Can Open Models Keep Up With OpenAI and Anthropic? | Nathan Lambert Transcript, AI Summary & Key Points

00:00 When I think about model progress, it comes down to different forms of capital and that is compute, data, and talent. >> How do you see that playing out in terms of like who's going to win in the next couple of years? >> The Chinese are so good at building models. To say that China is anything but equal on talent, you're going to have a hard time. What they don't have is compute.

00:15 >> Most people regard distillation as sort of like the main lever for them to do this. >> People distill my products every single day. They take it, they strip it down to its raw parts. That's called competition. Previous post training was much more reliant on the supervised fine-tuning and now this has morphed a bit into environments where it's RL heavy but it's just like hard to map this type of post training which is you get the completions and the score goes up to what is going on in reasoning today.

00:40 >> The discourse now is just so much fear on like how much the frontier is just hidden and also breaking away. The AI capability plot doesn't necessarily worry me, but the AI capability plot in the pretty much non-existing preparation of cross public cyber security and all these things like so Nathan really happy to have you here. Uh thank you for joining in person.

01:01 Super timely given you're launching Trillium which I'd love to talk about. um but more sort of on on a macro level um I think we chatted a lot about open source and that's you know a lot of what you care about and uh what I sort of um or what we've been hearing a lot in the public discourse is like hey we have to pace the frontier you know the frontier this the frontier that and I think an arc of the stuff that I'm really trying to ask e experts about is like what is in the frontier what is it even like what are we

01:28 pacing what's there and then more importantly kind of threading it with open source and open weight models like can they keep up with the frontier? And I think this matters very much um and is is something that you're an expert on. So maybe to to sort of kind of start on that arc, what's gotten to the sort of how you would break down what's gotten us uh to the frontier for both closed uh labs and open um source models?

01:53 you know, how does it break down in your mind maybe first at a higher level of pre-training, mid-training, post-training, how post- training has evolved, which is something you've, you know, done a great amount of education to the public on. >> When I think about model progress, it comes down to like different forms of capital that you've put into them and that is like compute, data, and talent.

02:14 And all of the labs have various amounts of this. And I think the the closed frontier labs of open eyes, Anthropics, Google's have clearly had the most >> of these and what we see is like the different abilities of the labs to organize these forms of capital and like making the org chart work for building models is really hard which is why you have companies have to do complete resets from the bottom up and then it actually kind of works sometimes like companies like Meta is like totally reset and they're like on a

02:39 positive trajectory again >> and like I think this I'm very interested in these human aspects of like why competition drives the frontier and like why the org chart is so important because like we see these things and >> this comes into play in the open closed >> dynamic as well where it's like the closed >> companies the the front the true frontier labs which I if somebody wants to work at the frontier of AI I mostly tell them to work at anthropic and open AI which is the easiest bet I think >> um Meta Google um

03:06 SpaceX AI like I think you still might be there but it's harder to see from the outside >> and like these companies have just taken the most ambitious bets the earliest in order to get there and that's why OpenAI has done things like kickstart the reasoning model revolution and I think that kind of breakthrough and like that beautiful plot with the inference scaling laws it's like peak open AI it's like they're really good at getting this plot that just changes how everyone thinks about everything >> and on the other

03:34 side it's like enthropic has been like very focused on the types of things the models are going to be able to do and then their reward was this cloud code explosion with their focus on coding for so along and there's these subtly different cultures and I think those two companies in particular have been the most forwardlooking and the most willing to spend egregious amounts to make a step towards a direction that people think is impossible.

03:57 The Chinese companies came into this because they want to have a company that matters and if they were to build their models and try to sell an API from China, nobody would know about them. Like nobody would know about Moonshot AI. Very few people would know about JIPU and there's a very long list after this. This is a very practical decision by them.

04:15 They're following the discourse. They know there's going to be a huge market. They know there's a huge demand in the US for paying for these things and it's very practical. So therefore like releasing open models is a way to >> shortcut this and get into the conversation and I >> I think there's a lot of complexity and like why do they do this? Is the government involved and and and not like not the case.

04:35 They're just like trying to build a startup and it's not it's not to say that they will not get there. Like >> the Chinese government is does things that would blow people's minds in the US. And I think that like that makes me think that the Chinese labs have a much more serious approach to security because like >> things happen in China that are much more drastic responses to companies doing things in the US.

04:56 And I don't need to go down this rabbit hole, but this is just like between the balance of open and close. It's just like these companies in China are trying to catch up. This is a very different dynamic than what I said about OpenAI and Enthropic that are trying to create the new paradigms that are trying to like make true RSI happen so that their company wins with the best models forever if you get there first.

05:16 these boom ideas >> and you just have to fill the market where there is a void and then open weight models are trying to do this and >> I think over time there will be these tiers of open models where there's the best closed models which you can demand the best margins for and then there's the near frontier models which is a step behind this and those will be open because just like some sort of supply and demand there as you get access to a different market that might be less profitable but the for example like if you

05:51 have private enterprise data or so on and these tiers are kind of natural evolutions of companies that are trying to do things and I think we were talking about the frontier and it's just like >> if you go back to capital like they have less capital they are going to make less money and like that's okay but I think this drive to catch up is very strong and once you know once anthropic and openai show you the type of model you need to build it's a lot easier to then be like oh we just need to do this thing and >> they

06:18 buy similar data sets. It trickles through the data economy where open AI and anthropics spend way more >> to build out this function of a data industry and then those data companies a year later might sell them to everybody. they might sell them to they need to maximize their return on their investment and they had some contractual period >> and I think there's a lot of dynamics where fast following is going to be very doable and it we've seen the cycle proceed for a while like longer than many people would expect I

06:46 think when chatbt was released we talked about consolidation the industry and predicting it and I think we're just starting to see a much more >> active mergers and acquisitions market but I thought this was going to happen two years ago and the open model companies have lasted much longer than I thought. >> No, there's so many interesting threads to pull on here.

