Steve Hsu: Well, we, we're in a weird situation right now where the Trump administration really wants to destroy the universities. I say this with some knowledge from having spoken to pretty high-level people in Trump world. and the basic idea is just that if you're on the right, you think universities are machines for churning out enemies of, of the right.

Thomas Fink: Right.

Steve Hsu: And so, you know, and I argued with them. I said, "Well, science is not really the main problem here," and yet it, in my opinion, is maybe the thing which, in the university, which produces the most benefit to society. So you should, you should carve out science and engineering and not destroy them, even if you have a problem with other ideological things going on at the university. But they didn't accept that. They just said, "Yeah, Steve, that's, that's too idealistic. We're just gonna kill the whole thing." And so

Thomas Fink: Right.

Steve Hsu: they really are trying to kill the whole thing. Now, now they're not gonna succeed completely, but it's just it does make it a, a not good time for American university-based science, I would say.

Welcome to Manifold. My guest today is Dr. Thomas Fink. He is a theoretical physicist educated at Caltech and Cambridge. He's also the founder of the London Institute for Mathematical Sciences. Thomas, welcome to the show.

Thomas Fink: Hi, Steve. Thanks for having me.

Steve Hsu: It's great to have you here. you've done something very original in creating the London Institute for Mathematical Sciences, which I guess from now on I'll just call LIMS. Is that what you guys call it?

Thomas Fink: Yeah, LIMS for short.

Steve Hsu: Okay. So you've done something very original in creating LIMS, which we'll get to. Maybe we can just start with your biography. Just tell us a little bit about yourself.

Thomas Fink: So even though I I live in London, I grew up in, in America. I'm, I'm American.

I grew up in Texas, San Antonio. I wasn't first interested in physics until I was a teenager, but I was interested a lot in coding and mathematics and symmetry I think you've heard you and others sometimes talk about what was the magic that kind of led you, got you excited.

For me, it was Roger Penrose's book, "The Emperor's New Mind," which I just thought was, you know, extraordinary because, you know, in high school you're learning like mechanics and electromagnetism. But Penrose, who I think has a real sort of knack for you know, getting people excited, I thought that was incredible.

From there I went to Caltech. I was at Caltech. I was an undergrad.

Steve Hsu: Yeah, it's funny that you mentioned Penrose's book. I, I actually really enjoyed that book when it came out. I'm, I'm quite a bit older than you, so I think it came out when I was in grad school or something. The book that played a similar role for me was Gödel, Escher, Bach when I was in high school.

And, but I'm curious, just a quick question, like do you feel like Penrose's thesis in that book is actually correct? Or is it just something that sparked your interest in.

Thomas Fink: The one that human minds can do non-computational things. Yeah. You know, I've, I've seen him, argue that in other books. And, and I've seen sort of con-controversy on that. I mean, I suppose I'd put it this way. There are some real sort of dead spots in science, like, you know, things that we, we really don't understand.

You know free will, physics seems incomplete. I think we don't have a thermodynamics of life, of, you know, self-reproducing or, you know, self-replicating systems that learn. His work leads me to believe that there's probably some fundamental parts of physics we don't know about.

And I wonder what that might tell us about, you know, life, about, you know free will. So in that sense, I think he may be onto something.

Steve Hsu: Yeah. Well, it's definitely, it definitely raises a lot of questions, I think, that still need to be explored further. Well, yeah, I won't get into the book 'cause that's like maybe the subject for another discussion.

So after that, you did a PhD at Cambridge in condensed matter theory. Is that right?

Thomas Fink: Yeah. So at Caltech, I, you know, got into a, a course early on called Physics 11, which was sort of an experimental course set up by someone called Tom Trombara, who's became head of physics and math and astronomy there.

And it was for a handful of students who wanted to do, you know, get stuck in genuine research problems. So, you know, I spent some summers there doing research projects you know they call them SURFs there, Summer Undergraduate Research Fellowships. And I particularly, you know, got into what I would we would now call statistical physics.

At the time, I just thought it was like, you know, discrete dynamical systems. Stephen Wolfram's cellular automata I found very exciting, and some of his work, you know, extending some of that. I went into stat phys, statistical physics for my PhD. And yeah, I went to, went to Cambridge to the Cavendish Lab, which is largely an experimental place, but they have a, like a one part of it, which is theory, theoretical work, so theory of condensed matter and, and a lot of stat mech.

Steve Hsu: One of the things I was interested in in your bio is I think after that, did you actually become a CNRS, so that's the French National Research Ad-Administration. Did you actually become a CNRS staff member?

Thomas Fink: Yeah. That's a funny story in the sense that I, I really didn't know what I was getting into.

I got a junior fellowship a bit like yourself. Yours was at Harvard, but mine was in Cambridge at Gonville and Caius College. And that was great because it gave you a huge opportunity. You know, you didn't have a, a supervisor. It was right after my PhD. It gives you a lot of freedom to pursue some things.

And I, I thought I'd I'd put that on pause for a year, you can do that, and I went to Paris to ENS to work with someone called Bernard Derrida, who's a, a stat phys guy. And, you know, some of the other scientists there were saying, "Oh, there's this thing you should apply for. Like, you know, just, just submit the application."

And this was a CNRS post. Anyway, I was at the ITP, I think it's in Santa Barbara. It's Kavli now, I think. Anyway, the Santa the Santa Barbara Institute for Theoretical Physics, where you can spend like an extended stay. And I got a call or I got an email saying, Can you come for this CNRS interview?"

So I went there and I said, "Look, I'm not gonna try to speak French but here's what I'm doing and here's what I'm, you know, what I'm excited about research-wise." And to everybody's surprise, I got this, and people were a bit shocked, like, "How did an American get, you know, a French CNRS post?" So it turns out these things are tenured track.

They're not tenured track, they're just permanent positions. You're classified as like a, a civil servant. And you know, the salaries aren't super high, but you have a lot of freedom, and if you want, you can teach to augment your salary. Anyway, having gotten this, I thought to myself, you know "Do I wanna spend the rest of my life in Paris?"

And it was around this time that I, that I got to know London because I never lived in a big city in my life. Well, I lived in the outskirts of San Antonio, then Pasadena, California, then Cambridge. But Paris was a big sort of eye-opener to big city life, and I liked that a lot.

