Hello, I'm Andrew Mayne and this is the OpenAI Podcast. Today my guests are Kevin Weil, head of OpenAI for Science, and Alex Lupsasca, who is an OpenAI research scientist and professor of physics at Vanderbilt University. We're going to be discussing how AI is impacting science, an upcoming research paper, and where science might be headed in the next five years.
Maybe the most profound way that people are going to feel AGI in their lives is through science.
With ChatGPT I can just launch it in that direction, in that direction, that direction.
The acceleration that is going to come from these tools is going to change science.
So you're running the OpenAI for Science Initiative. Could you explain what that's about?
Yeah, the mission of OpenAI for Science is to accelerate science. So the question is, can we help scientists do the next, say, 25 years of scientific research and scientific discovery in five years instead. Science underpins so much of what we do and how we live. And if we can make progress go faster by putting our most advanced models into the hands of the best scientists in the world, we should do that. And that's what we're trying to do. You could ask like why now? Why didn't we do this a year ago? Why aren't we doing this a year from now? One of the big reasons is we're just starting to see our frontier AI models being able to do novel science. So we're starting to see examples where GPT-5 can actually prove new things. Maybe not yet things that humans could not do, but things that humans have not done. So these are like these little existence proofs of GPT-5 being able to break out past the frontier of human knowledge and into the unknown. And if there's one thing that I've learned from now, you know, a year and a half or so at OpenAI, it's that you go very quickly from the model can't do something, to the model can just barely do something, and it's not great at it yet, but you see these early examples. And then, you know, six months later, twelve months later, all of a sudden you couldn't imagine doing this thing without AI. And I think science is in that initial phase where we're seeing real acceleration for scientists that are using AI. Sometimes novel, you know, not yet maybe large breakthroughs, call them small breakthroughs. And that just says that there's so much potential in this space.
We've seen examples of let's say AI helping with mathematical proofs. Could you give me an example of how it might do things in some other areas like physics or whatever kind of things we might see in the short term?
Yeah, I mean we're seeing examples every day and they're across the range of sort of the scientific frontier. You see examples in mathematics, in physics, astronomy, life sciences, like biology. Alex, I mean, you've worked on some of these. Maybe it's a good time to talk about some of the physics stuff that you've seen.
Yeah, I think coming back to Kevin's point about how this is a special time, that's very much how I feel as well, because I started the year 2025 thinking, yeah, ChatGPT is cool. Like everybody, I used it when it came out and I thought it's a great chat bot, but I was sure it would take a very long time before it would become really relevant for my own work. So I started the year I would say as an AI skeptic because I like to see evidence before I'm convinced of something. And I saw people using it to help in their writing. And I started to use it for that as well. It's very useful for proofreading, but I thought, oh, it's gonna be a while before it gets to do the special stuff that I'm really a specialist at. You're like black holes, right? Like black hole physics, exactly. And I had this experience early this year where I was trying to find this magnetic field solution that describes what happens around a pulsar, which is a rotating star with very powerful magnetic fields. And I was going for this very particular solution. I had to solve a partial differential equation. I was able to identify that solution as an infinite sum over products of special functions called Legendre polynomials. And if you won't go to physics grad school, this is the kind of thing that you spend a lot of time getting familiar with. And I also like these puzzles. And I was playing around with the sum and I felt like there should be a simple formula that it evaluates to. And I thought, okay, I have this friend who has ChatGPT o3 Pro, which I didn't have access to at the time. And I thought, okay, I'm just gonna send it to him and see what comes out of it. And he sends me back this output. It thought for 11 minutes, which at the time I had never seen it do because I was using the free version, which doesn't think for as long. And it gave this beautiful answer where it was able to understand what the sum was and break it down into pieces that it could tackle. And then it had to go and find the special identity that was published in one paper from the 1950s in a Norwegian Journal of Mathematics. And so it understood what the problem was and it knew about this random identity that was just the thing for the job and it used them. And it gave this beautiful output. And at the end, the answer was wrong because it made a silly typo. It added an extra factor in front, it's almost kind of like a human making a silly typo at the end. But it was very easy to check the derivation and I went through it and I realized, okay, there's this extra factor, but aside from that, it did the work. And that really sent me reeling because I thought, okay. I would say that's a uniquely human ability. I thought that's something that makes theoretical physicists special. You know, now in 2025, clearly they're capable of doing things that I would consider amazing.
