Building the Automated AGI Lab: Core Automation's Jerry Tworek and Rohan Anil

If I play football, for example, it looks very, very close to reinforcement learning. I can get ball a lot of times and every time I adjust it a little bit and I see if it roughly matches what I want it and I'd order some self-reinforcement head money. When I learn mathematics, it's very different type of thing. It's like reading about hard concepts and thinking about them very deeply inside my head until things click and until I have them connected. And both of those in some way are learning from experience. They are just very different. We probably are spending the most compute that, than ever, on learning from experience. But reinforcement learning is not the end of learning from experience and there will be better approaches that researchers will be coming up in the coming years on how to help to use that data. Jerry Rohan, thank you so much for joining us today. The two of you are the founders of Core Animation, one of the hottest Neo Labs in terms of scope right now and before studying Core Animation, you led some of the most important research projects of the AI era. Jerry, you were VP at OpenAI where you worked amongst other things on running the strawberry and reasoning teams. And Rohan, you were one of two of the four pre-training leads at Gemini and before that led a lot of the fundamental AI research at Google Brain and were the fix it guy across Google and then Anthropic and so between the two of you, you've seen more than your fair share of what the world looks like in terms of doing frontier research. And so I'm very, very excited to dig in. Let's start with you, Jerry. You tweeted a very spicy take recently. The first step to replacing Transformers is appreciating deeply how far they were able to carry us. Is that a eulogy for the Transformer? What does that mean? Thank you very much for inviting us here. So yeah, I feel like a lot of my interviews these days is explaining my tweets and what did I mean. But appreciating Transformer means like understanding what it does. Well, I'm not solving the problems that it is solving well. You have to focus on its weaknesses. You have to understand good parts and bad parts. And it's very easy and a lot of the work what people are doing in the architectures is trying to make Transformers cheaper and trying to make Transformers more efficient. I very rarely see people thinking about how do we make Transformers more powerful, trying to do more expressive. But seeing someone weak parts and seeing someone strong parts are almost almost almost the same thing. It's just understanding the shape of Transformer a little bit more. Well, I think right now we are in this stage. We got to really, really good at training really, really big models. We mastered two algorithms. We mastered pre-training at a large scale. And we mastered reinforcement learning at a large scale. And I'm asking myself a lot what is next in machine learning. I think at this moment what the bottleneck is two better models and two smarter systems as the architecture itself. It is this moment to revisit the train we've been riding for the last six years of trying to add more and more parameters to essentially two of the same. Operations, which is MOE and attention. And when I am thinking about what we are today and what we are doing, I am thinking a lot about what codecs and what cloud code are doing for us. And I'm really, really appreciative of those systems and of the coding and of the workflow automation and of the systems of the products that we have today. That we essentially have built over those six years of scaling. And I think this is the first step of thinking. What is the, if we want to work on the replacement, we need to like see where we are, what problems we have solved, to like start seeing what the next stage is, what problems we haven't solved yet. What kind of are we missing? And this is this is kind of whenever whenever I use codecs and I am successful at a task. I also start thinking why didn't I try to push that thing harder? Whenever whenever I come to work, there are a lot of things I do with codecs, but I still come to work. I still ask it to do certain things for me. And I'm always asking myself, why am I even needed there? Why is a core automation as name and its concept is we want to be automating tasks? And why are those things not yet automated? Why is not codecs doing everything for me? And this is the question of like the research where we want to go. And with that research, I'm trying to think what kind of models, what kind of systems do we need, what kind of qualities do we need with that that we don't we don't have today? And that's what I'm thinking a lot these days. And you have the starting premise of the architecture is the issue, which I think is a contrarian point of view. So what led you to that point of view? What did you see that made you think the architecture was the issue? It's fundamentally what is the issue. It's like it comes back from the from the previous implication. What I think is the issue is that the models are being trained in the lab and are being deployed in the real world. That is that that is the fundamental fundamental tension that is that is there. And like a bit of a bit of my disappointment came comes from my my personal story. Whenever we were starting the research and progress on scaling up reinforcement learning at OpenAI, I basically believe that scaling up reinforcement learning is a necessary stepping stone on a path to AGI since since I started like working working at OpenAI. And I was always reinforcement learning maximalist. I was always believed this is what we need to focus on. This is what we need to do. I've seen LLAMs being scaled up to to to higher and higher levels