[
  {
    "i": 0,
    "speaker": "Demis Hassabis",
    "text": "Actually, the story is pretty complicated in the sense that the leading labs at the time, so us included, and actually both DeepMind and the Google Lab, Google Brain, had large language models."
  },
  {
    "i": 1,
    "speaker": "Demis Hassabis",
    "text": "In fact, the Google Brain side of it had a product, a very early one, it was called Lambda in those days."
  },
  {
    "i": 2,
    "speaker": "Demis Hassabis",
    "text": "And there'd been a conscious decision not to release that, and it was obviously in retrospect probably the wrong decision, but you can sort of see, and I wasn't involved in that because we were at the time just supposed to be working on research, right, because DeepMind, prior to that, prior to Google DeepMind being formed,"
  },
  {
    "i": 3,
    "speaker": "Demis Hassabis",
    "text": "we were just sort of running independently mostly as a research organisation. So, but I think that's an interesting thing of where you can be sometimes too close to your own technology. So, at building that technology, those early language models, they were very interesting, but they were very flawed."
  },
  {
    "i": 4,
    "speaker": "Demis Hassabis",
    "text": "Even today's ones, very impressive as they are, have things like this. Things that we all know when we use the chatbots, they can't do very well. They make mistakes, they hallucinate, and those early ones were doing that a lot."
  },
  {
    "i": 5,
    "speaker": "Demis Hassabis",
    "text": "So, you can imagine, and I can understand that I wasn't involved in the decision making, but you can sort of think about, you know, Google's known for accurate information."
  },
  {
    "i": 6,
    "speaker": "Demis Hassabis",
    "text": "That's what Google is for, Google Search is for, organising the world's information. That's its bread and butter, that's what all of us come to it, Google Search, for."
  },
  {
    "i": 7,
    "speaker": "Demis Hassabis",
    "text": "So, I think it's sort of doubly tricky to think about putting out a bot that, you know, 20% of the time, gives you the wrong information. And so, I think there's a clear issue there for a company like Google."
  },
  {
    "i": 8,
    "speaker": "Demis Hassabis",
    "text": "Whereas, obviously, a start-up, especially someone with someone like Sam leading it, Whereas, obviously, a start-up, especially someone with Sam leading it,"
  },
  {
    "i": 9,
    "speaker": "Demis Hassabis",
    "text": "you know, it's like, go for it, you know, you only live once, YOLO it. And it's fine, right?"
  },
  {
    "i": 10,
    "speaker": "Demis Hassabis",
    "text": "And the thing is, to be fair to us as well, from a research perspective, no one knew how big that was going to get in ChatGPT, including OpenAI. They didn't know."
  },
  {
    "i": 11,
    "speaker": "Demis Hassabis",
    "text": "And it's obvious, I mean, Sam's admitted that, but it's also obvious from the name. ChatGPT is a programmer's name. You can, you know, just like, you know, they would have named it something catchier if they knew how big it was going to get, right?"
  },
  {
    "i": 12,
    "speaker": "Demis Hassabis",
    "text": "It's not a catchy name. Although, you know, it's obviously synonymous with AI now. So, you can see, and I think it's, they've talked about, they just thought it was going to be a research experiment."
  },
  {
    "i": 13,
    "speaker": "Demis Hassabis",
    "text": "And obviously, they could take more risks with, given that they were a start-up, with reputation and other things, if it makes mistakes, so what? They don't have any existing products that rely on reliability."
  },
  {
    "i": 14,
    "speaker": "Demis Hassabis",
    "text": "So, the thing that caught everyone by surprise, I would say, including them, was actually people finding use for even these early language models that, even though they had all of these flaws, you know, people still found, obviously, very creatively, once millions of people use it, oh, you can summarize things, can help with homework."
  },
  {
    "i": 15,
    "speaker": "Demis Hassabis",
    "text": "It didn't matter if it made some errors, right? Those weren't to do with information seeking. And that's what, these were new use cases that hadn't existed before. And obviously, there was this, turned out, there was this huge demand for even those kinds of early use cases."
  },
  {
    "i": 16,
    "speaker": "Demis Hassabis",
    "text": "And that's what caught everyone by surprise in terms of the technology being mature enough to come out of the lab and be put into the general public's hands."
  },
  {
    "i": 17,
    "speaker": "Speaker 1",
    "text": "And one of the upshots was that for Google, Google Brain and DeepMind merged. You now run all, essentially, all AI at Google."