Yeah. I think one of the cool things, so you've got examples like Alex's where it was probably not something that he like he could have done it himself over you know eventually, but GPT was able to do it faster. That's acceleration on its own. And there's something qualitative about that even as well, because if you can explore, instead of exploring two paths over the course of a week, if you can explore 10 paths in parallel in, you know, an hour, all of a sudden there's a lot more ideas that you can try. And that's also acceleration. We also see examples in like literature search, which you don't think of as maybe like deep scientific innovation, but it's really important to be able to understand, you know, has somebody worked on this problem before? And if so, is there something I can learn to speed up my own work? So, and we've seen interesting examples where there was one, I might get the details of this wrong, but we were talking to this researcher, and he was saying he was exploring this particular idea in like high-dimensional optimization. And he was like, man, you know, this thing I'm working on, it's interesting, but somebody must have worked on this before. I can't be the first person to have had this idea. I just can't. But I can't find any examples. And then he had given it, he'd sort of given a description of what he was working on to GPT-5. And GPT-5 found an example from, I think it was like economics or something, a completely different field that used completely different terminology, so no keyword lookup would have ever worked. GPT-5 did sort of a conceptual level literature search. Found somebody's PhD thesis in German. So also a completely different language, you know, it was like basically lost to time, but this person had done really interesting, sort of related work that helped him in his research. And so, you know, that's another area. So you can talk about the acceleration that comes from just like novel proofs and GPT-5 being able to do something on its own or guided by an expert. But there's also these examples of acceleration in calculations and literature search and all of them contribute to accelerating science.
Yeah, and the exact same thing happened to me. I was trying to derive this property of black holes, and I got this equation that described this phenomenon I was after, and it had a three derivative term, which is pretty unusual. And I looked at it and I recognized it's something called the Schwarzian derivative, which is a special thing that appears in math. And I thought, hmm. Wow, this is really strange that this would show up. And I just copy pasted the equation into ChatGPT and said, have you seen this before? And it said, oh yes, this is the conformal bridge equation. I had no idea what a conformal bridge was at the time. And it said, oh, just look up this paper. And that was amazing because it turns out that this equation that showed up in my work, had already been studied in some other works, and I've heard from a lot of colleagues doing research in physics that there's a lot of that going on. And at the forefront of knowledge, everything becomes so niche that it's very hard to know the latest details in neighboring fields. And GPT is an amazing help with that.
Yeah, that's another thing that we've heard from professors and researchers that we've talked to is there's so much you have to be so specialized today. And so sometimes it gets hard to explore an area outside of your main area. There's one particular mathematician we were talking to said, you know, one of my last papers, I knew there was an area that I wanted to go follow it off in this direction. But it wasn't my specialty and it would have taken me a long time and I just kind of ended up feeling like, you know, maybe that's not the most efficient place for me to spend my time. Now with GPT-5, I'm gonna go back and explore that because I've got a coworker effectively, a collaborator who has read just about every scientific paper that's out there and is a pretty meaningful expert on just about any topic you want. And I think I'm going to be able to go explore these adjacencies in a far better way with ChatGPT than I could have on my own. And so that's also a fascinating new take, right? It can help you go deeper, like you were saying, and it can also help you go more broad.
Literature search is pretty interesting because like one of my weird hobbies is I like to go back and look at when was some early scientific discovery made that didn't get utilized till much later on. Yeah, famous one was carbon filaments, you know, when Thomas Edison spent all that effort to try to find it, and it'd been published in like 20 years before. Of course, you know, Dewey Decimal System was invented that year, so you can't blame him. Other things like silicon as a semiconductor. You know, if somebody would reading the literature, we might have had that five to ten years earlier. Ability to replicate DNA, that had been published like 10 or 12 years earlier before somebody figured that out. And then the shotgun technique we use for DNA, you know, understanding, you know, figuring out like the DNA sequencing. That was first published in like 1982, but at that time there weren't supercomputers that could run it. And that's exciting just to think of just having a really good tool that can search through all this stuff and pull up these answers you have.
