you, but makes it crystal clear that it is it is makes it crystal clear that it is it is makes it crystal clear that it is it is permitted it is permissible. Uh and and permitted it is permissible. Uh and and permitted it is permissible. Uh and and I think, you know, I think, you know, I think, you know, the licenses maturing, right? To fit the the licenses maturing, right? To fit the the licenses maturing, right? To fit the use case better should be extremely use case better should be extremely use case better should be extremely positive signal uh for for the way that positive signal uh for for the way that positive signal uh for for the way that the ecosystem is thinking about open the ecosystem is thinking about open the ecosystem is thinking about open models. models. models. Uh to to the fact where even even the Uh to to the fact where even even the Uh to to the fact where even even the lawyers are on board. lawyers are on board. lawyers are on board. >> know how much lawyers cost? >> know how much lawyers cost? >> know how much lawyers cost? >> [laughter] >> [laughter] >> [laughter] >> Yeah. >> Yeah. >> Yeah. >> Yeah, it's uh >> Yeah, it's uh >> Yeah, it's uh as we move on to the, you know, kind of as we move on to the, you know, kind of as we move on to the, you know, kind of the third topic which here is of course the third topic which here is of course the third topic which here is of course optimization, um something that you optimization, um something that you optimization, um something that you know, we've implicitly said, but haven't know, we've implicitly said, but haven't know, we've implicitly said, but haven't said it quite explicitly yet is uh said it quite explicitly yet is uh said it quite explicitly yet is uh effectively that I I think that for a effectively that I I think that for a effectively that I I think that for a lot of people there's this preconceived lot of people there's this preconceived lot of people there's this preconceived notion that when you're deciding to use notion that when you're deciding to use notion that when you're deciding to use an open model for whatever the use case, an open model for whatever the use case, an open model for whatever the use case, um there there are the tradeoffs that um there there are the tradeoffs that um there there are the tradeoffs that come in the form of performance at the come in the form of performance at the come in the form of performance at the benefit of getting things like, you benefit of getting things like, you benefit of getting things like, you know, maybe data sovereignty and so know, maybe data sovereignty and so know, maybe data sovereignty and so forth. Um, but what we are now talking forth. Um, but what we are now talking forth. Um, but what we are now talking about is that with the with the right about is that with the with the right about is that with the with the right customization and optimization, customization and optimization, customization and optimization, depending on the the use case that depending on the the use case that depending on the the use case that you're and the harness that you're you're and the harness that you're you're and the harness that you're applying it to, you can actually exceed applying it to, you can actually exceed applying it to, you can actually exceed and build the model against the tool to and build the model against the tool to and build the model against the tool to to get better performance than even to get better performance than even to get better performance than even frontier models. Um, I'd love to hear a frontier models. Um, I'd love to hear a frontier models. Um, I'd love to hear a little bit more about the cuz I think little bit more about the cuz I think little bit more about the cuz I think that uh another thing that we would that uh another thing that we would that uh another thing that we would probably agree on is that the the probably agree on is that the the probably agree on is that the the current level of intelligence um, current level of intelligence um, current level of intelligence um, already has so much uh left to diffuse already has so much uh left to diffuse already has so much uh left to diffuse into society. And so, where are those

into society. And so, where are those into society. And so, where are those areas where that diffusion is is is areas where that diffusion is is is areas where that diffusion is is is happening in the specific industries? I happening in the specific industries? I happening in the specific industries? I know, for example, things around uh know, for example, things around uh know, for example, things around uh again, kind of fundamental pieces of again, kind of fundamental pieces of again, kind of fundamental pieces of infrastructure, whether it's like infrastructure, whether it's like infrastructure, whether it's like browser use um and how you can start to browser use um and how you can start to browser use um and how you can start to train models to to be able to use uh you train models to to be able to use uh you train models to to be able to use uh you know, the the internet better when know, the the internet better when know, the the internet better when looking at a computer and so on and so looking at a computer and so on and so looking at a computer and so on and so forth. So, what are the what are the forth. So, what are the what are the forth. So, what are the what are the some of those examples to where you some of those examples to where you some of those examples to where you think that the uh post-training of open think that the uh post-training of open think that the uh post-training of open models will uh models will uh models will uh see new use cases basically unlock see new use cases basically unlock see new use cases basically unlock compared to just paying full price for compared to just paying full price for compared to just paying full price for the the frontier models? the the frontier models? the the frontier models? >> Yeah, I think I like like I can start on >> Yeah, I think I like like I can start on >> Yeah, I think I like like I can start on this. Like this. Like this. Like I think the the power what that we've I think the the power what that we've I think the the power what that we've seen with a lot of different customers seen with a lot of different customers seen with a lot of different customers is is really kind of this idea that like is is really kind of this idea that like is is really kind of this idea that like if you want to make a specific use case if you want to make a specific use case if you want to make a specific use case work, like we can take the example of work, like we can take the example of work, like we can take the example of like if you want to figure out like a like if you want to figure out like a like if you want to figure out like a way that agents can actually automate way that agents can actually automate way that agents can actually automate your text. Like the the the most like your text. Like the the the most like your text. Like the the the most like concrete way you can do it today really concrete way you can do it today really concrete way you can do it today really is like build an RL environment for that is like build an RL environment for that is like build an RL environment for that use case. Like train on it and then use case. Like train on it and then use case. Like train on it and then deploy it into production with those deploy it into production with those deploy it into production with those users, right? Like let's say with a a users, right? Like let's say with a a users, right? Like let's say with a a million accountants that then now use million accountants that then now use million accountants that then now use this agent to ultimately get it towards this agent to ultimately get it towards this agent to ultimately get it towards full autonomy. It's a bit like almost full autonomy. It's a bit like almost full autonomy. It's a bit like almost like Tesla's levels towards full like Tesla's levels towards full like Tesla's levels towards full autonomy where like you kind of need to autonomy where like you kind of need to autonomy where like you kind of need to deploy it into like do the last mile of deploy it into like do the last mile of deploy it into like do the last mile of actually like training for that specific actually like training for that specific actually like training for that specific use case, but then also use case, but then also use case, but then also um deploying it to those specific users, um deploying it to those specific users, um deploying it to those specific users, right? So, like there's a reason why right? So, like there's a reason why right? So, like there's a reason why like like like uh chatbot isn't good at self-driving uh chatbot isn't good at self-driving uh chatbot isn't good at self-driving because like it's not trained on that.

