DeepMind's Pushmeet Kohli on AI's Scientific Revolution

Pushmeet Kohli leads AI for Science at DeepMind, where his team has created AlphaEvolve, an AI system that discovers entirely new algorithms and proves mathematical results that have eluded researchers for decades. From improving 50-year-old matrix multiplication algorithms to generating interpretable code for complex problems like data center scheduling, AlphaEvolve represents a new paradigm where LLMs coupled with evolutionary search can outperform human experts. Pushmeet explains the technical architecture behind these breakthroughs and shares insights from collaborations with mathematicians like Terence Tao, while discussing how AI is accelerating scientific discovery across domains from chip design to materials science. Hosted by Sonya Huang and Pat Grady, Sequoia Capital

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Speaker 00

  So I went to a biology conference and after I gave my talk, a biologist approached me and he said, Pushmeet, I have been working on this protein for the last 10 years.

• 0:15Speaker 00

  And I had collected so much lab data to characterize this protein, to figure out its structure, but somehow this had, this has alluded, like this has evaded all kind of investigation and we still didn't know the structure. But we had all this data. If we knew the structure, we could sort of validate it very quickly. I ran AlphaFold 2. It gave me the structure. It perfectly fit the answer.

• 0:43Speaker 00

  I've been working on this for 10 years. Wow. What do I do next?

• 1:05Speaker 01

  What happens when AI stops just answering questions and starts asking them? In this episode, Pushmeet Kohli discusses DeepMind's AlphaEvol, a breakthrough evolutionary AI system that discovers entirely new algorithms. Pushmeet reveals how coupling language models with evaluators creates something unprecedented, AI that can tackle decades-old math problems and generate human interpretable code that outperforms expert-designed solutions. Pushmeet shares stunning examples of AI uncovering hidden mathematical truths and explains why we're witnessing the emergence of a new scientific method, one where AI doesn't just accelerate discovery, but transforms which problems we can even attempt to solve. Enjoy the show.

• 1:46Speaker 01

  Pushmeet, thank you so much for joining us today. We've all been eagerly waiting for the moment that AI is capable of making novel scientific discoveries. Do you think that AlphaEvol is that watershed moment?

• 1:57Speaker 00

  Yeah, so it's certainly a key sort of milestone. What we have shown is that you have AI sort of model, a large language model, when coupled with a harness is able to discover new algorithms. And, and not only that, it's basically able to view new mathematical results, which have been studied for many, many years.

• 2:23Pat Grady?

  You used the words "when coupled with a harness." Can you tell us more about that harness?

• 2:28Speaker 00

  Yeah, so if you now go back to AI models, like the history of AI for science is very long. We have a number of different models that have tried to do scientific discovery. Like, one of the key models in this category is AlphaFold, right? Which is the prototypical example of what can be achieved by AI in science. We released AlphaFold 2 at the end of 2021, and it won the Nobel Prize last year. So the impact of AI in science is very well understood.

• 3:09Speaker 00

  Now, the question is whether LLMs and foundational models, how can they impact science? In around two years back, we had an agent called FunSearch, in which we took an LLM and we coupled it with an evaluator. And the evaluator allowed the LLM to figure out when it was making new conjectures on sort of coming up with new ideas to solve problems, whether they were hallucinations or whether they were brilliant insights. So essentially, in this particular case, hallucinations were great

• 3:49Speaker 00

  because some of those hallucinations were, in fact, brilliant new insights that nobody had thought about. So this is where the harness comes in, that you have this evaluation function and essentially a search protocol associated with the LLM that together is able to come up with completely new discoveries that are really impactful.

• 4:14Speaker 03

  You mentioned FunSearch. Could you say a word on the difference between the results you all accomplished with FunSearch versus AlphaEvolve?

• 4:21Speaker 00

  Yeah, so FunSearch was our first instantiation of taking a large language model and trying to see if it can discover new algorithms. The models at that time were weaker, right? And the type of search that we were trying to do, we had not sort of explored things much further. So what we asked the LLM to do was essentially try to complete a small function.

• 4:52Speaker 00

  And see if it can do that much better. And surprisingly, it was able to discover completely new algorithms that mathematicians had been trying to study for a long time. But the limitation was that the mathematician or the researcher had to give a template in which the algorithm should be found. With AlphaEvolve, we have removed that restriction. AlphaEvolve is not just sort of searching for a few lines. It's basically looking at whole sort of whole algorithms themselves, right? Very, very large pieces of code and sort of optimizing them over a long period of time.

