Ep 88: Unpacking DeepMind's Quest for SuperIntelligence with Demis Hassabis' Biographer

Sebastian Malaby: Demis has a Nobel Prize. Sam, um, didn't finish his first degree. Therefore, Demis doesn't take Sam very seriously.

Host: Do you think he feels like or will have to become more of a public figure than he is today?

Sebastian Malaby: Demis, in some ways, is too sensible.

Host: What was the moment? I guess your view of him kind of shifted the most.

Sebastian Malaby: He would sometimes erupt in these conversations I was having, and he would start banging the table and saying, maybe if we approach science the right way, we understand more about nature, we will be getting closer to something that we could perhaps call God.

Host: Sebastian Malaby spent over 30 hours with DeepMind CEO and co founder Demis Hassabis in preparation to write his book, the Infinity Machine. I got to sit down with Sebastian to get all his reflections on this experience. Uh, it was fascinating to talk about some of the stories that Sebastian reported on in this book, including the fact that Reid Hoffman offered a billion dollars to demis to spin DeepMind out of Google at some point. We talked about Demis's relationship with Elon and how that's evolved over the years, as well as what he thinks of Sam, Dario, and the general relationship between all the AI leaders in the space today. And we hit on Sebastian's reflections on demos. What motivates him, what his blind spots are. Just a fascinating conversation with someone who spent unprecedented time with someone who's really, uh, at the cutting edge of AI today. I think folks will really enjoy Sebastian's candid takes here. Without further ado, here he is. Well, Sebastian, thanks so much for coming on the podcast. Really appreciate it.

Sebastian Malaby: Thank you. It's great to be with you.

Host: I feel like the Infinity Machine, uh, and your kind of story of DeepMind and Demis Hasabis is, uh, has captivated a bunch of folks in the AI world. I know that I inhaled this book. I think I finished it in like 24 hours. It's just a gripping story and you got this just incredible access to Demis. I think, you know, uh, the sitting in this pub for multi hours at a time talking, uh, about all sorts of things, just, uh, a fascinating, you know, opportunity, uh, and window into definitely one of the most interesting people and companies, uh, in this AI race today, and so appreciate, uh, you writing the book and appreciate you coming on the podcast to talk about it. So our goal today is to use your reporting to understand the people, decisions, and dynamics that shape the AI race that we're living through. And so maybe to start, I feel like one of the interesting themes in your book is whether this race that we have today between the labs was inevitable or could have gone differently if different players or decisions had been done. Um, I guess I'm curious, after doing this work, do you think there was another path or was this kind of inevitable?

Sebastian Malaby: I think it was inevitable. I think when you have this sort of supremely strong technology, there's going to be multiple labs in multiple countries that are just desperate to try and build it. And we know the China stack is pretty strong. And so despite the lack of semiconductors, um, they were going to have a go at it, and they are actually doing pretty well. And then clearly in the U.S. uh, and then in a few other countries, you've got Mistral in France, Cohere in Canada. There was bound to be many players. Just as the technology is too sweet for it to be only interesting to one team. What's strange about this whole debate, though, is that's not how people saw it. Ex ante that, you know, when Demis was starting DeepMind, he really hoped that he could avoid the race dynamic. It seemed naive in retrospect, but that's what he hoped.

Host: One really compelling anecdote you had was, I believe Demis in his interviews would say, uh, you know, he was interviewing candidate, and he'd say, look, at some point we may be really close to AGI and we'll go, uh, fly to a bunker and figure this all out as kind of like one team. Would you be willing to, you know, get on that flight and do that? You know, I think certainly part of the founding origin of Anthropic is also this belief like, hey, we need to be the team that, like, gets, you know, as we get to the precipice of AGI, we're the ones figuring this out. You know, clearly a key part of starting all these companies. Do you think these folks, I mean, particularly Demis, like, do they still believe this, or how is their kind of thinking evolved? Uh, having seen the way things have played out these past years?

Sebastian Malaby: No, I think Demis has swung from one extreme to the other. You know, he began by thinking there could be a singleton scenario, just one lab. And by that he really meant deep mind in himself. Um, uh, to the opposite extreme, where he now sees that there's a very crowded field and therefore that it's almost pointless for one lab to pursue safety by itself. Because if one lab is safe and then the other ones aren't, it doesn't make the world safer. So he really has shifted to seeing this as a collective action problem that only a, uh, government can solve.

Host: Because one thing I thought was really compelling for your book is obviously there was this early AI Safety summit that Demis and the team put together and they're sharing, you know, the progress they've made and why they're worried about it. And you know, some people in the room like Reid Hoffman, Elon Musk take that information and then they're like, oh, this technology really is starting to work. We should ourselves potentially, uh, do something around that. You know, do you think, uh, obviously to, to get Safety to work going forward, you need these different companies to agree to share things and collaborate. Uh, I imagine that left a bad taste in Demis's mouth. How do you think he and other kind of key players in this space think about, you know, potentially sharing what they're doing, uh, given that, you know, past experience?

Sebastian Malaby: Yeah, I mean you're totally right. That was in 2015, summer, um, of 2015, they have this summit at SpaceX. Elon Musk is hosting it. And the idea is he's going to be brought into the tent by DeepMind, he's going to be part of their efforts. He's going to be, you know, chairing this sort of safety oversight board. And therefore he wouldn't set up a competitor. And of course at the end of 2015 he did set up a competitor OpenAI. Uh, and so that kind of really drove home the reality that we're going to have a race. And I think if you ask now, okay, so what about future collaboration? I think from Tembis point of view or from any of the lab leaders, frankly, um, you can't trust the other guys. Uh, and therefore the only way you get trust is if you have a government enforcer that comes along and say, look, here's the rules for everybody. There's going to be a level playing field. You're all going to have to abide by some sort of safety, slow down pre testing of models before you release them, all that stuff. And you all have to do it. Um, and then the reaction of course will be yeah, but what about the guys in China? And that's why ultimately I think this has to be a US China collaboration. You know, remote though that prospect may seem to many listeners right now.

Host: I mean, do you think Demis and others. I think that's actually realistic. I mean obviously there's, it, you know, is such a pressing problem, but you know, it doesn't seem like, uh, folks have the most confidence in governments and certainly, you know, intergovernment collaboration, uh, across the world, like it intellectually makes sense that that is kind of the clear way to solve this problem. But do you think they think there's actually a probability that this does happen successfully?