07:05 Um I mean starting with actually your comment commentary on the um organizational differences between the top two frontier labs. You know we can maybe table that or or go on through that thread. But then of course also kind of explaining are you explaining sort of how the open um Chinese uh models or open labs in China have really just chosen that as much more of a distribution practical commercialization strategy uh rather than hey you know obviously out of altruism though it's great that the researchers maybe maybe

07:35 for better or for worse were were motivated for from you know sharing the beginning but at the end of the day it perpetuated because it was an excellent excellent commercialization uh commercialization strategy. Um, so the other thread maybe to pull there is sort of, you know, I think most people regard distillation as sort of like the main lever for them to do this so well and kind of increasingly uh well because um the lag between the open source models and the frontier models have only shortened uh with time in the

08:01 last few years. Um now it's like maybe only two to three months. Um, and this has also pushed the uh, Frontier Labs to maybe want to uh, close off the APIs um, close off their best models to do first party services. I mean, one probably to capture more value because it's always, you know, possible that way, but then two also because it's harder to distill, not impossible, but much harder to distill.

08:23 Um, maybe on those two threads because they're both very interesting. So, we'll try to touch on it. Um, let's start with the with the is the distillation story really more of a caricature and then triangulous labs since you've actually met with them and you've studied this a lot more. Uh, is it sort of like they're allrounded like great at pre-training also have their own, you know, good post-training uh, recipes that they may or may not have revealed.

08:47 Um, and where do you see the levers kind of accelerating? >> Yeah. So, the Chinese labs are very technically excellent. They are more comput starved I think. >> Yes. like without a doubt they have a comparable talent pool to draw on. I think there might be like a slight edge based on who was where when AI really got going. But to say that China is anything but equal in talent is going to you're going to have a hard time because they just have so many people being trained at such a high level.

09:12 And what they don't have is compute. You talk to people there and you're like, "Oh, I heard an open average open AI researcher has 5,000 H100 equivalents that they can access at any time." and then like their jaw just falls to the floor. They're just like I can't comprehend. And this is like a top like a a top pre-training researcher at one of the labs that you would know and you have this interaction a few times >> and like the compute thing I think is the biggest difference and distillation plays in as it is a way to

09:41 effectively spend more money to get what is compute equivalent like outputs from a language model are a function of compute and if you know the outputs you want it's the same as having compute to generate with a with a model and I've seen the stages of distillation and how I see it is that previous post- trainining was much more reliant on the supervised fine-tuning and other like this RHF era which is um where post- training started and you could get most of your performance just by you do this next token prediction

10:14 loss on the outputs of a very strong model. It was very established that you take the best model you can, you generate completions, and you train on it, and your model's going to be pretty good so long as you have access to the right prompts. Prompts have always been a key to post- training. I remember at AI2, like with Hannah, who I co-led post training there with, we would always say like prompts are the key unit of post- training.

10:34 It's like you need to get the prompts and you can do different things with them at different stages of training. And now this has morphed a bit into environments where it's RL heavy. But it's just like hard to map this type of post training which is you get the completions and the score goes up to what is going on in reasoning today. Mhm. >> And based on the public literature, it's very hard for me to be like distillation is a major factor because for example in the fully open literature of people trying to do open

11:02 post-raining recipes, all the open model builders are stuck using like a really outdated Quen model >> because you can't figure out how to get the teacher right. My hypothesis is that there's some like mid-training and something in the common pre-training data that's open that makes it hard to do this thing which is get reasoning traces from the absolute best model and do SFT and the numbers go up.

11:21 So like if I were to have access to Claude's reasoning trace when I was building I'm actually like claude opus quad mythos whatever I'm not sure that it would have actually made our model a lot better. >> So this is like my evidence grounded in the literature why I think distillation is hard. Um, and then like how do you take the outputs of a model and build an RL environment to do a big RL run?

11:45 Because then you have the Xiai of the world doing this week-long open RL run or they're like here's what we did and then they've dropped the model and it's like where exactly does distillation go into this. They're they're clever. They know how to do marketing well. There's some questions of like did they do this ahead of time because they did their RL run and then they dropped the model and then their tech report said they had an extra post training stage after RL.

12:05 some questions but like mapping the established distillation literature to that is very hard but there are question marks that make me support the labs a bit which is like these quad proxies and these model proxies in China where there's a ton of tokens that are used and there's intermediaries and they have tons of accounts and even cases where Kimmy was serving claude in their products to users and I'm just like don't really understand how this fits into distillation as a common technical tool in post- training and

12:37 like yes we should watch it too like there's other weird stuff with Chinese companies using the outputs from open AI and entropics models at a vast scale >> so I think we have a lot more questions we need to answer and I think most reasonable people when they're like what's the gap from distillation are like there's very big error bars but when I say that distillation helps them by a few months it's based on this scientific literature that >> we just like that's the information we have we should do more open research

13:04 and enable this so that we can study it. But like I don't know how you can come to a different conclusion that it's like not a huge effect in current post training based on this. >> And that's just like the Chinese are so good at building models. Well, I think there's a bit of cope where people are like only we are good at building models and it's like >> there are so many companies there's like 20 to 30 companies that are building good models.

13:27 >> You can like read. Yeah. Yeah. >> And it's like you just >> Yeah. And like that's hard for me to get over. And I think that there was this very prominent paper that's like stealing reasoning traces from closed models. And >> I know that the Chinese companies are way ahead on doing this since the AI safety community later published it. Like I think every Chinese lab has a team that knows those sorts of things.

13:49 And this goes back to what you're saying about like they don't need to offer the API. Mhm. >> And I think there's some characteristic of reasoning models that generate these thinking tokens where it's easy to just knock them out of the API so you see the reasoning token and then you can use that to train your model. >> And so long as they offer these APIs, >> there's going to be no way to stop that type of distillation.

14:11 >> But if they do make it so it's only in products and things where you have to sign enterprise agreements or whatever, it'll be way harder. And >> I support the labs doing that if they want to secure it. But it's like they have to secure their products if they want to. Like I don't like I don't think knowing your customer would be good for anybody.

14:26 I don't think it'd be good for their business. But also I don't know why you should get the government involved in a terms of service violation. It's like not even like I don't even think terms of service are a signed contract. It's just like a I don't think you can sue them in the same way. You just can cancel their access. >> Right. >> So that's why I'm not that >> sympathetic.

