First Steps Toward Automated AI Research — Richard Socher, CEO Recursive AI

AI Engineer · 20 min · 169 sentences · from YouTube's caption track

Each timecode opens YouTube at the start of that sentence. Line anchors (#s42) are the cue ids in the WebVTT, and every line carries its start and end seconds. All transcripts has every talk, and the whole corpus as one file.

00:01[music]
00:12All right.
00:13Hello everyone.
00:14Really excited to be here.
00:15It's a big room.
00:17Very uh very cool conference so far.
00:20Uh I want to talk to you today about something that's been on my mind for many many years.
00:25This is actually the first time I I talk about it, sort of my version of going to Mars.
00:29Um and that is the Eureka machine.
00:31A machine that will eventually invent pretty much all future inventions for humanity.
00:38Uh and the way we're going to get there is uh by taking a step back and thinking about what else has given us a lot of really incredible inventions,
00:47uh namely evolution, and how that leads us to automating research uh and pushing the scientific frontier forward.
00:55And this is a joint work with a lot of uh amazing folks uh at Recursive.com uh and even some uh folks at AIX Ventures.
01:04And some of these slides are uh actually inspired by uh and taken uh partially from one of my co-founders at Recursive, Tim Rocktäschel.
01:12So, uh why do I talk about evolution and why is it so important?
01:17Uh I think basically evolution is this like open-ended process that has gotten us to a lot of different things that we really like.
01:25Uh it started in biology, it's moving to science, technology, and eventually AI.
01:30And I think it can inspire us in a lot of different ways to build better AI systems as well.
01:37In fact, uh whenever we take out and there's this famous saying, "Whenever I fire a linguist, my accuracy goes up."
01:44Uh I think that's true for machine translation back in the day.
01:48And it may be true that we should fire all the AI engineers uh and that that are here uh and have them mostly manage an actual AI engineer that is AI and works on AI.
02:02Uh and so that may be uh one of the conclusions of this talk.
02:06Uh and I think most of us are going to be excited about it cuz it means that we'll all become managers of such an AI rather than having to do the nitty-gritty ourselves.
02:16All right.
02:16So, let's start with evolution, right?
02:18The really, really big picture, 3 and 1/2 billion years or so.
02:22Uh this is kind of the incredible process uh that has led from, you know, simple bacteria and plants and fish and amphibians and so on to, after many billions of years,
02:35us.
02:36So, I That's That's a good starting point.
02:38That gives us some indication that evolutionary processes can do pretty amazing things, right?
02:43But now, let's zoom in and uh go maybe down to a few million years.
02:49There, we can also see how in a very first primitive ways, technological evolution uh has basically increased the world's uh sort of product uh in terms of monetary value.
03:02It's a little bit harder to estimate in the beginning, but we can see these sort of sequences of exponentials.
03:08And most exponentials eventually become S curves.
03:10They flatten out, but humanity has done pretty well by basically developing uh many of these very basic technologies, hunting, farming, but then also thinking about science, the scientific method
03:23um in the early days of the Enlightenment, and then of course the Industrial Revolution.
03:27So, now we can zoom even further, uh and no worries, we're eventually going to get to Nanochat and actual auto research and and what we're doing.
03:34Uh it's a very, very quick zoom.
03:37Um and now we can zoom down to last few thousands of years.
03:41And what we're seeing there is that with more technology, we were able to sustain more people, right?
03:47So, when we're working on pushing that frontier forward, uh we're very certain that that will lead to more human flourishing, right?
03:54And especially in the last few uh, hundred years, we're seeing this incredible explosion in the population of people because of technology.
04:04And the evolution uh, that it brings.
04:07And in many cases, that evolutionary process is run by us, so it's sort of conscious.
04:12Uh, but there are sort of interesting uh, inspirations that we can take from that as we're thinking about the evolution of AI in the next cycles.
04:20Uh, in fact, and I might not agree with everything with Marc Andreessen, but uh, he is very smart and we agree on a lot of things.
04:27Uh, and so I think he wrote this really great uh, techno-optimist manifesto in which he, I think, correctly points out that the only perpetual source of growth for the entire economy,
04:38a lot of people worry about AI taking jobs and things like that, but the truth is it will very, very likely increase uh, the economy massively and that will benefit
04:46uh, benefit a lot of us.
04:48And so the perpetual source of growth is technology.
04:51Uh, in fact, we can go even further and say that there's no material problem, and again, it's not sort of psychological problems and things like that, but no material problems
05:00uh, that cannot be solved with even more technology.
05:03Right?
05:04For problem of starvation, we invented the green revolution, darkness, light, uh, cold, indoor heating, heat, air conditioning, and the list goes on.
05:14So, I think we can kind of realize that this evolutionary process has been going on for a very long time and continues to make a huge amount of progress.
05:23In fact, the progress is so fast that there can within one lifetime be a major, major shift.
05:31Right?
05:32If you're born in 1900, uh, then 3 years, when you're 3 years old, the first human ever was able to, thanks to the Wright brothers, kind of have sustained
05:42motored flight.
05:43And then about 60-ish years later, in 1969, humans flew all the way to the moon, right?