  },
  {
    "i": 18,
    "speaker": "Speaker 1",
    "text": "And obviously, you've not just closed the gap, but on some metrics are leading in terms of some of these models that you are putting out with Gemini. That in itself, I mean, we probably don't have the time to go into the intricacies of that merger."
  },
  {
    "i": 19,
    "speaker": "Speaker 1",
    "text": "But that in itself, Sebastian, you were saying that a merger of Google Brain and DeepMind stands out to you in terms of just in terms of a corporate merger as something of a standout moment."
  },
  {
    "i": 20,
    "speaker": "Speaker 2",
    "text": "Yeah, I mean, the conventional wisdom, if you ask a business school professor about mergers, is that, you know, they're very difficult because there's always cultural differences. It's painful to smush two different organizations together and so forth."
  },
  {
    "i": 21,
    "speaker": "Speaker 2",
    "text": "And in the case of, you know, Google Brain, which was doing AI research in Mountain View, and then DeepMind, which was doing AI research in London, you're putting together two teams which had an eight-hour time difference between them, which doesn't make it easier."
  },
  {
    "i": 22,
    "speaker": "Speaker 2",
    "text": "And they had a little bit of a record, which, you know, I hope you don't mind me saying this, there was a bit of a record of competition between the two because to some extent, if DeepMind got compute resources to, you know, run some tests and build a model, the people in Mountain View might see that as at the expense of their experiment they were trying to do."
  },
  {
    "i": 23,
    "speaker": "Speaker 2",
    "text": "So there was, at least on the level of kind of engineering gossip, you know, quite a lot of back and forth about, you know, this bad blood. So here you are, you've got to do a merger. You've got eight time zones of difference."
  },
  {
    "i": 24,
    "speaker": "Speaker 2",
    "text": "There's a bit of a record of competition and you're in the middle of this, knock down, drag out, capitalist fight, right, with open AI racing ahead and Google's got to catch up. What are the odds of this succeeding?"
  },
  {
    "i": 25,
    "speaker": "Speaker 2",
    "text": "I think most business school professors would have said, nah, not going to work. It worked."
  },
  {
    "i": 26,
    "speaker": "Speaker 2",
    "text": "Like two and a half years later, at the end of 2025, DeepMind has an AI model, Gemini 3.0, which beats the ChatGPT rival on the leaderboards, on these independent metrics of performance and accuracy and so forth."
  },
  {
    "i": 27,
    "speaker": "Speaker 2",
    "text": "So in two and a half years, it was a total turnaround. It's like, I think this will go down not merely as a story about a frontier technology, but as a sort of case study in business mergers."
  },
  {
    "i": 28,
    "speaker": "Speaker 1",
    "text": "And this is helping us, this merger and being under Demis and the ambitions around AGI are helping you to get to that long-term ambition that's driven you since you were a student, which is AGI and super intelligent machines."
  },
  {
    "i": 29,
    "speaker": "Speaker 1",
    "text": "So I want to get to the title of the book, belatedly maybe, The Infinity Machine, what is Demis, what is the Infinity machine, as you understand it?"
  },
  {
    "i": 30,
    "speaker": "Demis Hassabis",
    "text": "Well, I mean, I think it's a great title for the book. And I think actually Sebastian mentioned it in the opening."
  },
  {
    "i": 31,
    "speaker": "Demis Hassabis",
    "text": "The way we look at problems, and it started really with AlphaGo and then onto AlphaFold is, I think we found, and this is what AGI should be, a kind of general solution that can be applied to many, many problems."
  },
  {
    "i": 32,
    "speaker": "Demis Hassabis",
    "text": "And the way I look at both Go and protein folding, which ostensibly look really different type of problems, is you're effectively navigating, there's a needle in a haystack solution, but the haystack is unfathomably large, right?"
  },
  {
    "i": 33,
    "speaker": "Demis Hassabis",
    "text": "So, for example, in Go, there are more possible board positions than there are atoms in the universe. So, you know, that means you can't brute force and enumerate every option. There's just not enough time in the universe."
  },
  {
    "i": 34,
    "speaker": "Demis Hassabis",
    "text": "And the same with proteins, even more complicated. An average size protein, you know, has 10 to the 300 possible conformations, possible shapes. So, you can't find the right shape by looking at every single shape."