because like it's not trained on that. because like it's not trained on that. It's not deployed into that context, It's not deployed into that context, It's not deployed into that context, right? And like I think it's the same right? And like I think it's the same right? And like I think it's the same even for these specific like knowledge even for these specific like knowledge even for these specific like knowledge work use cases, where it's like if you work use cases, where it's like if you work use cases, where it's like if you want to have the perfect like financial want to have the perfect like financial want to have the perfect like financial agent, it's much more likely that you'll agent, it's much more likely that you'll agent, it's much more likely that you'll be able to get there if you have like be able to get there if you have like be able to get there if you have like our environment for that use case, if our environment for that use case, if our environment for that use case, if you deploy it into production, for you deploy it into production, for you deploy it into production, for example, as a bank, right? Like to example, as a bank, right? Like to example, as a bank, right? Like to millions of customers, than if you're millions of customers, than if you're millions of customers, than if you're there's like one got model chatbot. Like there's like one got model chatbot. Like there's like one got model chatbot. Like And I think this is sort of like what And I think this is sort of like what And I think this is sort of like what we've seen now with a lot of verticals we've seen now with a lot of verticals we've seen now with a lot of verticals and customers that like um there's like and customers that like um there's like and customers that like um there's like a huge unlock there to um really go into a huge unlock there to um really go into a huge unlock there to um really go into this like specialized domains, post this like specialized domains, post this like specialized domains, post train on them, deploy into them, and train on them, deploy into them, and train on them, deploy into them, and then continuously learn from production then continuously learn from production then continuously learn from production traces. So, we work with like some um traces. So, we work with like some um traces. So, we work with like some um also big AI natives on things like also big AI natives on things like also big AI natives on things like computer use, where ultimately having computer use, where ultimately having computer use, where ultimately having like millions of of traces from like millions of of traces from like millions of of traces from production data really can help you to production data really can help you to production data really can help you to to um continuously improve uh those to um continuously improve uh those to um continuously improve uh those those agents. And I think this kind of those agents. And I think this kind of those agents. And I think this kind of applies to almost every single domain, applies to almost every single domain, applies to almost every single domain, and I think it's sort of the the white and I think it's sort of the the white and I think it's sort of the the white pill for like the AI application pill for like the AI application pill for like the AI application builders and and the AI startups to builders and and the AI startups to builders and and the AI startups to actually have a a huge opportunity to actually have a a huge opportunity to actually have a a huge opportunity to build kind of their modes and and to get build kind of their modes and and to get build kind of their modes and and to get to this data flywheel to this data flywheel to this data flywheel um of like specialized models even in a um of like specialized models even in a um of like specialized models even in a broader sense. Like just going after broader sense. Like just going after broader sense. Like just going after like let's say computer use agents, like let's say computer use agents, like let's say computer use agents, right? And I think um yeah, this is right? And I think um yeah, this is right? And I think um yeah, this is something something something where I think um we're just seeing a lot where I think um we're just seeing a lot where I think um we're just seeing a lot of like movement especially now with of like movement especially now with of like movement especially now with like all models catching up to the like all models catching up to the like all models catching up to the frontier.

frontier. frontier. Um and I think the other piece is like Um and I think the other piece is like Um and I think the other piece is like optimization, where like I think um like optimization, where like I think um like optimization, where like I think um like GM is a great example or like also like GM is a great example or like also like GM is a great example or like also like Trinity and NeMo Triton is like you have Trinity and NeMo Triton is like you have Trinity and NeMo Triton is like you have the whole ecosystem sort of like driving the whole ecosystem sort of like driving the whole ecosystem sort of like driving down the cost and optimizing it further, down the cost and optimizing it further, down the cost and optimizing it further, right? Like like we very heavily work right? Like like we very heavily work right? Like like we very heavily work like very closely with all the teams like very closely with all the teams like very closely with all the teams here, but then also um like deeply also here, but then also um like deeply also here, but then also um like deeply also with Nvidia and with teams like ViLM to with Nvidia and with teams like ViLM to with Nvidia and with teams like ViLM to really drive down uh the cost and and really drive down uh the cost and and really drive down uh the cost and and make the for example inference and make the for example inference and make the for example inference and training for models like GM or like training for models like GM or like training for models like GM or like models like Trinity and NeMo Triton models like Trinity and NeMo Triton models like Trinity and NeMo Triton extremely efficient, so you can extremely efficient, so you can extremely efficient, so you can basically drive down the cost like basically drive down the cost like basically drive down the cost like further and further. And I think this is further and further. And I think this is further and further. And I think this is something you don't obviously get with something you don't obviously get with something you don't obviously get with the closed APIs, where like they have the closed APIs, where like they have the closed APIs, where like they have like a huge margin on top. Like they like a huge margin on top. Like they like a huge margin on top. Like they might drive down the optimization, but might drive down the optimization, but might drive down the optimization, but then might not pass through those then might not pass through those then might not pass through those savings. So, I think in general, like savings. So, I think in general, like savings. So, I think in general, like the open models are only getting through the open models are only getting through the open models are only getting through the open ecosystem like more and more the open ecosystem like more and more the open ecosystem like more and more efficient like efficient like efficient like and and cheaper and cheaper to run and and and cheaper and cheaper to run and and and cheaper and cheaper to run and train on. So, I think there's like this train on. So, I think there's like this train on. So, I think there's like this element as well. element as well. element as well. >> I think too like >> I think too like >> I think too like a couple things that I a couple things that I a couple things that I I want to make sure we're very clear I want to make sure we're very clear I want to make sure we're very clear about is you you like most people about is you you like most people about is you you like most people probably do not need frontier level probably do not need frontier level probably do not need frontier level intelligence for like 90% of their intelligence for like 90% of their intelligence for like 90% of their tasks, right? Like tasks, right? Like tasks, right? Like like not to say that you're not doing like not to say that you're not doing like not to say that you're not doing cool smart stuff. Not to say that I'm cool smart stuff. Not to say that I'm cool smart stuff. Not to say that I'm I'm sitting here trying to do not cool I'm sitting here trying to do not cool I'm sitting here trying to do not cool smart stuff, but like a lot of the time smart stuff, but like a lot of the time smart stuff, but like a lot of the time these models are just overkill or they these models are just overkill or they these models are just overkill or they have like this really smooth have like this really smooth have like this really smooth you know you know you know capability horizon that means they're capability horizon that means they're capability horizon that means they're they're also quite good at chemistry, they're also quite good at chemistry, they're also quite good at chemistry, but like most people are using models to but like most people are using models to but like most people are using models to do one or two things very well.