Sebastian Malaby: Well, you know, I think they ought to believe that it could happen. And we do have the Food and Drug Administration and people compet complain that it's slow and all that, but it does have expert reviewers who look at clinical trials and say, is this drug safe or not and can it be released? And if you can do that for pharmaceuticals and you should do it, AI models that can be more destructive on a grander scale than a drug certainly need to be reviewed ex ante. And I think in Britain actually you have an example of a government AI safety institute that's pretty effective, has some high class scientists. I mean the chief scientist there is Jeffrey Irving, who was at DeepMind originally, did a lot of the post training on the early models. He's a really serious researcher, was also at OpenAI, by the way, in a close collaborative. Darius. Um, so you've got serious people there and they have found vulnerabilities in systems in the past and quietly shared them with the private labs that have failed to find the same vulnerabilities. So it shows that you can aggregate some technical expertise inside these public institutions because that number of folk feel motivated and public spirited enough to go work for the government. Um, so I don't think we should give up on that.

Host: From the outside, it's hard to tell. You know, it's almost like such a dispiriting story of, you know, Demis kind of being idealistic and then realizing this dynamic isn't going to play out. And so it's hard to tell, you know, I guess reading between the lines in your book, whether he's kind of just resigned to this dynamic or is found newfound hope in like the actual possibility of, you know, intergovernment collaboration. Because obviously even the examples you give just within a single country. Right. And certainly not at the scale of, of, of coordination across countries.

Sebastian Malaby: Yeah, I mean, what, what he says, I mean the very last line of my book is I'm, um, optimistic still. And that's a quote from Demis trying to, you know, persuade me. I've been, I'm pushing him in, in, in my interviews with him. Are you really optimistic? Because it seems like we've got a race dynamic, you know, in, of course, January 2025, you have deep seat come out and that really signals the advent of the Chinese labs into this space. And you know, I'm writing for through 2025 into the end of 2025. And I'm saying, are, ah, you still optimistic, Denis? And he says, yes, I'm optimistic still. And I think the reason for that insistence, apart from like, what's he going to say, is also that, you know, when something kind of dramatic happens, the world does coalesce and take action. You know, if you look at Covid, the reaction to Covid, I mean, governments on a national basis really did stuff that people didn't expect beforehand. Um, you know, so in a crisis, people do react. When you have something like, you know, trade, where protectionism has a pretty slow and subtle effect on economic performance, um, you don't really get a political reaction to try and fix that. And that's what we've seen in the last few years. But if you have an acute thing like Mythos, uh, an anthropics model comes out, you know, and uh, people suddenly get worried. You see, the US government pretty much do a 180 from a laissez faire position to, oh, well, we better control this kind of position. And I think that shows the shock effect is extremely important in determining whether you get government action or not.

Host: Switching gears, obviously, a key part of, you know, in Throughline through your whole, uh, work is just the story of DeepMind and its ultimate relationship with Google. And it's just an incredible, uh, story. And so I guess as I recall, you started this or you kind of got demos to agree to this, like right before ChatGPT. And so obviously post, uh, AlphaFold and some really interesting things, but certainly the world, uh, evolved, uh, in a bunch of interesting ways from the start of your project. People talk about Demis and DeepMind all the time, you know, before you published this book, but you'd kind of done the research, what do you think the popular narrative got, like most wrong about Demis, DeepMind, the way people would talk about, you know, the two.

Sebastian Malaby: Well, the extraordinary thing to me is how people basically discounted Demis. I mean, you know, I've been interviewed a bunch of times since my book came out a few weeks ago. And you know, often people can't pronounce his name. They say Demis, they say Hassabis instead of Hassabis. Uh, they kind of barely know who he is. You know, my publisher, uh, Penguin Press, was, was figuring out what to do the, how to do the COVID design, and they wanted to use this picture, but at the same time they didn't think people would recognize him. So you couldn't sell books based on some, you know, person.

Host: Nobody Recognizes at this point, I assume Sam, Dario, they'd be fine for uh, a cover of a book.

Sebastian Malaby: Yeah, yeah, totally.

Host: Elon.

Sebastian Malaby: Yeah, Elon for sure. And um, Demis, you know, is the guy who founded the OG lab in 2010 way before anybody else really created the model that then OpenAI copied later. I remember going to see Dario when I was doing my research and you know, he said yes, this was the original figure and he's got the whole AI for science space to himself, which was true at the time. I think it's less true now. But uh, so I think people just underestimated how important um, he, he is. Um, they underestimated Google, uh, as a company because they thought innovators dilemma, they were too slow. OpenAI went first with a model with uh, a, with a chatbot, um, and now they've got this massive brand effect, uh, OpenAI does and so nobody can catch up. Well that proved wrong. And you know, as of late 2025, uh, the Google DeepMind models, Gemini 3.0 were better on the leaderboards than the adversaries. Of course since then we've seen a big anthropic surge. But you know, the point is I think people were too quick to crown um, OpenAI and Sam Altman as the winner and underestimated, ah, both Demis as a person and Google DeepMind as a company.

Host: It is interesting obviously like since you're reporting, you know, I guess in 2026 you've had these rise of these, you know, coding agents and uh, you know, certainly, you know, Anthropic and Claude code were the first there and OpenAI with Codex was right behind. It does feel like Google continues to do like amazing science and you know, they do, you know, they always do well on the leaderboards, but for whatever reason it struggles in like the zeitgeist and like actual usage on the product side, if these two monumental moments were, you know, ChatGPT and consumer products and then cloud code and the coding agents, Google doesn't really have a super competitive product in either of those spaces just from a usage perspective. Not the products themselves are good, the models are good, but they haven't really figured out the latter part. And I'm wondering kind of what you make of that.