  },
  {
    "i": 35,
    "speaker": "Demis Hassabis",
    "text": "You're going to have to do something much smarter. And so, what we generically have built with our AI systems, at least applied to science, is using deep learning, building neural networks, give them the data that you have on that particular domain."
  },
  {
    "i": 36,
    "speaker": "Demis Hassabis",
    "text": "So, Go game or, you know, protein structures that are known. And it's got to create a model of how that domain works. And then, once you have that model, you can apply a search process, but in a smart way."
  },
  {
    "i": 37,
    "speaker": "Demis Hassabis",
    "text": "Instead of having to search everything. So, in a Go game, instead of, you know, you have 100 possible moves in a position, you don't search all of them, you just search. The model tells you which ones are the most likely, the most fruitful ones to look at."
  },
  {
    "i": 38,
    "speaker": "Demis Hassabis",
    "text": "So, you can spend your compute budget on very useful alternatives and only look at the ones that are likely to be fruitful. And so, that way, you can actually cut down these enormous search spaces into something that's tractable in, potentially, in a very fast amount of time."
  },
  {
    "i": 39,
    "speaker": "Demis Hassabis",
    "text": "And so, this is a very, I sort of realize this is a very generic solution to many problems if you can couch it in this way. This massive combinatorial space, you're trying to find the needle in a haystack, and the final thing you need is an objective function."
  },
  {
    "i": 40,
    "speaker": "Demis Hassabis",
    "text": "You need to have a clear goal you can hill climbing against that the search is going to do and the model is going to help you find."
  },
  {
    "i": 41,
    "speaker": "Demis Hassabis",
    "text": "And in games, and why we used games at the beginning, apart from that I love games, is that it's very convenient to train your AI systems on."
  },
  {
    "i": 42,
    "speaker": "Demis Hassabis",
    "text": "In fact, before Go, we used Atari games as the earliest measures of how our systems are doing because they have clear objectives, win the game or maximize the score. But in science, there are also clear objectives."
  },
  {
    "i": 43,
    "speaker": "Demis Hassabis",
    "text": "You know, minimize the energy in the system, for example, is a very clear one that often you find in biology and chemistry and physics."
  },
  {
    "i": 44,
    "speaker": "Demis Hassabis",
    "text": "So, as long as you can couch the problem in that way, and most of the time, in a surprisingly large number of times, you can, then these types of methods that we are using can be very, very effective."
  },
  {
    "i": 45,
    "speaker": "Speaker 1",
    "text": "We're living in a moment where we're thinking about AI in warfare, whether it's in Ukraine or in Iran. We're thinking about maybe some of us, Anthropic and Mistral and cybersecurity."
  },
  {
    "i": 46,
    "speaker": "Speaker 1",
    "text": "The clear positive is that guiding light that you've had throughout your life, with scientific discovery and specifically AlphaFold 2, AlphaFold 3, Isomorphic Labs."
  },
  {
    "i": 47,
    "speaker": "Speaker 1",
    "text": "Is that where... When we think about the risks, benefits, balance for AI and AGI, is it very clear for you that it's medical?"
  },
  {
    "i": 48,
    "speaker": "Demis Hassabis",
    "text": "Because I believe it's the ultimate tool for science and broadly construed. So, science and medicine. I think the number one thing we can apply AI to is curing terrible diseases."
  },
  {
    "i": 49,
    "speaker": "Demis Hassabis",
    "text": "So, number two on my list would be helping with the environment and finding new renewable energy sources. And you can see that's what I care about because that's what I personally do. I'm not going to be done with my time other than building the core technology, right?"
  },
  {
    "i": 50,
    "speaker": "Demis Hassabis",
    "text": "Things like AlphaFold, Isomorphic Labs and all of our... We've had a science group, an AI for science group for nearly 10 years now."
  },
  {
    "i": 51,
    "speaker": "Demis Hassabis",
    "text": "Some of our rivals are starting to set up science teams to use the AI, but we've had one for 10 years, including... That's where AlphaFold came from."
  },
  {
    "i": 52,
    "speaker": "Demis Hassabis",
    "text": "And so, we work on fusion, we work on weather prediction, we work on climate models, material science. So, there's almost no area of science that I think AI can't... help with."
  },
  {
    "i": 53,
    "speaker": "Demis Hassabis",
    "text": "And I think if we do that right, in the next 10 years, we'll have almost a second renaissance, a new golden era of scientific discovery. And ultimately, you know, I want..."