• 5:37Speaker 00

  And secondly, FunSearch, our original sort of model, used a lot of function evaluations to discover, to make these new discoveries. AlphaEvolve can work with many fewer function calls. And it can basically, by looking at fewer proposals, it can discover new algorithms much more quickly.

• 5:59Speaker 03

  Can you tell us about the role that the evolving Gemini models play in the capabilities of AlphaEvolve? And I think I saw on your blog post, you have both Gemini Flash and Pro involved in the in the harness. What is each responsible for?

• 6:13Speaker 00

  Yeah, so I think, see, we have been evaluating, as Gemini improves with the various sort of generations, it is becoming much, much better at its understanding of code. Now, if you have a proposal generator which can understand code much more effectively than it gives, it generates proposals which are not only syntactically correct, they are also semantically trying to solve the task. And then you are sampling what are the different ways in which the task can be solved. So as the baseline model, Gemini's abilities to perform coding improve.

• 6:58Speaker 00

  Our sample effectiveness in searching for the right solution on these very hard maths and computational problem becomes much better. So, so if you want to search in a large space, there are two sort of elements. One is the speed of how you can generate these proposals. And then the speed at which you can evaluate those proposals, right? Like, so first, how quickly can you say, can you give me a new sort of candidate algorithm? And then secondly, how quickly can you evaluate whether the algorithm is any good or not? And both things are really important.

• 7:40Speaker 00

  And the fact that you have these variants of Gemini, like Gemini Flash, which can do that very efficiently and very quickly, this is really important.

• 7:50Pat Grady?

  I know Alpha evolve is more of a broad domain model than some of its predecessors. How broad is it? What's in scope? What's out of scope?

• 7:58Speaker 00

  Yeah, so Alpha evolve essentially allows you to search not only in terms, not only in terms of the size of what you can search over, right? You can now discover whole new algorithms, but it also is extremely general in its ability of thinking about algorithms in various different languages. So not only can it sort of search in C++ but it can also do it in Python. It can also sort of do in Verilog, which is what the language is for define describing chips, right, in chip design.

• 8:38Speaker 00

  So, so the generality of, of Alpha evolve is in its ability to search for these large algorithmic spaces, but also in different syntactic and semantic representations, right? It's not restricted to a particular language like Python, but it can sort of do that search across many different types of languages and many different types of tasks. The only expectation it has is, is that you have a function evaluator, that you can quickly evaluate whatever proposal there is and say how good it is.

• 9:15Speaker 03

  It seems like the rough cognitive architecture, so to speak, of generating a bunch of algorithm candidates, evaluating them, and then, I guess, evolutionarily deciding which ones to keep and then going forward from there. And it seems like it roughly mirrors the scientific method. Is that intentional?

• 9:33Speaker 00

  Yeah, so I think there are also, like, if you think about it, there is another sort of agent that we released earlier this year, which was called Co-Scientist. And in Co-Scientist, essentially, what you had was Gemini playing the role of the whole scientific academic process. So Gemini playing the role of a hypothesis generator, Gemini playing the role of the critique, Gemini playing the role of sort of ranking, different sort of reviewing those ideas and ranking those ideas and then editing those ideas. So it was Gemini playing all these roles in a multi-agent setup.

• 10:19Speaker 00

  And these were all sort of Gemini models prompted differently to play different roles. And very interestingly, this combined multi-agent system came up with behavior that went much beyond a single Gemini sort of model's answer. So it was able to give much, much, much better proposals and new ideas compared to a single sort of model.

• 10:49Pat Grady?

  What's the intuition behind why that works?

• 10:51Speaker 00

  Yeah, so I think it is something that is still being studied. But it is a fascinating sort of thing. The one thing that I actually sort of noticed is that, especially with regards to sort of Co-Scientist,

• 11:07Speaker 00

  You would run sort of co-scientist on a particular problem. And the very first answer that you you might get might not be very different from the baseline Gemini sort of model.