Sebastian Malaby: People use Gemini more than we realize. Right. It's sort of bundled into Google search at this point, the AI mode thing. But I take the broader point which is that both of these sort of seminal moments in um, proving out, you know, the consumer experience, which were chatgpt, as you say, on the consumer side, and then more recently with coding, um, neither of those came from Google, DeepMind. And I think what that perhaps shows us is that, you know, partly because of Demis's personality and his intellectual formation, which was PhD in neuroscience, this very broad study of what intelligence might be, a very broad approach, therefore, to building artificial intelligence, there's this. There's this kind of let's try everything approach to AI research. Whenever there's like two different paths you could go down, they say, well, we'll do both, and if we can find the third path, we'll probably do that too. They're very hedged. Right. Whereas I think Anthropic, um, got to coding because it was willing to take a more concentrated bet. Um, it never went into the whole field of generative video, uh, just never had a SORA equivalent. Um, and at the same time, OpenAI, being a startup, didn't have the reputational baggage that it had to protect that Google has, and so was willing to put out a chatbot that hallucinated a lot at the beginning.

Host: It honestly has a lot of echoes of the Transformer itself. Right. And Basically, I think OpenAI making the bet of going all in on, uh, scaling Transformers, which were obviously invented at Google, and Google pursuing a bunch of different paths. And then to your point, OpenAI, uh, started to do a bunch of things and anthropic deciding, hey, we're just focusing on coding. Have you seen kind of a shift in, uh, Demis and DeepMind's focus? I guess, given what's played out and given this kind of, uh, it's such a tension, right? As a polymath, as someone who's so interested in many facets of AI, to focus on a ton of different things versus the thing that's happening at the moment, obviously OpenAI, I think, famously has kind of consolidated down and said we were doing too much. Now we've got to really hone in, uh, and catch up to Anthropic on coding models. You know, how do you think Demis and, and the DeepMind folks think about this at this moment?

Sebastian Malaby: I think, you know, they do have this tendency to try and do everything. Um, as I was saying, I don't think that's been cured. I mean, you know, I was chatting actually just last week with, uh, one of the young scientists from that. From that shop. And, um, you know, that was very much the tenor of what he was telling me. It's like, you know, we always make multiple bets. We never really go hard down one avenue and. And you Know from a corporate point of view maybe this makes sense. I mean if we think about um, what happened with the consumer side, the sort of general chatbot model. Yeah, they were late. They were late both in trying to develop it after the Transformer thing dropped in 2017 and then they were late again in productizing behind ChatGPT. But then they caught up by 2025. So maybe if you have very deep pockets like Google and a very deep pocket technical bench and tons of human talent and amazing amounts of compute, you know you can afford. They're not doing Apple, right? Apple is the absolute extreme where they just say whatever, we're not going to do this AI thing, we'll just charge people to put it on our iPhone. Uh, that's an extremely hands off position towards AI. Google is much more in it. Um, but it doesn't mind being a couple of years behind because it figures it can catch up.

Host: And I wonder at some point whether that does shift. Obviously you've allowed $2 trillion plus we'll see what they're end up being worth. Companies or at least in the, in the prediction markets seem to be worth that. Uh, you know, come, come up uh, and gain a lot of territory and you know, I think it'll be really interesting to see going forward. Um, you know, whether that, that focus or that like diversity of different things they're pursuing, um, you know, ends up, uh, ends up continuing. You know you talk a lot about, in the book about you know, the governance of DeepMind and Google and like the way these two organizations work together. Um, and I think actually one of the most interesting parts I thought was this DeepMind was considering spinning out of Google. There was like some financing behind that. They ultimately chose not to. Why don't you think they ended up you know, deciding to, to spin out at Google and I guess in retrospect, was that the right decision?

Sebastian Malaby: Yeah. So there was this secret plan called Project Mario to spin out and um, Demis was not so happy when I kind of discovered about it from other sources. I had to speak to the general counsel of uh, Google DeepMind at one point who tried to persuade me I couldn't write about it. Other people delete me the documents and I had them and you know, definitely had good information. It was all true and I wrote it. So whatever. Um, and the point of this story is actually partly it goes back to your safety thing that um, Demis really, really wanted to get safety oversight over the Google DeepMind models or just the DeepMind as it was then. Um, and Google Corporate in the Mountain View wasn't doing that. And so he had to have a credible threat of spinning out. So he went to Reid Hoffman. Reid Hoffman pledged a billion dollars to finance a spin out. And Demis used that to kind of pressure Google, um, although, uh, I don't think he ever mentioned that Reid Hoffman had done that. But he pushed Google knowing that he had this back pocket option. And I spoke to a lot of the advisors who were in the room, um, when DeepMind was figuring out whether to go with a Reid Hoffman spin out option. And there were those who said, look, you should spin out because then you'll be an independent startup and you'll have all the incentives and the kind of, you know, alertness and the flexibility, the agility that go with that. You can, you know, give all your team, you know, very high powered financial incentives, um, linked to your performance and it'll be awesome and you should spin out. And then Demis's view in the end was, you know, it's, it's legally, it's going to be, there may be a court battle over it, you know, whether we have the right to do that. Um, and I just want to do science. I want to not be distracted by legal fights. I want access to tons of compute. I'm staying in. And so that was the call he made. Was it right, was it wrong? Well, I mean it did lead to him getting a Nobel prize, right, in 2020. You know, it's a small matter of a Nobel prize, uh, in 2020. So a year after he gave up that fight with his parent company, he ships, uh, Alphafold, uh, the protein folding prediction model and that gets him the Nobel Prize.

Host: The kind of way you just framed it was more like, hey, this was part of negotiating leverage to get some of the safety stuff. I mean, how seriously do you think the DeepMind team was considering this at the time?

Sebastian Malaby: Well, I think they always, they never quite got to that decision point in their own heads. They wanted the plan B spin, uh out option because why not have that? And they weren't sure how to then use it in the negotiation with Google. And I think they determined never to tell Google explicitly that they had that, but a kind of hint, uh, that you know, there might be something in the background. And if you don't do what we want with safety oversight, we may do something you don't expect kind of thing. But you know, ultimately they never really waived that threat explicitly and they never exercise a threat even though Google didn't give them the safety oversight that they wanted. So in the end, I mean, in some sense, it's a story of the naivete of the. Of the founders, both Demis himself and Mustafa Suleiman, his co founder, who's now at Microsoft. I mean, they went into this, um, you know, not quite sure what their end game was. And in the end, they never really used that spin out. I mean, they not. In the end, they didn't use the spin out option. And so in some sense, spending three years going back and forth to Mountain View with this was a waste of time.

