AI for Atoms: Liam Fedus on Periodic Labs and Revolutionizing Materials Engineering

Source material

Transcript

The Dandepires were talking with Liam Fettis.

Liam is one of the co-creators of chatGPT, which I think almost everybody uses at this point.

He was the VP of post-training at OpenAI and before that was at Google Brain, where he worked on a variety of really early AI innovations.

Liam will be telling us a bit about periodic labs, his company which is focused on building an AI foundation lab for atoms.

In other words, how do we impact the physical world, material sciences, chemistry, etc.

Using AI.

Very exciting topic and excited to be talking with them today.

Liam, thank you so much for joining us today on Apprires.

Yeah, thank you so much for having me this great to see you.

Yeah.

So maybe what we can do, I think you're doing incredibly interesting things in terms of alternative types of models, specifically for material sciences, for the physical world, effectively what you're building is an AI foundation lab for atoms, which I think is fascinating.

That's right.

But maybe we can start with this a little bit more of your background, you know, I think you were VP at OpenAI.

You worked on one of the first trillion parameter models, ever, etc.

Could you tell us a little bit more about just like what got you here?

Yeah.

So even further back, I was a physics major in undergrad, spent some time doing dark matter research.

We had a apparatus that was directly sensitive to dark matters direction, so it was very interesting.

Sorry, but I'd love to come back to this, but why are there so many physicists in area right now?

So you look at Dario, Marty, who runs anthropic, of course, yeah, you look at Adam Brown, and Google, you look at a variety of people, and they all kind of have these physics backgrounds.

Yeah, my old manager, Joshua, who's a physics and an anthropic, yeah, why do you think that is?

I think it's a great way to think about the world.

It's like very principled, very like hard-nosed, scientists, very careful, and I don't know, I think it's such an incredible field.

You have such high leverage in computer science in AI, and so I think a lot of physicists we're seeing that, particularly in like high energy physics, after the discovery of the Higgs, I think a lot of high energy physicists were sort of looking for what's next.

Ultimately, it becomes bottlenecked on the new apparatus for pushing the energy frontier.

And I think a lot of physicists were looking at their skillset and looking at the progress, skillset, and saying like, yeah, I think I could be a huge contributor elsewhere.

This has been fascinating to see like string theorists and people working on the goals and all sorts of effects, like kind of moving into AI, absolutely.

It almost feels like we're recreating them in hand project or something, except now what we're seeking is, you know, different forms of intelligence, so yeah, that's very exciting.

I'm just going to have perspective.

Sorry to interrupt, say, you know, you studied physics, you worked on duck matter, that's right.

And then I was basically, and then in grad school in physics, I was always gravitating towards the machine learning problems.

I was looking at particle reconstruction, and it's seeking an effectively machine learning problems, but it felt, if I really wanted to push frontier of machine learning, I should be in, you know, computer science, so ended up at Google Brain was overlapping with the first year residents there, absolutely remarkable group of people, remarkable period for Google Brain.

I mean, as a era of when there's the creation of like, just should be the training strategies, texture of experts, the transformer, it was a really rich period in that history, and it was a fun kind of like Cambrian era where people were really pushing the frontier, which is like a handful of GPUs, really small collaborations.

The field was a much, much earlier, and I think there's a lot of diversity and entropy in the research, and it was very fun.

So it's kind of late, 20, 10s or so, something like that.

This was 2016, 2017, so Google Brain at that point was so small, and eventually was sub-tuned by DeepMine, or combined with DeepMine.

So that Google for many years mostly used to doing architecture work, so it was really pushing sparsity that allows for more efficient serving of models of scale, and just really pushing the scale of what we could do towards late 2022, really became excited about the creation of products, the technology was getting very compelling, and so I ended up that opening eye with some other Googlers as well, but I do work on specific way to opening eye.

Well, so the goal was, we need to come up with some productionization of GPT4.

So we, opening eye had GPT4, it was pre-trained, and there were some like, um, Leverof post-trains on it, and there's questions about like, how do we turn this incredibly powerful model into products?