do one or two things very well. do one or two things very well. And open models let you And open models let you And open models let you choose those one or two things and then choose those one or two things and then choose those one or two things and then make the model just very good at those make the model just very good at those make the model just very good at those things at the expense of at the expense things at the expense of at the expense things at the expense of at the expense sorry of almost everything else. And sorry of almost everything else. And sorry of almost everything else. And that that is that that is that that is great. I mean that's exactly what we great. I mean that's exactly what we great. I mean that's exactly what we should be doing, right? should be doing, right? should be doing, right? To to to use this model that is hyper To to to use this model that is hyper To to to use this model that is hyper generalized and able to you know perform generalized and able to you know perform generalized and able to you know perform well across like 90 different axes is is well across like 90 different axes is is well across like 90 different axes is is dope and cool, but it is not really you dope and cool, but it is not really you dope and cool, but it is not really you know know know using the model effectively. It makes using the model effectively. It makes using the model effectively. It makes sense for someone who is trying to sense for someone who is trying to sense for someone who is trying to ensure that everyone can use this one ensure that everyone can use this one ensure that everyone can use this one endpoint to do their task, but it makes endpoint to do their task, but it makes endpoint to do their task, but it makes much less sense when you're a person much less sense when you're a person much less sense when you're a person who's trying to do that task yourself. who's trying to do that task yourself. who's trying to do that task yourself. What what what was just said about What what what was just said about What what what was just said about efficiency is also deeply true, right? efficiency is also deeply true, right? efficiency is also deeply true, right? I mean I mean I mean the idea that you are all here at a the idea that you are all here at a the idea that you are all here at a local AI summit. Presumably you are local AI summit. Presumably you are local AI summit. Presumably you are running AI locally. Presumably you would running AI locally. Presumably you would running AI locally. Presumably you would like it to be faster and better. And like it to be faster and better. And like it to be faster and better. And presumably many of you are quite uh, presumably many of you are quite uh, presumably many of you are quite uh, quite cracked engineers, right?

quite cracked engineers, right? quite cracked engineers, right? This is a whole room of people who is This is a whole room of people who is This is a whole room of people who is going to contribute in some small part going to contribute in some small part going to contribute in some small part to making the ecosystem just a little to making the ecosystem just a little to making the ecosystem just a little bit faster, just a little bit more bit faster, just a little bit more bit faster, just a little bit more efficient. And while it's true that efficient. And while it's true that efficient. And while it's true that closed companies can, uh, afford to hire closed companies can, uh, afford to hire closed companies can, uh, afford to hire great amazing teams of people, as we saw great amazing teams of people, as we saw great amazing teams of people, as we saw with Linux over the uh, whole time that with Linux over the uh, whole time that with Linux over the uh, whole time that it's existed, right? Uh, Linux is the it's existed, right? Uh, Linux is the it's existed, right? Uh, Linux is the thing that runs the internet, it runs thing that runs the internet, it runs thing that runs the internet, it runs networks, it runs all of these services networks, it runs all of these services networks, it runs all of these services that, uh, that require it to be hyper that, uh, that require it to be hyper that, uh, that require it to be hyper optimized in a way that I think you can optimized in a way that I think you can optimized in a way that I think you can only get when you have people who are only get when you have people who are only get when you have people who are trying to run as resource constrained as trying to run as resource constrained as trying to run as resource constrained as possible. And, uh, possible. And, uh, possible. And, uh, all of that to wax poetic and say all of that to wax poetic and say all of that to wax poetic and say this idea that like local AI and open this idea that like local AI and open this idea that like local AI and open models in the most efficient version of models in the most efficient version of models in the most efficient version of the of the model ecosystem, uh, is is the of the model ecosystem, uh, is is the of the model ecosystem, uh, is is necessary to do it in the open. I think necessary to do it in the open. I think necessary to do it in the open. I think it's, in fact, not possible to do it it's, in fact, not possible to do it it's, in fact, not possible to do it behind closed doors cuz you're shutting behind closed doors cuz you're shutting behind closed doors cuz you're shutting too many people, uh, that could make too many people, uh, that could make too many people, uh, that could make that one small contribution, uh, that one small contribution, uh, that one small contribution, uh, out of the room.