Host: I'm wondering, in the course of the conversations that you, you know, that you had with Themis, I mean, obviously we talked about that, the AI Safety summit already and kind of realizing, yeah, that was not ideal. Are there any other things that, like, he regrets, I guess, about how the whole, you know, the whole past decade has unfolded?

Sebastian Malaby: You know, uh, what does he regret? I'd say that he would certainly not regret, um, spending all that energy on AI for science.

Host: I love that anecdote in your book, by the way, that, like, the day they won AlphaGo, AlphaGo won the, you know, beat the best player in the world. He was already on to Bio and someone picked it up on a mic, which, uh, I, Which I just thought was talking, uh, about an ambitious and, uh, always moving, uh, leader.

Sebastian Malaby: Right, right. He rests on his laurels for about 10 seconds and then he says, the next thing we're going to do is we're going to solve protein folding. Yeah, that was amazing. But I think, you know, the serious point here is that he not only got a Nobel Prize out of it, he also views this, I think correctly, as absolutely central to the whole acceptability of artificial intelligence in society writ large. If AI doesn't deliver, you know, clear benefits for humans, uh, and it's just like, lots of job disruption, which may be good for productivity, but it's kind of painful for people on the receiving end. I'm not sure that politically, AI will really be rolled out with some massive. Without some huge backlash. So I think both from. It's good to advance science. I like winning a Nobel Prize, but also, AI can't succeed without this kind of science stuff. I think he doesn't regret that for a second. Um, does he regret not being faster onto chatbots? Yeah, of course, it would have been better to understand, like ilya Satskaeva did. OpenAI the moment the transformer dropped, Ilya is like, jumping out of his chair, running down the corridor, going to find Alec Radford saying, hey, we're going to build a language model based on this transformer architecture. I mean, on the day the paper dropped, because he had this prepared mind. Ever since his PhD, he had been thinking about how to deal with sequential data like text. And so when the transformer paper came out, he immediately saw the significance. And of course it would have been great for, uh, DeepMind if they had had that same perception that quickly. And it took them instead, you know, two, three years to get there. And Demis, you know, frankly, still has a bit of a, ah, blind spot about this. You know, when I say to him, well, you were sort of three years late, he goes, no, no, no, no, we had chinchilla, we had these other models. And I'm saying, yeah, yeah, but you didn't release those till the end of 2020. And Ilya by then had been working on it for three and a half years, so you were late. Uh, but he really resists that, and I think that is because he does regret that.

Host: Well, I thought it was interesting too. You know, the way you kind of put it in your book is that, you know, part of the, uh, resistance, maybe the transformer architecture was this. You know, he obviously comes from such a deep neuroscience background and this belief that, like, the path to artificial general intelligence has to flow through something that looks much more, you know, like the way humans learn and the human brain, which obviously was the rl world, uh, and less of just, you know, how can you really understand what it is to be a human by consuming the Internet. And I think your point was actually, it turns out, uh, you can understand it way more than one might have anticipated from scaling these, uh, these models.

Sebastian Malaby: Yeah, and by the way, I think there may be some vers of that same dilemma with robotics. In order to train robotics and simulation, you need these real world simulations. And probably video is one important tool for building those simulations. So potentially, if DeepMind gets the right bet on the right pathway to much better robotics, it has what it takes. Um, but we'll see, uh, whether, you know, there's some version of what, what went wrong with language. Uh, and the transformer model is that GEM has just underestimated the importance of language to super intelligence. And, um, then maybe, you know, to get robotics right. You know, is it, for example, something that you're going to do by building physical robots and having them crash and break their arms, um, you know, in some kind of physical lab?

Host: Uh, but it's an interesting, you know, point because actually, you know, it's many parallels in the robotics space where there's you know, 10, 15 different approaches that people are trying and I imagine if you're Google and like given the way DeepMind's approached other things, you'll kind of put your hand in a bunch of these approaches and meanwhile you'll have startups that were like we're all in on you know, X approach or Y approach. And it'll be interesting to see how that plays out because you can certainly empathize today when the recipe is not clear. Uh, it's, it's not particularly irrational for, for DeepMind to keep their options open, uh, but also leaves you vulnerable to someone that, that really goes all in on the transformer or on code or whatever the next, you know, equivalent is in the robotic space.

Sebastian Malaby: 100%. And I, and I, I think here actually there's a larger fascinating question about whether venture backed innovation beats sort of hyperscaler, you know, tech behemoth, um, you know, AI approaches.

Host: Because I have a biased opinion on that one.

Sebastian Malaby: Ah, yeah, yeah, well that, that's fair but, but I think, but, but you know, I'm less biased but you know I've written about a hyperscaler now but my previous book was about venture. I kind of done both. But I think it's kind of an interestingly balanced um, thing because you know, this is a very capital intensive project. Um, you need very deep pockets. If you look at OpenAI's ability to fight Google over the long haul on generative AI. You know, I wrote in the New York Times in January that I thought OpenAI had a 50% chance of going bust by next year.

Host: Is it still 50 by the way?

Sebastian Malaby: Yeah, I mean, you know, I said fail. I mean, I, I, what I mean is it sells itself as a disc at a discount to, to some hyperscaler. And I think, you know, I've been impressed by some of the um, you know, expenditure cuts they've done. You know, canceling SORA was a smart move. They've done some impressive cost cutting. Um, so I give them credit for that. On the other hand, of course, um, the sort of rumors around Sam being replaced by Brett Taylor are pretty widespread and there's just a general sense of you know, the leadership is damaged goods. Yeah, I think on net I'm still around 50, 50 that they fail by kind of next summer. Um, and that's not because they don't have great tech. Because they do. I mean the tech is great. It's just the business model, it's the problematic. And it's problematic because you're up against Google, which just has unlimited amounts of cash to spend you into the ground. And so in one sense, um, this looks like a platform shift that benefits the incumbents. On the other hand, as we've been discussing, when you have the luxury of unlimited amounts of talent and compute and money, you tend not to make these strategic concentrated bets. And if you're up against a bunch of different venture backed startups that each of them makes different, um, highly concentrated bets, then maybe one of those bets work out.