05:51So, that within one lifetime, humanity went from like no one can fly for a very long time other than sort of gliding down a hill or something, no one can really fly to we all fly to the moon, right?
06:04And so, for us, I think, what that means is we're probably, and I sometimes say this, we're like too late to explore Earth, we're too early to explore the stars,
06:14but we're right on time to build an AI that could actually do what flying did for some in one lifetime due to intelligence.
06:24We can build and move from AI being worse at everything that we do to possibly being better at any specific task that we do.
06:32Right?
06:33And that that will probably be our our 60-year time frame, and because everything moves faster, it might only be 30 years or so.
06:40So, then, uh there's an interesting connection between technology and science and theory, right?
06:45Like sometimes the application comes first, and then we develop the theory later, and then improve at the technology.
06:51Sometimes the theory comes first, and from that we can build new kinds of technologies.
06:56And so, it's very helpful to think a little bit about the philosophy of science, and no better uh to be inspired there than Karl Popper, wrote that just like in other types of evolution,
07:08when we choose a theory, we also choose one that is best uh in competition with other theories.
07:13Of course, you need if you wanted LLMs to do that, they need to find them, you need web search, for instance.
07:19Um but, uh in the theory that best holds its own, uh it's one that, just like evolution, has a certain natural selection process, right?
07:28It proves itself, uh and there is also a sort of survival of the fittest uh going on in scientific theories.
07:36And uh in fact, uh a lot of science, according to Popper, is basically us proposing a new theory hypothesis or explanation or description and then subjecting it to rigorous empirical testing.
07:49That is the essentially evolution evolutionary pressure of scientific theories.
07:58And basically that was a very short run through so the history of open-ended evolution which hopefully makes us all realize that more science lead to more technology which will lead to more growth which lead to more human flourishing.
08:13And so that then begs the question does it make sense for us to try to just scale up and spend a lot of our resources as humanity to scale up scientific discovery
08:23in order to lead to this flourishing.
08:26When when you double click into that you kind of realize which Stanisław Lem already realized a long time ago that the exponential growth of science will actually be at some point halted by the lack of people working on it, right?
08:39There's so many niche subfields now in all the different areas of science that is very hard to get a million people to work on that particular thing.
08:48And so as a result of this incredible widening of the scope he says the number of people focusing on any single section of it has decreased.
08:58And that then leads us to really thinking about how could we automate this and automate scientific discovery and that then leads us to what I call the Eureka machine.
09:09This is basically our attempt at trying to build a machine that automates the process of scientific discoveries.
09:19And in fact I like in a couple months I'll have a book coming out on on this exact idea and so I'll just give you a super high-level highlight of how such a Eureka machine could be built for basically everything from physics, chemistry, biology, neuroscience,
09:33medicine, economics, astrophysics and so on.
09:36And there are essentially four pillars that are all extremely important to this machine.
09:41One is, of course, you have to understand what knowledge is already out there, what things humanity has already invented.
09:50You have to get all the scientific measurement data into, as in the second pillar, this machine.
09:57Then, for things that you cannot yet measure, we don't yet know, you should try to then build simulations.
10:05Anything you can simulate, you can verify, and you can then solve with AI.
10:10And if all else fails, or at the very end of these processes, you still need to have some kind of physical industrial like lab that actually can run real experiments in the real world.
10:22And on top of all of this, you'll have a basically an agent swarm that will deal with all of these different sources of knowledge and data and experimentations and and rewards.
10:35And in terms of, you know, the foundational model of knowledge, of course, we also, you know, basically is is a good example of how every single technology we've built so far, especially
10:48in AI, but also before that, the internet, browsers, GPUs, and so on, we can rethink, and there are a lot of startups possible in rethinking every single one of the layers of technology
11:00as infrastructure for superintelligence.
11:03And at you.com, for instance, we work on web search for LLMs, right, and agents and so on.
11:09And that actually is quite different, right?
11:12Agents can read thousands of very long snippets, rather than just 10 blue links with like a very short snippet.
11:18And so, you can rethink each of these different layers of technology that we've built for people, and rebuild them for AI in order to use them as tools to then build
11:31a superintelligence.
11:33Now, that is essentially uh the sort of why.
11:37Like like we want to build superintelligence in order to automate science.
11:43Uh and to me, that will be the next big step function change uh in in humanity uh and technology as we know it.
11:53Now, how do we actually build it?
11:56Uh I think the best way to build it is to have it built itself, right?
12:00We moved as a field and especially natural language processing for instance, which I've worked on for many years.
12:05We moved from not having linguists.
12:08This feels like ancient, you know, BC uh history, uh but before ChatGPT, um we we moved from having linguists tell us a bunch of things about language and then training statistical models on top of that.
12:20And when we allowed neural networks to actually automate learning those features with word vectors and uh other neural network architectures and back-to-back uh and end-to-end learning and backpropagation, we basically
12:32uh were able to get much bigger improvements.
12:35Uh then we did a bunch of architecture engineering.
12:38Now a bunch of people at least are working on a unified architecture, uh but even that unified architecture has a lot of manual processes.