through GPT-3 to GPT-4. And we're still doing very little RL. And I had this this internal belief that the moment we start scaling up RL, we'll solve everything. We'll we'll be able to solve solve all the problems. And we eventually started to solve scaling up RL. I was I was just in there. I was in the center of it. I was thinking here we are. If you ask Jerry in 2024, when do we get AGI? I would say 2025 will be will be that year. This is where we solve everything. And I saw us training model after model. This model was getting better and better. All the benchmarks course were going up. And did we also solve all the real world tasks at that moment? Unfortunately, unfortunately not. And we still have work. And I realized there was this bit of distinction as all the benchmarks that we are evaluating our models. They were essentially the same thing as we were training the models. Like all the evals and training tasks are the same sides of the coin. But the real world distribution and real world tasks as much messier, much murkier, much more, much more different art training data didn't really replicate the real world use cases. And despite us basically maximizing all the tasks. If you see ask anyone training models. Hey, what is one of your main issues? I don't have hard enough tasks. I don't I don't have what to train our model on yet. We are still not covering the entirety of the real world distribution. From that, my conclusion is we need to have models that learn a test. I need to have models that learn with users on their data, on the real world task, on the real world distribution. And there when you ask, why don't we have that today? Why are transformers not learning anywhere? And there are essentially two types of learning that we could be doing at test time. We could be doing end context learning essentially of transformers, which is it doesn't have fundamental problems of catastrophic forgetting. It doesn't have that issue. It is pretty data efficient. So that that is great. But it's not very scalable. We only can have so much of it is limited and has some more, even more of mechanical limitations of what actually are you doing when you when you build context, but maybe we can we can come back to it later, but we have in context learning, which is very limited and very, very small amount of data. Whenever I'm using codecs roughly after around 20 minutes of usage, I need to I need to compact it and move and move it afterwards, which is not that much, not that much data. We have all we can learn as far for 20 minutes is not that much. And the second thing is fine tuning. We could try to continuously fine tune our models, but then those have the issues of catastrophic forgetting. We have issues of very low data efficiency. And neither those are very solvable. Neither of those are very easy to to find ways people have been trying. If there was were easy to solve some already solve it. So my personal belief is we need to find an algorithm that we can we can metalar. We can express on the architectural layer that can represent how does how does learning look like. How does learning look like that can work on much, much longer horizons? Do you expect the architecture will look transformer like because my my understanding from the from the chiefs states is that, you know, open I had been trying the scale of reinforcement learning for a long time. And it wasn't until the transformer came about that it seemed like there was an even kind of scalable prior on the world upon which to even scale our L. And so how do you even go about trying to think about scaling up this this new regime? Yeah, it's a great question. I think those two things happened at the same time, but if anything that that happened there was mostly about the economics because technically you can scale up LSTMs. Just no one really, really dared to go in that direction and they did scale much, much more poorly. They are scaling in a scaling loss paper. There is presented a comparison of LSTMs and transformers. And fundamentally the scaling loss of transformers was better. There is a world where we never invented transformers and we would be scaling LSTMs and we would be having some models. But because they would be much more expensive to train and much less impressive as a product, we would have just a source experience and maybe no one would be able to convince people to spend as many dollars training that those gigantic LSTMs because we wouldn't get the market return. The majestic thing about transformers, which goes back to, why do we have to appreciate transformers? So deeply is that transformers are economically valuable. They're training them. The cost of training them is lower than the revenue that they generate, which is magic of machine learning. It's not guaranteed by itself. But for LSTMs it probably wouldn't be that way, which made it happen. But many ways you can scale most of the architectures. I think a lot of reasons why people didn't scale things before was because researchers before opening I had a lot of reluctance to scaling. It was often seen as unscientific and research in algorithm was providing how do we become more and more efficient? How do we further seeing compute budget get better and better result? And it was a bit of a contrarian bet by OpenAI at that moment, try to say, hey, we don't care about better and better algorithms. We care about more and more scalable algorithms. And how do we pour more and more compute and get better results, which OpenAI was criticized repeatedly by many people in the community for a long time? But thanks to that, we have the models that we have today. And I think there are tons of architectures that can be scaled up. And I am part of the core automations mission and our belief is that a lot of architectural research