Host: But it is a game of very thin margins. Even if you take Anthropic, I think there was always a question about whether they would be able to raise enough capital to stay in this race or basically have to sell or be deeply embedded with a hyperscaler, which they are to some extent. And it felt like it might be headed on that track with either Google or Amazon. And then right at the moment where you might have wondered, hey, can they keep raising the amount of capital that's required to stay in this game? The coding model sit. And they hit in a huge way. And now Anthropic is the hottest company in the bay. But it's a really interesting counterfactual. If that stuff had hit six months later or 12 months later, it certainly wasn't inevitable, uh, that it was really going to start working then. Um, yeah, a lot of this stuff, I mean, it seems you can write a very clean narrative in retrospect, but it's not an uh, inevitability, uh, that things play out that way. You know, I want to go back to something you said on Demis, which was, you know, him, uh, you know, obviously focusing on science and this kind of belief that, hey, we have to show that, you know, AI can have beneficial, uh, you know, outcomes for society or else the world will turn against it. And one thing I'm struck by in particular, you kind of alluded to it right at the beginning that you couldn't put demos on the COVID because people wouldn't recognize, you know, who he was. And in many ways, like, the people that the world associates with AI today are Sam and Dario. And they're kind of, you know, uh, taking a lot of the oxygen in the public conversation about, uh, how people think about AI. And I'm wondering, one, how Demis feels about that and two, over time, do you think he feels like or will have to become more of a public figure than he is today to shape that conversation? Obviously he's deeply passionate about this stuff, uh, in a way that's more, uh, analogous with how he Feels, Yeah, I

Sebastian Malaby: mean first of all, to be clear, I mean the compromise that my publisher, uh, struck was to have him on the COVID but to kind of fuzz it up so that he looks like a kind of enigmatic geek and if you don't recognize him, it's still a cool picture. So that's what they did. But yeah, I think to your question, you know, Demis does understand that um, he needs to be out there more. He's good at kind of self narrativizing in one way, which is that, you know, he's the kind of retrospective story of his journey up until where he is now is something that he's been good at communicating. He spends more than 30 hours talking to me. And of course there is a reason for that. Uh, but also there was the documentary the Thinking Game, which a lot of people have seen and before that there was a documentary about AlphaGo which didn't get made by mistake that they actually brought a documentary team with them to Seoul when they played that game against the sole Go champion. So, so there is, there is this retrospective storytelling which he's good at. What he doesn't do is prospective. He doesn't like Sam, you know, has, or at least in his heyday had the ability to um, kill the attention on a new um, DeepMind release simply by putting out a couple of tweets.

Host: They would always go one day before, like they know something was going to happen.

Sebastian Malaby: Yeah, and when we say go one day before, we mean, you know, just trail something on X, just put out a couple of posts and not even release something. But because Sam's following on X is like five, six times, Demis's uh, the echo chamber is so much stronger that he could really dominate the narrative even when the other guys were releasing the product. And that's a problem both in terms of product adoption for DeepMind, but also like talent recruitment, um, controlling the narrative does matter and I think they kind of get that uh, over at Google DeepMind, um, whether they quite have the ability to fight it, we'll see. I mean what, you know, what they don't do is, is do what ah, Darius does, which is pick a, you know, public fight with the Pentagon, um, you know, unreleased mythos, uh, all these things which just, you know, immediately turn you into a massive celebrity. And Demis in some ways is too sensible to pick a public fight with the government. And so maybe that makes him less notorious and uh, you know, maybe it's, it's a kind of byproduct of strategic caution. That he's a bit less uh, front of mind in the media.

Host: I'm curious like what you ended up learning about the extent to which that does or doesn't attract researchers. Right. Obviously a huge part of the battle here is the battle for talent. You've had, I mean Google has incredible researchers. You've had folks like, you know, uh, like Jack Ray, who you talk about in the, in the book, who left, then came back, then left again. Um, you know, how do you think about or in talking with folks, the types of people that are attracted to DeepMind, the extent to which they, you know, do or don't need to change their hiring brand to kind of stay uh, on par with the others.

Sebastian Malaby: Yeah, it's a great question. I think it sort of um, overlaps with what we were saying about the rivalry between the, you know, deep pocketed hyperscaler and the, and the venture model. Um, because I think, you know, for Jack Ray, when He left UM, DeepMind and went to OpenAI, he explained to me that, you know, this was because there was a concentrated bet on language models and OpenAI and that's all they cared about at the time. And that's what Jack was doing. And so he wanted to be there. And I think, you know, that would be true today in robotics. There'll be certain researchers who are really great and they want to go somewhere where there's one concentrated bet which they believe in themselves and then they will just work all out to be part of that team and to realize what they think is the right path. Whereas if you put a robotics researcher, uh, in this big, careful, diversified lab and say we really believe in you, but we also believe in these other five bets we're making at the same time. You don't feel so good about it. You know, there's a kind of passion piece that's, that's missing.

Host: Well, what do you make of the decision? You know, obviously the Alphafold work eventually got folded out or spun out into, you know, Isomorphic. Right. In a different company that kind of, I mean Alphabet still owns a bunch, but is a separate effort, like does that make sense or is that a path forward you think will be repeated?

Sebastian Malaby: That's a great question. I mean, I think um, Isomorphic does have that effect. I think the spin out was also about trying to be able to do partnerships with big pharma partners and maybe that was a bit easier in a spun out format.

Host: Just because they wouldn't want to work with Google directly or something.

Sebastian Malaby: Yeah. And maybe also just um, the sense was that this by itself as a freestanding thing could become so big. Uh, and there'd been this history of, um, Google and DeepMind separately doing AI for health. And health has got its own politics, its own kind of, you know, very long lead times. You know, it, they, they felt it just needed to be in a different home. Um, so they did that. But I, I think you're raising a good point that there might have been an extra reason to do it, which is that you, you give people the option of working somewhere where they are the bet, uh, their thing is the thing. And I think that's a super important recruiting tool. I also, you would note, I mean, it's probably, you know, fairly well known in the Valley, I guess, but, you know, Anthropic's churn is very low relative to everybody else. And I think that shows you that if you, if your kind of identity as a leader is like Dario, where you're kind of out there and unfiltered about your extreme concern for safety, uh, and your extreme concern for responsibility and social impact, you write these long essays, you know, it, it probably means that some people would never join you because they think you're a bit wacky. But the ones who do join absolutely love you and believe in you. And there's this kind of amazing loyalty which is kind of special.