out of the room. out of the room. >> I I think that that >> I I think that that >> I I think that that it's important to state, too, that I it's important to state, too, that I it's important to state, too, that I don't think any of us agree that or or don't think any of us agree that or or don't think any of us agree that or or or of the mind that closed models or or of the mind that closed models or or of the mind that closed models or frontier, you know, what OpenAI and frontier, you know, what OpenAI and frontier, you know, what OpenAI and Anthropic, just to name names, you know, Anthropic, just to name names, you know, Anthropic, just to name names, you know, Anthropic and others are doing it is, Anthropic and others are doing it is, Anthropic and others are doing it is, uh, not extremely beneficial uh, not extremely beneficial uh, not extremely beneficial or that not don't use them. or that not don't use them. or that not don't use them. Um, I I I certainly, uh, I use those Um, I I I certainly, uh, I use those Um, I I I certainly, uh, I use those models near every day. It's it's just models near every day. It's it's just models near every day. It's it's just that it's where does it fit in the, uh, that it's where does it fit in the, uh, that it's where does it fit in the, uh, in in the future of this ecosystem. Um, in in the future of this ecosystem. Um, in in the future of this ecosystem. Um, and just like, you know, Chris is and just like, you know, Chris is and just like, you know, Chris is alluding to, you can think of open alluding to, you can think of open alluding to, you can think of open models and self-hosted or kind of owned models and self-hosted or kind of owned models and self-hosted or kind of owned intelligence or LLMs as like the Linux intelligence or LLMs as like the Linux intelligence or LLMs as like the Linux layer, which you're beginning to see layer, which you're beginning to see layer, which you're beginning to see kind of take place. Linux runs kind of take place. Linux runs kind of take place. Linux runs enterprises, you know, it runs the enterprises, you know, it runs the enterprises, you know, it runs the cloud. We're seeing a very similar thing cloud. We're seeing a very similar thing cloud. We're seeing a very similar thing take place with hyperscalers and neo take place with hyperscalers and neo take place with hyperscalers and neo clouds and providers like Fireworks, clouds and providers like Fireworks, clouds and providers like Fireworks, Together, Base 10's models. Together, Base 10's models. Together, Base 10's models. Um, Um, Um, but just like Macs are one of the best but just like Macs are one of the best but just like Macs are one of the best ways to get work done individually in ways to get work done individually in ways to get work done individually in the same way that maybe using open AI the same way that maybe using open AI the same way that maybe using open AI and chat GPT is the best way for you to and chat GPT is the best way for you to and chat GPT is the best way for you to do the vast majority of simple check my do the vast majority of simple check my do the vast majority of simple check my email, help me rewrite, you know, check email, help me rewrite, you know, check email, help me rewrite, you know, check for grammar, those kind of things. It's for grammar, those kind of things. It's for grammar, those kind of things. It's accessible, it's easy, and for accessible, it's easy, and for accessible, it's easy, and for you know, your average consumer, an you know, your average consumer, an you know, your average consumer, an individual, it's pretty cheap. Um, if individual, it's pretty cheap. Um, if individual, it's pretty cheap. Um, if you're using like the $20 a month plan.

you're using like the $20 a month plan. you're using like the $20 a month plan. In the same way that you know, Microsoft In the same way that you know, Microsoft In the same way that you know, Microsoft helps helps helps the the world of medium size to large the the world of medium size to large the the world of medium size to large businesses run on Microsoft and Windows businesses run on Microsoft and Windows businesses run on Microsoft and Windows because because because you know, they're not as you know, they're not as you know, they're not as expensive as getting everybody a Mac, expensive as getting everybody a Mac, expensive as getting everybody a Mac, and you're going to see a similar world and you're going to see a similar world and you're going to see a similar world play out there for for some closed and play out there for for some closed and play out there for for some closed and open open frontiers themselves. So, it's open open frontiers themselves. So, it's open open frontiers themselves. So, it's all an ecosystem, you know, I don't want all an ecosystem, you know, I don't want all an ecosystem, you know, I don't want to give the impression that to give the impression that to give the impression that you know, I think anytime you log into you know, I think anytime you log into you know, I think anytime you log into chat GPT or Claude that that you're chat GPT or Claude that that you're chat GPT or Claude that that you're committing a sin, only that as you are, committing a sin, only that as you are, committing a sin, only that as you are, you know, this is a a conference for AI you know, this is a a conference for AI you know, this is a a conference for AI builders, AI engineers, builders, AI engineers, builders, AI engineers, as you're looking at the best way to as you're looking at the best way to as you're looking at the best way to engineer your product or your service, engineer your product or your service, engineer your product or your service, that there is another layer you can go that there is another layer you can go that there is another layer you can go down into and it's becoming down into and it's becoming down into and it's becoming way more accessible than it used to be. way more accessible than it used to be. way more accessible than it used to be. >> I I think that's a great point, and I >> I I think that's a great point, and I >> I I think that's a great point, and I think that the relationship between think that the relationship between think that the relationship between closed frontier models and open models closed frontier models and open models closed frontier models and open models will be one that is it's constantly will be one that is it's constantly will be one that is it's constantly there, right? I think that we have there, right? I think that we have there, right? I think that we have it's never you'll never get the it's never you'll never get the it's never you'll never get the headlines to apply nuance and say that headlines to apply nuance and say that headlines to apply nuance and say that both will coexist and gain more usage both will coexist and gain more usage both will coexist and gain more usage and are going to be useful to everybody, and are going to be useful to everybody, and are going to be useful to everybody, but that is kind of the de facto state but that is kind of the de facto state but that is kind of the de facto state that that will not only currently exist, that that will not only currently exist, that that will not only currently exist, but will continue to exist. Um, I want but will continue to exist. Um, I want but will continue to exist. Um, I want to to to spend the last few minutes here to spend the last few minutes here to spend the last few minutes here to really really really give the audience something that only give the audience something that only give the audience something that only you guys potentially can answer. Often you guys potentially can answer. Often you guys potentially can answer. Often times I reflect about my time at Nvidia times I reflect about my time at Nvidia times I reflect about my time at Nvidia and I think I I feel as though I have a