I liked the bars, I liked the people. And so that's when I would start visiting London quite a bit.

And it was also around that time, you know that some sort of question marks have opened up in the career trajectory of young researchers, you know in particular This idea that, you know, every time you get promoted, you, your life tends to get a little bit worse.

You, like, you get punished for, for doing, you know, for doing better science. Punished in the sense that at every career progression, it seemed to me and to some of my, you know, friends, physics friends, that each stage, you know, reduced by a significant fraction the amount you could spend on working on cool problems and the joy of insight.

Steve Hsu: Yeah, I feel the same way. I think in terms of research freedom, I think when I was a postdoc, I had no responsibilities, no boss. I could do whatever I want. So that, those are, those are the ideal circumstances, and as you become a professor and you end up leading a research group and having to apply for grants, it just sucks, sucks the life out of you.

I think another thing that just happens to coincide at the same phase of life as that is people tend to get married and have families, and that also sucks a lot of energy away from research. Of course, you know, family life is great for its own reasons, but, you know, it could have a negative impact on your research.

I think these are all, like real things that happen. I didn't, never really knew that much about the French system, but I always romanticize it thinking like, "Wow, isn't this great? You get a CNRS position and you're just set for life. You can work on whatever you want. The French positions, French system should be fantastic.

There should be all these people that are just free to produce really creative work and not be pressured about getting promotions and tenure.Probably the reality is quite different than that idealized picture.

Thomas Fink: I think the CNRS is a good thing. The idea is you don't have to teach, and you'll get sort of a low base salary, and if you wanna teach you can get more.

It was partly that which, you know, led me to start thinking, why aren't there, institutes, you know systems outside of the university sector where you can do research?" Now, you know, at, at this time, I was convinced that maybe, maybe there's something fundamental inextricably important the link between doing science and teaching science, creating knowledge and passing on knowledge.

Now, in retrospect I think that's a frozen accident. I don't think there's a deep reason that they should coexist, but one thinks, well if there was a better system, we would have figured it out by now. I much more believe that much of life is a frozen accident, kinda like the QWERTY keyboard, right?

You think it's there for a reason. It's actually there for a bogus, a silly reason, to stop mechanical logjams in typewriters. I really wondered, you know, why is it that knowledge creation and knowledge transmission, you know, teaching and research, why do they need to be bolted together?

Why aren't there places where research takes precedence? And, you know, the more you look into that, I think you find that universities are thousand-year-old machines for passing on knowledge. And, and I think we've only in the last 100, 130 years, we've bolted on knowledge creation research onto those systems, and they weren't really designed for it, right?

To my mind, that tells us that we're, we're really far from optimizing organized discovery. I think there's a lot of scope for better systems.

Steve Hsu: Yeah. I agree with you completely. Maybe you've talked to Michael Nielsen about this. He also has thought pretty deeply about these issues.

You know, I think that in my own experience, two people that had very different opinions about this kind of thing, just to take two examples. One, you know, one is Richard Feynman, who, you know, he could have been at the Institute for Advanced Study where they don't have any teaching duties and they don't have to deal with students if they don't want to.

But he said that was where people go to become, you know, mummified or entombed. Mm-hmm. And he felt that the interaction right with students you know, to my benefit when I was a student at Caltech stimulated him. So every time he was forced to deliver some lecture and it seems like a real sort of tedious thing to write some elementary lecture for some kids, it, it forcibly stimulated him to reconsider the ideas again in a kind of possibly fresh way, and he felt he needed that, so he never wanted to give up teaching.

Thomas Fink: Mm-hmm.

Steve Hsu: The other person I knew really very well as a researcher is a guy called Sidney Coleman, who interestingly, he had many, many more students at Harvard than Feynman ever had at Caltech. So Richard Coleman was one of the, the, you know, if you look at his tree of descendants, PhD students trained, it's huge, and a number of really, really great theorists.

But he never liked it. He always, he always said, "You know, this is a, a, a duty that I have to perform, but in the Boston area, there are so many good mature theorists who are already professors for me to work with. I would if I could, I would just rather do that than deal with all these students." So yeah, I can see huge variation in, you know, what are the optimal systems for producing breakthroughs or really good research. And, and also like, I think something we'll get into later in the conversation as well, all this is gonna change too because of AI. And so I definitely admire someone like you who you know, you've literally created your own institute, right?

To maybe test some of your theories about how these things should work. Curious, how, how did that happen? So you, you were at I don't know how long you were in Paris. You were saying you're getting familiar with London. how did it happen that you ended up creating your own institute?

Thomas Fink:

I sort of bounced this idea around. You know, I remember I went to the head of the theory group and he then became head of the Cavendish, someone called Peter Littlewood. I think he then went on to become a head of maybe Sandia, one of the national labs.

Steve Hsu: Yeah, I know him. He's at, he's at Chicago and Argonne now

Thomas Fink: Okay

Steve Hsu: I think. Yeah.

Thomas Fink: Right. Maybe it was Argonne. So, okay, you know Peter Littlewood. So he

Steve Hsu: Yeah

Thomas Fink: he was my PhD internal examiner. And I, you know, would sometimes go to him for, you know, advice or we didn't do very similar research, but I was interested. I remember him saying I said, "Look, Peter, crazy question.

Why aren't there places where outside of universities where we can just focus on research?" 'Cause there's a lot going. You know, universities wear many different hats. They wear the hat of, you know, teaching and research. They wear the hat of, you know, in parentis locus, like forming the character of young people.

They wear the hat of arts and sciences, you know, theory experiment, you know, law schools, medical schools, sports teams, you know. And I said, you know, "Why, why is this?" And he said, "Look..." I mean, he, he said, I would "I suspect it's really hard to have to constantly eat what you kill and live off you know, research grants."

You know, he was a bit wary of this. Those turned out, turned out to be prophetic words, but nonetheless, you know there were several of us another junior fellow, a couple you know, people who had recently finished their PhD, or at least in the last five or six years.

We thought there should be something different. It turns out with in the CNRS, like many positions, you can moonlight twenty to twenty percent of your time. So I thought, "Okay, I'm gonna moonlight. You just have to tell them." I told them, "I'm gonna moonlight in London." it wasn't particularly mysterious.