Host: I mean, another kind of key character in, in your work that we haven't talked about is, you know, David Silver, who obviously played a key role in, in a bunch of the reinforcement learning work and was one of the early, uh, collaborators with themis on a bunch of things. You know, he obviously, I think, you know, since publishing your book, he recently left DeepMind, right. To start another company. I'm curious what you made of that.

Sebastian Malaby: I think it kind of fits your narrative here, actually. I mean, in the sense that he was this, he is this intensely determined believer in reinforcement learning to a point where I think most of his colleagues at DeepMind ended up thinking it was just too much. Um, you know, he was the hero when, uh, first of all, the Atari, you know, game playing System, uh, that DeepMind rolled out 2012, 2013, you know, made these incredible breakthroughs. And, you know, that was almost like imagenet but for early, uh, agents. And David Silver was a key person in that. And then there was AlphaGo, and then there was AlphaZero, which I was, you know, pretty much all learning from reinforcement learning. And no, uh, you know what, there was deep learning, but it was, but it was, it was heavily skewed towards the reinforcement learning side. And, and then came this paper from David Silver and rich Sutton, his PhD supervisor. Um, called, what was it called? Um, um, Experiences Enough. I think it was something like that. And, and basically learn from experience, learn from reinforcement learning, don't learn from data. Was, was the, was the message of that paper. And David is, is just very, very hard over on that vision that, you know, learning from data is inferior because the data includes mistakes. Right? The kind of, if you train, if you take the go analogy, if you train on a human go players, past games, even if it's expert games, those players don't have perfect understanding of go. And you need to get beyond that to have super intelligence. And so you need to learn, the machine needs to learn from its own experience, not rely on the kind of crystallized knowledge of humans passed on through text or other kinds of data. Um, and he is such a believer in that that it's all agents for him, only agents. And they have to learn from themselves. And I think, you know, Demis told me once, or actually more than once, that, you know, that kind of approach may ultimately win in some future when you already have AGI and now you're just sort of like perfecting it into even greater superintelligence. Um, because ultimately, yes, you know, if you, if the machine learns from its own data that is purer and uh, ultimately will get you further. But to try and get you to that AGI, you need to bootstrap yourself with existing data. And that's what the whole language model revolution showed us. That for a long time there was much, there was almost no reinforcement learning, unless you count reinforcement learning from humans.

Host: It didn't work if their base models weren't strong enough, except for these specific domains. And then what's fascinating obviously about this moment in time is the combination of large pre training LLMs and then reinforcement learning on these verifiable domains. But it's just interesting that he left at this time because obviously it feels like generally reinforcement learning is back in vogue as the way to improve, um, these models. And so it almost would have made more sense a few years ago, uh, when reinforcement learning was more out of favor. Uh, I thought it was interesting timing that it happened now.

Sebastian Malaby: Yeah, I think you're right. But it sort of reflects the human lag between feeling something and acting on it. Right. I think he'd been feeling for a while, frankly, that, um, he was swamped in a big organization which didn't fundamentally want to put more than a sort of small amount of its chips on the reinforcement learning table, uh, and then even within reinforcement learning, you know, his vision of how it should be done differed from some other people's. And I think he is kind of a classic, uh, startup sort of person who wants to be in a small organization where his vision is the vision, um, because, you know, he's extremely visionary on reinforcement learning. And so it makes sense for him to go do his own thing.

Host: One thing that we've kind of talked about a bunch here is just like, in the end of the day, there's a few people here that run these labs and have intensely like, personal histories and relationships. And I think the, obviously the Sam Dario relationship has been incredibly well documented. Uh, Elon, Sam, all coming out, uh, in this current court case. I think Demis's relationship with each of them is probably less well understood, uh, by the general public. Um, and I'm wondering if you could just talk a little bit about that, uh, and the extent to which those relationships exist, his kind of feelings on the two main, other protagonists of the space right now.

Sebastian Malaby: So, um, Demis relationship with Elon is very interesting. Um, Elon tried to buy him, right? Yeah. Uh, but just going back, I mean, you know, the origin of this whole thing is that they were both funded by the same VC shop Founders fund, right? Where, you know, elon was the SpaceX guy and uh, Demis is the DeepMind guy. And they both, in 2012 or thereabouts, get invited to an LP off site and they both present and then they get talking to each other. And you know, Elon's always very competitive.

Host: That would have been a pretty valuable offsite to pay attention at.

Sebastian Malaby: Like, you know, well, I've got the most important technology in the world because, you know, even if the world is screwed up by your AI, um, we can all move to Mars, be a multi, multiplanetary species. I've got the most important thing. At which point Demis says, yeah, but if you think you're going to be safe on Mars, remember that my AI will be able to conquer space flight and it will just follow you to Mars, so then you won't be safe after all. And then there's a silence. And then Elon goes, hm. And then the next thing he, he says, well, I'd like to invest in your Series B. And he writes a $5 million check into the Series B. And then as you say, he wants DeepMind, um, at the start of 2014 to prevent it from being sold to Google. And, and Demis just like waves them off and there's this crazy story where Luke Noc, the Founders Fund partner, who is sitting on SpaceX's board, you know, sees Elon at a party in LA and they agree that they've just got to stop DeepMind from being sold to Larry Page. He's a transhumanist. You can't trust him. So they go up, um, to some closet upstairs in this party and they Skype Demis in the middle of the night in London and say, you got to sell to us, not to them. And, you know, sell it to SpaceX, sell it to Tesla, do something. But just don't sell to Google because they're evil. And Demis, like, you know, no, no, Google's got the compute. I'm selling to them. Goodbye, good night. And he puts the phone down and then Elon goes nuts, right, and starts calling Demis, uh, an evil genius, which is a reference to a game that Demis worked on, um, when he was a video game designer. Uh, and, you know, vilifies Demis and is obsessed with this. And I think this has come out again a bit more in the trial that we've just been having recently, you know, the, the Elon vs Sam trial and the kind of obsession that Elon had with Demis as the evil genius that had to be counteracted. So that's the history there. Um, Demis is kind of very keen to say that these days they can run fine. And I haven't heard that directly from Elon, but I kind of feel that probably water under the bridge, Elon has moved on to fighting with Sam. Um, it probably is true that he's okay with Demis now. Um, but, um, uh, that's the history there. And then with Sam, it's just such a complete difference in personality and background if you compare the two of them, right? So Demis has a Nobel Prize. Sam didn't finish his first degree. Therefore Demis doesn't take Sam very seriously. Right. He doesn't have a college degree, let alone a PhD or a Nobel Prize. And he just, you know, to Demis, Sam, um, and I think there's some truth in this, frankly, that, you know, Sam is the sort of skillful, ultimate embodiment of the Silicon Valley network, who knows how to fake it until you make it to tell the premature truth, who is just a master at raising money, uh, at leveraging his connections in the Valley. And, you know, that's all great, but, you know, it's not the same as being, um, a serious scientist and you can't trust somebody like that. And I think, you know, look, I think people have come around to Demis's view to quite some, you know, some extent with Sam. Um. Um. But Denis always felt that. And so I don't think he ever liked Sam.