14:43 But I am like I like to listen to the like the national security community is loud. And but it's like there is a directional truth to that with where AI is going and how influential it is that you have to balance many >> many things in your head. >> I I wonder what your response is. So I I heard I think Jensen said in some literally either tweet or interview maybe >> great put it put it in put it in the show.

15:09 >> People distill my products every single day. They take it and they they strip it down to its raw parts. They test everything. That's called competition. >> Do you think that's fair game? You think distillation is not robbery? >> That's called competition. >> I'm very sympathetic cuz I think there's a lot of calls from the AI safety community that we want to have like US China dialogue as we're getting to such capability levels.

15:30 And if you're going to ask the company to just like >> start hitting their AI industry with slaps on the wrist and more >> downstream of distillation. So you won't regulate distillation with policy from Congress. You will probably just do federal action to make their life annoying. Yeah. Yeah. And at the same time when we're talking about how do we manage frontier risks where you have closed models in training going all over the web and accessing things that they shouldn't and we have valid worries about like what will

15:59 happen with openw weight models where this like hasn't been harms yet but it's like why like why aren't people using open weight models like why did they go to closed models and get around the safeguards and like >> we want to be able to communicate with the Chinese tech companies and everything about this and I just think like you're going to cause more harm than good if you do anything about distillation because they're not going to stop.

16:20 It's all it's like they're gonna keep doing it like >> right right and then and and the open source community benefits and therefore the startup community benefits a lot and we can go into you know a lot of it's not just like hey China USA it's actually like there's you know and we can actually talk about like I am sympathetic obviously or try to steal man the um frontier lab position which is like you have just unprecedented amounts of capital that's required to build these frontier models right and it's it's kind of

16:44 alarming slash not very um uh conducive to to this uh flywheel continuing if if the um through distillation or other kind of distillation related methods. Um open source can catch up so quickly. And so you know and what I mean by this flywheel is just kind of a very simple picture of you know you train the frontier model by putting in a lot of money to get compute and then they park this compute.

17:10 of course open anthropic are very guilty slash you successful in this >> whichever way you want to like look look at and there is a huge compute you know uh kind of uh um sort of um lack of compute available because there's actually like not just compute in the chips but there's also power etc. So like all of this scarcity is kind of you know bottleneck until 2028.

17:30 So this is all to say that it's a successful strategy to park that which leads to actually a lot of um revenue because once they get to the frontier because they understand you know scaling is the one thing that kind of gives them or the science of scaling is what gives them a handle over how how well um that can translate like into a predictable revenue stream.

17:50 So they get the revenue and because of that revenue and they're able to raise more money. It's just unprecedented like beautiful flywheel and this happens for both open AI and anthropic. Um, of course we need exponential amount of resources. And so we've we've gotten to the point where we're like, okay, we need to IPO. We need to go to the public markets.

18:05 And so anything that threatens this flywheel from functioning well um is is a huge threat to just the entire like being at the frontier. Um, and so >> the entire AI industry realistically >> and that and exactly that actually. So like, you know, maybe not just kind of being sympathetic to their version of like trying to, you know, uh, eek out capitalism for themselves.

18:26 It's it's also just like, hey, if if they don't do well in the IPO, that's actually bad for everyone else. And so I I I want everyone, you know, I want them to do well. Um, on the other hand, you know, it's like the Chinese open- source models are a huge benefit to the startup ecosystem. I mean, we see this all the time. Um, just kind of metrics I remember from like Decagon and Harvey who was trying to do a lot of fine-tuning.

18:51 um there's like a huge amount of uh startups doing post training as a service of various degrees and trying to productize that and and it actually works because you can get to the PTO frontier with these you know Kimmy K3 or pick your favorite like GLM um open source model. So, I guess like weighing these two uh things, how do you um I I guess there's so many questions can fall from that, but I guess one um how do you see that playing out in terms of like who's going to who's going to win in the next couple of years

19:21 versus, you know, what what and and then also relate that to, you know, why you're doing obviously in open source. So I think open and closed models are on different time scales of exponentials and the closed labs are so integrated where they have efficient inference. They can serve the model at scale when they're ready to because they have all these resources.

19:42 They can plan and they get to these markets first that are very high value. We have seen this with coding and this has been the story of the year where the revenue has just gone zero 100 billion and likely to grow. But the open model labs partially this is just like an economic issue being the fast follower is that you don't have this flow of information and you don't have this flow of revenue.

20:04 So the open model you have open model builder then you have the open model inference companies and you have the open model post- training companies and like there's information from those customers which are like things you need the models to do and it's like you can't like that information doesn't just flow back to the model builder very easily and they also don't have the economies of scale to serve as many users to serve it as efficiently to have as much margins on influence on inference and these are just like it's

20:30 going to take a lot longer but to the extent that things can be commoditized. It seems like companies really want to have private data. There's a lot of things that they just cannot send over an API and we could talk about the like um zero data retention or not and so on. feel like these it takes a long time to figure out if a company has internal expertise or they figure out how to build an internal evaluation that actually matches to revenue and then they figure out how to fine-tune or then they go find >> they go

21:00 find Thinking Machines or Prime Intellect and they're like we could make money in mass if we figure out how to fine-tune a model that solves this and like they have these customers but there just so many more companies involved that it takes so much longer but at the end of this process they have real ownership which is model weights no terms and conditions that they can run on their hardware and they figured out they can make an ROI on.

21:19 So this is why I just think the open model economy is it takes way longer to build and I'm less concerned about this being kind of bubbly where it's like this could eat into the market of the really fast takeoff as people figure out how to what is the cheapest model that accomplishes a task and how do you compete on margins >> and I don't really know like I don't really know what's going to happen.