• 11:24Speaker 00

  And but what happens is even as you sort of increase the amount of computation over sort of

• 11:33Speaker 00

  this is you're not talking about just a few minutes or a few hours, but even days. As the whole multi-agent system sort of looks at the solutions and then refines them and sort of tries to sort of rank them, it just becomes much, much better. So why why that might be happening? It might be that the proposal, there is deep insights or there is some sort of intuitions that are buried in the tail of the distribution. Right? And then somehow Gemini's ability to evaluate which sort of proposal, which idea is better,

• 12:15Speaker 00

  is much better than its capability to come up with a sort of a new idea. It's the same sort of thing in computer science, right? Sometimes we are able to find if sometimes we know whether a particular solution is correct or not, but it's very difficult to come up with a solution. Right? So it's it's the same sort of thing appearing again in this multi-agent setup that somehow the agents working together are able to extract many more impactful results.

• 12:47Speaker 03

  It seems like the architecture of kind of, you know, generators and verifiers. It seems like that paradigm is is being echoed across the broad AI space, whether it's, you know, very general models or very specific, kind of like AI systems for very specific applications.

• 13:03Speaker 03

  Is that fair that that's sort of the consensus architecture right now? And do you think that'll be the thing that people continue to to push and scale?

• 13:10Speaker 00

  Yeah, so I I think there is going to be more work in agents, right? What we are seeing is basically the very start of research on agents. Yeah, whether in Alpha evolve, you you had a generator coupled with an evaluator. The generator was a neural network, a foundational model, an L L M, and the evaluator was even hand-coded, right? But together with an evolutionary sort of search scheme, you were able to sort of get these much more effective results. In Co-Scientist, you didn't have just one agent.

• 13:54Speaker 00

  You had multiple agents working in a shared memory. Now, like, what is the optimal agent configuration? Like, this is still an open research problem.

• 14:06Speaker 03

  Super interesting. Are the results that you're getting, are they different from the ways that humans would derive them? And I'm I'm kind of thinking of the the AlphaGo Move 37 stuff. Like, are the methods different? Are the results? How do they compare to the ways humans would think about them?

• 14:22Speaker 00

  So let's go back to the sort of original sort of motivation for why we started even working on the first iteration of using L L Ns for algorithmic discovery, which was FunSearch.

• 14:37Speaker 00

  So a few years back, like, as you know, DeepMind has spent a lot of has done a lot of work in using A I systems for searching over large spaces, right? We we have done a lot of work in building agents which have been trained using reinforcement learning, which can deal with many complex challenges, from the game of Go to playing StarCraft, which are quite complex challenges.

• 15:10Speaker 00

  Set ourselves a challenge that can we take the same kinds of models like the AlphaZero family of models which were extensions of or sort of extensions of what we had done in Go and the development of AlphaGo. Can we use the same types of models for discovering new algorithms? And we came up with a new sort of agent called AlphaTensor, which was particularly sort of focused on finding solutions for the matrix multiplication problem. And we found that this agent was able to improve

• 15:54Speaker 00

  over the past known results which had stood for fifty years. But the key question sort of remained, can you do something better? And secondly, can you sort of come up with a solution that is more interpretable?

• 16:12Speaker 00

  At the same time, like when we were looking at practical problems in Google, like how do you schedule jobs in a data center?

• 16:22Speaker 00

  Now, there are, there has been a lot of work on coming up with new algorithms and these heuristics have been designed by some of the best researchers and engineers at Google and because they have a huge amount of impact in terms of computer utilization. If you use a typical reinforcement learning agent on this kind of problem, you might get better results, but it might come at the cost of interpretability. Because now you have a neural network deciding which workloads go to which computers. And if something breaks, then you don't know, how do you debug this thing? So what engineers would really prefer is instead of giving them a neural network,

• 17:06Speaker 00

  you gave them a piece of code that they can interpret and they can run. And this was essentially the motivation, can we now use L L M's instead of searching in the space of of specific algorithms like we had done, matrix multiplication algorithms like we had done in Alpha Tensor, or coming up with a neural network policy to directly solve the problem, can we come up with an agent which can search in the space of programs and come up with a program that solves this hard problem. And the benefit, of course, will be interpretability, that you can see the sort of code, you can see what its properties are and so on. And that's

• 17:49Speaker 00

  what happened. We found sort of programs that not only were effective, but when the experts actually saw those programs, they could sort of recover insights. So for instance, we, one of the maths problems that we had looked at for Fun Search was called the Capset problem. This is a problem that Ted and Tao, one of the famous mathematicians, he is very interested in. And we collaborated with this mathematician, Jordan Ellenberg at NYU. And when we looked at the program that Fun Search had produced, he found that there were certain symmetries that were in the problem that had not been recognized before. And

• 18:34Speaker 00

  somehow, the program, like Fun Search, the agent had discovered those and was utilizing those to get a better solution.