Yeah. And I think especially some of the most interesting research now happens at the intersections of two fields. And again, it's hard for one person to be an expert in two fields, let alone three or four or five. And sometimes it's tough for humans to collaborate. You don't necessarily find the right person, the person doesn't have infinite patience. And here with GPT, you have now the option to have a collaborator that will work 24-7, has infinite patience, you know, has read substantially every scientific paper written in the last however many years. And so it's just a new kind of collaboration that is its own form of acceleration.
You think about like Claude Shannon's wife was a mathematician and how much that helped what he was able to do. And I think we forget how much collaboration really is a factor of that. But I would say some people hearing this might go, yeah, but it couldn't spell strawberry last year. Yeah. It couldn't do math. So why are we gonna have it do, you know, science?
Yeah. So actually I don't even know if I've told you this, my own sort of origin story with appreciating what GPT-5 could do, or in this case it was, I think, oh this was almost a year ago, so it was o1 preview maybe. But I was meeting with this guy named Brian Spears, who's a physicist at Lawrence Livermore. It was in DC and we'd never met before. So I didn't know sort of what to expect. I thought maybe I was gonna go in and be talking to him about what was new and what he could do with o1 Preview and why he should give it a try. Little did I know. I sat down and he immediately took control of the conversation and said, let me tell you what I can do with your models and like these are the most amazing things for science and this is gonna change the world and he was like, okay, let me take you through this. And he opened up his laptop. And you know, he works on fusion, right? Lawrence Livermore was the first to do large-scale fusion with positive energy, like super exciting. So he's like, all right, we're gonna take a fusion example. And first I'm gonna start with the undergrad version of this problem. And so he shows me this conversation and he's like, all right, so you've got you know a copper rod and we're gonna bombard it with super high pressure waves. What happens? And you know he's like it. So he answered and o1 preview gives a good answer. It's like okay, cool. So it got the undergrad problem right. And then now let's ask the graduate version of this. Now what happens inside the rod itself as you're doing this? And you know what needs to be true in order for it to generate these certain kinds of shock waves? And he goes through and he's like, okay, so got that right. All right, now let's ask the postdoc level question. All right. Now let's ask the and at this point I'm like, you know, despite having a physics background, I'm just following along for the ride because he's beyond anything I can do. Like, all right. Now let's ask the you just joined Lawrence Livermore and you, you know, kind of question. You've gone through your postdoc, you're a nuclear physicist, and he keeps going. And o1 preview keeps getting the answer right. And then he's like, all right, now let me ask you the you've worked at Lawrence Livermore for twenty years question. And it goes and it gets it right. And then not only that, but it suggests that the only way to go forward is to use these set of simulation tools that are like partially classified or that only Lawrence Livermore has. It's like, you know, I don't have access to these, but if you did, you would want to use these tools. And he's like, look, nothing in here that nothing that I just showed you was something that I couldn't do. But it would have taken me days. And certainly not everybody at the lab can do this. Like the acceleration that comes, that is going to come from these tools, is going to change science. And so I went from like sitting down with this guy who I thought maybe I was going to be sort of talking to him about the value of AI, to him just completely blowing my mind about the potential of AI. And this is a year ago. This is o1 preview, you know. We've come leaps and bounds since then. And the thing that I always try and remind everybody, the AI models that we're using today, as good as GPT-5.1 Pro is. These are the worst AI models that we will ever use for the rest of our lives. And when you think about that, the fact that we're here just implies that the future is very bright.
How have your colleagues been using these tools?
Yeah, there's a lot of different usages, I think. Literature search, here's what I'm working on, does it connect to any other thing? And this is something that we spend a lot of time on as scientists, just understanding when something new shows up in our work, how it connects to other things. And okay, my own experience that made me become AI pilled, I think.
This is the reason you came to OpenAI.