And we're all spit-balling ideas, like writing-bought, coding-bought, you know, very natural at the time.

Some of our least interesting ideas were a meeting-bought, so just in a Google meet, take notes, and then send out like to do it after.

But John Shulman was very opinionated, he's like, we think we should keep it very general, let's do a chat-bought.

And that became a large part of the effort for those few months.

That's what I call, yeah.

I say it worked on Chatchee PT, that's right.

And obviously, I felt like that was kind of the starting gun of this whole AI revolution, or at least in terms of people's awareness, like I'd started investing in the area beforehand.

Right.

But it seemed like almost as a secret up until Chatchee PT came out of, and suddenly everybody realized that there was this powerful technology available.

Yes.

How did that lead you to materials and atoms, and, you know, the physical world again?

I know that was sort of your starting point in terms of economics, but what brought you back, given how much is being transformed right now through language?

I think just the inevitability of connecting these systems to the physical world.

The opinion that I and others held as part of the periodic was you are not going to see this kind of acceleration in science and technology, unless you start connecting these things to the physical world.

Science ultimately isn't sitting at a room thinking really hard.

You have to conduct experiments, you have to learn from them, you have to interface with reality.

And the creation of Chatchee PT in late 2022 was a, you know, important technology, but it's still far too weak.

We couldn't have done periodic on technology of that era.

I think over the next few years, past that, we saw ever improving models, we saw reasoning, I think like test time inference became really important.

That led to more reliable, error correction, more reliable, tool use, and we see like the rise of coding agents and other agents.

And I think those were foundational technologies necessary to then connect these systems to the physical world.

Like it was just not impossible with like the AI technology of 2022.

I guess the other thing is missing from the physical world is data, or at least data of the season.

So you look at something like the big foundation models on the language side and they're basically trained on the internet as a major corpus, it's augmented and all sorts of ways with other data sources.

How do you think about that for what you're doing where you're trying to model atoms in the physical world and how all that stuff kind of works?

Yeah, so experiment.

I mean, so we have simulation, physics simulations, and we have experiment.

And I think exactly as you're pointing out, ML systems are good on the data you've trained among, on the task you've trained them to do.

I think sometimes there's like this mythology of AGI, SI, RSI.

And I think we see increasingly powerful systems, but they do become limited if they don't have access to the raw data to actually make informed systems.

How much do you need?

And so I know that there's some data skill related research and other things in terms of how you kind of hill climb towards like a really good model, how many experiments do you need to run, or how many data points do you need, or how do you think about the diversity of data points you need to generate?

I'm a little bit curious, like what does that actually look like tangibly?

So there is some generalization from the existing models, so we don't need to reproduce a system that can understand and write English or write code.

So we're kind of like leveraging, and I'm using it up in search for that, or close first models, or some of these are combination, for example, like periodic spend zero effort on improving coding models, we're incredibly impressed by Codex, Cloud Code, and so that's been a huge accelerator for the company.

But focus on machine learning efforts where the existing frontier is not sufficiently good for us.

I think going back to the data question, we're leveraging, call it, order tens of trillions of tokens that went into open source models, and that's given us like very like foundational understanding, but once we start moving into specific discovery areas, chemical spaces, we can see a very high level of sample efficiency.

So the system isn't starting as like a randomly initialized neural net.

It has a strong prior on the world.

And so where does that prior come from?

Oh, just like in terms of that, just general, just like you know, papers, the internet as you're pointing out.

Yeah.

However, that's insufficient.

One of the engineers on our team was looking at a reported material property, and it was just sort of extractive values from literature, and it was really interesting to see the reported value-spanned many orders of magnitude.

And so you train a ML system on that, and it's like, well, the best you can do is all of the distribution, but you're no closer to like a ground truth.

And that's where experimental data comes in, where you don't have a grounding in that.

So, but really important, it's not just like a pool of data, it's this interactive, closed-loop system that is so powerful.