• 18:44Speaker 03

  Can you say a word about, you mentioned working with Terence Tao and other famous mathematicians. Is math considered the gold standard for, you know, testing and benchmarking if these models are generating novel scientific results?

• 18:58Speaker 00

  Yeah, so math certainly sort of has some properties which are very interesting, right? The fact that it's very precise. Like you can, you know, whether the property that you are looking for, whether you have found it or not, right? You know, the matrix multiplication, for matrix multiplication, how many multiplications you require. Like for a four by four matrix, what was known was that you can do it with forty nine sort of multiplications, that's by Strasser, and we showed that you can do it by forty eight. So that's a very precise result. Right, there is no sort of arguing about that.

• 19:42Speaker 00

  Um, so it it gives you a very crisp way of evaluating how well you are, how well you have done. Um, and there is no sort of uh uh like RLHF needed in terms of human feedback, whether this was a nice result or whether this was a nice output or not. Uh, and you don't need to rely on a sort of an LM- LMSIS code. You just know that you're better.

• 20:08Speaker 03

  Yeah. Okay, so then when you go from the, you know, beautiful, pristine uh environment that is math to the real world, you know, it seems like you all have found a lot of real world applications and data centers uh in the Verilog world. Uh, could you say a little bit about, you know, where, which applications you expect Alpha evolve to be most impactful for?

• 20:29Speaker 00

  Yeah, so wherever uh sort of you can find a good function evaluator, wherever you can find an evaluator, where you can say, I really trust this evaluation scheme. If you give me a program, I can tell you very concretely how good it is. If your problem satisfies that that setup, then you can use Alpha evolve. Because Alpha, unlike a human sort of programmer who can try ten things or hundred things or a thousand things, Alpha evolve does not like it, it's, it does not, it does not sort of uh, it can go on and on and on and on. Right, it can come up with very counterintuitive sort of

• 21:12Speaker 00

  uh uh strategies to find, to solve that problem. Some things that you might not have ever imagined.

• 21:20Speaker 03

  Can you have humans be the function evaluators? Or does that not work?

• 21:24Speaker 00

  Humans, humans can be the function evaluator. It's, it's, it's a bad, uh, it's, it's a question of sort of scale. Right, like how much, uh, how many can you sort of evaluate? And whether you, whether you can. Evaluate the property of the program effectively. So at scale and with the right level of accuracy.

• 21:46Pat Grady?

  How do you do that? Do you build that into the application itself? So that there's a human in the loop evaluating as it goes? Do you do that offline separately before the application is produced? I guess, how do you do that? Or how do you imagine people doing that?

• 22:02Speaker 00

  Yeah, so I mean, like, so we haven't used a human in the loop for AlphaEvolve, right? Most of our evaluators were programmatic evaluators, right? But imagine a hypothetical scenario where AlphaEvolve was told that you have to solve this maths problem and come up with a new algorithm to solve this problem. And suppose it came up with many different types of problems, many different kinds of solutions, which are all equivalent in performance, okay? But then, which one is the best? It's the best is the one which is not only sort of very effective on the on the problem, but is the most elegant according to a mathematician or the most simple to understand, right?

• 22:47Speaker 00

  And that's a very subjective human thing. Like simplicity or interpretability, like we don't have a sort of crisp definition of it. It's it's it depends on it is grounded in the human observer.

• 23:03Speaker 03

  At what point do you need to pair kind of what's happening in the digital world to any kind of physical world stuff? I think in your blog post, you mentioned that you could see AlphaEvolve being useful for, for example, for material science. Do you need to be able to connect to a real world laboratory to kind of get any of that feedback? Or do you think all of this can kind of happen in the algorithmic domain?

• 23:23Speaker 00

  Yeah, that's a very good question. And I think this goes back to how much do you trust the evaluator? If you if sort of if your evaluation was based on a computational sort of method and the computational method was perfect and you completely trusted it, then you don't have to. Then you think, well, I believe the computational model. The computational model says that the solution that AlphaEvolve came up with is satisfies these properties. Job is done, right? But if you don't believe that the computational model is the perfect.