I met Mark Chen, who works here at OpenAI, he's chief research officer, and he gave me a challenge. He was very proud. He said, you know, why don't you just give it a hard problem? And I thought, ha, you want a hard problem, okay. And so I gave it this question. Right. So I had just found these new symmetries of black holes, which is something that doesn't happen that often. And I'd written up a paper that came out in June on the archive, and I was very happy about that. And I thought, okay, well let's see how GPT Pro handles this new question. And so I gave it the equation. And I didn't say that it has some symmetries. I didn't give it a leading question. I just said, what are the symmetries? And it thought for five minutes and it said, no symmetries. And I go, ha. It's not there yet. Still better than the AI. And Mark Chen is visibly crestfallen. He goes, okay, well just give it an easier question then. And so I think, okay, I'm gonna give it the warm-up baby version of the problem, which is find the symmetries of this equation, not in the full black hole space-time, which is complicated, but in the flat space limit with where the space-time is empty. And hit enter, it thinks for you know nine minutes, and it comes back with this beautiful answer. Oh, this equation has conformal symmetry, which is the correct thing, and here are the three generators, and it was very beautiful. And you know this version of the equation it probably has been studied, I'm sure has been studied many times over the decades. So I don't know what he did exactly, but he came up with the answer. And I thought, okay, this is very good. This is a great outcome. And then Mark said, okay, well, but now that it's been primed on the warm-up example, try again in this instance of chat the harder problem. And I thought, okay, let's go. And so we give it the hard problem again. Hit enter and it thinks and it thinks and that was the first time I saw it think for so long. I think it took 18 minutes and it comes out with this beautiful answer that was completely correct. And that blew my mind because I had been working on this for a very long time. And I would say that that calculation is at the edge of my abilities. I think it's something that, you know, very few people could have done the way I did it. And so I was really shocked because you know you spend years of your life training to be best in class or something and finding symmetries of black holes and these kinds of equations. That's my jam. And I thought, okay, so I guess that just happened. And it really sent my mind reeling. And I was a little bit shell-shocked for a few days, and then I just couldn't stop thinking about it. And after that I realized, okay, I have to become involved in this because to see this capability emerge into the world. Like right now and not to be involved with this just seemed crazy to me.
I was gonna I actually think you made a really important point in the middle of that around the fact that you gave it the hard question. It didn't get it right. You gave it an easier question, it got that right, and then you were able to give it a harder question, it got there is still, you know, as excited as we clearly are about the future here, there's also a very real sense, like when you're giving GPT-5 or any of these AI models a problem that's on the frontier, that's at the limit of their capabilities, they tend to still be wrong a lot. Right. Kind of like any human would be at operating at the level of at the frontier of their capabilities. And it takes you know, it isn't just automatic yet. Hopefully in the future it will be, you know, enter in any hard question and the model answers it. But today there's a lot of back and forth. And the people that are best, the researchers that are best at getting the most out of the models have a sort of patience to go back and forth with them. I think that's natural. It's probably the way that you would work with any two people operating at about the limit of their capabilities. But I think it's important especially for folks listening to this who are doing research with the models to know that it isn't just one shot and it always works. There really is a back and forth and sort of a patience that it takes. And one of the interesting research problems that we're spending a lot of time thinking about is how we help people with, yeah, how we sort of help reduce that cognitive load. Because when you're working on a problem, say the model has a 5% pass rate on some problem. So technically the model can get it right once out of 20 times. But it's really at the frontier, so it's not going to get it right nearly, you know, even close to every time. If you're sitting inside ChatGPT and just entering in this question, you're gonna have to enter it in, you know, what, ten times before you have the odds that it's gonna get the right answer? And that's most people aren't gonna do that and so there's a whole host of problems that the model can solve that people probably try and are like, oh after three tries, it didn't get it right, so let's I'll move on. The model's not good enough yet. And actually it is, but it's just very hard to tell apart low pass rate problems from problems that are too hard. And I think that's actually a really important thing for us to help researchers and mathematicians get past. Because the most interesting problems right now are going to be the ones where the model has a very low but non-zero pass rate. Those are going to be the hardest problems that the model can solve, the best ways that it can help accelerate science. And so that's a really interesting research problem that we're taking on to try and make that a little more automatic, a little less grunt work. But for now, I like putting in the time and really going back and forth with the model does yield results.