I thought, okay, well, how do we, we gotta create a nonprofit. So how do you do that? Well, first you gotta, like, create a company. Then you gotta, like, get an accountant. Then you apply in Britain to the Charity Commission and say, "This is our public good to disseminate knowledge." So it was really just, you know, start at the beginning.

Now in Britain, bizarrely, this was totally unprecedented. The number of independent organization, research organizations in the physical sciences is zero. Actually zero, 'cause the government has a list of them. All, all places that can get UKRI funding. UKRI is like NIH, NIH and NSF combined.So we knew we were unusual.

And, you know, there's a lot of interest in this now from various scientists, and my, my sense is it's, it's often the very best scientists who are most passionate about spending a lot of time in their research. Not always, but I think the ones who are very passionate tend to be really good scientists, really good physicists, in this case.

One thing we realized, it would be hard to get research grant funding in our early days from Britain because they're gonna say, "What are you guys? You guys are jokers. You can't just like rock up and, you know, create a, a, you know, a framework for this thing." But, but ex- you know, foreign countries, you know, they don't have the bandwidth.

You know, if you get an EU grant, you know, and you're someplace in Poland, Brussels in isn't just doesn't have the bandwidth. As long as you're nonprofit, you tick a couple of boxes, you get it. So we set this up. I was there 20% of my time. A few other physicists would spend one day a week of their time.

We found an office which was 400 pounds a month. It was sublet, a single room from like a telecoms company. But you know, we had it and then, and then I had a DARPA grant when I was in Paris with my CNRS post, and I was able to transfer the DARPA grant to LIMS. And I remember the program manager is someone called Ben Mann. It was called Fundamental Laws of Biology, and he was such an inspirational program manager. And I'd written to him since. I said, "Ben, I don't know if you knew how bold this was to allow that transfer, but thank you for doing it." And it was Simon Levin. I don't know if you know him. He's at Princeton. He was a program.

He was, he was the head of this particular DARPA project. Anyway, that was, that was the beginning. So we had a grant and we had a, you know, we had a room, and we had three or four one day a week scientists.

Steve Hsu: Wow, that's, that's amazing. You know, I, I had I have a friend who's at I don't I shouldn't out him, but he's at UCLA, and he's, he's pretty unhappy about, in particular, like the woke diversity stuff there.

Thomas Fink: Mm-hmm.

Steve Hsu: And as a way of protesting, his plan is to create a non, a not-for-profit also, and run all his grants through the not-for-profit so he doesn't have to pay UCLA overhead because he doesn't support what the UCLA administration is up to.

Thomas Fink: Right.

Steve Hsu: So he had a very similar idea to yours. And, and I've heard other people also, other research scientists moot this kind of idea. I don't know very many that have carried it through. Here, it's a pretty painstaking process, I think, to get approved as a not-for-profit, but maybe it only takes a year or something like that. But anyway, yeah, it's a great idea. And I guess it's grown quite a bit, right, since, since what you described?

Thomas Fink: That, what I just described happened 15 years ago. It was an it was an idea 2011. We got this, this DARPA grant, then we applied for a Department of Defense. It'sone of the agencies. It's called DTRA, Defense Threat Reduction Agency. But the thing is these defense, US defense will fund basic science.

Then we applied for EU funding. EU spends maybe 100 billion a year on multi-country, multi, you know, partner projects like, you know, three or more countries getting together. We got a few EU grants. We had to register with Brussels. And eventually, we convinced the British government to recognize us as what's called an IRO, independent research organization.

What that means is that you're allowed to apply for government funding, research funding. In our minds, we thought the hard part would be creating a good scientific brand, you know, getting, doing first-class research and getting first-class people. That would be the hard part. We thought the easier part would be raising funds for this because there's such a gap in the market, right?

Of course, and, you know, so many people think that why would universities in Britain have not just a kind of a monopoly, a 100% complete monopoly on, in the physical sciences? You know, biomedicine, you have exceptions to this. Like here you've got the Sanger Institute. You know, you, you've got various things in America, but outside of biomedicine, you know independent physics is rare.

Steve Hsu: Mm-hmm.

Thomas Fink: Anyway, we-- the government, agreed, to recognize us. But, but back to my point, we thought the hard thing would be creating a good brand, scientific brand, easier to raise money. It turns out that it was the opposite. You know, we were, you know, we, we early on in our board we had the head of Caltech Physics.

And, you know, we got you know, one of the top physicists in, in, Luciano Petronaro sort of a one of the top physicists in, in Italy. Vice President of the Royal Society joined. we were attracting really good scientists, but, you know, raise funding was tough. Funding was really tough and, you know, it, it was, it was painful.

It was really painful. The nice thing, and I see this sometimes with people who are creating startup companies, the nice thing about not growing too fast, not getting too much money too fast, is that you stay really lean. You know, you, you really think carefully about that next hire. Like, do we really need a science writer or do we need, you know, a, you know, a coder or a graphic designer or digital designer or we do we need another physicist?

And, you know, I don't know it's one of those mysterious things in life, but that suffering, I think, kept us very lean and pure and, and I think shaped a lot of our organizational routines to this day. So very little bureaucracy. A real focus on meritocracy, just hiring the best people.

It helps that we're in theoretical physics and we've also have some pure mathematics, because that keeps us out of the political radar. We make a point of avoiding anything that's not that we would do, you know, things concerning energy or sustainability. It's a bit on too applied for us. But, you know, very few people wanna get, you know, up in arms about number theory black holes and, you know, emergent phenomena.

Steve Hsu: So I, you know, the, the dichotomy between, you know, how hard it is to create a quality institution versus how hard is it to get stable and predictable funding, I would've predicted the opposite of what you said, what you originally thought, that actually the, it, the hardest thing is to get this, the funding.

You know, I think for an organization like yours, in order to get someone to take a, quote I don't know if you have, to what extent you have, quote, "permanent positions," but if you wanna get someone to take a permanent position at LIMS instead of taking a permanent position or tenured position at an ordinary university, usually you have to have a big endowment or you have to have some way of, you know, reassuring that person that of the stability and the perdurance of the organization. And not so hard for post-docs 'cause post-docs are, are in that precarity anyway, and they're only temporary positions, so they don't care if you're gonna be around in five years because

Thomas Fink: Right.

They're not gonna be around in five years. So I'm curious how that has all played out for you. Like, what fraction of your staff are really quote, "permanent?"Right.