happened at two small scale for too long time. A lot of people are trying to say, hey, let's try to first try our architecture on a small data set, on a small computer regime, and then see where it scales only after you prove itself. But for example, when you do work on reinforcement learning, you know that to get to any interesting results, you need certain level of compute to even see the capabilities in the model. Reinforcement learning needs a baseline of ability to only to only start working. So where I am coming from, probably there are many architectures that need a baseline of compute to even start doing anything, anything interesting, anything useful. Can I see there maybe a touchy question, please do? If you need a baseline of compute, that sounds like a job that would be well served inside of a big research lab. Why start a company to go do this? It's a great question. And it's, I think in many ways, it's likely a timing thing, timing issue market is right now in a very specific place where the biggest and the most successful labs by coincidence or by fate are probably in the most competitive market fights ever right now, which makes them not very keen on trying different paths, trying alternatives. If transformer is profitable and if you can spend more efforts and more resources scaling transformer to win in the next quarter, it's very hard to put this a lot of attention and a lot of energy to work on something that will, that will maybe better or maybe, or maybe will redefine the field in a year or two. So actually, I think the biggest labs, and I talked to basically all of them, don't have that much interest in trying the alternatives to transformer, and the labs that are not the biggest are doing whatever they can to do what they, what the most successful labs are doing, and everyone is trying to try the same coding agent. If you look at the last week's releases, everyone is trying to release a coding agent right now, and I think we need different paths and different approaches here. So that's what the niche in ecosystem we are trying to fill in. - And Rohan, you were at brain when the transformer was invented. Do you agree with Jerry's eulogy for the transformer? - Yes, in some sense. Like, once the first when the transformer is ashyish, you know how many of this came up with it. I had like worked on my work on online distillation around the same time we presented it at the same internal research conference. It wasn't a big deal internally. There was only a few people who actually got it. A lot of people were like, oh, it's like, yeah, it's another work, and people were finding ways to, and it was also very focused on, at least the original work was very focused on a real problem, which is translation. So they saw like the Beatles DM on translation, and it took opening, I mean, internally at Google, there was definitely at least like, you know, and a few others were definitely interested in scaling language models. I think it is until GPT-2 and GPT-3 that we saw the benefit of transformers working quite well. At least the way I think about architecture is how do we spend computation? And transformer is one way, very efficient way to spend computation. But now that I look at the industry, it's a lot of our computation is inference time and spending it on tokens. Let me ask this question. Like, if I want to optimize for a better architecture, I want to look at both pre-training and RL together. And I would like to find architectures that spend computation much better than current chain of thought token generation. In order to give a much better overview, I think of like pre-training as built the transformer with certain context length. And RL comes in and it's like, well, that's not sufficient. I need more computation. Let me do it while adding one token at a time. This is quite inefficient from inference perspective. You're doing one token at a time. So most of the solutions have been finding to do better ways of speculative decoding. So it's like a band-aid to a problem that we've picked something that can only generate one token at a time, so autoregressive decoding. There is problems with the transformer in terms of how do we spend the computation for the longest time. I think most of the world was training very large dense models. And it took the industry roughly two to three years to get to refine the architecture to what we now take for granted was not obvious to a lot of people, sparsity and a mixture of experts, and getting good training efficiencies with them. So then you can ask what's wrong with the transformer? Well, it's the computational depth is poor. How do we increase computational depth? And just posting that question opens up like 20 new directions on how we can modify the mechanism to incorporate it. So I see like to do work like this, it takes time. And usually like fundamental research in the past have taken like five, six years to land into industry. And it's largely from organizational, knowing that it is important. This is the bet. Like just like Jerry had the inner belief that RL is needed. Absolutely did not have that belief at Google. I was a pre-training maximalist. You guys are good fit. Thanks. All right. And then so that inner belief and second is, you need your architecture to run efficiently on hardware. A theoretically optimal architecture is not useful to anyone. It is something when it comes into practice. So you need the research inception to getting it productionized and getting kernels and everything written, the end to end loop. And there's only few places right now which have integrated teams doing that. And I think we have built the team in a way that puts the experts together, not in different silos, that like we are accelerating on having everybody look at the problem holistically