21:41 I think there's it's too early to know. >> It's it's it's so yeah it's it's so hard. I mean it's definitely something that's always top of mind thinking about like there's thing there are things where scaling up general reasoning just as a general model is so uh it's going to beat out any specialized model and we see obviously examples in math and that's probably the most pure domain that you can see this um but also you know even in say robotics you know with with Astra um and and I for maybe uh however degrees of

22:14 like true Truthy the signal is it just doesn't sound like they've actually tried to ascend in robotics is they did I think this is a cycle of like six to eight months ago they started hiring people and buying a lot of data on robotics >> they did >> and then the model magically six months later is really good at it but I think that the >> but I actually well the other other you know more contrary things I've heard is that it's actually coming out of computer use um because it's you know because computer use is like

22:39 some level of spatial reasoning and then getting to have good blender you know demos is is another extension of that. Um, so you know, we can believe some version of that. But either way, I guess it's like it maybe if it's like really specialized, you see it's it's the capability gain is really from like trying to get better in a certain area. But like in math, I don't even know if that fell out of trying to specialize in math.

23:07 I mean, I think the whole thesis there was just like let's just scale up reasoning and actually got better in in in natural language. It got better than any specialized model in math. I mean, Anthropic wasn't even trying to be good at math, you know, open AI, you know, certainly I think, you know, they hired a bunch of mathematicians, but they also weren't trying to like use specialized, you know, kind of like lean data set to get better at it.

23:27 Um, or at least that's not um story. So, anyways, this is this is all kind of fodder for the view that like it's kind of hard to compete with the frontier models because eventually they start ascending so much that you know you you get uh good enough in respective um categories. With that said, I think actually there's also very competing like a counterforce of data that you're seeing which is like the open- source models that are already good enough uh allow for fine-tunes, you know, to your point of like these slow

23:56 percolation diffusion of these uh um very specific benchmarks and and data sets that are only available in these, you know, companies that you're being applied to. And it's just it's just good enough and it's actually we're starting to see um people adopt that. I mean in the most maybe pure regime where you would expect frontier models are good is um I think cursor published something over the summer where like hey we just you know we had this long horizon task of writing rewriting SQL light and turns out using two

24:21 frontier models you know one for planner one for um execution is not as good or is equivalently as good but far more costly than just using composer with maybe opus or you know fastra whatever as planner and that I think that's actually great evidence that the open-source finetunings are very good or good enough and has surpassed um intelligence hungry tasks.

24:47 Um there's a long way of asking what evidence do you find is most compelling or you know pieces of evidence for various versions of this stable state of where the frontier wins versus open source. I'm a big proponent that a lot of the frontier progress in domains that are not fully verifiable is by buying data that is high quality and giving the models worlds to act in and he'll climb on.

25:12 And I think there's two time there's two things. We have the math and code thing and we'll we we don't know how big the coding market will be and then there's kind of everything else which is still they're targeting this knowledge work this biology healthcare legal and all of these things are not as verifiable and I think is much closer to the story of opening I bought a lot of robotics data they're probably going to tell us it's generalization magic poof people love poof generalization um and but realistically it's

25:40 probably some of A some of B but a lot of it comes down to data when I talk to people at Frontier Lab jobs. There are some people that have amazing jobs. There's some people that contribute by doing a very narrow thing, which is you make a data set and you figure out what the behavior should be for your model in that area and you figure out how to make the data that helps get that data.

25:57 You get that behavior and you figure out what data you need to buy >> and that's going to take I think we're going to have years of this across different sectors >> and that's very capital intensive to buy all of this data. But that's not just like poof magic RSI happens where I do see in this math and code and some like distributed ML engineering like it's very verifiable.

26:18 You know what your cost is and the AI models are going to drive that >> down at a really magical rate >> and that's going to just have a lot of surplus benefits and we can do research faster. You could look I'm very interested in like I think the efficiency of models is going to drop multiplicatively over years. So we've seen OpenAI have drastic price cuts and it's like what does it mean for GPT Soul or Astra and these bigger models to reduce in price 2x every eight months for the next 3 years?

26:46 Like you start getting >> you start doing this math and it's like it's 1% of the cost and whatever like four years and it's like >> are they going to able to have higher margins due to Jeban's paradox there or their revenues going to decrease because the market fills up? There's just like all these really weird dynamics, but fundamentally like AI progress is about data and about writing better software to serve your models and those are very different worlds.

27:14 Yeah, I mean at least definitely I agree with you that observing from the outside there's a lot of um data RL environment companies that are you know doing very very well um selling both to the top two frontier labs but an increasing kind of longer tale of labs uh as well. Um so to your point I think a lot of the capability gain has really come from just scaling that up.

27:37 Um and it's not an easy task. I mean maybe there's actually very curious on your opinions between uh the the data collection and environments that are built by hand versus the ones that are you know done in sim there seems to be like different versions of like what could be you know what what kind of price they command etc. They all have good things and bad things.

27:57 And I just think like there's a lot of talk of like like we've talked about distillation, but the other side is like the Chinese company buys the RL environment that they use six months later. >> Oh, entropic. I bought this. Now there's a premium. >> The good data will do this. >> But the data that people are buying is pretty flawed. Like >> I would say that the method is you buy in mass and then you figure out what works and it's not 100% hit rate.

28:23 >> Yeah. And I think I personally think that this RO environment explosion where there's just so much opportunity to make money is have a lot of people that like like cut corners and this can induce things like reward hacking. >> So it's like the models are so smart that if you put them in a world and they're like my job is to solve this task but they didn't give me the like you have a test file and like the test file actually has an empty file instead of the instead of the maze that you have to navigate and they're

28:48 like the model is like well I should still figure out how to get a high reward anyways. And I think a lot of this behavior is downstream of this supply chain >> and I think it'll fix it. But it's interesting to see the demand from the labs to eliminate reward hacking and all these company researchers spending so much time on it >> and then all the data companies like buy our data and I think proportionally spend less time on model behavior and more time on world building and revenue and somehow these two parties need

29:16 to meet together to actually make higher quality data. >> Yeah. Well, a lot of the labs are actually taking it in-house partially because they don't want >> Yeah. >> You know, data leaking. Not that, you know, like lit literally and figuratively speaking. It's like, you know, signals leaking of what they're doing, but they're trying to send on. But despite that, I think there's a tremendous amount of growth in um how much uh they're purchasing.