Steve Hsu: And how did you reassure them? Yeah, go ahead.

Thomas Fink: So that's, you, you hit the nail on the head with that question. And that was a big thing for us. So how do you... You know, first of all, people want or very, particularly scientists, are very brand conscious.

Like am I, am I at a good place? Am I surrounded by really good people? I think the, the best scientists wanna be by, surrounded by people better than them, so they're like stretched up upwards. And also can I attract really good post-docs and, you know to work with me? There were two things, two things that I think helped us get over that hump.

One was that the very good people you know, who feel they can be sort of snatched up you know, and are willing to, were are willing to take a bit of a gamble. One of the, the first person. We don't call it tenure, we offer it we call it an open-ended position.

Because one of the things we realized is, you know, we, we have staff and we have scientists. We're probably sixty percent scientists, forty percent staff. And we made a real point of having first-class staff, like world-class science writers. Our science writer worked at The Economist before and you know, world-class fundraiser or accountant or graphic designer.

And we don't segregate them. They work side by side. And this blew my mind because Agmo, I'm sure you're used to this as well. When, you know, usually in, in science departments, there's like first-class citizens are the scientists and second-class citizens are the staff, and the staff are perhaps not first rate and maybe they might be, you know, second or third rate.

And there, there's often division between the two. Like, "Oh, you're the, you're the enemy. He thinks you're amazing." And it's like, "Oh, you're, you know, you're the, the, the, you know, grants officer who's always trying to block my applications or slow me down." Anyway, we found when you have A players on both sides, we actually it's really fun to work with super talented staff.

The, the reason I mention this is that, you know, none of our staff asked for tenure. We just said, "Okay, you've got a job here." And they thought, "Okay, well, I'll stick around until I don't wanna stick around or maybe you guys don't want me around." And there's the key thing, "And I know I could be employed somewhere else."

Right? So there's this sense of transferability, like or, or sort of a, a non-brittleness in, in the real world. But in academics, there was a feeling like, "Oh, if I lose I better hang on to this," because particularly after a certain age, well, maybe it's very difficult to get a position somewhere else.

But so, so early on, you know, we one of our first what we call open-ended or long-term, you know, permanent open-ended positions was someone called Yang He who's just been. He's probably been with us about, you know, six, seven years now. He, I think, is somebody who very much believed in what we believed in, and I think he could immediately just get a post somewhere else.

You know, he's in, he's in great demand, and he's very inspirational. But probably the, the, the thing that affected us the most is when the Ukraine war broke out we sort of inadvertently ended up becoming the biggest importer of Russian brain drain. So of course, you remember everyone around the world is talking about helping Ukrainians.

But we said, "Look, there's gonna be a lot more Russians." I mean, they're just more Russian theorists because it's bigger. They have a great tradition of physics and mathematics. And you know, Ukrainians are certainly men it's hard for them to get out of the country. And we, we quickly raised three million pounds from two Russians business leaders.

Well, one is two Russian business leaders, businessmen and in London and, and we created this program and we did it really fast. We did it like in seven weeks we were advertising for positions. Now here I'm gonna say shame on the universities 'cause the universities never had the courage. This is at a time where it's unclear what is our view on, on Russian scientists as opposed to, you know, Russia itself.

They don't have the courage to create these programs for Russians. So, you know, the Russians would tell us, "Yours is the biggest program. It's you guys or Israel." And so we, we had we've raised enough funding for 12 positions. We've had eight Russians, four Ukrainians. Those were four-year, four-year posts, so long enough for them to get indefinite leave to get their British passport if they want it.

And that meant we were able to get talent we could not otherwise get. You know, the, the analogy I sometimes use if, if you want to the only way you're gonna get Einstein out of Princeton is if America has a civil war. And you know, we had somebody from the Russian Academy of Sciences, which is, you know, the equivalent of the US Academy National Academy. So that was an interesting thing. You know, when there's a crisis abroad, great talent flees and, and open your doors

Steve Hsu: You know, when the Cold War ended, there was a huge flood of Russian scientists. You know, literally all the top ones, I would say, came to the US. So people like Migdal and Polyakov and people like this. And I had the unfortunate I was in the unfortunate situation, I finished my PhD around the time when this huge influx was still in, in place, so it was very tough for Americans to get jobs

Thomas Fink: Right.

Steve Hsu: When all the top jobs went to the top former Soviet people. I'm a little sad that the, our system here wasn't nimble enough to really take advantage of the, in this sense, of the Ukraine war because that is what happened last time. We had a flood of really top Russian and Eastern European scientists. Yeah, but it sounds like you, you were, you were alone in, in doing the right thing.

Thomas Fink: Well, you know, we, we were wary of a sort of a backlash because, you know, the, the, there was certainly here, but I think around the world there was a feeling like you, you should have nothing to do with Russia.

The public sentiment, what are you doing bringing funding, throwing money at Russians? Aren't they the bad guys? But our approach was to say, "Well, look at, turn to history. Turn to the 1930s," you know which more or less to my mind created American scientific dominance was the 1930s import of talent.

In Britain you had the Academic Assistance Council which, you know, was bring, you know, scientists over. And you know, as you mentioned in the end of the Cold War, you had a big flux of scientists. So I think, you know, history tells us, and this is something we, we storified. We wrote, went to the press and said, "Look, this is what we're doing and this is why it's important."

History tells us that great nations open their, open their arms to talent when it's on the move. I think it's a bit sad that, you know some countries have, you know, have demonized not only the state but also individual scientists or said you can't collaborate with certain people or we won't publish joint papers with certain people.

Steve Hsu: Actually speaking of that, it's the situation with China is getting very bad in the sense that, you know, for example Tsinghua University offered a visiting professorship to me and I was told if I took it I would be in big trouble. You know, I would never be able to get a grant in the United States and I would be blackballed and the IRS would audit my taxes.

Thomas Fink: Wow.

Steve Hsu: And so I, I just said, "Well, I'm not taking it." Even, you know, I might visit you guys

Thomas Fink: Right.

Steve Hsu: And give a talk. I think I'm still allowed to do that, but I'm not allowed to visit for a year or something like that, so. And, and arguably that's the, I would say from my own personal experience, that's currently the biggest concentration of student talent anywhere in the world actually right now. It's too bad, too bad politics gets involved here.

