谢尔盖·布林：谷歌创立与 AI 未来

Sergey Brin (00:00:00): They wondered if there was a better way to find information on the web. On September 15, 1997, they registered Google as a website. One of the greatest entrepreneurs of our time. Someone who really wanted to think outside the box. If that sounds like it's impossible, let's try it. Speaker 2 (00:00:17): He took a back seat in recent years to other Google leaders. Brin is now back helping Google's efforts in artificial intelligence. Sergey Brin (00:00:24): I feel lucky that I fell into doing something that I feel really matters. You know, getting people information. David Friedberg (00:00:48): No introduction needed. Well, I I Sergey Brin (00:00:52): just agreed to this last minute, as you know. I don't know where you pulled up that clip so fast. David Friedberg (00:00:56): You guys. The team is amazing. Sergey Brin (00:00:57): Kind of amazing. This is kind of amazing. Yeah. I thought David Friedberg (00:01:01): Sergey, Sergey just had like, he asked to come check out the conference, and I was like, definitely. Like, come hang out. Sergey Brin (00:01:08): I didn't actually understand, to be perfectly honest. I thought you guys just kinda had a podcast and, like, a little get together or something, but yeah. This is kinda mind-blowing. Congratulations. Thank you. David Friedberg (00:01:20): Well, I'm glad you came out. Yeah. Thanks for doing it. Sergey Brin (00:01:22): Feeling a little bit shy, but yeah. David Friedberg (00:01:24): Wow. But thanks for agreeing to chat Sergey Brin (00:01:26): a little bit. David Friedberg (00:01:26): We're gonna talk for a little bit. So this was not on the schedule. But, I thought it'd be great to talk to you given where you sit in the world as AI is on the brink of and is actively changing the world. Obviously, you know, you founded Google with Larry in 1998, and, you know, recently it's been reported that you've kind of spent a lot more time at Google working on AI. I thought maybe and and a lot of industry analysts and pundits have been kind of arguing that LLMs and conversational AI tools are kind of an existential threat to Google search. That's one of the—and I think a lot of those people don't build businesses or they have competitive investments, but, you know, we'll leave that to the side. But there's this big kind of narrative on what's gonna happen to Google and where's Google sitting with AI. And I know you're spending a lot of time on it, so thanks for coming to talk about it. How much time are you spending at Google? What are you working on? Yeah. Sergey Brin (00:02:19): Honestly, like, pretty much every day. I mean, like, I'm missing today, which is, you know, one of the reasons I was a little reluctant, but I'm glad I came. But I think as a computer scientist, I've never seen anything as exciting as all of the AI progress that's happened in the last few years. Thanks. No. But it's kind of mind-blowing. When I went to grad school in the nineties, you know, AI was kind of like a footnote in the curriculum almost. Like, you're like, oh, maybe you have to do this one little test on AI. We tried all these different things. They don't really work. That's it. That's all you need to know. Yeah. And then somehow miraculously, all these people who are working on neural nets, which was one of the big discarded approaches to AI in, like, the sixties, seventies, and so forth, just started to make progress. A little bit more compute, a little more data, a few clever algorithms. And the thing that's happened in this last decade or so is just amazing as a computer scientist. Like, every month, you know, well, all of you, I'm sure, use all the AI tools out there. But, like, every month, there's, like, a new amazing capability, and I'm, like, probably, you know, doubly wowed as everybody else is. Yep. That computers can do this. And so, yeah, for me, I really got back into the technical work, because I just don't want to miss out on this, as a computer scientist. David Friedberg (00:03:59): Is it an extension of search or a rewriting of how people retrieve information? Sergey Brin (00:04:06): I mean, I just think that the AI touches so many different elements of day-to-day life, and, sure, search is one of them. But it kind of covers everything, for example, programming itself. Right. Like, the way that I think about it is very different now. Like, you know, writing code from scratch feels really hard compared to just asking the AI to do it. Right. You know, sorry. David Friedberg (00:04:37): So what do you do then? Sergey Brin (00:04:39): Actually, I've written a little bit of code myself. Yeah. Just for kicks, just for fun. And then sometimes I've had the AI write the code for me, which was fun. I mean, just one example, I wanted to see how good our AI models were at Sudoku. So I had the AI model itself write a bunch of code that would automatically generate Sudoku puzzles and then feed them to the AI itself and then score it and so