29:37 Do do you think that's probably one of the strongest levers? Can you kind of qualify how much that will kind of increase? Do you expect that to increase uh based on spending? Um or are there other you know because for instance if it can be done well in sim I think maybe that isn't more of an argument to that being absorbed by labs and it you know in particular and then >> all the companies are doing a mix of like all the companies are selling synthetic data where they're using language models to modify or build the

30:07 world or generate permutations of it and like if a human is guiding this process it's still synthetic data. So like there's very few people that are going to like sit there and lender and like we're going to make a little world for our model to act in. Like all the companies I think are moving far faster than that, >> right? >> And they buy them because they the models are getting better from them.

30:28 >> Like I I still have denial about it. I'm like these numbers are crazy. The billions of dollars research budget on environments. I'm like >> they're only doing this because it is helping the models and they see the signal. And I think like my general observation is that >> the data industry has a lot of lowhanging fruit on improving quality and improving just like OBSAC.

30:45 So like if you look into the details, you can find weird things with like data companies that use contractors overseas. Those people are through an intermediary. They don't know they're selling it. They just upload it to GitHub. There's a lot of stuff out there that if you poke around the data industry, you find weird rough edges. >> Yeah. >> And it's just very early.

31:03 And that's like it's it's the one thing is like the numbers are too big, the quality is bad, but on the other hand, the labs are improving very fast and know how to onboard it. So I think this will take a few years to transition of like really pulling the the winners into higher quality, higher ROI, you takes less resources when you have these systems for doing better and then the labs keep working with them because they get even more training signal and have to do less internal filtering >> and I I think this is going

31:29 to just take years. How do you think about then using environments and like your purchasing decisions as part of the nonprofit that you're starting? >> I think most of our work will be making open- source environments for tasks that are very high leverage. So I think of this work that I advised Udub or Udub and AI2 state and Hamish Iverson on this T-Max work which is you use a few it's for terminal tasks which are like super popular at terminal bench whatever number we're on and this is fully synthetic data where you

32:00 prompt a frontier model to generate a type of problem and you use a few different prompts to generate diversity and then you use model rollouts and other things to find bugs and lamb and eliminate the ones with bugs. And this process is not magic. It's like you read every data point, your data will get better and you can do this. But there's just not that many institutions that are investing what it takes thousands or tens of thousands of dollars to make these environments well and then release it because the profit

32:27 because the profit motive is so high. >> So high. I was joking with the author like you could probably sell this data for a million dollars and then release the paper and they're like nobody like we care about doing the right thing. But it's just like oh my god these open resources are the fun foundation of >> so much study and it's like you release some RL environments with some RL rollouts people can do research on say like a val awareness you you have the generations from the model people can look for different

32:51 behaviors if you've done RL on it they can see how it goes over time and this is just the the units of infrastructure that PhD students spend a year looking at and produce >> increased knowledge and the tech industry is the place where value is captured. captured from CS research all CS research the benefit occurs to the tech industry in like 10 years and I just think like this is why more people supporting open that if they're going to work if you're going to work in the industry for 10 years like your peers are going

33:20 to be the ones that benefit >> fossil fuel let's not deplete it all >> but it's just like the opportunity to short-term win is so high the opportunity to make what is $100 million like nine figure returns for your business within five months of founding a startup with your friends is just like these are people who may have done more open- source work or these are people who left their PhD or something and it's just like pulling back which is why we came up with this name trillium for our lab which is like trillium is a

33:46 spring ephemeral flower which it blooms before the canopy has grown its leaves so it's like winter there's no tree leaves on the trees these trillium flowers bloom and they're like essential for spreading nutrients through the cycle of growth that comes through spring and summer and this is where I feel like the AI industry is doing so much, right? It's like so much is going right, but it's just like we just need a little bit more life to like spread this out and make things so new people can come in like there's so

34:14 much attention and just like do research, do education. >> Yeah. >> And then as this scales and becomes more impactful is like you have these open things which become the standards of study for such important problems which is like how do you understand sandboxes like how do you define what loss of control is? How can you see how this changes in a RL round that is a week on a thousand GPUs or 100,000 GPUs?

34:34 And it's just like >> make these places so that people have more trust. And I think this pays so much dividends to everybody that stands to profit more than I do. And it's just like >> hard to get people to take that leap. And like I've been rewarded so much for kind of stumbling into this area and now I get so much reward. they get to do things like this which is fun and it's like now is the best job because I try to create space for more people to like grow their independent thought and build this infrastructure.

35:05 >> Yeah. And maybe one like kind of naive way that it kind of perpetuates is just because it's open source, people get to you you might get more of a distributed contribution from people who are just so you know it's like they're motivated not for just trying to sell to the labs themselves but because hey it's out there they can share their work and then you actually have a very high caliber amount of contributors uh to the project.

35:27 I mean that that would probably be the the naive hope of how that >> Yeah, it's a new puzzle with AI agents where people just throw things at you. Like I have some re random research repositories for research papers and I'm like people open the most crazy GitHub issues now. So >> what do you mean? >> Like just like like fake research results. They open an issue and they're like >> oh my god >> and they're like I found this thing in your data set.

35:48 And I'm like, well, if you looked at the paper, we said that in like the second sentence, and it's just like a lot of language model output. And that's a new thing for the open community to >> go through. >> But this is a solvable problem. Like like there are people out there that want >> to contribute to this stuff because they >> right not just clout chasing in another, you know, dimension, but like actually, hey, it's out there, you know.

36:10 I mean I I would just hope because it is published and open or you know and and available you can actually just get like PhD students or other researchers who are like okay I do this on my day job but maybe I can you know still be able to contribute. Yeah, hopefully from the Frontier Labs. >> Like I see people like this and I was like great somebody from like Eastern Europe was helping me improve the code for my book that's just right here and it's just like doing it for the right thing and like >> we talk and like

36:36 there are people that do it and I think a lot of people just take like they don't have the time or the >> stability to take the long time it takes cuz when you're trying to get into a new field you know nothing. You have to spend like weeks reading and knowing nothing and then you send an email and nobody replies >> and you just like it takes you weeks to know the context to like ask a good question.