• 24:03Speaker 00

  Characterization of reality, then you want to make sure that you sort of validate that result in the real world, right? And you see whether that assessment of the evaluator was indeed correct.

• 24:21Speaker 03

  As Alpha evolve becomes more and more successful, as Gemini becomes more and more powerful, what do you think happens to these domains? And how will the, you know, the human scientists and engineers working in them adapt? So for example, if you take chip design as an example, you mentioned these models are getting very good at, you know, creating new, generating verilog, creating new chip designs. Does that mean the, the role of a chip designer goes away, changes? Like, how do you think that this changes the world?

• 24:48Speaker 00

  Yeah, so I think that's, that's again, sort of a very interesting question. I'll give you the example of what happened with Alpha Fold. So we started working on this problem of protein structure prediction. So for those of you who don't know, like proteins are the building blocks of life. They are the Lego blocks of, of life. And for many, many decades, scientists have been trying to figure out what is the shape of proteins? Because if we understand the shape of proteins, we understand how they function and we can use that to sort of develop new drugs to treating a sort of the most, a sort of challenging diseases on the planet.

• 25:33Speaker 00

  We can develop better sort of enzymes and so on. Now, in 2021, as I sort of mentioned, we released Alpha Fold 2. Before that, you used to take, or even a single protein, sometimes one to five years to find the structure of a single protein, and it might take a million dollars. And there were some proteins which are so notoriously hard that people had been trying to study them for almost one or two decades and had not found the solution. And which is why only thirty seven percent, roughly thirty seven percent of the human proteins, their structure was known. So after we sort of released Alpha Fold 2, I went to a biology conference.

• 26:18Speaker 00

  And because Alpha Fold 2, with Alpha Fold 2, we could find the structure of all protein, not just human proteins, all proteins on the planet. And we made the structures available to everyone on the planet. So I went to a biology conference and after I gave my talk, a biologist approached me and he said, Pushmeet, I have been working on this protein for the last ten years.

• 26:47Speaker 00

  And I had collected so much lab data to characterize this protein, to figure out its structure, but somehow this had, this has alluded, like this has evaded all kind of investigation and we still didn't know the structure.

• 27:03Speaker 00

  But we had all this data. If we knew the structure, we could sort of validate it very quickly.

• 27:09Speaker 00

  I ran Alpha Fold 2, it gave me the structure, it perfectly fit the answer. I've been working on this for ten years. What do I do next?

• 27:22Speaker 00

  So what, what has happened after Alpha Fold 2? What happened is basically suddenly it did three things.

• 27:32Speaker 00

  It first advanced structural biology. What was not possible earlier, it would take a synchrotron and six months and a million dollars is now done in a second. Right? So it really advanced what was possible. Secondly, it accelerated it. And thirdly, it democratized it. Like that particular scientist working in Latin America or South Asia or Africa on some neglected tropical disease had no chance to sort of figure out the structure of their protein. They did not have the funds or have access to instruments that could find them the structure. Now they have access to those things, to like the any sort of parasite

• 28:14Speaker 00

  that they're working on. So what do they do? They are now working in this new model where structures of proteins are not hard to get, they are everywhere. And so they are working on the next set of things, like how do you now use that knowledge to treat diseases and design better drugs? And I think the same thing will happen with the with Alpha Fold. When you have these agents which can go beyond human abilities in solving these problems, then the question becomes which problems do we solve?

• 28:49Speaker 00

  What are the important characteristics of a chip that we need to improve on? Right? Like we want to make it much more efficient, much more sort of so that it requires less cooling, it requires sort of less expensive construction mechanism, it's more fall tolerant, many other things. You can make the problem more and more sophisticated because now you have more sophisticated systems to sort of optimize them.

• 29:17Speaker 03

  Well, I have you. Something I've always wondered. The Alpha Fold results are phenomenal. And the story you shared with us is really impactful. Do you think that it's caused an inflection point in the kind of, you know, availability of new drugs? Or have there been other but- are there other bottlenecks now that are just, you know, we're faster at one part, but unfortunately, everything else is just hard. So we're still slow overall?