Thomas Fink: I think that is a shame. You know, I one of, one of the things of not one of the advantages of not being a university, you know, where the government doesn't have any control over us, you know, we don't get subsidies. I mean, we can apply for grants, but is that, you know, we, we have a certain quite wide range of possibilities in how much we engage with different countries.

I guess in America, it's coming from a federal at the federal level.But I think that's sad. And you know, you see this time and again, you know, throughout history. I think G.H. Hardy said, you know, he was outraged that after World War I, a conference wouldn't invite German, a German scientist. So it's a problem we repeatedly see.

Steve Hsu: Yeah. So tell me a little bit about your future plans for Limbs. Like, where, where would you guys like to be in five or 10 years?

What, what are the main challenges you're trying to surmount right now?

Thomas Fink: So, you know, one of the things we, we think a lot about is well, two things. One is how do you, defend and execute basic science?

So by basic science I mean, you know, fundamental science. That's the kind that's not done with regard to how useful it may be, but because it, it's, you know, it's curiosity driven. It seems you know, you're spotting the most, intriguing or magical patterns because it seems as though, you know, fundamental science, curiosity-driven science seems to lead to the, to the biggest and most radical breakthroughs.

The other is thinking about how do you, how do you fund research, particular fundamental research? And because, you know, we, we don't have student fees. We're not like a, a British university, which is a bit of a hybrid between an American private and state university. We don't get government handouts.

You know, we our view is either you change the organizational model, right? Right now we have two organizational models. We have a 400-year-old for-profit company with shares, and we have a 250-year-old charity. You know, charity is a British London invention. Within the current model, I don't think we have enough organizational structures 'cause I don't think there are good incentive structures right now for, for science, individual scientists.

We've turned to, creating an endowment, and we've been quite, you know, quite aggressive about and we're relatively young. We thought the only way to survive the sort of vicissitudes or fluctuations in governments, governments' attitudes to-towards science is to have a big endowment. Part of the inspiration of this came from the n-just former Caltech president, Tom Rosenbaum.

For the 125th anniversary Caltech, you know, they I think they were a little bit, m-quite a bit concerned about fluctuations in American science policy. You know, how much different governments wanted to, you know, throw, you know, fund it. And they thought, "We just need to go to town on our endowment." So they raised three point four, I think, billion dollars, which is, is a pretty big amount.

We got to know him and, and their lead fundraiser. And they came to visit us. They were in, in London.

And we, you know, we that really led to us to thinking about what is the role of the state? How much can you trust this? By the state, I mean the government. How much can you trust the government?

You know, capitalist democracies have quite, you know, have, have short-term changeovers in government, and different governments can view science funding differently. I don't know if you, if you have thoughts on this, particularly given what's going on and has been going on in America's science funding.

Steve Hsu: Well, we, we're in a weird situation right now where the Trump administration really wants to destroy the universities. I say this with some knowledge from having spoken to pretty high-level people in Trump world. and the basic idea is just that if you're on the right, you think universities are machines for churning out enemies of, of the right.

Thomas Fink: Right.

Steve Hsu: And so, you know, and I argued with them. I said, "Well, science is not really the main problem here," and yet it, in my opinion, is maybe the thing which, in the university, which produces the most benefit to society. So you should, you should carve out science and engineering and not destroy them, even if you have a problem with other ideological things going on at the university. But they didn't accept that. They just said, "Yeah, Steve, that's, that's too idealistic. We're just gonna kill the whole thing." And so

Thomas Fink: Right.

Steve Hsu: they really are trying to kill the whole thing. Now, now they're not gonna succeed completely, but it's just it does make it a, a not good time for American university-based science, I would say.

Thomas Fink: We had a, a big discussion internally, and then we did a public debate, on you know, what, what should Britain do in light of some of the, you know lack of jobs for American scientists or American scientists looking. And our, and our view and of course, I, I'm American, so you know, I, I'm on the American side as well as the British side.

You, you know, I'm American first. And it's not to say we'll steal your top talent, but I think that, you know, Britain, Anglo-American scientific relationships have been strong for a long time. They wax one way a little bit, then the other way a little bit. And we've, you know, we've, we've brought in some American scientists and, you know, one who's particularly, particularly talented on the back of this.

But again, you know, I, I think that this leads us to the question, you know, of, you know. I think it's, it's happened in the past and will happen in the future. To what extent should we rely on, on governments? Or perhaps we shouldn't be surprised when governments, switch moods, you know, drastically or at least somewhat.

So that's that's why, you know, we set out this endowment. We set out a, a, a target. We called it our exceptional campaign or E6 after the Exceptional E Group. £60 million, which is $78 million, which is the dimension of E6. So it has a, a little fun twist. And you know, in, in March, we raised 20 million from a, a former, you know, somebody who studied math and computer science and then went into fintech called Ben Delo. And so we're, we're looking to complete that, you know, finish that 60 million.

We've got 40 million to go. But even though that's not big by American standards, in Britain, it is considered quite audacious for a place like us, just, you know, a small place to say we really do need to create a war chest of funding and be, you know, long-term stable and, and whilst we invite government support and we think many often government will invest in, in fundamental science, I think we have to not be overly reliant on it.

Steve Hsu: Yeah. I think the general pattern though, I mean, Trump, current Trump administration excepted, the general pattern though is almost all countries that aspire to compete internationally, economically, and technologically want to fund research and development. Now, it might be in a, in a way that isn't ideal, right?

It may not be the, the structure that most encourages really deep original thought and things like that. But, but just in terms of like putting some percent of GDP into funding R&D in the university sector and in, at institutes like yours I think, I think that's a generally like growing kind of trend.

I mean, especially in, in developing countries like China and other places, India, I think there's a pretty strong commitment. They get it that they can't keep up.

Thomas Fink: Right.

Steve Hsu: With other countries unless they spend. So I don't know. In my lifetime, I would say the, the overall level of resources for science and, and the, the scale of science has just gone up.

So, so I'm not pessimistic in that sense. You could argue like, yes, universities are not the best places for research to be centered in the like in the United States, most basic research is happening at universities, and you might say there are structural reasons why that's far from ideal. I, I think I, I agree with that 100%. but pretty hard to change that at an institutional level.

Thomas Fink: I had very happy times at three universities that I spent time at, and and, and I often try to you know, make clear I'm, I'm not against universities. I think universities are magical places.