36:59 >> Yeah. >> And now with language models, most people just push a button and they're like, >> "Oh, interesting. I read this. I'll send it off." And it's just like >> this >> Oh, yeah. >> kind of like sitting in it is the thing that people don't do as much. And that is the key to like becoming into AI is that like you just have to focus on it. You have to write some code by hand and you have to do this for months and then you have to wait six to 18 months for you to get noticed.

37:24 It just takes a long time and like all the AI agents I think just like frazzle that and makes it harder for people. >> That's what the mathematicians complain about too. That's all it's like um >> so it's a hard time to be a PhD student but people are doing great work in academia still and but it's hard. >> Yeah. But I agree with you. I think it's like solvable as the agents kind of perpetuate some version of the problem.

37:47 can also help solve and and identify an end quality filter. Um, okay. So, environments you're going to release open source versions of that help precipitate maybe more people to contribute and and reveal that. I think that's like really I mean especially given how little incentive there is to release this instead of just get um you know fast uh revenue from from the labs.

38:05 Um, another area I think I read on your announcement post is RSI, which is, you know, quite um I think I mean we can first talk about what is actually recursive about it or how scary it is or whether we're going to have takeoff. I think you have a very grounded opinion on that. Um uh but you know another thing I want to kind of talk about is I think RSI maybe related to how much it's kind of hyped is is a reason why we are afraid Frontier Labs will pull away.

38:29 Um, in fact, it's also a reason why I've heard not just once but several times that people are one convinced that their own jobs are going to be automated as a researcher. >> Yeah. Yeah. Like I've gone through the last few years, some of my best friends, I'm like, you are some of the best people in the world at this. And then they go to the work and they're like, >> Nathan, I'm I'm struggling like what am I going to do?

38:51 Because to your point like a lot of what they do is verifiable, right, for um but because of that, you know, there's actually quite a lot of other non-verifiable or harder out of distribution tasks such as um uh uh was alignment and things like that. And so actually because I mean one argument I heard was because there's such a difference between how well RSI is doing for model capabilities versus alignment which is much more out out of distribution maybe creative work that gives people an ease on how you know maybe

39:22 there is actual dangers etc etc. Um so anyways because of this I'm very interested in hearing from you like how because I think it's really important that we do have an open-source um kind of front and in and showing how RSI is being done how much maybe then we can you know understand the capabilities of like if in within a lab they have a better model what what are the chances they actually pull away just better understanding calibrating um so yeah would love to hear more about that that >> yeah so I think this we

39:45 won't be the only people doing this but essentially like RSI has so much mystique like we're using models for everything and like what we can do as a open lab and a nonprofit and I think other companies like a Luther or hugging pace or open Athena which is Marin and Stanford they will get into this good marketing line for them is and me I'm doing this and it's like we can share what tools we're using we're going to build an open source stack for SFT for mid-training we're going to use SkyRL miles all these things you

40:14 know and we have our task we have our configs we know what type of compute we're running we could work with cloud and GPT and all these models and we'll just like we can store the trajectories for what we did. We can tell you what our tasks ever use linear or whatever these popular things are. We'll say what our tasks were and why we chose them and whether models work on them or not.

40:35 And you just do this for a year of a lab and you're like here's >> what we did and then you just check in every six months and you're like has it changed? And and it's just like to be able to have the full stack and not need to go through these classifiers and like try to like say what the task is or vaguely and so on. It's just like people that can share what automation is happening should do this and we have a different distribution of tasks.

40:59 But I think that the research is similar like other things like >> the AI research that I've done at AI2 and we'll do at Trillium versus what people do in some sects of acne and the labs. It's not that different. You have more resources. You scale it up and it's just like so good for the world when everybody's talking about this to just just show people like >> this is what Nathan asked his model to do and this is what it got stuck on.

41:25 And I think that will help a lot of like literal policy makers like you're like, >> "Oh, okay." >> It's like, "Oh, that's why it's very much >> like that's what it got stuck on." Like that's the bug that it put in its code. >> Like the bugs and the catchups will will reduce over time. But I think >> I think the jaggedness and some of these fundamental things of the model is like some more things will take off.

41:48 Like I said efficiency earlier, like I do think the models would be really good at lowering serving costs. These companies have billions of dollars on the line and the models are good at it. They're going to make a lot of money on making serving more efficient and tweaking architecture in order to go into that. >> But it's not necessarily like how do you solve knowledge work?

42:06 Like Claude's not going to cla's not going to like you're not going to be like Claude make tell yourself to make better PowerPoints like like what what is >> environments or some other you know >> like it's going to take a while. >> Yeah. Yeah. No, I I love the the specified goal of like let's just demystify this because I think a lot of why the rhetoric gets so heated around RSI just like it sounds >> I think AI nonprofits can do a lot of things like this like it's hard for academia to be like >> as a professor 15% of

42:34 my time and 40% of time of my grad student is going to be documenting what we work on. But there's like such a great like policy and community service of like here are our tasks. People can use this data to make evaluations or whatever. And it's just like just like build things that have >> rare information so that when people put more high focus work into it, it just has more value for all sorts of things that you can't predict in the future.

43:01 >> It's honestly just also very timely because the kind of, you know, it's not from one person, but several people that I know from the labs are just like, "Hey, I'm going to get automated." I mean to to your point to your friends like I I will definitely as a person sorry will get automated in a year and so that probably means there's going to be less people who actually are exposed to what's being done on that particular you know part of the org um uh chart and so um >> there's a related academic there's an open RSI

43:25 thing which is a bunch of academic institutions I'm not affiliated with it they just launched like last week so I was like yeah >> they sent me a few emails I'm like open RSI is a thing they're trying to do this as well like it doesn't need to just be >> one group there's a lot more. >> It's good to have different approaches doing this because you get you need the >> different ways of looking at it and different results.

43:45 >> Yeah. Whole point of it being open. The other thing I also heard was from more kind of like you know Frontier Lab Goss is just like the actual methods that are being used in the labs are not that it's not that they're unknown. It's just sort of like what are the proportions that need to be used that are successful and this is partially because they have the capital to scale it to experiment.

