Sam Altman x Nikhil Kamath: How to Win When AI Changes Everything | People by WTF | Episode 13

0:00Micheelynn 说限量 Donellaerdad What's up?

0:21What's up? How did he do it? He said when Nikhil is. Nikhil? You've told him five minutes? He has two minutes, not... Yeah, two minutes. Everything looks good. Just the monitor, the main monitor, went off. Everyone who's done can leave.

0:43Hi Sam. Hey Nikhil. How are you? I'm good. Sorry I'm late. I got caught up in getting ready for the launch tomorrow and lost track of time and excitement with the final results. No worries. I'm guessing it must be really hectic, right? It is a very hectic day.

1:13I have the model and I've been playing with it a little bit. How is it different, Sam? I'm not an expert at this, so yeah. There's all these ways we can talk about, it's better at this metric or it can do this amazing coding demo that GPT-4 couldn't. But the thing that has been most striking for me is in ways that are both big and small, going back from GPT-5 to our previous generation model is just so painful. It's just worse at everything. And I've taken for granted that there is a fluency and a depth of intelligence with GPT-5 that we haven't had in any previous model. It's an integrated model, so you don't have to pick in our model switcher and know if you should use GPT-40 or O3 or O4 Mini or any of the complicated things.

2:01It's just one thing that works. And it is like having PhD-level experts in every field available to you 24-7 for whatever you need, not only to ask anything, but also to do anything for you. So if you need a piece of software created, you can kind of do it from scratch all at once. If you need a research report on some complicated topic, it can do that for you. If you needed to plan an event for you, it could do that too. Is it more agentic in nature in the sense that sequential tasking is one step closer to it? Because I was trying that. It's much better at things like that. The robustness and reliability has greatly increased and that's very helpful for agentic workflows.

2:48So I'm very impressed by how long and complex of a task it can carry out. So we did a call a couple of weeks ago when I was bugging you about what sectors and themes to invest in for the next decade. So I don't want to talk about that too much. I thought we'll keep today about first principles and how the world is changing by virtue of all that is changing in the world that you dominate. So the very first thing I want to start with is if I were a 25-year-old boy or girl living in Mumbai or Bangalore in India, I know you've said a bunch of times that colleges are not holding on to the place of relevance they might have had when I was growing up.

3:38But what do I do now? A, what do I study? If I'm starting a company, what kind of company do I start? Or if I were to even find a job, what industry do you think has some kind of tailwind? I'm not talking 10 years down the line, but even as close as three to five years down the line. First of all, I think this is probably the most exciting time to be starting out one's career, maybe ever. I think that 25-year-old in Mumbai can probably do more than any previous 25-year-old in history could. It's really amazing what you can do with a tool like this. I felt the same way when I was 25. The tools then were not as amazing as the tools we have now, but we had the computer revolution and a 25-year-old then could do things that no 25-year-old in history before would have been able to.

4:33And now that's happening in a huge way. So whether you want to start a company or be a programmer, or go work in any other industry, create new media, whatever it is. The ability for one person to use these tools and have great ideas and implement them with what would have taken decades of experience or teams of people is really quite remarkable. In terms of particular industries, I am very excited about what AI is going to mean for science and the amount of science that one person will be able to discover and the rate at which they can do that. clearly it's transforming what it means to program computers in a huge way.

5:15And people will be able to create completely new kinds of software at a huge new scale. Definitely for startups, if you have that idea for a new business, the ability for a very tiny team to do a huge amount of work is great. But it feels like this is just now a very open canvas. people are limited to a degree that they've never been before only by the quality and creativity of their ideas and you have these incredible tools to help you realize them is there anything in particular you want to say about GPT-5 then I can ask you questions around that GPT-5 does feel to us like it's going to be another big step forward in how people use these systems the level of capability the level of robustness, the reliability and the ability to use this just for a lot of tasks in life to create software, to answer questions, to learn to work more efficiently.

6:17This is a pretty significant step forward. Each time we've had one of these, we have been amazed by the human potential it unlocks. And in particular India is now our second largest market in the world. It may become our largest. We've taken a lot of feedback from users in India about what they'd like from us, better support for languages, more affordable access, much more. And we've been able to put that into this model and upgrades to ChatGPT. So we're committed to continuing to work on that. But every time we've had a major leap forward in how the capability we can bring to users, we've been amazed by what those 25-year-olds go off and do in terms of creating new companies or learning better, getting better medical advice or whatever else.