• 29:42Speaker 00

  No, it's it's it's so it has it has speeded things up, but I think there's a Want us to understand that drug discovery is a long process. Now, what are the, what are the roadblocks for drug discovery? First, you have to understand the target. You have to understand, here's a protein in the body that I need to bind. Because this protein is somehow involved in the disease. So if I can somehow bind something to this protein and change its function, it will have an effect that can sort of treat the disease. Like first, you have to come up with that conjecture. Then you have to say, okay, now I have a target protein. How do I develop a drug? How do I develop a small molecule?

• 30:27Speaker 00

  Or another protein then binds to it? So for that, you needed to understand the structure of the protein. Which other proteins that it interacted with? How did it interact with this molecule? This would take a significant amount of time, sometimes two years. Now that process is dramatically sort of accelerated. Now you can do it in sort of a few weeks or a month, a few months. That took you multiple years sometimes. But that's not the end of the story. After that, you need to now clinically validate it. So you have to go through phase one trials, phase two trials, phase three trials. You have to think about toxicity, all these other sort of things. So what Alpha Fold did was take one blocker away,

• 31:12Speaker 00

  made the overall timeline faster, but there are other sort of blockers, which our new generation of AI for biology models are hoping to accelerate and make much faster. So we have taken a big step, but we need to take a few more big steps.

• 31:29Speaker 03

  What domains do you think will be most lucrative for this, this family of, of models?

• 31:34Speaker 00

  I think the question is, what is, I mean, the answer to your question is basically what domains do you think are important for society? Because AI is going to accelerate everything. It's going to accelerate healthcare. It's going to accelerate the sort of the ability for us to develop more smart systems.

• 32:00Speaker 00

  From healthcare to sort of material science, like if you think about uh are the history of our civilization. We even describe our civilization in the sense of there was uh first we were sort of cave dwellers and then we went into the Stone Age and then we sort of uh uh went to the Iron Age and then the Bronze Age and now depending on who you talk to, you are either in the Silicon Age or in the Plastic Age. Whether you are optimistic or feeling a bit sort of sad. Um but if you take a step back and you think about what has humanity achieved? What we have achieved compared to any other species is the ability to transform energy, to leverage energy.

• 32:44Speaker 00

  Right? We have been able to leverage energy and do big things with that power. Now, if you can come up with, say, a new room temperature superconductor, that completely transforms your ability to handle energy.

• 33:00Speaker 03

  Hmm.

• 33:01Speaker 00

  Right? What changes will it bring about in society? They are hard to predict. If you can, if you can deal with energy in that way. Right? If we can unlock fusion.

• 33:14Speaker 00

  Right? And energy becomes so cheap every like like if you think about geopolitics, if you think about economy, a lot of it is about energy. Right? And suddenly if energy sort of uh goes down to zero, what will be the impact on on on the economics of the whole thing? Similarly, like if you think about like coding and if you have these agents which can code, what does that mean? If everyone can sort of code, like intelligence uh sort of is completely ubiquitous, everyone has access to all these different things.

• 33:52Speaker 00

  So there will be dramatic changes and everything will be impacted. So uh from from materials to energy to sort of coding to healthcare.

• 34:01Speaker 03

  Really cool. Do you think we're gonna have a fast takeoff moment moment for scientific discoveries? Do you think we're at the ramp of one? You think we're already there?

• 34:09Speaker 00

  I think we are in the, we are living through the middle of it. We are just, we don't have, like when you're in the middle, you don't really see it. But I think we are already in that era of AI accelerated scientific discovery.

• 34:22Speaker 03

  What do you see as the biggest bottlenecks going forward?

• 34:24Speaker 00

  I think sort of two elements.

• 34:29Speaker 00

  One is um validation, bringing bridging the gap between the digital and the real world. Right? How do you validate some of that? So that is one sort of key idea, right? And really sort of capturing what is important for the problem, right? And and the second is sort of the other bottleneck is uh how do you make this technology accessible?

• 35:00Speaker 00

  You can build the most sophisticated technology if people don't know how to use it. Then then you will not have the the impact that you want, right? Alpha Fold two, the it was not just impactful and transformative because it had very high accuracy.

• 35:22Speaker 00

  Because even if like it it had, it was quite accurate, but it was not perfect, right? But and suppose it was accurate on ninety nine percent of the of the things that it predicted. It it's definitely not at ninety nine percent, probably at the ninety or ninety five percent mark. So but suppose even it was accurate at ninety nine percent.