and I think I I feel as though I have a and I think I I feel as though I have a clear vision outside into clear vision outside into clear vision outside into there's definitely still a fog of war there's definitely still a fog of war there's definitely still a fog of war out there, but I have a vantage point out there, but I have a vantage point out there, but I have a vantage point that many people don't have and you guys that many people don't have and you guys that many people don't have and you guys because of the positions you are in as because of the positions you are in as because of the positions you are in as well. And so well. And so well. And so what is the the thing if we're looking what is the the thing if we're looking what is the the thing if we're looking forward towards AI engineer World Fair forward towards AI engineer World Fair forward towards AI engineer World Fair 2027 2027 2027 that you think that you think that you think if you were to make a a bold prediction, if you were to make a a bold prediction, if you were to make a a bold prediction, let's say, let's not be conservative let's say, let's not be conservative let's say, let's not be conservative around the intelligence in in the open around the intelligence in in the open around the intelligence in in the open source source source and ground it it was some frame of and ground it it was some frame of and ground it it was some frame of reference. What do what do you think we reference. What do what do you think we reference. What do what do you think we can look forward to by by this time next can look forward to by by this time next can look forward to by by this time next year? year? year? >> I think one key aspect obviously that >> I think one key aspect obviously that >> I think one key aspect obviously that people are closely tracking is like people are closely tracking is like people are closely tracking is like sort of like the just like capabilities sort of like the just like capabilities sort of like the just like capabilities of open frontier models and I think of open frontier models and I think of open frontier models and I think they'll they'll keep being very close to they'll they'll keep being very close to they'll they'll keep being very close to the general frontier potentially even the general frontier potentially even the general frontier potentially even like what like what like now the the like what like what like now the the like what like what like now the the speed or like the of those close speed or like the of those close speed or like the of those close frontier models like slowing down. I frontier models like slowing down. I frontier models like slowing down. I think like they'll they'll catch up even think like they'll they'll catch up even think like they'll they'll catch up even more and I think the the most concrete more and I think the the most concrete more and I think the the most concrete thing that I think will be very exciting thing that I think will be very exciting thing that I think will be very exciting is like seeing the world move from sort is like seeing the world move from sort is like seeing the world move from sort of chatbots and now coding agents to of chatbots and now coding agents to of chatbots and now coding agents to like just general knowledge worker like just general knowledge worker like just general knowledge worker agents. I think over the 12 the next 12 agents. I think over the 12 the next 12 agents. I think over the 12 the next 12 months right like to see more and more months right like to see more and more months right like to see more and more of like kind of like everyone across of like kind of like everyone across of like kind of like everyone across every every every knowledge worker domain like adopt knowledge worker domain like adopt knowledge worker domain like adopt agents in their agents in their agents in their workflows which I think like developers workflows which I think like developers workflows which I think like developers have with coding agents have probably have with coding agents have probably have with coding agents have probably done better than any other domain in the done better than any other domain in the done better than any other domain in the world, but I think we'll we'll see over world, but I think we'll we'll see over world, but I think we'll we'll see over the next 12 months like the next 12 months like the next 12 months like a lot of like domain specific like a lot of like domain specific like a lot of like domain specific like knowledge worker agents, but then also I

knowledge worker agents, but then also I knowledge worker agents, but then also I think domains like computer use agents think domains like computer use agents think domains like computer use agents and I think others will take off. Like I and I think others will take off. Like I and I think others will take off. Like I think in a similar way that like coding think in a similar way that like coding think in a similar way that like coding agents have taken off. I think we'll agents have taken off. I think we'll agents have taken off. I think we'll just see almost like in some ways you just see almost like in some ways you just see almost like in some ways you could say like almost like the could say like almost like the could say like almost like the general intelligence for like the general intelligence for like the general intelligence for like the knowledge worker in digital domain knowledge worker in digital domain knowledge worker in digital domain before then hopefully maybe moving on to before then hopefully maybe moving on to before then hopefully maybe moving on to the physical and I think very concretely the physical and I think very concretely the physical and I think very concretely like I think we'll like in in 12 months like I think we'll like in in 12 months like I think we'll like in in 12 months I think it's pretty likely that we'll I think it's pretty likely that we'll I think it's pretty likely that we'll have like better than fable metals level have like better than fable metals level have like better than fable metals level capabilities and open models. And I capabilities and open models. And I capabilities and open models. And I think this is like a huge opportunity think this is like a huge opportunity think this is like a huge opportunity uh to ultimately enable a huge crop of uh to ultimately enable a huge crop of uh to ultimately enable a huge crop of new like AI startups and companies. So, new like AI startups and companies. So, new like AI startups and companies. So, there's some ways like you want to there's some ways like you want to there's some ways like you want to almost like write the the levels of almost like write the the levels of almost like write the the levels of capabilities. Like to some extent like capabilities. Like to some extent like capabilities. Like to some extent like cursor really only took off when like cursor really only took off when like cursor really only took off when like Opus was good enough to do coding, Opus was good enough to do coding, Opus was good enough to do coding, right? So, it's like this is when like right? So, it's like this is when like right? So, it's like this is when like cursor inflected. And I think we'll see cursor inflected. And I think we'll see cursor inflected. And I think we'll see like hundreds of these inflections for like hundreds of these inflections for like hundreds of these inflections for like startups getting started like now, like startups getting started like now, like startups getting started like now, like in over the next year or two. Um like in over the next year or two. Um like in over the next year or two. Um once like open model and and I think once like open model and and I think once like open model and and I think we've seen this literally a month ago we've seen this literally a month ago we've seen this literally a month ago with like GM 5.2. I think like with like GM 5.2. I think like with like GM 5.2. I think like uh it was I think legitimately one of uh it was I think legitimately one of uh it was I think legitimately one of those moments when people were like, those moments when people were like, those moments when people were like, "Okay, this is now like similar to the "Okay, this is now like similar to the "Okay, this is now like similar to the Opus inflection point. Feels like an Opus inflection point. Feels like an Opus inflection point. Feels like an inflection point for open models to be inflection point for open models to be inflection point for open models to be like extremely strong and ultimately like extremely strong and ultimately like extremely strong and ultimately enable a ton of new businesses. And I enable a ton of new businesses. And I enable a ton of new businesses. And I think this will only continue like think this will only continue like think this will only continue like >> Uh not so bold prediction is that uh >> Uh not so bold prediction is that uh >> Uh not so bold prediction is that uh Primin Electin RC are going to have a Primin Electin RC are going to have a Primin Electin RC are going to have a combined valuation of a trillion combined valuation of a trillion combined valuation of a trillion dollars.