Once you have the experimental data, you can look through it, you can look for aberrations, you can look for patterns, you can look for consistency with simulation data, with literature, and then that helps drive the next set of experiments.

So it's not just a pool of data, it's just very active-loop.

I see.

And then, how do you think about it versus the data?

So I look at something like alpha-fold, or some of the protein-folding related models which are amazing, right?

If you think about it, I used to work as a biologist, and we would, you know, a crystal structure would take years if it happened at all, because you wouldn't necessarily, certainly, if you could crystallize the specific protein and of certain reagent conditions in a way that would be performant for actual, you know, crystalloxylography and everything, or NMR, whatever a pretty took for structure.

And then, sort of alpha-fold comes out, and you can just arbitrarily model anything.

Right.

I'm a protein-world, which was, you know, amazing, as a breakthrough.

But it was a very specific data set that already existed that had lost in lots of structures over-downed over decades of work.

How hard do you have to bootstrap that for every single material's domain, or do you choose specific ones that you think can then generalize?

We have seen internally the greatest advances where we have an abundance of data in some space, and that is led to the highest rate of acceleration internally.

But I think you can think of different levels of generalization, and for systems that are strongly governed by quantum mechanical effects.

There's some generalization there, I say.

But like, if you produce a system that has modeled quantum mechanical objects really accurately, it's not really helping much on, like, fluid dynamics or, you know, another kind of like level of abstraction.

And so the generalization we're seeing is quite good, but there's almost like the first principles you can.

How does that sound interesting?

So you could do, like, here the basic steps of chemical synthesis.

Here's quantum mechanics, here's different aspects of how atoms interact in general, or vendor wallforces, or things like that.

Absolutely.

That's interesting.

Yeah, that's cool.

And then from architecture perspective, is there anything unique that you're doing or interesting or can you talk a little bit about how you're actually constructing some of these models on top?

Yeah, so language models are incredibly powerful, it's a very natural interface, and to we continue to use these.

But we think about them almost as like an orchestration layer, so that's sort of a co-pilot assistant, but also like a system that can direct experiments.

And it's almost, it's orchestrating other specialized models as well.

So we do construct neural nets that are specially designed for atomic systems, where there's like some symmetry awareness, and those have much lower latency, and they've been like fine-tuned for that.

And so basically, you kind of think of this like orchestrating layer that can, in just literature, can go through our experimental data, can go through different modalities, but they can also use specialized neural nets as tools, as reward functions, so it's like an overall system.

Okay.

Yeah, that makes a lot of sense.

Yeah, I've seen a lot of people architect those sorts of approaches, even for things like customer support, or other areas, it seems like it's the common architecture that's emerging as you're doing these different use cases of animals.

That's right.

Yeah.

But transformers have been very powerful.

Yeah.

Yeah, that's really cool.

The language world, one of the things that was pretty unique about it, and it's the reason that I think these companies like OpenAI and Thropic and others are growing so fast, is it just plugged into a very big domain of human existence, which is all language.

And all language means enterprise software and enterprise interactions, and it means consumer behavior is basically how we interact with the world.

Yes.

It seems like there's a little bit more of a leap for other areas.

So for example, in robotics, there's really interesting things, different types of robots that exist in the world, but the footprint of that is quite limited relative to language.

And the same seems to be true for material sciences.

So how do you think about where you're going to commercialize this first or if you're going to work with or with a specific domains of products that you're working on first?

So we've begun working very closely with scientists.

We've treated periodic as our customer zero, and seeing how can we transform how this field of science is done.

But there's huge opportunities across all of these industries, all these enterprises that are interfacing with the physical world.

People who are bottlenecked by materials engineering, process engineering.

And again, those are kind of the same natural interfaces where engineers are asking questions about their data, they're trying to find aberrations, they're trying to debug machinery, they're trying to get to a better formulation.

It's actually a quite universal thing as well.

And so we've kind of created our little testing ground internally.