Host: I think it was the economist that, in talking about your book, you know, said, it's really like a test of its, like, great man theory of history. But I'm curious. We've talked a lot about these four protagonists. Um, and I'm wondering, after having done all this work, was all of this inevitable, or to what extent did it matter that it was Demis and maybe some of these other folks at the helm of the companies? Um, yeah, I'd be curious your thoughts on that.

Sebastian Malaby: I once wrote a book about Alan M. Greenspan, uh, super powerful player in global economics, um, but ultimately unable to stop the financial bubble of 2008. And I called the book the man who Knew because he understood that bubbles were dangerous. That's, in fact, what he read his PhD about. He was obsessed with bubbles blowing up, but he couldn't stop this thing from happening. And I feel that one could have almost used the same title for the Demis book. Right. I called it the Infinity Machine, but it could have been called the man who Knew, because Demis has known from the beginning that this thing is dangerous. But as the leader of one lab, even a very powerful, rich lab, even he, with his stature as a Nobel Prize winner, he understands that it's dangerous. But what can he do? Because if he makes his own lab safe, it doesn't stop the other guys from being unsafe. Um, and so I think there is a sort of this inevitability to the race dynamic. And it does matter who is leading these labs. I mean, clearly, you know, Dario, um, changed the narrative on AI Safety first by fighting with the Pentagon in public, and secondly, and more importantly, by the way, that he did the Mythos release. Um, so that's a clear example. On the other hand, you know, uh, Sam, um, decided to release ChatGPT even though it was hallucinating. That was a choice. It didn't have to do that. And that completely colored the way the AI race, uh, played out. And Demis, in a more understated way, um, told Rishi Sunak, the UK Prime Minister at the time, hey, we should have a global AI Safety summit. And, you know, that's what happened. And there was one in Bletchley park, and the Chinese came, and that was kind of the beginnings of International conversation on A.I. ah, safety. Um, so I think Demis is more behind the scenes in what he does. But yeah, I think it matters, the personalities of these leaders, but it perhaps, you know, it's not the only thing that matters. There are underlying forces as well.

Host: Well, before we wrap up, I definitely, you know, given that you spent so much time with Demis, uh, I think there's a few just questions about that process. I was, I was really curious to dig to, you know, you spent, I guess, three years meeting him at the same pub. Um, and I'm wondering across those series of conversations, like, what was the moment? I guess your view of him kind of shifted the most.

Sebastian Malaby: There were lots of surprises, but one of them is the depth of his conviction about discovering the deep mysteries of science. And this is actually a kind of spiritual conviction which I had no inkling of before. But, you know, he would sometimes erupt in these conversations I was having. They went on for two, uh, hours each time. So we already could, you know, get deep on stuff. And he would start banging the table and saying, look, this table, it's a mystery. It's a mystery span. We don't understand it. Like these atoms jumping around and there are gaps between them, and yet the table is solid. And why is your laptop able to think when it's just a bunch of sand and copper and. And you know, what's going on here? Why is the world set up like this so that it functions? It's a m. Mystery. We have to understand there must be some sort of intelligence behind it. This can't just be coincidence that it's like this may. Maybe it's like God. Maybe if we approach science the right way, we understand more about nature, we will be getting closer to something that we could perhaps call God. Now, I had no idea that he would feel that way, but I think he does. And it explains why he nonetheless pushes forward to develop this super strong AI, which he knows to be dangerous. It's because it's a kind of quasi spiritual quest for him.

Host: It feels like for a lot of people in the space, there's like a religious or spiritual element to, uh, getting to AGI, um, and it's really interesting to see play out. Um, and certainly I thought that came through really clearly in your work. Um, I guess it feels like you had such a wide ranging and open book. Were there things that Demis wasn't really willing to talk about or areas that you would have liked to put in the book, uh, but didn't end up coming through?

Sebastian Malaby: Yeah, he was very clear at the start that, um, he would talk A lot about himself, but he wouldn't talk about his family. And so I left that out. Um, and, um, there are passing references. He's, you know, he married his girlfriend from Cambridge. They're still married, they have two children. It's all very normal. Um, um, but I basically don't talk. I mean, most bike m when I wrote about Greenspan, uh, going to speak to his many, many, many ex girlfriends was one way of understanding the guy. Um, and actually brought out the human side of him in a good way. Um, and I, you know, I didn't really go there with Demis. Um, another thing is, you know, he didn't want to talk about fights between himself and Sundar Pichai and the leadership of Google and Mountain View. I did write about that because other people told me about that. And so I get into that quite a lot. Um, so, you know, his preference that this should be left out was not, um, honored. Um, uh, you know, he didn't really want me to talk about the way that he fired his co founder, Mustafa Suleiman M. And he would sort of say, I didn't really fire him. You know, there was a process, some sort of inquiry into bullying done by my general counsel and so forth. And you know, that was kind of the trigger, but it wasn't the deep reason why Mustafa was pushed out. Nothing happened at DeepMind unless Demis wanted it to happen. And I think he just decided it was time for Mustafa to go because they disagreed on too many issues. So, yeah, there were some things that he didn't want included and sometimes he won that preference and other times he didn't.