I think that they shouldn't be the only game in town, meaning there should be a significant alternative career path. And I feel that, that, that healthy competition would do a lot to force universities to say, "Okay, we're gonna have to, like, treat this person in a, in a better way or, or they're gonna bail, and they're gonna go to this other place."

I think it would for you know, force them to think a little bit about administrative bloat and, you know, like, okay, we're not as competitive as, as we should be. So I think that my guess is having alternative models. One question, and I raised this with the with the chief scientific advisor to the British prime minister who he was on our board for a while, is, you know, why is there so little variation in how science gets done, right?

It's, you know, if you go to a physics department around the world, it's pretty identical, like almost down to the décor. Places like, you know, the Institute for Advanced Study or Santa Fe, I think they more or less are imitating the conventional models of, of academia.

So I think that's puzzling. Why are there so feworganizational models or, or organizational cultures? Now, you know, maybe, and this is something that I'm, I'm wondering a lot myself, maybe the, the rise of AI in helping us do fundamental physics, pure mathematics might have an effect on that

Steve Hsu: Yeah, I agree.

So yeah, maybe we could turn to that since we're almost at the hour here. Okay, you could call me an AI optimist. You know, although I actually think there's a financial bubble right now. Setting that aside the actual progress in AI capabilities and then the impact of those capabilities on academic, say, mathematics or theoretical physics, I think is gonna be huge.

It's actually probably still underrated by most, professionals in the field. And I think it will drastically change the way that we want to organize ourselves to produce research. You could even take the extreme view that most of the best results, say, five or 10 years from now, or 20 years from now, if you like, are gonna be produced by AIs, and the problem will be humans trying to understand what the AIs are producing.

Terence Tao gave a very nice talk on this at the most recent International Congress, which was in, I think in Seoul. You know, he said we, we've switched to the problem where verification of proofs is the problem because AIs can often produce these really beautiful, complicated proofs, and we need to-- humans need to understand these proofs, decide whether we believe them, and then also then canonize or propagate the ideas of the most important results that the AIs are producing.

And if you notice the way he described the problem, it became a problem of how does the community of humans adapt to the presence of these AIs which can produce you know, at, at great volume, great scale proofs, results. And so then it I mean, I don't think he himself realizes, but, but the, the whole prescription he gave in his talk was about how the human community should adapt in this new circumstance, and it was less about what will the true frontier, the true frontier of mathematical knowledge, if you take the union of what the AIs, quote, "know" and what the humans know, what is the, what is the frontier of that union and how will that progress? He was, he was not spending a lot of time discussing that. He was just discussing how are humans gonna adapt to this new world. So I kind of agree with him in, in, in how I think this is gonna play out.

Thomas Fink: I'm certainly not a doomer. I'm, you know, I'm an optimist in the sense you mentioned. I'm also, I suppose you know, I something of an AI skeptic in the sense that I prefer to replace the word AI with automation. And you know, I mean, will have different forms of automation radically affected, you know, civilization?

At various sort of, you know, there have been jump discontinuities. you know, is it the case that, you know, at the prime of our lives, it just so happens that the defining moment of civilization, I mean, the prior distribution against that's quite strong. You know, I'm sure we're not the first person to say, like, it's, it's before me and middle age, and it's after me middle age.

It's totally two different worlds. At the same time, I, I think that there's a, I think there's exciting opportunities for working with AI. I my sense is that whilst AI can get us to the next level, you know, like, oh, I need to prove this thing, and what are the consequences of this theorem?

I'm less convinced that it'll sort of, you know, hyper escalate and solve mathematics. Partly because I think, well, there are an infinite way to progress mathematics, right? Just like there are an infinite way to progress technologies. I like to think of, a, a sort of a definition of a good theorem I like is G.H. Hardy's, a good theorem is one that enables future mathematicians to use it as a stepping stone to create yet more good theorems. So it's forward recursive. Now, I think AI's bad at that forward recursive problem, and it's, you know the future of mathematics, to use Wolfram's concept of computationally irreducible.

And you know, you know, perhaps it's, you know, the mathematics that the, the direction we choose is linked to what I call engineerable insights, human, you know, feelings, senses of, you know, of aesthetic qualities, but also it's linked to physics. It's linked to the universe. The mathematics that we need to kind of, I don't know, take charge of the universe.

So I'm, I'm equally not, not a skeptic, although I, you know, I myself am blown away. I mean, just before you were our call, I was, you know, getting ChatGPT to sort of, you know, extend a proof and simplify it. You know, I've been wrestling it with for days, right? With the same, you know, chat. So

Steve Hsu: Yeah, I think that's how most I think for outsiders, they don't realize that's how a very large fraction of physicists are working now in pretty tight collaboration with the, the models that are strongest at what we do. I didn't wanna suggest that AI is gonna quote, "Solve math." I mean, math, it's infinite, right? And physics may not be completely infinite, but, but I don't think we're in danger of AI solving every problem in physics or finding the ultimate theory of everything for us or, or building the experiments that we need to do to, to check these theories. But it is still possible to me that the role that human intelligence plays gets smaller and smaller and, and, you know, the, the true frontier of knowledge on planet Earth about some of these things lies not in wet, mushy brains, but elsewhere.

That's the big transition I think, at least some people think we can already see the contours of it happening. It doesn't mean the subjects will be exhausted, far from that. But the, the subjects may not be exhausted, but human brains might not have a good idea of where the boundaries are, most fruitful directions.

And, and that might shift to AI. Now, I realize that current AI doesn't seem like it would produce the situation that I just described, but you have to allow some extrapolation in goodness or capabilities growth of the AIs in, say, the next five years or something, 10 years, 20 years, et cetera.

Thomas Fink: Well, let me say first, my feeling has for a while been that we, we have too many scientists. That comes from this exponential growth model post-World War II that, you know, a, a scientist has, you know, a few PhD students at any one time, so he ends up, say, having 30, 25 students. Well, but that's crazy. I mean, m- my sense is you should probably have three over your entire career. One, you know, doesn't quite make it and, and the other goes to industry and one replaces you.

So this exponential growth seems to me a problem. I wonder if what we're seeing in development of AI, which is, shall I get a PhD student or shall I just keep using ChatGPT puts pressure on us to reduce those numbers.