from end to end. So that's like where I'm quite bullish. That's why I'm here. The current mechanisms are quite poor. And if you leave it to the world, I am afraid that it will take us like a much longer time horizon before we replace the transformer. And I think a lot of folks are already complaining a lot on token costs. And that seems like as someone-- We're not complaining. I mean, in terms of like, yeah, exactly. I come from the Google mindset where we had to serve billions of people so like finding more efficient architectures that fit have a deadline on latency and the number of tokens that you can serve. So when I look at that, the amount of the world that can use like Frontier Tech is very little. And we-- someone or some group has to like accelerate and make this better. And we are taking that shot at doing that. The current technology just scaled up is still only relevant to like a subset of humans. And this is the bet you're making. So one of the things I heard you say was the problem with transformers is the computational depth is poor. If that's the crux of the issue, tell us, what does that mean? Why is that the case? How do you fix it? I can give you like one insight. Like most transformers that we train are quite shallow. It's at most like 100 layers deep. Depth is like, it's called deep learning because you want a deeper representation. There has been experiments on going into depth where no one has actually shown us learning extremely deep representations. Chain of the art reasoning and RL to do chain of thought by model itself is one way to increase computational depth because every token you add, you add like one more pathway. And there has been, so then you can get out of like this bottleneck. that the pretrained architecture has set you up on, you can only do a number of layers times sequence length. Now you can increase the sequence length and you get much stronger results. You can do inference time scaling. Now, the issue with inference time scaling is that models now have to produce more tokens to get better results and that's very one token at a time. And from this, you can see like you can directly address many of these things and this is like a subset of work that we are looking at making this much more efficient. What's your forecast for the transformer based architecture? If it's not the end state, how far can it get us? When do we start to see it topping up? I think it all comes back to what we are, what we are training transformers for and what we can do with them. We're doing pre-training, which is very good at distilling all the knowledge from the internet into transformers and then we can arral them for, which is we basically can bake all the workflows that we want into a transformer. So what we're transformer is like capped out is we have all the knowledge of humanity in the model together with the relationships and how do they work together, how they can be combined. And basically any task that we have training data for, we can put into the model and this can be gigantic model trying with a lot of compute on all the data in the world. And then if we ever stop training that model, what would happen, the question worth asking often I'm thinking about transformer, what would happen if opening an ontropic stopped training models and we got a transformer we have today and say, this is it, this is the best model we have. Months pass, your spouse and the model is getting less and less useful. Maybe the lab really recorded all of every human on earth, what they were doing and what their tasks were and their environments and put them in the model, put them in an learning environment. But then what happens if anything of that changes? If there are new events in the world, if those events have new relationships between them, if there are new types of tasks, if there are new code bases, new tools to use, transformers are getting a lot of their usefulness and value through the things that are valuable have to be present in training. And when they are not, they suffer or there's some ability to adapt, but it's not very big and not very flexible. So in my mind, this is kind of the level where the transformer's top, which in many ways, what I think is a tool to use for us. If we kind of know if there's human who knows the limitations of a transformer, they can schedule that model, they can write a prompt of what is what is the task that you want. And by doing the training we are doing, you can get very successful in that and then you task the model fails, you can add it to the training data and the model can succeed. But that loop has to go for the lab training the model for you. And if the model that fundamentally needs to be trained in the lab, how much do you think of it that this is the goal or you would want to be able to update the model somehow, I'm not having to go back there. - Have you read the Rich Sutton and David Silver have this paper, "The Age of Experience?" Have you read it? I'm curious how much you agree or if you have any different opinions. Where do your opinions diverge? - Reinforcement learning is not a particularly new approach, particularly, particularly new thing to do. So in some way, in some way, age of experience, I think always has been there and people have been criticizing a bit pre-training 'cause pre-training very clearly is, 'cause this is the other way of looking at the models, which is like we have static data that data is mostly generated by others, although I have this personal view that pre-training today is largely distilling other models, other models into the new model because most of the tokens in the internet are coming from AI, but there are clearly pre-training, which is behavioral cloning, which is mimicry, which is compression of internet data. But reinforcement