And now we're sufficiently excited about the tech we've been building and to see this acceleration for advanced manufacturing more broadly.

And as your model going to be developing materials for other third parties is a developing or a materials that you then sell in the market, because it almost reminds me a little bit of a biotech model.

Yeah, we're in biotech, you can either partner with a big pharma, and then effectively help them create a drug and take a royalty on it, or you can build your own drugs.

How do you think about that in the context of what you're doing?

Thinking about us ourselves as an intelligence layer for these companies.

So you can think about system or record control plan for different experiments and getting to solutions, but like you're saying, there is a very interesting aspect of some breakthroughs here could have really high value, and it might be more akin to a discovery model like we've seen in biotech and elsewhere.

But starting, thinking about it just as a software business, have you ever had the diamond age?

Very fast.

Yeah.

Have you ever had diamond age?

No, that's the Neil Stevens book.

It's basically this book about it was written in the 90s, and there's two key concepts in it.

One key concept is there's effectively an AI tutor that's unleashed on the world, and it kind of teaches huge numbers of young girls, all sorts of skills, and it's a very interesting thing about AI education.

And then in parallel, basically this AI research scientist creates a primer for his daughter, and the Chinese steal it and clone it and distribute it across the country.

And because he built it for young girls, it's a very young girl in China has it.

Right.

So that's their reason.

China theft of IP kind of thing.

Right.

And then the other part of the book is about matter pipes into everybody's homes, and they all have 3D printers, and you download blue prints, and it just creates whatever you need.

Yes.

It's a physical world.

And some people start evolving different nanoboss city differently.

This is very advanced kind of AI plus materials, kind of future world.

Yes.

What is your vision or conception of what our world looks like in 10 years assuming periodic is successful?

Well, I mean, I think as you're pointing out, you're going from systems that aren't just writing essays, not just writing software, but to literally generating matter, and I think it has pretty profound implications to semiconductors, airspace, energy, and I think it's incredibly important for can we increase the pace of just like the physical development of the world?

I mean, we see how quickly the digital realm is changing.

Software engineering now looks wildly different than even six months ago.

But I think we see similar opportunities in the physical world.

Of course, like atoms are hard, and see we'll have some limits of physics.

But just because atoms are hard doesn't mean there's not an order of magnitude, or two to speed up, just making sense of huge amounts of data and getting to solutions more quickly.

Yeah.

So I think what we're trying to do is give humanity this agency for atomic rearrangement, synthesis, and we think it's going to just be a huge accelerator.

So I mean, if our physical world can keep up at some fraction to our digital world, I think life will just feel dramatically different.

Yeah, it's kind of the revolution that we're going to come.

I cut our minds and we have almost the materials equivalent of the agricultural revolution.

Yeah.

We suddenly had a massive spike in productivity of our body, and it seems like there's been all sorts of bottlenecks that have constrained us until now that you fix or trying to address.

That's right.

What aspect of the work that you're doing, you're most excited about?

The iteration with our between these groups of people, I mean, it's like, this is just irreduciably a multidisciplinary problem.

We have physicists and chemists working really closely with some of the top AI researchers in the world, working closely with some of the best engineers in the world.

And this multidisciplinary, like, close collaboration is just absolutely incredible because seeing firsthand how a field can fundamentally change.

People who have been doing research for, in some cases, decades, in a field.

And now seeing, like, oh, under these systems, under intelligent systems, it could look this very different way.

And I mean, I use, like, an analog to machine learning a lot, going back to the early Google brand days where the frontier is pushed forward by a few GPUs, and a few people.

Now you look at this era where it's really like industrialized, and there's dozens, hundreds of researchers working together with hundreds of thousands, millions of GPUs dictated and driven by scaling laws.

Everything is about scaling.

Even that predictability.

It's allowed us to put huge amount of capital into this field.

And I think the physical science is, physical engineering, will have a very similar property where we establish these scaling properties and bring that mindset.

And so periodic in this field is really thinking about how do we bring much larger scale sets of experiments to bear on this?