Host: There's been a ton of public discourse about your book. I feel like you've been on, you know, lots of folks have been talking about it, um, and I think it picked up on a lot of the themes we've discussed here today. Are there parts of the book or aspects of it that you think are like under discussed or you wish people would talk about more?

Sebastian Malaby: I think there are sort of interesting, um, takeaways for scientific innovation, um, how you do deep tech companies. Um, we talked about some of the shortcomings at DeepMind, the lack of concentrated bets and so forth. But on the upside, I think it's very interesting how he developed what he calls scientific taste. Uh, and what I mean here is that when he had his first, um, startup, which was this video game production company called Elixir, he basically blew it up by being too ambitious in terms of the product engineering. Like, how fantastic could the graphics be? Could you do kind of early reinforcement learning to make the characters more interesting. Um, and he drove his technical team over the brink in terms of how ambitious they had to be and they couldn't ship product on time as a result. And so that was a bad outcome. You know, he got some money out of that first company but it wasn't as much of a success as he hoped. But then you fast forward and he's doing DeepMind and he gets this moment in 2018 where you know, he's two years into the alphafold research and his alphafold team has produced the best protein prediction system in the world. And the boss of that team, Andrew Sr. Says, okay boss, you know, we've, okay Demis, we've done this. Now uh, let's declare victory and move on. Because we're not going to predict proteins precisely. Give me a break. That's impossible. Um, we're the best in the world, that's good enough. And Demis is like no, no, that's not the point. We want to predict proteins so that research biologists and medical researchers can use our predictions to build medicines and other fantastic breakthroughs. There's no point just being the best, better than the other labs. We want to solve this problem. And the guy who's running the team, Andrew Senior, says that's impossible. So Demis sits in the meetings of the team and this is where the scientific taste come in. He listens to what he calls, um, you know, he listens for the fluidity and the fluidity of ideas. If the research team are like bouncing possibilities of stuff they could look into off of each other, then that's fluent, um, and you should move forward, put more resources in, push harder. If there was kind of a silence and nobody had any good ideas, of course you should give up. And so he does that test. He listens to the team. They are fluid in their exchange of ideas and so then he basically switches out the project lead. Andrew Sr. Does something else. He puts in this other guy, John Jumper, who then goes on to become the co winner of the Nobel Prize with him. Um, and I think that evolution, uh, from blowing up his first startup to being right about protein folding holds a lesson for how you do this sort of frontier science inside a company.

Host: I love that story. I guess obviously, um, part of uh, this work really helps elevate and tell uh, the story of Demis more broadly, which may be an under known story throughout this work or in general, are there other folks in the AI ecosystem that you feel similarly deserve? Uh, uh, more light shone on them or an expose, uh, some kind of biography. Other undersung heroes in this kind of AI world.

Sebastian Malaby: Well, in my book, I do have this sort of pairing of Ilya Satskever and David Silva. One of them representing the deep learning tradition PhD under, um, Geoffrey Hinton in Toronto. And then on the other hand, uh, David Silver representing what I call the kind of Edmonton, Alberta tradition PhD under Rich Sutton, who was the sort of guru of reinforcement learning. And I think both of those individuals, and I have kind of embedded in my story about Demis a kind of mini biography of the two of them, kind of juxtaposing their two approaches. And there could be a fantastic double biography of the two of them.

Host: Certainly. Um, I think at various times as Google AI, uh, fortunes have seemed better or worse. Folks have speculated that maybe, uh, one day Demis will become CEO of Google. Uh, you think that's possible?

Sebastian Malaby: Yeah, it's possible. Um, there's a question about whether he would want to do that, because I think he does love being a little bit, uh, able to think about the science. And he does this double shift every day, meaning, you know, he eats dinner, uh, with his family and then goes back to his desk until 4:00am um, and that's when he's doing science. Um, now I think actually these days, a lot of that late time is spent doing calls to Mountain View and trying to coordinate with all of the many people who report to him out of Mountain View. Um, so already that science time is being eaten into. But if he became CEO, obviously there'd be nothing left of it at all. And I think he's genuinely conflicted about whether he would want that. But it seems to me that it all depends. Kind of like if Sundar were to decide that he's leaving, um, who would be the likely CEO, uh, replacement? If it wasn't Demis, and if it was somebody that Demis felt very comfortable with, he'd probably be perfectly happy not to do that. If he was not comfortable with it, uh, then he might kind of push his hat into the ring because he wouldn't. He wouldn't want to work for somebody that he didn't agree with.

Host: Well, uh, Sebastian, it's been such an interesting conversation. I always like to leave the last word to the guest. Uh, I think in this case, it's pretty clear where you might want to point people. But, uh, I'll leave the last word to you. Anything our listeners should, should take away further from this conversation. And also, please, please plug the book.

Sebastian Malaby: Well, look, I really enjoyed writing this book, it's my sixth book, and, um, spending, you know, more than 30 hours of demis Hasabis in the top of a British pub was really quite special. He would like me to point out that we were drinking coffee, not pints of beer, uh, but just simply the intellectual range of this guy who can riff about neuroscience, computer science, physics, biology, chemistry, you know, the history of movies, novels, science fiction. Um, it was just wild. And I, uh, do try to communicate that energy on the page. Um, it's the first time I've ever used the first person as a device in a book. Um, because I wanted to use that dialogue I had with him. And so having Demis riffer than me kind of ask him a question or just, you know, in that. Interspersed in what he's saying, this kind of what I'm thinking, um, to help the reader sort of understand it, uh, this was such an extraordinary experience that I was driven to innovate the craft of writing in terms of my own craft. Um, so, yeah, I enjoyed it. I hope you do, too. If you read it. Everybody out there, amazing.

Host: Well, thanks so much for coming on the podcast to talk about it.

Sebastian Malaby: Fantastic. Thank you, Jacob. It was great.

Host: I'm Jacob Efron, and this has been Unsupervised Learning, a podcast where I get to talk to the smartest people in AI and ask them tons of questions about what's happening with models and what it means for businesses in the world. As I hope is clear, I have a ton of fun doing this. It's a, uh, nights and weekends project, in addition to my day job as an investor at redpoint. But our ability to get these incredible guests on really comes from folks like you subscribing to the podcast, sharing it with friends. It's really what ultimately makes this whole thing work. And so please consider doing that. And thank you so much for your support and listening. We'll see you next episode.