learning is not something that people haven't been thinking and people haven't been doing. Reinforcement learning was used to solve by gum on backing the data, used to solve goes, darker after data, to solving programming right now. And every time it comes down to model writing its own experience and learning, learning from that experience, what is very clear? And what I think is interesting and what I think is still perplexing to people that reinforcement learning is not really the only way to learn from experience and there will be more, and there will be a little bit more of, I think, you can call it algorithmic, but essentially innovation of how we learn from experience, just because reinforcement learning is only one way to do it, it's a mathematical formulation, and especially right now how we are using it, it really likes those parallel rollouts for variance reduction and for comparing how the model does in parallel versions of the world, which is not how we learn from experience, we learn from our experience much more efficiently and much more. We use those in many, many, many ways. At some moment I've been trying to explain to people what brain does, how we learn. So there's one learning algorithm in a brain. I think there are multiple actually, and they work together, but if I play football for example, it looks very closely to reinforcement learning. I can get ball a lot of times, and every time I adjust it a little bit, and I see if it roughly matches what I wanted and I ordered some self-reinforcement head money. It's like reading about hard concepts and thinking about them very deeply inside my head until things click and I until I have them connected, and both of those in some way are learning from experience. So summarizing my thinking of the learning from experience is that we've been doing it for a while. We probably are spending the most compute than ever on learning from experience, but the reinforcement learning is not the end of learning from experience, and there will be better approaches that will researchers will be coming up in the coming years on how to use that data in a richer, richer settings. That's interesting. Rohan, I'm curious, since a lot of your work has been around optimization and efficiency, how do we get to a orders of magnitude more efficient, I guess, more compute efficient and more data efficient running algorithm? I had to start with measurement. I think pre-training as we define it right now is about compression. We look at public city and then measure how do we decrease the public city, and then we find that scaling and increasing parameter count and putting more compute is the way, and every time we increase a compute in log scale, we get this absolute more improvement in these metrics. This is, I think, this is fine for building the prior, but I think this is the wrong way to look at the problem. We should be looking at the end-to-end. What are we training these models for? Look at the outcome. Like, for example, I trained this model and give it to Jerry. Jerry will do RL and destroy all the public city metrics that I have created, right? So then it's sort of like, it was the best way we had so far to attempt to solve the problem, and I think the labs in everyone else have done a great job in producing intelligence that's super valuable and makes my work so much fun. But it was the bootstrap process to get there. We have to combine pre-training and RL together, and that's like where one order of magnitude improvement would come from, and that's like a training procedure. You can say it's a learning algorithm. In terms of optimization, my story is, I studied optimization at Google for logistic regression back in 2016, got nerd sniped by it, worked on some solvers for what we used to call Sible, which was the large-scale linear solver that was used at Google before neural network took off, and then replaced this. Then I asked myself, like, what do I want to work on with neural network? And it was quite clear, like, I want to understand the training algorithm and make it better. And then someone beneath CUP just showed up one day at my desk. It's like, I heard you already got it writing, optimization methods. We have this idea that, you know, like we worked out on a whiteboard, and it turned out to be the shampoo algorithm, can you help us implement scale, make sure it works at large-scale neural network training. So then as working on this, I think my manager, Yonghui Wu, who supported it throughout that until my end of my tenure at 2024, but largely like the community and most of the other people were not as excited by this idea. And for me, this is the most exciting thing, 'cause I was like, I'm putting in computation and making training better. This is the thing I have to figure out. I will spend as much time I would take to do it. And then people were making this assumption, oh, why, like, what's the upper boron? You could still use Adam, that's fine. Like, why are we, you could spend all the time on everything else, not optimization. But in some sense, optimization is like your model, you're optimizing it, you wanna optimize it better. Now, I'll connect it to some of the stuff that we talked about architecture. What has happened is that a lot of the work that we've done in architecture is to make these network strain. And in some sense, it's like two sides of this coin and optimization and architecture go together. You could have a stronger optimal. train a much harder to optimize model and get better performance or you can use a weaker optimizer on easier to optimize models and get decent performance. So these tradeoffs that appear all over and for me it's, I spend a lot of time working on it. I think we used it for Gemini 1.5 Flash and then like the community started getting like more