forth. Right. But it could just write that code. And I was, like, talking to the engineers about it and, you know, whatever. We had some debate back and forth. Like, I came back half an hour later. It's done. And they were kind of impressed because they don't honestly use the AI tools for their own coding as much as I think they ought to. David Friedberg (00:05:27): Right. So that's an interesting example because maybe there's a model that does Sudoku really well. Maybe there's a model that, like, answers information questions for me about facts in the world. Maybe there's an AI model that designs houses. A lot of people are working towards these ginormous general-purpose LLMs. Is that where the world goes? Some people, I think, refer—I don't know who wrote this recently. He said there's a god model. Like, there's gonna be a god model, and that's why everyone's investing so much is if you can build the god model, you're done. You've got your AGI. Whatever terms you want to use, there's this one thing to rule them all. Or is the reality of AI that there are lots of smaller models that do application specific things, maybe work together, like in an agent system? Like, what's the what what what is the evolution of model development and and the that how models are ultimately used to do all these cool things? Sergey Brin (00:06:23): Yeah. I mean, I think, like, if you looked 10, 15 years ago, there were different AI techniques that were used for different problems altogether. Like, you know, the chess-playing AI was very different than image generation, which was, you know, very different, than Yeah. Like, recently, David Friedberg (00:06:42): the graph neural net at Google that, like, outperformed every physics forecasting model. I don't know if you know this, but you guys published it. It was pretty awesome. Burst. Sergey Brin (00:06:51): Yeah. Right? But it was David Friedberg (00:06:52): like a totally different architect. It's a different system. It was trained differently, and it ended up Yeah. In that particular Sergey Brin (00:06:57): So there historically, there have been different systems. And even recently, like the International Math Olympiad that we participated in, we got a silver medal as an AI. Actually, one point away from gold. But we actually have three different AI models in there. There was one very formal theorem-proving model that actually did basically the best. There was one specific to geometry problems. Believe it or not, that was just a special kind of AI, and then there was a general-purpose language model. But, since then, we've tried to take the learnings from that. That was just a couple months ago, and tried to infuse some of the certain knowledge and ability from the formal approver into our general language models. That's still work in progress. But I do think the trend is to have a more unified model. I don't know if I'd call it a god model, but to have certainly sort of shared architectures and ultimately even shared models. David Friedberg (00:08:03): Right. So if that's true, you need a lot of compute to train and develop that model, that big model? Sergey Brin (00:08:14): Yeah. Yeah. I mean, you definitely need a lot of compute. I think, like, I've read some articles out there that just, like, extrapolate. They're like, you know, it's like 100 megawatts and a gigawatt and 10 gigawatts and 100 gigawatts. And I don't know if I'm quite a believer in, you know, that level of extrapolation, partly because also the algorithmic improvements that have come over the course of the last few years, maybe are actually even outpacing the increased compute that's put into these models. David Friedberg (00:08:49): So is it irrational, the build-out that's happening, everyone talking about the Nvidia revenue, the Nvidia profit, the Nvidia market cap, supporting all of what people call the hyperscalers and the growth of the infrastructure needed to build these very large-scale models using the techniques of today. Is this irrational, or is it rational because if it works, it's so big that Sergey Brin (00:09:12): it doesn't matter how much you. Well, first of all, I'm not like an economist or like a market watcher the way that you guys are very carefully, watch companies. So I just wanna disclaim my abilities in the space. I think that I know, for us, we're kind of building out compute as quickly as we can, and we just have a huge amount of demand. I mean, for example, our cloud customers just want a huge amount of TPUs, GPUs, you name it. You know, we just can't—we have to turn down customers because we just don't have the compute available, and we use it internally to train our own models, to serve our own models, and so forth. So I guess I think there are very good reasons that companies are currently building out compute at a fast pace. I just don't know that I would look at the training trends and extrapolate three orders of magnitude ahead just blindly from where we are today. David Friedberg (00:10:09): But the enterprise demand is there out there. You know, I Sergey Brin (00:10:13): mean, they want to do lots of other things. For example, running inference on all these AI models, applying them to all these new applications. Yeah. There doesn't seem to be a limit right now. David Friedberg (00:10:29): And where have you seen the greatest success, surprising success in the application of models, whether it's in robotics or biology? What are you, like, seeing that you're like, wow, this is really working? And where are things gonna be more challenging and take longer than I think some people might be expecting? Sergey Brin (00:10:50): Yeah. I mean, now that you mentioned those, well, I would say in biology, you know, we've had AlphaFold for quite a while. And I'm not personally a biologist, but when I talk to biologists out there, like, everybody uses it, and its more recent variants. And that is, I guess, a different kind of AI. But like I said, I do think all these things tend to converge. You know, robotics, for the most part, I see in this sort of wow stage. Like, wow. You could make a robot do that with just, you know, this general-purpose language model or just a little bit of fine-tuning this way or that. And it's, like, amazing, but maybe not, for the most part, yet at the level of robustness that would make it, like, day-to-day useful. David Friedberg (00:11:43): But you see a line of sight to it? Sergey Brin (00:11:47): Yeah. Yeah. I mean, it would be it's a I don't see any particular David Friedberg (00:11:51): But Google had the robotics business and then spun it out or sold it? Sergey Brin (00:11:56): We've had, like We had a lot of robotics. Robotics business. They David Friedberg (00:12:00): just weren't the timing wasn't right. Sergey Brin (00:12:01): Yeah. Yeah. Unfortunately, I don't know. I guess yeah. I I think that was just a little too early, to be perfectly honest. I mean, there was, like, Boston Dynamics, was called, Stark stamp. I don't even remember all the ones we had. Anyway, we've had, like, five or six embarrassingly. Yeah. But they're very cool, and they're very impressive. It, yeah, it just feels kind of silly having done all of that work, and seeing now how capable these general language models are that include, for example, vision and image, and they're multimodal, and they can understand the scene and everything and not having had that at the time. Yeah. It just feels like you were sort of on a treadmill that wasn't gonna get anywhere without the modern AI technology. David Friedberg (00:12:54): You spend a lot of time on core technology. Do you also spend a lot of time on product visioning, where are things going, and what, like, the human-computer interaction modality they're gonna be in the future in a world of AI everywhere? Like Oh, I what's our life gonna be like? Sergey Brin (00:13:08): I mean, I guess there's water cooler chitchat about things like that. David Friedberg (00:13:15): Shit. Care to share any? I Sergey Brin (00:13:22): trying to think of things that aren't embarrassing. Struggling. But, all friends. I I guess it's, like, just really hard to, you know, just forecast, like, you know, to think five years out because, you know, based on the base technical capability of the AI is what enables the applications. And then sometimes, you know, somebody will just whip up a little demo that you just didn't think about, and it'll be kind of Yeah. And, and, of course, then from demo to actually making it real in production, so forth takes time. I don't know if you've played with, like, the Astra model, but it's just sort of live video and audio, and you can chat with the AI about what's going on in your environment. David Friedberg (00:14:13): You'll give me access. Right? Sergey Brin (00:14:15): Yeah. I'll get, well, once I have access. I mean, I'm sort of sometimes the slowest to get some of these things. But it's, yeah, there's like a moment of wow, and you're like, oh my God. This is amazing. And then you're like, okay. Well, it does it correctly, like, 90% of the time, but am I really, like, is that then worth it if 10% of the time it's gonna make a mistake or taking too long or whatever? And then you have to work, work, work, work, work, work, work, work to get to perfect all those things, make it responsive, make it available, whatever. And then you actually end up with something kind of amazing. I heard a story that David Friedberg (00:15:01): you went in, you were on-site. I should have mentioned this to you before you came on stage to see if you were cool about talking about it, but here we are. And they're like a bunch of engineers showed you that you could, like, use AI to write code. And it was like, well, we haven't pushed it in Gemini yet, because we wanna make sure it doesn't make mistakes. And there was this, like, hesitation culturally at Google to do that. And you were like, no. If it writes code, push it. And you really and a lot of people have told