dollars. dollars. Uh that's all obvious. Uh that's all obvious. Uh that's all obvious. Uh but but but I think that Uh but but but I think that Uh but but but I think that this is going to be a huge year for um this is going to be a huge year for um this is going to be a huge year for um uh this is probably going to be the most uh this is probably going to be the most uh this is probably going to be the most consequential year for like the future consequential year for like the future consequential year for like the future of how AI gets distributed. Um of how AI gets distributed. Um of how AI gets distributed. Um the the the Fable uh and GPT 5.6 um the the the Fable uh and GPT 5.6 um the the the Fable uh and GPT 5.6 um you know, uh uh you know, uh uh you know, uh uh uh embargo, if you will, uh embargo, if you will, uh embargo, if you will, um um um has left a lot of open questions, uh you has left a lot of open questions, uh you has left a lot of open questions, uh you know, no pun intended about open models know, no pun intended about open models know, no pun intended about open models and and where and and where and and where um how this intelligence gets um how this intelligence gets um how this intelligence gets distributed. And at what capability distributed. And at what capability distributed. And at what capability level it starts to be uh politicized and level it starts to be uh politicized and level it starts to be uh politicized and and and kept back. And so, sovereign and and kept back. And so, sovereign and and kept back. And so, sovereign intelligence is going to be very intelligence is going to be very intelligence is going to be very important. I I think that important. I I think that important. I I think that um if you were to take, you know, the um if you were to take, you know, the um if you were to take, you know, the general population of AI um users, general population of AI um users, general population of AI um users, people that are using it every day. So, people that are using it every day. So, people that are using it every day. So, you know, upwards of uh a billion to two you know, upwards of uh a billion to two you know, upwards of uh a billion to two billion people if you, you know, include billion people if you, you know, include billion people if you, you know, include ChatGPT and Google and whatnot. Uh maybe ChatGPT and Google and whatnot. Uh maybe ChatGPT and Google and whatnot. Uh maybe 0.000001% 0.000001% 0.000001% have ever used an open model, You know, have ever used an open model, You know, have ever used an open model, You know, I think it it or you know, run it on I think it it or you know, run it on I think it it or you know, run it on themselves with with the multitude of themselves with with the multitude of themselves with with the multitude of different tools.

different tools. different tools. Um and Um and Um and I would hope that with the work that I would hope that with the work that I would hope that with the work that we're doing and the community's doing we're doing and the community's doing we're doing and the community's doing and the way that we're advocating and the way that we're advocating and the way that we're advocating um for open science and open models and um for open science and open models and um for open science and open models and open discussion, right? That's probably open discussion, right? That's probably open discussion, right? That's probably been the most frustrating thing about been the most frustrating thing about been the most frustrating thing about the last couple weeks is that all of the last couple weeks is that all of the last couple weeks is that all of these conversations around capabilities these conversations around capabilities these conversations around capabilities and who gets to use them and who doesn't and who gets to use them and who doesn't and who gets to use them and who doesn't have been happening behind closed doors. have been happening behind closed doors. have been happening behind closed doors. Uh my hope a-a-a-a-and I hope that I can Uh my hope a-a-a-a-and I hope that I can Uh my hope a-a-a-a-and I hope that I can predict that we will be able to have a predict that we will be able to have a predict that we will be able to have a a 10% to 15% of people that have ever a 10% to 15% of people that have ever a 10% to 15% of people that have ever used AI have used a model locally on used AI have used a model locally on used AI have used a model locally on their system and that that becomes a their system and that that becomes a their system and that that becomes a very important part uh of ensuring that very important part uh of ensuring that very important part uh of ensuring that you have access to what you need. you have access to what you need. you have access to what you need. Um and and so it's a prediction. Uh it's Um and and so it's a prediction. Uh it's Um and and so it's a prediction. Uh it's also something that I know all of us also something that I know all of us also something that I know all of us uh up here and and you out there are uh up here and and you out there are uh up here and and you out there are going to try to fulfill and I hope that going to try to fulfill and I hope that going to try to fulfill and I hope that uh we can continue to advocate for that uh we can continue to advocate for that uh we can continue to advocate for that because if we're if we're quiet, if we because if we're if we're quiet, if we because if we're if we're quiet, if we just let this uh things play out the way just let this uh things play out the way just let this uh things play out the way they are, uh open models will, you know, they are, uh open models will, you know, they are, uh open models will, you know, will will be put under the microscope um will will be put under the microscope um will will be put under the microscope um in the context of untrustworthy, unsafe.