interested in it. There was the soap paper published. We have like an entire literature of like shampoo so all like bath time, the things that you would use. And then it was quite clear like so that was like maybe a 2x improvement over what was happening. But even then if you look at shampoo it's quite weak in what it's doing. It's not using all the information that's available to you when you train. And as you use more and more information as part of training, you can get better improvement. And in some sense like your optimization algorithm defines what architectures you will discover. Like I have colleagues it's not very popular in the literature is only like maybe except like four people in the world care about it kind of ideas that are extremely interesting. Like residual connections have been extremely useful for training neural networks. There are folks who've now like gotten rid of them and learn deeper representations. But they needed a better optimization method. So like for me optimization methods and the question is just like how do we get this is combined with architecture and thinking about the problem end to end is where a lot of the computational efficiency is in it. And I also see RL as spending a lot of compute not as efficiently. And so like if you could spend it because you don't get much feedback and you're spending a lot more compute because you have to decord all this long chain of thought to get this one bit of information into the network seems quite inefficient and an easy target to get orders of magnitude on top of. I can go on talking about optimization. I love it. How do you think we'll over approach or surpass biological learning efficiency? I do not think so because I think we would need to change maybe like I was a strong statement at least with the hardware we have it seems pretty unlikely. Our biological like we have something as Jeff indents is model computation. So we built our own circuit as we grow up and we built our own learning algorithm with the hybrid and then we die and then we gone neural networks are quite different are hard the hardware stays the neural network stays but it's learning very inefficiently and you need a lot more of them and a lot of parallelism to get small amounts of information through. So until I think we design hardware to be much more like how humans operate maybe more analog figure out how to deal with analog circuits and to figure out how to do it at a correction figure out how to get information through it to be much harder view I think we are safe. Safe. That's an interesting way to put it. The idea that pre training and RL should be optimized and and seems like such a clear maybe obvious statement. Do you think the labs realize this and is it just hard for them to you know get rid of org charts and process to be able to make that come together or what stops the labs from being able to unify the two. I don't think it's as obvious because it's a completely again different optimization problem. You have a prior you're doing a lot you have higher variance your gains and then pre training as much large patch like more parallelism or compute for the unit of time that you can spend. So it is not an obvious thing for folks to combine these two training procedures until you think a bit more like why is it that the naive combination doesn't work. So that's one. The second one is if I pull like some of the best researchers in these labs they would say oh this mix ends we should probably explore it but it would be probably not in the top bucket because they have to train a model for the next cycle before it's like as Jerry said like there are companies now competing for release cycles because tokens are not sticky. So it's much harder to do a long slight like even long term research of six months in many of these labs in the end and in the end. So it seems like the core premises for core automation is you know you're starting you're starting at the lab at a time when no Sam has been talking about the AI scientist I think Dorios and talking about the AI scientist it seems like your job as researchers has actually fundamentally changed and you get you get to start the company native to that era and as a result you're maybe able to run a lot more experiments than otherwise might be possible. How admirable do you think the research job is and how are you guys approaching building your lab to be as I believe your mission mission is to be the most autonomous lab there is. Most automated lab in the world and to start with I think the automation version of automation by core automation is about giving each human maximum level of agency in some way it is we are not trying to really get humans out of the loop which is like one version to automate but it is about give humans ability to do the most with with their their amount of time. Whenever whenever you are walking you can get some distance whenever you get a bug you can go go walk a larger distance whenever you are a car you can you can go much much much much larger whenever whenever humans started farming they had to farm by hand and will work on a small plot of land when you have a machine you work on a much much larger plot of land. Personally I am both really great fan of the current coding agents and I'm very happy it's in some way it is what I've been working for many years both doing coding research and working on various versions of AI scientists inside of open AI and in the end I realized starting a company to realize that vision as this is one of the best ways to realize it because the way you can do research today is very very different because a single researcher can do much more in the end the speed of iteration the speed of research the speed of how quickly you can you can move through ideas and how quickly can get data