12:45And so, it's clear over and over again in AI that when we take out a manual process and we replace it with a learned system, improvements will follow.
12:54Uh and so, that's why I think uh we should try to build a supermachine by having uh an RSI that builds itself.
13:02And the beauty is that only now um AI can actually do this because AI is code and AI can code now.
13:13This this ability to really code in longer and longer time horizons has really only happened in the last like 6 to 8 months.
13:21And that now enables such an RSI to work on itself, to develop almost a certain sense of self-awareness of its own shortcomings and then fix those shortcomings.
13:31Uh and then once we have that machine that has gotten really, really good at doing research in AI itself, we can then use it to do AI research for a lot of other things
13:41uh in in other scientific fields.
13:44And so at a high level it's quite easy, right?
13:46We have three steps: ideation, implementation, and validation of ideas.
13:52That's true for basically almost every scientific field.
13:56And so uh to end maybe on some very specific examples, uh we have built this first kind of version of such a Eureka machine uh and we wanted to just show that it works on some small
14:10uh samples that a lot of people know uh and are aware of.
14:13And so we basically started uh with three things that show you and give you a very first glimpse of an and sort of simple proof points uh of what such a machinery can do.
14:25And that was basically better training, faster training, and and better kernels uh for for Nvidia GPUs.
14:31Um the first one, NanoChat, um I'm sure many of you have heard of it.
14:35A lot of people think that's already recursive self-improvement and it's kind of a weak form in the sense that usually when you do auto research, it's it's not recursive self-improvement,
14:45right?
14:45True recursive self-improvement is when you have an AI that has a sense of self-awareness of its own shortcomings, full access over everything uh in its arsenal from pre-training to RL training
14:58and harnesses and everything, and then actually updates that entire system in the next version of itself.
15:04Now you can also take such a system and just ask it to improve some other process, some other AI, like a small NanoChat run where you can train something in 5 minutes.
15:14And that is really exciting and it's an important milestone, but it's not actual RSI.
15:18So here we basically showed three examples of such an auto research system and what it can do and after a very very short time it essentially was able to outperform
15:31many different teams and teams that also use other AI research.
15:36So, let's double click into some of these.
15:39Nano chat is really exciting example.
15:42Basically, you train a very small chat model in less than 5 minutes and you basically want to have it get to the best possible bits per byte number.
15:57And so, the whole community had worked on this for quite some time and got to 0.93 and after training this for a little more than a day or two,
16:10we basically got it down to 0.91.
16:13Which is pretty exciting.
16:14Now, it wouldn't be that exciting if all it did was just find a couple of hyper parameters and tune them carefully, but it actually did find truly interesting novel
16:25ideas like hash bi-grams and tri-gram embeddings and tables for those and mixing that into various value paths of the intention through variety of learned gates.
16:37So, it actually started to doing more and more interesting things rather than just kind of tuning hyper parameters.
16:44Another one nano GPT speed run.
16:47Obviously, speed is very important.
16:49So, here we're able to work on this again, apply the system and after very short amount of time it got better than people working often together with the AI
17:00for over a year on on this very on this benchmark and made the whole thing another two seconds over two seconds faster at 70 seconds.
17:10And again, discovering very interesting ideas in the process.
17:16And then the third one is CUDA kernels.
17:18Of course, we all care about not burning through our GPU budgets too quickly and trying to be very efficient.
17:25I think in general it's actually kind of shocking how inefficient a lot of mixture of expert models still are running very large clusters that cost billions of dollars and then only have like 30% or so
17:35utilization.
17:36So a lot of work that's ongoing in the world to improve that and different fields or different groups of people are various different stages of that.
17:47But long story short, lots of different CUDA kernels are used during training and testing and here we basically again took that system and after a couple days it discovered
17:59better kernels than the leader boards best on the Nvidia benchmark website by again quite quite a sizable margin across all the different categories of those kernels.
18:13And while we are pretty good at AI and we actually in the team didn't have any particular CUDA kernel experts who just spent their entire careers writing good kernels.
18:25But still, you know, we do just enough to make sure and work together with Nvidia to make sure that there no reward hacks here and and other issues.
18:33But actually found that eventually these all checked out and were indeed pretty much all the different kernels found the best solutions there.
18:44And so with that I hope I could convince you that indeed RSI could be that next big S-curve an exponential that gets layered on top of previous exponentials and
18:57that should help us with not just AI but eventually science and then all of technology and then allowing many more people to flourish on our planet.
19:09And so maybe I'll end on this note here, which is a lot of people wonder how much longer I can go, right?
19:16Every exponential eventually flattens out and it's actually quite hard to know like when we even talk about exponential growth in the eye, what does that even mean?
19:24There are many different, I call them spaces of intelligence and we won't have time to go into all of all of these, but as soon as you actually try to define
19:32multiple different dimensions of each of these 10 spaces that make up this complex sort of volumetric uh thing that is intelligence, you'll realize that there's still so much more to go.
19:46Like on the upper bounds of intelligence, we're still astronomically far away from reaching those and across pretty much every single one of these dimensions and the spaces that they make up.
19:59So, if any of that is interesting and you want to help us build that, um we'd love to hear from you.
20:05Thank you.