me this story because they said, or, you know, I've heard this, that it was really important to hear that from you, the founder, in being really clear that Google's conservatism, you know, can't rule the day today, and we need to kind of see Google push the envelope. Is that accurate? Is that kind of how you've spent some time? Or Sergey Brin (00:15:45): I don't remember the specifics yet, just to be honest. Yeah. But, but I'm not surprised. David Friedberg (00:15:52): I mean, I guess that's the question for me. It's like, as Google's gotten so big, there's more to lose. Sergey Brin (00:15:59): I think there's, like, this yeah, I think there's a little bit of fear. I mean, language models to begin with. Like, we invented them, basically, with a transformer paper that was, whatever, six, eight years ago, something like that. And, oh, no one, by the way, is back at Google now, which is awesome. Congrats. And, yeah, we were too timid to deploy them, and, you know, for a lot of good reasons, like, whatever. They make some mistakes. They say embarrassing things. Whatever. You know? They're you know, sometimes they're just, like, kind of embarrassing how dumb they are. Even today's, like, latest and greatest things, like, make yeah. Really stupid mistakes people would never make. And at the same time, like, they're incredibly powerful, and they can help you do things you never would have done. And, you know, like, I've, like, programmed really complicated things with my kids. Like, they'll just program it because they just ask the AI using all these really complicated APIs and all kinds of things that would take, like, a month to, like, learn. So I just think that that capability is magic, and you need to be willing to have some embarrassments, and take some risks. And I think we've gotten better at that. And, well, you guys have probably seen some more embarrassments. David Friedberg (00:17:28): But you're comfortable with that. I mean, you have super voting stock. You're still, like I mean, you're comfortable with the embarrassments at this stage because it's so important to do this? Like Sergey Brin (00:17:36): I mean, not not particular on the basis of my stock. But I I, David Friedberg (00:17:40): But as a, you know, I Sergey Brin (00:17:41): mean, but I am comfortable. Your principal, I mean, I guess I just think of it as something magical we're giving the world. Yeah. And I think as long as we communicate it properly, like, saying, like, look, this thing is amazing, and we'll periodically get stuff really wrong. Yeah. Then I think we should put it out there and let people experiment and see what new ways they find to use it. I just don't think this is the technology you wanna just kind of keep close to the chest and hidden until it's, like, perfect. David Friedberg (00:18:20): Do you think that there's so many places that AI can affect the world and so much value to be created that it's not really a race between Google and Meta and Amazon? Like, people frame these things as kind of a race. Is there just so much value to be created that you're working on a lot of different opportunities, and it's not really about who built the model that scored the LLM that scores the best, that there's so much more to it? I mean, how do you kind of think about the world out there and Google's place in it? Sergey Brin (00:18:51): I mean, I think it's very helpful to have competition in the sense that all these guys are vying, and we were just, we were number one on LMSIS for a couple weeks, by the way, just now. And I think we're, last time I checked, we're still beat the top model. There's just so. Okay. David Friedberg (00:19:09): So you do care? Sergey Brin (00:19:10): Yeah. Not saying not back, but, and, and, you know, we've come a long way since, you know, a couple, whatever, years ago when ChatGPT launched, and we were quite a ways behind. I'm really pleased with all the progress we made, so we definitely pay attention. I mean, I think it's great that there are all these AI companies out there, be it us, OpenAI, Anthropic, you name it. There's Mistral. It's it's a I mean, it's a big fast-moving field, but I guess your question is yeah. I mean, I think there's tremendous value to humanity. And I I think if you think back, you know, like, when I was in college, let's say, and there wasn't really a proper Internet or, like, web the way that we know it today, like, the amount of effort it would take to get basic information, yeah. The amount of effort it would take to communicate with people, yeah. You know, before cell phones and things. Like, we've gained so much capability across the world. But the sort of the new AI is another big capability, and pretty much everybody in the world can get access to it in one form or another these days, and I think it's super exciting. David Friedberg (00:20:35): Awesome. Sorry we have so such limited time. Sergey, thank you so much for joining us. Please join me in thanking Sergey. Thank you.