7:09Is there anything that I could build on top of GPT-5 today as a 25-year-old in India that you think are low-hanging fruits per se that I should definitely look at? I think you could build an entire startup way more efficiently than you ever could before. Now, this is obviously, I'm biased because this is near and dear to my heart. But the fact that as a 25-year-old in India or anywhere else, maybe with a couple of friends, maybe just by yourself, you could use GPT-5 to help you write the software for a product much more efficiently, help you handle customer support, help you write marketing and communications plans, help you review legal documents.

7:54all of these things that would have taken a lot of people and a lot of expertise, and you now have GPT-5 to help you do all of this, that's pretty amazing. If I could push you to be a bit more specific and say, I get science, but what do I study? Say I've been studying engineering or commerce or arts or something like that. is there any specific thing I study in order to use AI to develop something in science? I think the most important specific thing to study is just getting really good at using the new AI tools. I think learning is valuable for its own sake and learning to learn is this like meta skill that will serve you throughout life.

8:43And whether you're learning engineering, like computer engineering or biology or any other field. If you get good at learning things, you can learn new things quickly. But fluency with the tools is really important. When I was in college or in high school, it seemed to me that the obvious thing to do was learn a program. And I didn't know exactly what I was going to use that for, but there was this new frontier that seemed very high leverage and important to get good at. And right now, learning how to use AI tools is probably the most important specific hard skill to learn. And the difference between people who are really good at it, really AI native and think of everything in terms of those tools and don't is huge.

9:28There's other general skills. Learning to be adaptable and resilient, which I think is something really learnable. That's quite valuable in a world that's changing so fast. Learning how to figure out what people want is really quite important. before this I used to be a startup investor and people would ask me what the most important thing for startup founders to figure out was and my predecessor Paul Graham had this answer that has always stuck with me for people to give to founders and it became the motto for Y Combinator which is make something people want and that sounds like such an easy instruction I have watched so many people try so hard to learn how to figure that out and fail.

10:14And then I've watched many people work really hard to learn how to do that and get great at it over a career. So that's something as well. But in terms of the specifics, are you supposed to take the biology class at the physics class? I don't think it matters right now. And if I were to build on top of that, and when you say learn to adapt and change and learn AI tools faster, is there a path? I'm just looking for a light that somebody could begin walking towards. How does one get better at the AI tools that are available out there? Well, one great thing to do. GPT-5 is quite good at helping to create small pieces of software very quickly.

10:56And much better than any model that I've used. And in the last few weeks, I have surprised myself how much I've used it to create a piece of software to solve some little problem that I have in my life. It's been an interesting creative process because I'll ask it for a first draft, I'll start using that. And then I'll say, hey, with this feature, it would be better. Or with this other thing, I'd be able to do something differently. Or I have started using it and realized that with my workflow, I really needed this. And by putting more and more of the things that I have to do in this kind of a workflow, that's been a very interesting way for me to learn how to use this.

11:36you mentioned Paul Graham and I was reading this letter or report he wrote in 2009 where he spoke about five founders to watch out for and I think you were 19 at the time and he mentioned you along with people like Steve Jobs and Larry and Sergey why was that you hadn't accomplished anything of note like they had? What did Paul see in you? And what do you think is the innate skill set that sets you apart from your peers? That was very nice of him. I remember that. I certainly, I remember the time feeling that it was very deeply undeserved, but grateful that he said that. I mean, there's like a lot of people who are starting great companies.

12:29We got lucky here in many ways. we've also worked super hard.

12:37Maybe the one thing that I would that I think we have done here well is to take a very long time horizon and think very independently. You know, when we started, it was like four and a half years before we had our first product and it was very uncertain. We were just doing research, trying to figure out how to get AI to work and we had very different ideas than the rest of the world. But I think that ability to sort of trust our own conviction over a long time horizon without a lot of external feedback was very helpful. Is that a V or an I thing? Because I'm speaking about when you were 19. Oh, sorry, I thought you were asking about OpenAI.

13:22Me when I was 19, I barely remember that. I don't know. I was like a naive 19-year-old that didn't really know what he was doing. This is not false modesty. My own self, I think I've done impressive things now, but my own self-conception in 19 was like deeply unsure of myself and very unimpressive. If the world were to be, the world of tomorrow were to be an AI kingdom of sorts, you're definitely some kind of prince. And I don't know if you follow Machiavelli, but he used to say a very interesting thing. He said that a prince should always appear, but not be religious, compassionate, trustworthy, humane and honest.