• 35:42Speaker 00

  The one person who got unlucky with their prediction and then spend the next sort of one or two years chasing a wrong prediction would then sort of say that I should not use it. I should not sort of use the prediction. So why is why is everyone using Alpha Fold? They're using Alpha Fold because not only Alpha Fold, not only is Alpha Fold good at making these predictions, which are accurate, but it's also very good in understanding the limits of its predictions.

• 36:15Speaker 03

  Hmm.

• 36:15Speaker 00

  When it makes mistakes, it basically holds its hand and says, I've made a mistake.

• 36:22Speaker 00

  So now, if it is making your prediction and saying I'm confident, like most of the time, it's correct.

• 36:27Pat Grady?

  Hmm.

• 36:28Speaker 00

  And that's great. This is something that the L L Ns of today don't have. They don't have calibrated uncertainty.

• 36:34Speaker 03

  Should we close out with some rapid fire questions?

• 36:37Speaker 00

  Yeah, sure.

• 36:38Speaker 03

  Uh, must read paper of the year.

• 36:41Speaker 00

  Uh, must read paper of the year. Oh, um, I would say Alpha evolve or coscientist. I like coscient, I like, like, uh, yeah.

• 36:51Pat Grady?

  Favorite algorithm nobody talks about.

• 36:53Speaker 00

  Oh, uh, the wake sleep algorithm. And very few people know about it, but it's essentially the idea. It's a, it's a paper from MIT, uh, from Kevin Ellis and Josh Tenenbaum, which sort of talks about, it's a way of sort of, uh,

• 37:14Speaker 00

  doing training where you find some exploration and then you somehow, uh, build the gist of it. Right? So think about, uh, sort of, uh, library construction. And analogy is library construction, right? You don't just want to write programs, but you want to also create the libraries that have common modules that will make all your future programs much easier to write.

• 37:37Pat Grady?

  Hmm. Very cool.

• 37:39Speaker 03

  Agree or disagree? Inference time compute will be the next major lag of compute scaling.

• 37:45Speaker 00

  Um, somewhat agree.

• 37:48Speaker 03

  Okay, say more.

• 37:50Speaker 00

  I, so I think inference time compute will be very, very important. I think also test time, uh, sort of training time compute will be equally important, right? We also, like, if you look at distillation, how powerful distillation sort of has been, right? So if these models have a, uh, have an ability to sort of understand and conceptualize what these models are able to do and come up with better inherent representations, then they just become much more effective in making predictions. Maybe their sort of uncertainty improves and so on. They become more efficient even.

• 38:33Pat Grady?

  Robotics, bullish or bearish?

• 38:36Speaker 00

  Um.

• 38:38Speaker 00

  I'm bullish about everything, so I have to say bullish. Like, I think everything will be sort of, will have an impact. So, but the question is basically near term or longer term, right? In the near term, it will take some, sort of getting robotics to work is challenging. But like, in the medium to long term, I think I'm bullish.

• 38:59Speaker 03

  Humanoid robots, bullish or bearish?

• 39:02Speaker 00

  We have constructed our world for humans, right? We, like, the human form. A lot of the, the, the non-natural world around us is made for humans, has been designed for humans, like, from an architecture perspective, right? Now, um, humanoids have the same form as humans. So they will fit in, in all these different architectures that we have built. Now, whether they are the most optimal thing, that is not clear.

• 39:39Speaker 00

  But they certainly sort of have an advantage that we designed everything for the human form. And now human, humanoids have the same form.

• 39:53Pat Grady?

  Future Nobel prizes in the sciences.

• 39:58Pat Grady?

  Will all of them be won by teams working with AI?

• 40:02Speaker 00

  No, I, I think we would be getting there, but I think like, humans are still winning Nobel prizes in the sciences. So, um, I think, um, I think there will come a point where AI will be indispensable. So it will be sort of humans and AI teams working together to achieving these amazing breakthroughs.

• 40:26Speaker 03

  Pushmeet, thank you so much for joining us today. These are really fundamental, really general results that you're pushing forward at DeepMind. And we appreciate you joining us to, to share more about how you, how you managed to do, do all this so far and what's ahead. Thank you.

• 40:41Speaker 00

  Thank you.