Well, it feels like we're at a moment kind of like when we went from GPT 3.5 to ChatGPT. 3.5 was a model, extremely capable model, but it was still effectively a base model. And I was a prompt engineer at the time and knowing how to prompt it, I could get great results for it. But it took all those little tricks to sort of understand the context. Then when we went to ChatGPT and we understood, okay, we know the kind of problems people are trying to solve. Let's make it a little bit easier for them to get there without having to do that. It feels like that's kind of where we're heading into a science though, that now that you have people like Alex explaining the problems you're trying to solve and what you're doing, that we may see like a big acceleration with this.
I think it's probably just a characteristic of any question that's on the frontier of or sort of at the limit of what the models can do. And back with GPT 3 and early versions of 4, the questions that were at the limit of what the model can do were much more basic. Now they're questions of you know scientific research, but you still, when you're operating at the frontier, the pass rate will be low. And so you gotta kinda like there's value in sticking with it and trying a few different things and taking the parts that it gets right and refining them while telling the model where it got other things wrong.
In this example I mentioned, it needed a warmup, but the warmup was the obvious warmup that you would do as a human. Right. Because actually when I was attacking this problem, I wasn't thinking about the black hole case first. The flat space limit was the obvious place to start and that is where I began. And so I think the models are actually really good, but we could get better at making them think of the warm-up problem themselves so they can go there directly. But more generally, I think there's this thing we have to bear in mind, which is that as scientists, our role is to push the edge of knowledge. There are things that are just beyond the edge. And our goal is to bring them before the edge of knowledge by understanding them, but this edge is very jagged. So there are very basic questions about the universe, like why are there three dimensions of space? Or, you know, what happened at the Big Bang? These are things that everybody wants to know the answer to. And yet, even though there's simple questions, there's really nothing intelligent to say about this. We just don't know. They're very hard problems, actually. And then meanwhile, there are these very hard questions that you would think we wouldn't be able to answer at all, to which we have extremely detailed answers. We can predict the electron dipole moment to, I don't know, 12 decimal places, something crazy. So the edge of human knowledge itself is very jagged. And it takes many years of graduate school to learn where the edge is. And I think what we're finding with these AI models is that the edge of their knowledge is also very jagged. So you mentioned, you know, there's some basic questions that the models can't answer. That's true. At the same time, there's some very hard questions that they're very well suited for already today. And I think what's exciting is that their edge of knowledge is very jagged in a way that's different from ours. So obviously as time goes on, I think the edge of ability for these models is going to keep expanding. But as long as it expands in a way that is slightly different from our edge, that's also really interesting because at the intersection where it can go farther than us or we can get ahead of it, that's where a lot of interesting things are gonna happen, I think.
Yeah, and human and AI together are exactly much more powerful than human alone or AI alone.
Exactly. I want to explore that a little bit more, but first tell me about the research paper.
Yeah, so we've talked a bunch about these anecdotal examples that Alex has gotten from the time that he spent with his colleagues that we see coming in across Twitter, you know, on a semi-daily basis at this point. And we wanted to sort of bring them together and just write something, publish something about that that lays out the current sort of state of GPT-5 with respect to science. And so what we've got, it's a handful of collaborators from inside OpenAI and I think eight or nine academics from beyond our walls. Across a bunch of different fields, math, physics, astronomy, computer science, biology, materials science. And the paper is something on the order of twelve sections, each one highlighting a different way that GPT-5 is accelerating their work. The goal was not to be, you know, hypey and say everything is solved. It's, you know, it's really to say, look. Like this is what works. This is what doesn't work. Here's what I tried. In many cases, we're sharing the full share links, the conversation, so you can see the back and forth that the scientist has with the model. And it's meant to be kind of a moment in time to say, this is where we are today. And I think we'll look back in six months, twelve months and we'll probably be much further and that'll be exciting. But even where we are today, we've got a section in the paper on a bunch of different examples around literature search. A section of the paper with a bunch of different examples around acceleration, whether it's calculations and other things like that. And then a section where we actually contribute four or five new non-trivial results in mathematics. And a couple of these are small, a couple of them probably could have been papers on their own. And so you go from kind of the mundane but very pragmatic and real bits of acceleration to the more sort of profound GPT-5 actually pushing past the current frontier of human knowledge. And so we're super excited about this paper. It's, you know, I think there'll be a lot more to come. We're not the only lab doing great work, by the way. Google has been doing this for a while, and I have a ton of respect for what Demis and the team have done with AlphaFold and more. I just think we're at a really exciting time. You know, ideas in science often have their moment when you have multiple people coming with the same idea, whether it's quantum mechanics like Alex was talking about or the light bulb. Right now it's very clear that AI is just beginning to change science. And it's going to be an exciting few years.