  },
  {
    "i": 54,
    "speaker": "Demis Hassabis",
    "text": "The reason I started on this journey is I wanted to understand the nature of reality, all the big questions in the world, nature of consciousness, what is time, all of these things, which are, you know, incredible deep mysteries."
  },
  {
    "i": 55,
    "speaker": "Demis Hassabis",
    "text": "And it's amazing, we... I find it amazing that we all just go about our daily lives, you know, without thinking too much about this, sort of, distracted with what we do. But for me, it just stops me dead in my tracks all the time."
  },
  {
    "i": 56,
    "speaker": "Demis Hassabis",
    "text": "And this is what I think about late in the night, in the small hours of the night, is how is it we don't even know what time is? This is crazy, like, we're just going around doing things. You know, it's crazy to me."
  },
  {
    "i": 57,
    "speaker": "Demis Hassabis",
    "text": "And then, of course, well, what do we do about that is the question, because I guess I was sort of born with this insatiable curiosity and also almost haunting thing about, like, this is a deep mystery here in the universe."
  },
  {
    "i": 58,
    "speaker": "Demis Hassabis",
    "text": "It's like an incredible puzzle. That's how I feel about the universe. How are we going to solve it? And then I guess that's why I got into AI early is, like, I tell you what we need."
  },
  {
    "i": 59,
    "speaker": "Demis Hassabis",
    "text": "We need some help, even the best scientists. You know, I used to read all the biographies of the top scientists, all my favorite heroes from Feynman to Turing."
  },
  {
    "i": 60,
    "speaker": "Demis Hassabis",
    "text": "And, you know, they're all really smart, but there's a limit to how much we can comprehend, I think, with just unaided by our human minds, even with scientific method and other things. And so it was obvious to me that AI, it was a two-factor thing."
  },
  {
    "i": 61,
    "speaker": "Demis Hassabis",
    "text": "Not only would it be an amazing tool for science, but it would be an incredible scientific artifact in and of itself as an amazing engineering artifact and possibly help us to understand our own minds better, which is also one of the deep mysteries."
  },
  {
    "i": 62,
    "speaker": "Demis Hassabis",
    "text": "So that's why I would have done this no matter what. Like, today, it's become the biggest thing it is, but for me, it's the most fascinating question and the most fascinating technology there is out there. But it's dual purpose."
  },
  {
    "i": 63,
    "speaker": "Demis Hassabis",
    "text": "And we also knew that at the beginning of AGI, which is why we were worried about safety from the very beginning of DeepMind in 2010, and we still are now because it's general."
  },
  {
    "i": 64,
    "speaker": "Demis Hassabis",
    "text": "So it can be used for anything. And I think it's down to us in the industry to try and push it and advance more quickly the areas that are going to be, you know, great for society like medicine and energy."
  },
  {
    "i": 65,
    "speaker": "Speaker 1",
    "text": "Where are we right now on the balance of risks, do you think?"
  },
  {
    "i": 66,
    "speaker": "Demis Hassabis",
    "text": "Well, look, I think right now one of the issues with this ferocious commercial battle, so there's good... So, look, the competitive part of me loves that, of course. It's like, great, you know, it's the kind of top competition there is in the world."
  },
  {
    "i": 67,
    "speaker": "Demis Hassabis",
    "text": "But at the back, the problem I can't... So part of me, you know, I like that, I lean into that. That's what helps me lead in this situation."
  },
  {
    "i": 68,
    "speaker": "Demis Hassabis",
    "text": "But at the back of my mind, I've got this gnawing feeling that there's something much more important, much bigger than the commercial race, which is overall getting AGI safely over the line for humanity. And to make sure that the benefits fully outweigh the risks."
  },
  {
    "i": 69,
    "speaker": "Demis Hassabis",
    "text": "And, you know, I'm going to try... We're only one actor in this now. There's, you know, five or six leaders and there's China as well and the Chinese labs. And so I think in the next few years, the story's still to be written, like, how this is going to go."
  },
  {
    "i": 70,
    "speaker": "Demis Hassabis",
    "text": "But I do think there needs to be more cooperation and coordination at an international level, ideally, although that's very hard with the geopolitics as it is today, around safety topics. And debates around, you know, the benefits versus the risks."
  }
]