And intelligent systems have enabled us, automation has enabled us, and you really need both an improvement to automation where you can soon become create bottlenecks in intelligence.

And I mean, the scientists very much feel this where they're not used to working at that level of throughput, and they just can't simply make sense of so much data.

It's interesting.

So I guess in terms of scale here, one of the real benefits, one of the things that's really benefited the frontier labs on the LLM side is just scale of capital and therefore scale of GPU and scale of data.

Of course.

Is this similarly a capital intensive area on your mind?

Yeah, we will require more capital.

GPUs are so extraordinarily expensive.

It was interesting as the compute cost relative to physical infrastructure is actually surprising where you know, so much money is spent on the compute that the physical infrastructure, sometimes is actually lower, but you know, has very large lead times, and there's intrinsic difficulty of having these well calibrated, well functioning physical systems, but from a capital perspective, it's primarily a compute cost.

Yeah, it's really interesting if you look up the cost of a Stanford post-doc for example, relative to a machine learning engineer, it's like such a big difference.

Absolutely.

You know, my takeaway is that many people working in science, particularly in academic center setting are very under compensated, relative to sort of their societal value.

Absolutely.

And so I always like it when companies kind of help bring people into the end of the fold in terms of both human impact, but also that ability to do things that real scale and really do things a different way.

So it must be very exciting for the people on your team.

Yeah, I mean, some of the scientists who join us are among the best in the world, and it's been absolutely incredible working with them.

Yeah, I mean, it sounds like you've built such an amazing interdisciplinary team.

Are there specific roles you're actively looking for right now or key things that you really want to hire up?

Absolutely.

So on our site, we have decomposed the world into bits and atoms.

It's at least tax on me, but on bitside, we're really thinking about mid-training pre-training roles from the AI side, always more infrastructure roles and on atom-side, like control engineering, system engineering, but also now thinking too about being a spanning out with like product engineering.

So yeah, across the market, yeah, it's really cool.

So I think one of the things that everybody is really thinking deeply about or is excited about right now is AI, AI, AI, sort of these advanced systems that are as good as humans or better than humans, at different things, where are very generalizable in terms of their abilities to do abroad, swath of things.

How do you think about that?

But then the context of what's happening over the overall foundation model curve.

Because obviously we're very integral in terms of the development of some of these systems.

And then how do you think about that applied specifically to some of the areas you're working in?

I think one fallacy is thinking about intelligence as a scalar.

We've consistently seen these systems have a very odd spikiness.

And it's actually possible to architect a system that is world-class on some math domain, but then you could do some perturbations to the questions and actually degrade it substantially.

So it's like a bad high school student.

And so there's this like odd spikiness to these systems.

So basically it can make a system that's like a genius at one thing and not very good at a bunch of other stuff.

And I guess the point I was making is those fields can actually be quite adjacent.

So like sometimes a generalization can be non-intuitive.

But one way I think about recursive self-improvement is really kind of akin to neural architecture search from roughly 10 years ago.

And I think there's a very clear path for software engineering.

So these systems have become so incredibly impressive on this on this domain.

As a result of huge amounts of data, really cheap, verifiable environments.

Like you can check unitesical from feeling the passing with just a few CPUs, basically instantaneous.

There's no domain expertise gap between an AI researcher or software engineer.

And obviously this will become and is becoming a larger contributor to the next generation of the system.

When do you think it just flips into, we just, everything is machine self-improvement versus human director or needs a lot of human intervention.

So do you think that's two years away, do you think it's five years away, do you think it's ten years away?

Well I guess like building on what I was saying is I think there's a domain caveat to that.

So rolling forward that software engineering self-improvement, I think you're going to have a system that can write complete repositories, identify bugs, refactor code, but it doesn't suddenly understand biology.

Sure, right.

It's just like there's a domain gap there and knowledge.

But even beyond that, there's sets of strategies done in software engineering that differ from scientific or engineering strategies.