14:14do you think this particular projection of i've heard you i've watched a lot of your interviews of late and i've heard you say repeatedly that you're not formidable and use words like that this projection of humility is it apt for the world we live in or the world you're walking into

14:41I'm not sure it's apt for either one but when I was 19 to go back to what you said I assumed that the people running these big tech companies really had it all figured out and there were adults in the room and somebody had a plan and there were these people that were way different than me that understood how to do things and the companies were running very well and there was not much drama and everything was being handled by the adults. And now that I'm supposed to be the adult in the room, I will tell you, I think no one has a plan. No one really has it all working smoothly. Everyone, or at least I, am figuring out as they go.

15:29And you have things you want to build and then things go wrong and you bet on the wrong people or the right people and you get a technological breakthrough here or you don't there. You just keep putting one foot in front of another and you put your head down and work. You have these tactics that some of them become great strategies, some of them don't. You try something and the market reacts in some way or your competitors react in some way and you do something else.

15:59And now my conception is everybody is kind of figuring it out. as they go everybody is like learning on the job and I don't think that's like false humility I think that's just the way the world works and it's like a little it's a little strange to be on this side of it but that is what it seems like. I'm not even speaking so much about the authenticity of the humility or not but more again from the lens of somebody starting to build for tomorrow because that's who we speak to. Is that a good image to project into the world? Does the image of humility really work today? You mean should someone project that image?

16:47I mean, I certainly have a very negative reaction to people who are projecting certainty and confidence when they don't really know what's going to happen. And the reason is not just because it's annoying, which it is, but also because I think if you have that kind of a mindset, it is harder to have the culture of intellectual openness and to sort of like listen to other viewpoints and make good decisions. A thing I say all the time to people is no one knows what happens next. And the more you forget that and the more you're like, I am smarter than everybody else. I have a master plan. I'm not going to pay attention to like what my users are saying or where the technology takes us or how the world is reacting.

17:39But, you know, I know better and the world doesn't know as much. I think you just make worse decisions. so having that kind of open mindset and the sort of like curiosity and willingness to adapt to new data and change your mind I think that's super important like the number of times we have thought we knew something to get smacked in the face by reality here has been a lot and one of our strengths as a company is when that happens we change what we're going to do where I think that's been really great and a big part of what's helped us succeed. So maybe there are other ways to succeed and maybe projecting a ton of bravado into the world works, but the best founders that I have watched up close throughout my career have all been more like the sort of quick learning and adapting style.

18:41And you probably know more about this than most, because of your role at Y Combinator. I have a lot of data points on it at least, yeah.

18:51When we met in Washington a couple of years ago at the White House, I remember when we were speaking and you went somewhere, I was speaking to your partner, and you guys had a kid. We did. And how is that? It is my favorite thing ever. I have nothing that is not a cliche to say here. but it is the coolest most amazing most like emotionally overwhelming in the best ways and hard ways to experience every everything everyone said about how great it is how intense it is how it's like a kind of love you didn't know you could feel it's all it's all true I have nothing to add other than I strongly recommend it and I think it's been really wonderful it's amazing so I ponder on this a lot Sam kids, why people have kids and also questions like what happens to religion and marriage tomorrow can I ask you why you had a kid?

19:51Family has always been an incredibly important thing to me and it felt like

20:01it felt like the most and I didn't even know how much I underestimated what it was actually going to be like But it felt like the most important and meaningful and fulfilling thing I could imagine doing. And it has, so far, so early, exceeded all expectations. Do you think it's the biological need to procreate? I don't know. This seems like a thing that is so deep it's difficult to put into words. But I feel confident. And everyone I know, looking back on their life, who has had a great career and had a family, has either said, I'm so glad I took the time to have kids. That was one of the most important things I've ever done, the best things I've ever done.

20:51Or they've said, that was by far the best thing I've ever done. That was way more important than any of the work I ever did. And I was willing to take the leap of faith that that would be true for me, too. And it certainly seems like it will be. And if it is just a biological hack, I don't care. I'm so happy. But there's a sense of responsibility and family is the word that keeps coming to mind that is just really great. The world seems to be having lesser kids. And do you have an insight into the future of marriage, religion, and kids? Yeah, I hope that creating family, creating community, whatever you want to call it, will become far more important in a sort of post-AGI world.

21:42I think it's a real problem for society that those things have been in retreat. I think that that just feels strictly bad to me. I'm obviously not sure why that's been happening, but I hope we'll really reverse course on that. And in a world where people have more abundance, more time, more sort of resources and potential and ability, I think it's pretty clear that family and community are two of the things that make us the happiest. And I hope we will turn back to that. as societies get more affluent if one were to buy into the mimetic desires of people we all tend to want what other people want not necessarily what other people have if we all had more do you think we would still want more if we all had enough?