in the context of untrustworthy, unsafe. in the context of untrustworthy, unsafe. Uh and as much as there's work and and Uh and as much as there's work and and Uh and as much as there's work and and um vitriol and weaponized terms being um vitriol and weaponized terms being um vitriol and weaponized terms being out there out there out there uh advocating for that, we need to be uh uh advocating for that, we need to be uh uh advocating for that, we need to be uh combating uh as much of that if not more combating uh as much of that if not more combating uh as much of that if not more uh with the reasons that it deserves to uh with the reasons that it deserves to uh with the reasons that it deserves to exist. exist. exist. >> Yeah, I just >> Yeah, I just >> Yeah, I just uh I could not uh I could not uh I could not uh plus infinity what the last part what uh plus infinity what the last part what uh plus infinity what the last part what Lucas said more. I think this is going Lucas said more. I think this is going Lucas said more. I think this is going to be the most consequential year for uh to be the most consequential year for uh to be the most consequential year for uh open intelligence that uh will open intelligence that uh will open intelligence that uh will a-a-a-a-at least for from where I sit a-a-a-a-at least for from where I sit a-a-a-a-at least for from where I sit determine the future of uh of a summit determine the future of uh of a summit determine the future of uh of a summit like this, right? Uh I think it will has like this, right? Uh I think it will has like this, right? Uh I think it will has a potential to look very different in a potential to look very different in a potential to look very different in two radically opposed ways. two radically opposed ways. two radically opposed ways. Uh as for bold predictions, I think that Uh as for bold predictions, I think that Uh as for bold predictions, I think that we will not be needing to go to an API we will not be needing to go to an API we will not be needing to go to an API for for for most of the tasks that we all do each most of the tasks that we all do each most of the tasks that we all do each day with AI. I think it's likely to day with AI. I think it's likely to day with AI. I think it's likely to assume that you'll be running a model assume that you'll be running a model assume that you'll be running a model that is sufficiently capable in let's that is sufficiently capable in let's that is sufficiently capable in let's call it day-to-day work on your on your call it day-to-day work on your on your call it day-to-day work on your on your MacBook within the year. It's already MacBook within the year. It's already MacBook within the year. It's already extraordinarily close, so not maybe not extraordinarily close, so not maybe not extraordinarily close, so not maybe not that bold of a prediction to be honest that bold of a prediction to be honest that bold of a prediction to be honest with you. I also think that we're going with you. I also think that we're going with you. I also think that we're going to continue to see to continue to see to continue to see models models models become the the future of AI, so not become the the future of AI, so not become the the future of AI, so not model, right?

model, right? model, right? Swarms of or specialized Swarms of or specialized Swarms of or specialized systems of models, I think you're going systems of models, I think you're going systems of models, I think you're going to be to be to be increasingly important. increasingly important. increasingly important. And lastly on the open model front, I And lastly on the open model front, I And lastly on the open model front, I think we're going to see some very large think we're going to see some very large think we're going to see some very large architecture shifts, especially as we architecture shifts, especially as we architecture shifts, especially as we start to crack things like diffusion start to crack things like diffusion start to crack things like diffusion models for text a little bit more to to models for text a little bit more to to models for text a little bit more to to get us models that are better suited for get us models that are better suited for get us models that are better suited for the the hardware that we have in our the the hardware that we have in our the the hardware that we have in our houses. And then last meme one, I think houses. And then last meme one, I think houses. And then last meme one, I think you're going to buy you're going to buy you're going to buy computers with agent operating systems computers with agent operating systems computers with agent operating systems on them instead of traditional operating on them instead of traditional operating on them instead of traditional operating systems. systems. systems. Similar to like you buying a Spark Similar to like you buying a Spark Similar to like you buying a Spark preloaded with with Hermes or whatever. preloaded with with Hermes or whatever. preloaded with with Hermes or whatever. I think that's that's likely to occur. I think that's that's likely to occur. I think that's that's likely to occur. >> I also predict that come September when >> I also predict that come September when >> I also predict that come September when the next iPhone comes out, you're going the next iPhone comes out, you're going the next iPhone comes out, you're going to get a lot of texts from family to get a lot of texts from family to get a lot of texts from family members asking about this magical new members asking about this magical new members asking about this magical new Siri. So, a lot of people who have not Siri. So, a lot of people who have not Siri. So, a lot of people who have not engaged with AI are about to in a very engaged with AI are about to in a very engaged with AI are about to in a very real way. real way. real way. And the the response to that's going to And the the response to that's going to And the the response to that's going to be very very cool. And so, just like be very very cool. And so, just like be very very cool. And so, just like when Deep Seek came out, I'm sure a lot when Deep Seek came out, I'm sure a lot when Deep Seek came out, I'm sure a lot of y'all got questions about what's this of y'all got questions about what's this of y'all got questions about what's this Deep Seek thing?

Deep Seek thing? Deep Seek thing? There'll be another one in September, There'll be another one in September, There'll be another one in September, get ready for it. get ready for it. get ready for it. >> Yeah, I think like this might be >> Yeah, I think like this might be >> Yeah, I think like this might be actually one of the kinds of crunch actually one of the kinds of crunch actually one of the kinds of crunch almost like unlock stuff like I think almost like unlock stuff like I think almost like unlock stuff like I think combination of like basically open combination of like basically open combination of like basically open models getting good enough as well as models getting good enough as well as models getting good enough as well as like the on-device compute getting like the on-device compute getting like the on-device compute getting strong enough to serve the current of strong enough to serve the current of strong enough to serve the current of like today's frontier models, right? like today's frontier models, right? like today's frontier models, right? Like in a year or two. Like Like Like in a year or two. Like Like Like in a year or two. Like Like basically if you can like run all posts basically if you can like run all posts basically if you can like run all posts like at a decent speeds on your like like at a decent speeds on your like like at a decent speeds on your like phone or laptop, I think the majority of phone or laptop, I think the majority of phone or laptop, I think the majority of humanity will probably like run local humanity will probably like run local humanity will probably like run local models. Like and and I think this models. Like and and I think this models. Like and and I think this probably applies more to the consumer probably applies more to the consumer probably applies more to the consumer than to the heaviest like enterprise than to the heaviest like enterprise than to the heaviest like enterprise agents, but I think agents, but I think agents, but I think it seems pretty likely to me that like it seems pretty likely to me that like it seems pretty likely to me that like there will be some flexion point even there will be some flexion point even there will be some flexion point even then like almost like similar to a new then like almost like similar to a new then like almost like similar to a new platform shift where like platform shift where like platform shift where like you can almost like tap into the local you can almost like tap into the local you can almost like tap into the local compute of a phone or laptop and then compute of a phone or laptop and then compute of a phone or laptop and then like start a next generation of almost like start a next generation of almost like start a next generation of almost like AI-enabled applications without like AI-enabled applications without like AI-enabled applications without like that that can ultimately really like that that can ultimately really like that that can ultimately really like leverage the local compute of like like leverage the local compute of like like leverage the local compute of like device on-device compute. device on-device compute. device on-device compute. >> You can run a 4 billion parameter model >> You can run a 4 billion parameter model >> You can run a 4 billion parameter model on your on your phone right now that is on your on your phone right now that is on your on your phone right now that is way more useful than GPT-4 was when it way more useful than GPT-4 was when it way more useful than GPT-4 was when it came out. And I think it's important for came out. And I think it's important for came out. And I think it's important for us to continue to to to focus on how do us to continue to to to focus on how do us to continue to to to focus on how do we best utilize that in the most we best utilize that in the most we best utilize that in the most meaningful way possible meaningful way possible meaningful way possible as well as chase as well as chase as well as chase the newer capabilities that will come the newer capabilities that will come the newer capabilities that will come from things like, you know, drug from things like, you know, drug from things like, you know, drug discovery and and discovery and and discovery and and and scientific exploration. It's It's and scientific exploration. It's It's and scientific exploration. It's It's going to be a a fun couple years.