So you're not operating under, it's not like decision making under uncertainty to the same degree.

It's like very, very valuable and that's driven so much of our work.

So in that domain, I think it's happening nowish and I think we'll see the same thing too for AI research.

That's a slower outer loop because now the experiment isn't just checking some unit test passing, but it's checking what was the scaling property did this model converge, what's the generalization of the system?

That requires GPUs, that requires many hours of experiments, but I think that will also will, and those are all e-values that people use today as they're looking at existing models.

And so they do have that utility function, that feedback loop that can be driven by self-learning.

That's right, that's right.

But again, the connection to these things to the physical world, it's going to be so critical because both of the systems are being trained in a closed loop against that domain.

So it's a closed loop for doing software engineering.

A closed loop for doing AI research, and that's the premise of periodic.

Like we need to have these closed loops of actually doing science of actually doing engineering.

And these two, I mean, these two domains are how I think the rest of the world will go with some delay.

And this is again like the foundational technology that we're interested in.

Do you think you need sufficiently good robotic systems in order to have that closed loop for what you're doing?

In other words, do you need something like pie or skills or something else to work in order for periodic to hit that escape velocity in terms of a closed loop system?

No, but it's a huge accelerator.

The goal for periodic is to generate high-quality data, diverse data.

And automation is assistance to that.

So right now, we employ people as well, and we have autonomous parts that are just very reliable.

If you had a dexterous humanoid who could wander into an unstructured lab and make sense and follow instructions reliably, that would be a huge accelerator right now.

The automation of physical systems is requires a very careful design and it's slow, but I think with improvements in robotics, it's just going to accelerate this.

But already the reliability of the hybrid systems is sufficient to produce a huge amounts of reliable data, but it's just going to accelerate us pretty quickly.

Yeah, one of the reasons I ask is I'm going to start in this company color, and we built our own liquid handling remmoning systems, right?

We'd buy liquid handling robots that then we'd have to adjust them dramatically.

We had like cameras that would use ML the monitor, the rest of them, and sort of make adjustments.

We had a 3D print parts to decrease vibrations on the platform because we were dealing with such small volumes of liquid, right?

And so there was enormous amounts of customization, versus just having, and the firmware for it was awful and ready against that was painful.

Yeah, versus just having like robotics a similar work like a modern system, in all the ways that you'd conceive that.

Right.

As that's the reason that I was asking is if you really want to do hot-u-put experiments, you need these underlying systems to be able to do all the liquid handling and to do all the titration of stuff and all the rest of it.

Yeah, that's right.

I mean, I think it's right now we're using almost like more like off the shelf for robotics.

It's like very simple, very commoditized, not doing like a huge amount of innovation on that front.

But again, like as these more general robotic systems come to be like hit this reliability threshold, it's going to be a massive accelerator for spinning up new labs as well.

Yeah, you've seen such a wide range of different things happen in the AI world since indeed you're working with.

I guess at this point about a decade ago, and so you were there during the birth of the transformer model, you were there for the birth of Chachi PT.

What are you most excited about outside of periodic over the next few years in terms of what's happening with AI?

I mean, of course, robotics.

Again, I'm just so excited about the interface of AI systems with the physical world.

And we're approaching one angle of that, which is science engineering.

And we need that data in order to make those advances.

But simply just agency and control of the physical world via robotics is going to be transformative.

So I'm very excited about these interface layers.

I think that's going to be such a massive opportunity.

Because I mean, you know, how many software engineers are there in the world versus people who are like the physical world?

And there's just labor shortages everywhere.

So yeah, I think it's me a very interesting decade.

Oh, amazing.

Well, thank you so much for joining us today.

Yeah, thank you so much.

This is really really good chatting today.

Yeah.

Find us on Twitter at no priors pod.

Subscribe to our YouTube channel if you want to see our faces.

Follow the show on Apple Podcasts, Spotify, or wherever you listen.

That way you get a new episode every week.

And sign up for emails or find transcripts for every episode at no-fire.com.