Steve Hsu: So first of all, I agree with you. You know, we obviously this exponential growth caused by too much reproduction I think is isn't necessarily a great thing.

Thomas Fink: Right.

Steve Hsu: Although one could argue that the second order effects of having these people trained to think like you do, and then they go off and work at an AI lab or they work in a semiconductor company or a software company.

Thomas Fink: Mm-hmm.

Steve Hsu: Some people would argue that-

Thomas Fink: Yeah

that's the greatest benefit to society of, of what we do, actually. Hmm-hmm

Steve Hsu: But yeah. I think personally I didn't, never liked having too many students. I like to have a very, like, close working relationship with my students and not have very many at a time. I really like that, and the problem I have now is that the, the, the productivity I get, like if there's a particular research problem I want to explore, first thing I do is start discussing it with the models and then we can go very far.

Even though if they're far from perfect, we can still go very far in, in, in pursuing this thing. And to get a student up to that level is extremely the amount, the the differential, the difference in work factor for me to get the student to the point where they could even participate in that conversation that I'm having with the AI is, is very daunting. And so I think a lot of professors are just

Thomas Fink: Right.

Steve Hsu: You know, focusing because their main goal is producing good research or, you know, just curiosity, they're focusing more and more of their energy into interactions with AI and less and less of it into interactions with PhD students. And I, I just think that's happening everywhere.

Thomas Fink: Right. Yeah. I mean, obviously we don't have PhD students. We do have postdocs. I've noticed something similar. I mean, at the same time, I think I'm like you, I like to have quite intimate relationships with junior scientists that I'm working with.

And, you know, and it's, it's as much a, you know, ideally it's a two-way relationship. You know, I'm sort of pushing things and they say, "Oh, well, what about this thing?" And then there's a degree of, of, of mentorship. You know, it's not just utility you know, research utility. I and colleagues of mine, I think we, we try to be quite careful because, you know, I think they, you know, if you don't get the right ones, well, a, a good, a good junior scientist or collaborator postdoc, you know, a great one is, is, you know, ten times more effective than a, than a, than a good one.

So I think it's certainly made us more perhaps more picky, of that. But also, you know, one of the, one of the sort of things we say to scientists who, who come here, I mean, because we're in some ways we're quite countercultural. We tell people, "You need to show up to work." Like, we, we tell our staff to show up to work, and we don't want to tell the staff to show up and not the scientists, because then you create all this weirdness and, and we just like having people around.

We're not a big place. But we also say, "Look, you are paid to get your hands dirty. We don't employ science managers. If you're not doing calculations, this is probably not the place for you." And now, you know, in theory tends to be a little bit less manager than experiment, which is often a lot of management.

But there are a lot of theorists who are just science managers out there and I feel it's a different type of art compared to, you know, those theorists even at a, you know, middle or more senior age are actually doing calculations. And but I feel that the system is so built, it's so baked into the system that you become a science manager, that the people who are now, who, you know, the, the most scientists think, "No, but I, I've, I'm, I was selected because I'm a science manager."

Though I think it's we're kind of stuck in this catch 22 of exponential growth of science and science managers. Not always, but often.

Steve Hsu: Yeah, I agree with your observation. One of my favorite quotes is a quote from Julian Schwinger about and you know, he was always calculationally extremely strong, one of the strongest maybe ever. But he, at toward the end of his life, kept saying, "You have to keep your hand in it. You have to keep your hand

Thomas Fink: Right

Steve Hsu: you have to keep your brain sharp and continue doing calculations. You can't outsource your calculations to your post-docs and junior colleagues. Otherwise, at some point it's, I think he says something, it's all over for you or something. So

Thomas Fink: Right

Steve Hsu: so he definitely didn't believe in the science manager role. Now, I'm not saying there isn't a role for that kind of thing. I mean, you know, people who really understand the field well and can think of, you know, maybe I should put this student on this, and put this student on this, and, and you know they can create a lot of productivity in the field. I wouldn't denigrate that. But I think the true theoreticians keep their hand in it as long as they can, and that was what Schwinger and obviously Feynman thought as well. But the difference now is that, like, all these, like, things about being a research manager, of like trying to understand your student, like how does my student think and what's the best way to explain it to him? What's the best way to motivate her? A lot of that isn't necessary as much, or it's just different with the models, because the models just come in and they're like "Well, you know, you thought maybe there was some application of graph theory to this problem, but you never really studied graph theory or not since undergrad."

Right. "And well, here I'll just summarize everything for you," and then it just suddenly gives you this huge thing. And like the main bottleneck is then me trying to absorb what the model's trying to tell me about the relevance of graph theory to my problem. And yeah, it's just a very, very different situation than, than I'm used to with dealing with humans.

Thomas Fink: Do you find, as I'm asking myself, that the you know, AI-generated, you know, calculations can cause us to get a little bit you know, stop getting our hands dirty and get lose our craft? 'Cause I think, you know, it's kinda like playing a musical instrument or painting, it's craft.

Steve Hsu: Yeah, 100%. I think the main difficulty is it's so easy to get the model to do the calculation and then just kind of slowly start to accept. 'Cause you notice like, or I notice like, oh, maybe the model's wrong about this. Let me sit down and try to check what it did, or maybe just check a special case or something right?

Thomas Fink: Right.

Steve Hsu: But then I realize like, yeah, its error rate is way lower than my error rate. And so

Thomas Fink: Right

Steve Hsu: then I start to default over to it, and then I start to use like sophisticated agent pipelines where I have, you know, Claude you know, Claude checking on GPT, and then I have DeepSeek checking on that.

Thomas Fink: Right.

Steve Hsu: You know, and then like I'm like, well, maybe theoretically that beats down the error more than my doing the calculation and helping out.

Then you might, as Schwinger says, then you, it's, game over for you, right? Because you're no longer

Thomas Fink: Right

Steve Hsu: able to like really participate. So yeah, I don't know how this is all gonna end.

Thomas Fink: Right. Well, it's, it's certainly, it's certainly moving fast. I think I was, I was quite skeptical around the time you did your, physics letters B paper. But I'm not skeptical anymore. But at the same time, I mean, but I am optimistic. I'm quite optimistic because I feel, you know, I find it quite joyful working with this and almost a bit addictive.I think there's a real sense of, of gaining insight. Now, I certainly have to go through, like, see the calculations.