44:06 Um so I think that's one heartening to know because like okay it's not like there's a big secret of what's going on. So people actually know the techniques but on the other hand you know capital is the one true barrier still because it's hard to actually know which of which of the techniques recipes uh work. Um >> that's why I'm not pre-training. I'm happy to be free.

44:26 I can use any base model but equal I'm I'm now get to benefit from competition. >> Yeah. I do you think that there's like a chance for any pre-training open-source effort? I know the capital requirements are just like orders of magnitude larger. So that is what's difficult. Um but maybe what are the values of that and then maybe we can kind of discuss potential commercial strategies that make >> pre-training science is pretty like that the tools are pretty accessible and you have what is a pre-training ladder which is

44:56 like the decision-making tool you have to test if your ideas scale and the pre-training data isn't growing at like like the environment scene like it's essentially so first approximation you train on the internet and like there's a lot of there's little there's little details and secret things. But >> but like that's not changing that much. The state-of-the-art architecture is and the GPUs are and that's like that changes and >> but it's not astronomically expensive to make a pre-trained model that is solid.

45:28 So I think like you don't have to spend a billion dollars to get a very good base model. The more you spend the better you will get. But it's like people keep making progress and I'm constantly surprised by a new lab just goes from scratch because they can. And some of this is on open tools. So like >> you can go off Megatron like a lot of new startups they're like oh what what should we train?

45:49 We train they train off the open source tool and they fork it and then they never release it again. This is what all the all the like even prominent open model labs have done. They take the open source code they make it their own thing and they just a lot of them don't push upstream. Mhm. >> So, I've always been surprised by this. So, I'm like not betting against new entrance because like it's not that you need 10 billion dollars to like have a relevant model.

46:15 The hard thing is like get a model that's good enough for people to use. >> Mhm. >> And to get a story where if you release it, you can do your next fund raise, which is like you have to get past this lock of like we need our first model to land. If like your whole company is based on models are the best and you're not in the ballpark and you're not releasing like you're on the clock and it's like I think of like thinking machines getting inkling out where it's like they were very transparent.

46:40 They're like this wasn't the best model. We're proud of it. Has some cool things. Yeah. >> And like they're now they've released their second model and they're getting going and they have Tinker has this like that's very good for them. They're like off the ground and it's just like >> that I don't think that's what they were do. Like they were building models but open models wasn't their first plan.

46:59 and then they got there and there's a lot there's rumors that like Meta said they'll release a new openw weight model again like that'll come there's a lot of companies out there that couldn't afford models. >> Yeah. Yeah. I know I'm definitely you know glad that thinking machines is and it makes so much sense for their business model um because they do post training um and you know with meta I think you mentioned this as well in some previous writings it just for them it's just such a good strategy.

47:23 I think it was also just kind of obvious when they they do it. It's like it's complimentary to my, you know, revenue streams. >> Mark wrote the best blog post on this. Like when Llama 3 came out, he wrote like the best like why open his path forward. >> Right. Right. >> Guy got Guy got a little distracted. Seems like he's coming back to >> Right. Right.

47:39 Yeah. Yeah. So I hope that they they keep on being motivated to do that. um getting to the frontier with pre-training. You know, I think there's a lot of help given that there's such good open source models to maybe, you know, somehow still kind of bootstrap from. I I keep on getting kind of conflicting signals as to like whether pre-training is actually as saturated or well understood.

48:00 I mean, maybe maybe to your point like there's a lot of like well-known techniques that work, but what's hard is just actually having access to the 20,000 30,000 chips. people are turning the crank on pre-training and pre-training became adjusted when super long context generation happened. So there's just way more return if you're if you can for example like get a sparse attention architecture to work as well and then your 100,000 token generation is 40% faster.

48:27 So I think there's a lot of co-evolution of the whole stack where they have their pre-training stack they realize that inference scaling is so much higher but even the fundamentals of pre-training keep improving where you just have a marginally better architecture returns on investment and then it kind of blurs into post- training where this mid-training where you start the reasoning behaviors you take them from your older models you do RL on them the seed behaviors like this cycle is much more complicated than

48:52 pre-training like I think it's if you don't have the right people it's a lot harder to get that off the ground than I think pre-training was like I think there's like I think like to do a controlled like let's just start pre-training our model seems a lot more approachable to me then I don't really know much about language models I'm going to figure out this whole post- training thing where it's just like the wild west.

49:12 >> Yeah. And that's why I was like when I left AI2, I felt like our post training was honestly falling behind just cuz our cycle time was not high enough >> and we're going into these agentic things and these are still some of the like >> leading open source post training recipes and >> to that's just like that's what I want to fix is like >> it doesn't take as much capital to make >> very big compounding gains on the state of the art and the open source because the bar is very low.

49:39 It's like it's not like it's not close. Yeah. Yeah. No, that that makes sense. It's just like a kind of efficiency of where you can put capital and make a huge difference. Yeah. Yeah. kind of related to the whole you know post- training side maybe you know pulling away from closed and open is the fact that the agents you know the type of harnesses that are being used in the closed labs are very different and slash you know actually trained and you know post-trained on to to you know to to be specifically good for um

50:08 their models are specifically good for those harnesses and so that in increases the mystique even more makes it very hard um how do you I guess one, do you agree? And then two, like how are you sort of hoping specifically to kind of address that? And things I mean are really examples like, hey, spin up like thousands of agents, tens of thousands of agents.

50:26 It's just like a very different workflow than what people are are obviously um exposed to. >> I think it'll be okay for open because I thought years ago that say the closed labs would be like Apple. There would be the integrated company where they have the model designed for the harness involved. It gives them um like kind of economy, scale, efficiency.

50:45 But it turned out that so many people have different harnesses for different things that they like to do that the models companies have been pressurized to make it so that their model can work in all sorts of different harnesses. And I thought it would be a weakness of open models that they have to be designed to train on all sorts of types of harnesses.