22:42I do sort of think that human demand desire, ability to play status games, whatever, it seems pretty limitless. I don't think that's necessarily bad. Or not all bad. But yeah, I think we will figure out new things to want and new things to compete over. Do you think the world retains, largely the world retains, the current model of capitalism and democracy in a way? Let me give you a scenario. what happens if a company X, let's say OpenAI, gets to the point where it is 50 % of world GDP? Does society allow for that? I would bet not. I don't think that will happen. I think this will be a much more distributed thing.

23:36But if for some reason that did happen, I think society would say, we don't think so. Let's figure out something to do here. the analogy I like most for AI is the transistor which was this really important scientific discovery that for a while looked like it was going to capture a ton of the value and then turned out to be something that just gets built into tons of products and services and you don't really think about the fact that you're using transistors all day long, it's just kind of in everything and all these companies make incredible products and profits from it in this very distributed way so I would guess that's what happens and it's not like one company is ever half of global GDP at one point I did worry that that might happen but I think that was a naive take right but do you think the odds of the world moving towards socialism go up or if something gets that large will it get nationalized and we become more socialist I don't know if something will get nationalized I don't know if the world will like officially turn towards socialism but I expect that social support or redistribution or whatever you want to call it will increase over time as society gets richer and as the technological landscape shifts.

24:55I don't know what way it's going to happen and I expect in different countries it'll happen in different ways. I think you'll see experimentation of new kinds of sovereign wealth funds, new kinds of universal basic income ideas, redistribution of AI compute. I don't know exactly what. But I suspect we'll see a lot of experimentation in society here. On universal basic income, I think WorldCoin was a very interesting experiment. Can you tell us a bit about what's happening there? The idea was, could we, we have all this AI coming, we really want to care about humans as special. Can we find a privacy-preserving way to identify unique humans and then create a new network and a new currency around that?

25:38So it's a very interesting experiment, still early, but growing quite fast. If AGI eliminates scarcity or scarcity by virtue of increasing productivity, could one also assume that it would be deflationary in nature? Capital or money loses its ability to return a rate of return, and capital no longer remains a moat in the world of tomorrow? I feel confused about this. I think if you look at the basic economic principles, it's supposed to be hugely deflationary. And yet, if the world decides that building out AI compute today is super important to things tomorrow, maybe something very strange happens with the economy and maybe capital is really really important because every piece of compute is so valuable

26:46I was asking someone at dinner the other night if they thought that interest rates should be minus 2 % or 25 % and he kind of laughed he's like well that's a ridiculous question and then he stopped and said actually I'm not sure so I think it's like yeah it should be deflationary eventually but I could see it being weird in the short term that's actually a very interesting thing to say do you suspect it would be minus 2 %? eventually but I'm not sure and maybe it's just like we're in this massive expansionary time where you're trying to build the Dyson sphere in the solar system and you're borrowing money at crazy rates to do that and then there's more expansion beyond and more and more.

27:29And I don't know. I find it very hard to see more than a few years in the future at this point. The conversation we were having a couple of weeks ago, I was doing more research on the sectors you had suggested. I think we agreed on older and sicker world. You also made a case for as discretionary spend goes up, gateway luxury brands might do well. Yeah. What happens to them in a deflationary world? Because the value of these purchases go down. Maybe not. In a deflationary world, some things can face huge deflationary pressure and others can be the sink for all of the extra capital. So I'm actually not sure they do go down in a deflationary world.

28:18I would bet they go up, actually. Yeah? I think so. Because the excess capital has to flow somewhere. When you look at classical economic theory, like Adam Smith and stuff like that, the Austrian school always spoke about the marginal utility of things. If you have one kettle at home to make tea, it has X in value. When you have two kettles, it still has some value. But when you have 20 kettles, it has no value. Do you think the world goes in that direction? Yeah, so 20 kettles doesn't help you. but even if you're only gonna you know spend two hours a day playing video games or whatever and that amount of time is fixed and so you don't need 20 hours worth of video games if that two hours of games gets better and more entertaining forever and ever and ever and just keeps getting more and more compelling that still has value to you and I think there are a lot of categories where we will find that people can just get much better stuff, even if they don't necessarily get more of them.

29:29I think we'll see this in a really big way. Do you think there's a use case for the wrappers that are getting built on these large models right now? Like I was in the US recently and I met Harvey, for example. What happens to a wrapper like that? Does it get innovated out by the model itself at some point of time? Some of them yes and some of them no. And it's like sometimes you can obviously predict when one is going to go one way or the other and sometimes it really depends on the choices the company makes down the line. The main thing I would say is using AI itself does not create a defensible business.