I then have to absorb what I think are the most relevant ones and kind of get that into my head and, and to this phrase I've been knocking around is engineerable insights. Insights which can be used as building blocks for new level insights, and I, I feel sometimes I get this convoluted proof that are like you have no sense of feeling for it from an AI model.

And I think even could AI use that as an engineerable, engineerable insight, right? It's so convoluted. I think engineerable AI is an, could be a relevant or engineerable concepts, you know, as in future embeddedness could be an important thing. And I think maybe we're at the early, you know, early days of, of that.

Steve Hsu: I'm often in a situation where the AI produces some long convo well, what seems to me this long, convoluted thing, and I'm definitely, I get the sensation I am the bottleneck. Like if, if I

Thomas Fink: Right.

Steve Hsu: If I knew what to write next

Thomas Fink: Right.

Steve Hsu: The thing would just proceed. Maybe it would simplify that thing in a very nice way, but it just takes me a long time to figure out what, what is in here, right?

And then I go away thinking like, "Wait, it, it, this part of what it said, I understand now. This part of it's still mysterious to me. I gotta take a shower and think about it in the shower," you know, or maybe something will click for me. But it's a very different situation than, we had before. And, and I can also ask it to unpack.

I can say like, "This particular paragraph, I don't, I don't quite understand the motivation for this."

Thomas Fink: Right.

Steve Hsu: Or what, how does that what? And then, but then it can explain, and it doesn't get impatient with me. You know, Ed Witten would get impatient with me, right? So or somebody. You know, they would be like, "What? This is obvious, Steve." They would say, "This is obvious. Why don't you get this?" But the model won't do that.

Thomas Fink: Right.

Steve Hsu: And so

Thomas Fink: the lack of shame is great.

Steve Hsu: Yeah. They don't shame you. They're not judging me, right?Maybe eventually it's calibrating like, "Oh, Steve, this is the Steve's level of capability. Let me try to make him better." It was it'll save us both some time.

Thomas Fink: Right.

Steve Hsu: The other thing you mentioned, which I think is really important, this engineerable, engineerable insight is I think what you said.

Thomas Fink: Mm-hmm.

Steve Hsu: I think what the models can use as an engineerable insight is different from us. So I can often

Thomas Fink: Right

Steve Hsu: paste, cut and paste huge slabs of LaTeX From one model into another model and, and their LaTeX for them is a very good, it's a, you know, it's a very good way for them to process the information. And often the second model will really be able to just take that in their context. You know, they have a million token context window now.

So they can take that in and it can drive them to suddenly, okay, yeah, that, that makes sense. Let me now proceed in this other way. It's, it's just incredible watching them do this kind of thing. So I don't know.

Which, which suggests maybe we have two different, you know, models for, you know, engineerable insights. I don't know, you know, how do you get those aligned? Yeah.

Thomas Fink: I sometimes think, you know, if, if we were to meet aliens, my thought is we probably wouldn't agree on the fundamental like, say, oh, that mathematics is wrong and, and that physics is right, but that physics is wrong.

I think we'd agree on the facts, but we'd have quite different levels of how far we we've advanced different fields. Like maybe we've advanced, you know, algebra and group theory far, much further, and they've advanced, you know, something else much further. And I, and I feel that there, you know, that, that's kind of, it's hard to know.

I don't think there's any optimal way to figure out how do you advance these things. If left to its own devices and say, you know I think you talk about this what's this recursive improvement you know, we we're on the foothills of that. if left to its own devices, do you think AI could kind of build mathematics upon mathematics in a meaningful direction?

Steve Hsu: I don't think it's at that point yet, but it

Thomas Fink: Okay

Steve Hsu: could be at that in my lifetime, I would say. but humans working with math with AIs, I think can push math ahead much faster than what we're used to. By the way, I So I think you said something I really agree with, which is that, so these AIs are an alien form of intelligence, and what constitutes engineerable insight for us may be different from them.

So that's, that's a key, I think, concept here. One of the things that we joke about, this is more of a kind of Silicon Valley AI bubble kind of thing that we discuss. It's kind of a little bit in the RSI context as well, but we often say things like, "Look, these apes, me included, my working memory is like 10 objects, 10 concepts." Like how

Thomas Fink: Right.

Steve Hsu: like at a dinner party, how can I even keep track of like who's at the dinner party and what they're saying to each other? Like it's like 10, right? Like, you know. or you know, you have some things interacting in some statistical system, like what, how many things can my working memory really keep track of? It's like of order 10. It's definitely not 100.

Thomas Fink: Right.

Steve Hsu: It's more than-

Thomas Fink: Uh-huh.

Steve Hsu: Their working memory is like a million. It's a million tokens that they can

Thomas Fink: Right, right.

Steve Hsu: They can reason effectively within that. So it's just a different kind of intelligence. And I, I don't know. to me, I feel like I'm living in a dream because as a kid, I never thought I would live long enough to see such powerful AI tools. You know, I grew up on "Star Trek" and, you know, "Star Wars" and

Thomas Fink: Right.

Steve Hsu: In the sci-fi era, we had great rocket ships and laser guns and stuff, but we didn't have great AI. And so I'm just to me, every morning I wake up and it's like, wow, I can't believe I'm getting to live through this. It's great.

Thomas Fink: Yeah. It's a, it's a bit like "Star Trek," particularly I remember the I think "Star Trek: The Next Generation," you know, computer. It's like, "Tell me, you know, you know, give me a derivation of this thing," and it seemed to work. So, so "Star Trek" is uncanny. You know, all, all we need now is to be able to you know, to, to beam people.

Steve Hsu: Yeah, transporter.

Thomas Fink: The last missing thing. The transporter. Yes.

Steve Hsu: Yeah. Great. Well, hey, Thomas, it's been fantastic having you on the show, and maybe we can follow up and do another episode at some point. I congratulate you for the thing you've built. It's really an incredible accomplishment.

Thomas Fink: Well, thank you for that.

You definitely deserve the praise. And next time in London, I would love to just drop by and meet some of your colleagues.

Steve Hsu: Oh, that would be amazing. You've got some big fans over here in, in particular Manifold.

Thomas Fink: That would be amazing. really enjoyed talking to you. You know, it was 10 times as much fun as as I could have imagined. So really nice. Thanks a lot for that, Steve.