going to be a a fun couple years. going to be a a fun couple years. >> Absolutely. If I were to try and >> Absolutely. If I were to try and >> Absolutely. If I were to try and summarize, I think that, you know, we're summarize, I think that, you know, we're summarize, I think that, you know, we're going to learn a lot more about how to going to learn a lot more about how to going to learn a lot more about how to these local models are are in built these local models are are in built these local models are are in built incredibly in the systems and their incredibly in the systems and their incredibly in the systems and their relationships as they, you know, relationships as they, you know, relationships as they, you know, interact with frontier models over the interact with frontier models over the interact with frontier models over the next panels, but this panel really shows next panels, but this panel really shows next panels, but this panel really shows that I think that we are at an that I think that we are at an that I think that we are at an inflection point to where if you think inflection point to where if you think inflection point to where if you think about how, you know, not even a short 6 about how, you know, not even a short 6 about how, you know, not even a short 6 years ago it was AI was really for the years ago it was AI was really for the years ago it was AI was really for the research crowd and not really many research crowd and not really many research crowd and not really many people cared about it. And then of people cared about it. And then of people cared about it. And then of course it came into the public course it came into the public course it came into the public consciousness with ChatGPT, consciousness with ChatGPT, consciousness with ChatGPT, um but now there's this next thing which um but now there's this next thing which um but now there's this next thing which is that open source is now really is that open source is now really is that open source is now really starting to enter the public starting to enter the public starting to enter the public consciousness, but very few people have consciousness, but very few people have consciousness, but very few people have touched and played with it touched and played with it touched and played with it and have had that aha moment. And it and have had that aha moment. And it and have had that aha moment. And it sounds like we have the potential to do sounds like we have the potential to do sounds like we have the potential to do that this year and sort of guide the the that this year and sort of guide the the that this year and sort of guide the the future wisely, but ultimately it's up to future wisely, but ultimately it's up to future wisely, but ultimately it's up to a lot of the builders in this room as a lot of the builders in this room as a lot of the builders in this room as well to to to leverage that and well to to to leverage that and well to to to leverage that and represent, um, you know, this important represent, um, you know, this important represent, um, you know, this important inflection point that we're in, uh, on inflection point that we're in, uh, on inflection point that we're in, uh, on the side that hopefully brings, uh, you the side that hopefully brings, uh, you the side that hopefully brings, uh, you know, intelligence, more intelligence to know, intelligence, more intelligence to know, intelligence, more intelligence to all of us, which is ultimately, I think, all of us, which is ultimately, I think, all of us, which is ultimately, I think, what everyone in this room would agree what everyone in this room would agree what everyone in this room would agree is is sort of the direction of progress.

is is sort of the direction of progress. is is sort of the direction of progress. >> And you know, everyone has said it on a >> And you know, everyone has said it on a >> And you know, everyone has said it on a panel previously, so I'll just also say panel previously, so I'll just also say panel previously, so I'll just also say it, which is that, uh, it, which is that, uh, it, which is that, uh, and and both of you have already said and and both of you have already said and and both of you have already said it, in fact. Uh, like you you guys are it, in fact. Uh, like you you guys are it, in fact. Uh, like you you guys are extraordinarily important to this goal. extraordinarily important to this goal. extraordinarily important to this goal. Uh, every one of you who is in this room Uh, every one of you who is in this room Uh, every one of you who is in this room and your friends and whoever what and your friends and whoever what and your friends and whoever what whatever communities you're part of, uh, whatever communities you're part of, uh, whatever communities you're part of, uh, without you guys, uh, we we will we lose without you guys, uh, we we will we lose without you guys, uh, we we will we lose the fight, right? So, thank you for the fight, right? So, thank you for the fight, right? So, thank you for showing up and, uh, I I can't wait to showing up and, uh, I I can't wait to showing up and, uh, I I can't wait to see what we all build together. see what we all build together. see what we all build together. >> And with that, thanks Vincent and Lucas, >> And with that, thanks Vincent and Lucas, >> And with that, thanks Vincent and Lucas, RC and Prime Intellect, and of course RC and Prime Intellect, and of course RC and Prime Intellect, and of course Chris from Nvidia. Chris from Nvidia. Chris from Nvidia. >> [applause] >> [applause] >> [applause] >> Always, always a pleasure.