51:00 But it ended up being pretty equilibriated or like it's at equilibrium where they all have to do it to some capacity. And then what we're seeing with this like agent swarms is that OpenAI and Enthrop are going to be earlier to it. They're going to know how to design their models. They're going to know how to design what do we call what do we call want to call a harness for a swarm.

51:18 I don't know like I don't know but there's going to be a word for it. It's not a harness. It's like a coordination system. But then eventually other people are going to want to use them. And then I think it's going to be like these cycles where the Frontier Labs like they push this out and they have this >> to date they've had huge returns on being first.

51:35 like what is the returns on Navier Stokes and all these things like we don't really know like we don't know they push the frontier >> in the assumption that there will be high returns and that this marginal intent marginal intelligence has like disproportionate returns and we don't know if that's always true but I kind of expect these cycles of like once you know the shape of how do you control 10,000 agents in one harness then the open model companies be like oh we can do this too >> but it's not like Timmy is going

52:02 to do that first for one they don't have the comput. >> Yeah, >> it's an astronomical amount of comput, but these things go through cycles. Nvidia's chips will get way faster. There's more comput built out. >> So, you're very convinced that like there's not too much hidden there that you can probably guess at what they're doing and then when it becomes relevant, let the open source models actually try to also post train on similar meta harnesses, whatever name.

52:28 And I also think like >> the the security at these labs in terms of information is not as high as they think. Like Silicon Valley is famous for like ideas traveling like what is it like one beer or something and it's like everybody chats after work >> except which is why SSI is so impressive. That's what's going on. >> That's the only past. But it's like we're all everyone's scientists.

52:49 If they made a breakthrough they're going to want to tell their friends about it. They might hold it. They might hold it for a few months, but like >> I don't think there the amount of lock like information security you would need is just so high that knowledge spreads. It's hard to keep this secret. Like what you what they can win on is putting more inference into their infrastructure which lets them serve at a lower price and make higher margins and then continue this over a longer time.

53:16 So I'm very confident in the frontier labs being able to serve the cheapest intelligence per per watt per scoral. >> They are going to be able to go to the places with the highest demand and have the like the cheapest models and that will be revenue creative like that will be good business for them. The question is like they can't do everything and that's where the rest of the economy fills in is like >> is it just coding that is profitable then?

53:47 Definitely not >> good case for frontier models because they will have the mo like they have integration. >> They'll do very well but it's coding >> is the whole economy of knowledge work is the market for things and then it might not be good for them because can they do everything at once >> or they don't need to though. I think there's going to be a lot of winners.

54:04 >> Yeah. >> Yeah. I mean that's what I I mean that's why we invest in the frontier models and in startups you know um not because we want to hedge but because I actually do think you know to kind of cooperate your view that there's enough >> harder tasks at the fringe that requires a very specialized data that's kind of out of the splash zone for getting to AGI or whatever you want to call it these days um um that makes it possible to have many many winners.

54:27 Yeah, >> I do like your RSI point as we talked a little bit about it which is just kind of say practically hey it's exponential cost and resource consumption and therefore there's a very obvious kind of limit to how much that um the Frontier Labs can pull away um if that's not too oversimplification of your view um anything else you kind of want to add there because I I do feel like the the the discourse now is just so so much fear on like how how much the frontier here is just hidden and also breaking away.

55:00 I think there's so much spend on inference and that in the short term we are learning how to um corral inference into performance gains and this is like a thing that we just had and like >> this is what we are seeing now is that like people are spending so much and >> like all the employees at the frontier labs are spending exponential amounts relative to the improvements but like we're figuring out how to do this and there will be a lot of gains but I don't think that it's necessarily going to be this like fast

55:32 takeoff when the models build themselves. It's human human bottlenecks. The all the scaling laws are exponential increase in compute for linear increase in performance and exp like that's just like it's hard for me to bet against this. I would do say that like we have Navier Stokes and I'm of the opinion that math is exceptionally well suited maybe like the scientific field that is best suited to this shape of intelligence and like intelligence can take many different forms and that's not to say that like if they do

56:05 like if an AI model does say discover something that's very different than the transformer that gives very different properties and unlocks a new type of intelligence like it is worth updating on this >> and worth being like okay like that is different but until that happens I'm like pretty happy with my evidence of like the companies are going to do well they're spending exponentially more to keep this remarkable pace of progress that some of the clawed system cards say this they're like it would take say like 10x

56:36 more progress to make this a fast takeoff and it's like for now they're helping us sustain the pace of progress and I think it's a good equil like it's a good equilibrium but I do think that like the labs have people saying the the science is like it's hard to overcome that and it's okay but it's like we don't know exactly what it'll go from here. >> Yeah.

56:56 Yeah. No, I I I think that's kind of com comforting because we know what the inputs and and lovers are here. It's just there is a very successful auto reggressive regime that we've been optimizing and that one just you know asks for a certain amount of like exponential resources which are clearly capped. Um and therefore open source hopefully you know keeps it strong as well.

57:16 just it wouldn't lag behind. >> There is dramatic amount of work to do in preparing for infrastructure. Like there's the AI capability plot which doesn't necessarily worry me but the AI capability plot in the pretty much non-existing preparation of cross public cyber security and all these things like >> there needs to be more preparation in the assumption that this is coming >> than there is and then that's kind of the thing that worries me more than say like RSI is coming and unlocking new risks like there are risks

57:44 that are here and that are going to continue to diffuse and many times have said close labs get there And then the open labs get there. It's like we're in this process for a lot of things that are going to cause some problems. And it's just like figuring out how to get society to prepare is very hard. >> Yeah. >> And I come of the opinion that we need more eyes.

58:03 We need more science so that society feels informed. >> Yeah. >> But there are other valid opinions. And like I do think that not fearing fast takeoff does not mean that you don't worry about risks. I don't like you. >> No, I think that that's completely Yeah. Maybe I'll I'll send in uh some people your way that are defected from the lab worried precisely about this.

58:23 Well, thank you so much, Nathan. This was a lot of fun and um yeah, I'm really excited to see what Trillium is um up to now that it's finally launched. >> Yeah, thanks so much for having me.