What advice do you have for students and grad students in the sciences? Because I hear people talk about like, oh, we're not going to need scientists anymore, which sounds absolutely crazy. It's not like the telescope got rid of the astronomer. It actually created the astronomer. How do you feel about that and what advice do you have?
Okay, I think first of all it's important to acknowledge there's a lot of anxiety in academia right now that is unrelated to AI. It has to do with lots of changes in the way that science is organized in this country and we're still going through these changes. I think that talking to young people, there's a lot of anxiety surrounding this. I actually think AI is a really exciting new tool that's coming, that's becoming available, that is going to help a lot because it's just going to make everybody just so much more efficient. As Kevin was mentioning earlier, when you work on a research project, oftentimes you don't know which way exactly to go. You know you're here, you want to get there, but there are different possible paths, different lines of attack. And the whole point of research is that from the get-go, you don't know which way to go. And one of the things that's really fun, actually fun with GPT is that you can just say, hey, I'm trying to solve this. Here's some ideas I have. You can upload some notes that you have or just describe it in a few sentences. And it's very good at getting what you're trying to do. And then you can just say, what if I approached it this way? Or what if I were to do it this way? And it can immediately go off and chart a path through the unknown, just signposting different potential avenues. And that actually saves so much time because, you know, I'm a human. I have limited time, energy. And when I'm gonna put in the effort to do a calculation, I spend a lot of time trying to prototype it and think ahead where it's gonna take me. And with ChatGPT, I can just launch it in that direction, in that direction, that direction. And it doesn't completely get everything right, but just having these signposts along the way is so helpful because then when you do go down the path yourself, you have somebody helping you along, it feels like. And I think that's just gonna make everybody faster, more productive. And you know, I already the young people that I meet are spending a lot of time experimenting with ChatGPT and figuring out its capabilities. And I think it's gonna be a boon for everyone.
You mentioned part of the idea of the paper was to say, okay, this is where we are now. Let's go look in six months. Let's talk. We're five years since GPT 3. We're five years from now, we're sitting down here. What are we gonna see?
Oh man. The five year question is so hard. Here's a crystal ball. Yeah, you know, I think I mean the exciting thing about this field in general is you look back twelve months and you're completely embarrassed by where you were twelve months ago. You know, the idea if I When GPT-3 launched, it was unbelievable. Right. I mean, it I'll speak for myself. It blew my mind the idea that AI could do any of these things. And then somewhere in around like GPT 3.5 and 4, the Turing test, which we had held up for like what, 75 years as the pinnacle of artificial intelligence research. Like oh man the world will be different when an AI can pass the Turing test. We just went whooshing by and like now we just don't talk about the Turing test anymore. And even you look back to the beginning of this year of 2025. And most people were writing code themselves. Most engineers were writing all of their own code. And the idea that you're writing it yourself. And now fast forward, and you've got like the idea that you would do really much of anything without leveraging Codex, Claude Code, GitHub Copilot, you know, any of these tools. They're all incredible. Is crazy, right? You're so much more productive with it. So just in 12 months, and I don't in 12 months software engineering has fundamentally changed. I think over the next 12 months, we're going to see profound changes in the way that science is done, you know, both in the stuff that we can do in silico, in theoretical physics and mathematics, and computer science. And I think we're going to begin to see it in the life sciences, in the physical sciences. That's over the next twelve months. I mean five years?