on your ideas is something something very very different and you can try to move the old structures around it and the teams workflows how how data is gathered or we can try to build like like you said you can try to build build relatively for it for processes that that like maximally empower each researcher and allow them to just to just iterate on their on their idea much much quicker we are we are here and we are trying to to to rebuild the full deep learning stack and try to think how we can do almost each operation differently what are what are various options and if we can execute at least even one of those experiments a day that's already a pretty good iteration speed versus anything that was done before and there isn't really like any any fundamental like lots of physics reason why not and maybe maybe one day we get to to ten of those a day maybe one day we get we got to hundred dose a day and fundamentally for that like search process optimization process we should be able to just just find things that work in a better deep learning setting and like go what what we are trying to do is like within like almost all of us we are we are a team that is very agent-pilled and automation-pilled we are we we're trying to do the next experiment like how far how far how far we can we can push those things and how far how how much an organization that tries to do as much as we can with a small team how far how far we can get we've done when will we know that we've reached AGI at some moment I used to say it's it's very much in everyone's heart whatever whatever they consider AGI opening I says the system that can outperform all humans in economically valuable work but it goes to my my my my previous statement what if what if open AI stops training models yeah with that with that would that would that still keep working when they still would that still keep the automation level the same the same or with that 2.3 and for me AGI is a model that can improve itself without human in the loop in any in any in any way that said that that's I think I think the moment where we can meaningfully talk about about AGI because that is in some way it is a sub definition of the previous one because improving AI model actually job that humans can do it is economically valuable work and it the family it the family is their case but removing humans from from from loops with models has been actually notoriously notoriously difficult so far we haven't come anywhere close to it it's very very hard for me to find any task where lm's were able to like get to get humans out in the loop we are we are the they the human lm hybrid is really really successful right now but all nams without humans not so much not at all and what I have seen there in 2024 and in early 2025 is that the current like puff doesn't doesn't get us there and I think we need we need some pretty serious research on why this company or some other to try to unlock how do we how do we make our models learn and and adapt it's a test I'm on a deeper level than they've been so far I feel like we've alluded to this throughout the conversation but it's kind of been one of these you know five blinded blind folk people trying to find the elephant things what is the grand master plan for for core other that you're willing to share i can share our six month pro map in some sense like building architectures as I said like wouldn't it's not just how good the architectures does it run well and can we get users including ourselves as part of the lab to use it right so then that's directly like we can do many things now, but the thing that is going to be difficult and think that we want automated away is kernel generation. So we would we have a set of hardware, GPUs, black wells that we have to train and run inference on. We will build the best model that we can to basically reduce the time from having a very cool idea that can make these bottlenecks go away in the architecture to having them run at the highest T-flops on GPUs. And in some sense like current coding agents plus humans can go a long way. But like an example of this is our QR kernel competition that we hosted with GPU mode. It's for running this very old linear algebra operation QR. It's used for optimization like shampoo line of work uses it many other places uses it and you want to run this efficiently on B200 node. If you use crossover for the shapes that we care about you get some efficiency and then a human plus some search loop can get you like something like 7x. But it requires the highest taste human like there's maybe three people in the world to and spend about $100,000 on these coding agents over a span of four weeks to get to a solution that's 60x faster. So these models today are nowhere close to getting that 60x faster kernel. And there is a real bottleneck. Now that was a single problem. It has like perhaps three different operators a work on this panel, do this matrix multiply, fold it back in and do this repeatedly. That's what the secure factorization of the matrix would look like. And if you give this problem to and topic models, open AI models, Gemini, it just wouldn't solve it. It just it's not our models are not even close to solving this problem. So for us, it's like something that we've talked about, something that we are getting close to as sort of like getting to that point because that's our inner loop to having more efficient architectures. Why kernels? Is it just because like maximized intelligence per flop of compute? In some sense, I've had like three projects two of them have kind of landed in the industry. So first is secondary methods. Kernels was a bottleneck because you have to run it well. If you were at a place like Google, you cannot spend 10x amount of compute and get a 2x with. So I could only spend maybe a budget of 20% and get the 2x when everyone's happy, but like straight. So like I think that's the market, right? You spend less than you get. So kernels ended up being a bottleneck there because most of the operations were novel that we haven't gotten a lot of people to look at it. There's only like two humans at Google who could write it, Russmas and Peter Hawkins, because it was deep XLA-Lolo code that you have to write, do make this work and then that took them two years to do. The other idea I had with Monomeco workers at that time was like replacing some of the parameters in a transformer with extra memory. And we called it engrammer engram memory. We worked on it in 2020. It's a we had versions of it internally deployed, not the big version, the smaller version, but there I needed like something that can accelerate sparse gathers and scatters as part of training. It required hardware change and a hardware making use of the hardware. It never arrived. I had conferences set up with the TPU team us and a bunch of others. We were talking about it and during COVID, like, oh, we're going to have it happen and it never arrived. I also was using TPUs at Anthropic while I was leaving just barely started the surface of being able to do it. But at the same time, six months before that deep seek wrote their engram, which is an improved version of adding more memory, showed scaling loss that, yeah, you don't need MOE, you could actually replace it with these engram embeddings. For me, that was like, oh, yeah. It was like a five-year thing and I was very happy for that. So much of the state space to explore isn't even possible if you're not writing kernels. Cornels and you need to be assisted in writing kernels or solve that kernel to have the highest performance. And the roof line is pretty high. So it's like the QR. If I use Qsauva's QR, I get some performance. If we use the competition minors QR, you get 60x faster. And that is a completely different playing field. Now it opens up an entire new set of algorithms you can apply. And in terms of training transformers, training optimizers, QR is so fundamental in analyzing the Agenda Composition to be for Agenda Composition and many other things. So it is a thing that I think also, if you think about it, only a few people have the skill set to. And they're very much not at the same place. It's like one person here, one person there. And it would be ideal if models had those abilities. Yeah. Maybe I'll summarize a little bit and talk from the high level of what we want. QR automation is a lab created to build models that continuously learn and learn from deployment. We believe, as I mentioned, that transformers are incapable of continual learning. There's no way how to put continual learning on transformers. So we know we have to find a different different architecture. Some of any way our quest is to find that new architecture, find that transformer replacement. And we want to build the most automated lab to do it. We want to be able to build experiments at scale the quickest we can iterate on them, try a lot of new architectural ideas, have strong priors of what we want to do to search the space of architectures efficiently to find to go to that, go to that place fastest than anyone else. That's what we want to do and all the work we are doing on kernels, on large scale training, on trying new architectural ideas is exploring that space. So you're going to experiment your way into finding a superior architecture. How will you know when you've found it? What are you looking for to say, aha, this is the one. That's a great question. Are there always two angles? In my mind, you know, experience, every successful research had a plot that shows something that other other places don't show there. There's one line that is a little bit bending in a different different way. And you're saying this is what you want, but at least it is my experience with research always has been that plot is already quite late in a journey where most of the time you already know what you want and already know what you are up to. I am a bit joking, but it's actually true that all the best plots in my life. I have them in a dream before, before actually they were real. I kind of knew what I was looking for. Just the question is, it's like when it actually clicks. If you know what I mean, because most of the time you kind of know what you are looking for, but you are not finding it. You try one thing and it doesn't work. Try second thing, oh, it doesn't, it doesn't work. But eventually all the right pieces fall into it and most of the deep learning systems are where I enter a cat. So usually you have to get five things right in a row for the for the main to start working. And then eventually you get the plot that looks like like like like you want and then and then you know. So so I haven't think what we are what we are looking for is systems that learn and test time. And if we see meaningful long term adaptability of our systems and like we are we're joking, but it's it's a it's a real we want to be evaluating our systems of our every day work, they get better at doing the work of core automation scientists each day. Yeah. We like co-onification as a team and see if the lab produces something better for the week and give them what you do when you get back. We'll see where I extend the vacation two times and four times. Until we are on permanent vacation. That is a beautiful note to end on Rohan Jerry. Thank you so much for joining us. You've both worked on and really really transformative work for where we are today. And I'm so excited to see you starting a lab on this new journey and very excited to see what you're able to come up with. Thanks for joining us. It's been great to be here and to chat with you. Thank you. [Music]