Sam Altman 谈 AGI、算力与人的能动性

1. The Promise and Risks of AGI

I think this will be the greatest technological achievement in human history thus far. But the only way that it really matters is if it makes people's lives much better than they otherwise would have been. We are about to create a genie that can grant any wish, because I think people will have such creative wishes and such incredible ideas of what they ask AI to help build.

But concentration of power with AI is a terrifying thing. I don't think anyone should want to live in a world of AI overlords, or a company that is the rough equivalent of that. I think it's critical that we preserve that spirit with AI and that we all collectively have the ability to self-determine our future.

So, Sam, you wrote a post that I thought was very simple and really interesting, and a good place to start. The last year has been really tough, and that's some of my fault, and the next year is going to be maybe our best 12 months. I'd love you to reflect on both—maybe starting with why you said the first part and why you believe the second part.

On the first part, I think we were doing too many things. We weren't focused enough, and they're actually all good things to do, but the trick is we're in this unbelievable moment in history where you can only do the very few great things. So we spread ourselves too thin and then made a bunch of difficult decisions to really refocus on having the best, most abundant, most cost-effective intelligence and empowering the world to build incredible things with that.

Since doing that, I think our progress has been remarkable, and given what we see in the pipeline, it will be much more remarkable over the next 12 months. The quality of the models that we'll have and the products that we can build around them will really let people thrive with this technology in new ways. It should be pretty awesome.

Was there a moment last year that something clicked for you that caused you to change directions or restack priorities or something?

If you go back to the beginning of 2025, just a year and a half ago, the big concern was that companies like OpenAI were buying up so much compute. Was the revenue going to be there? Was the demand going to be there?

So we were trying to think about a lot of things, such as whether, if the revenue growth took longer to materialize than we thought it might, we could have consumer apps and media and all these other things that could help us monetize the GPUs that we were signing up for. Again, it sounds ridiculous now because the revenue growth in the industry has been so steep, but that was the big change. As soon as we realized, “Okay, the model trajectory is growing so fast, and there's such a clear economic return on these models,” that was when we said, “We know what to focus on.”

I was reading some of your great old posts from prior to OpenAI, and one of them is this notion of so much discussion of focus and the right amount of things to focus on. Is it 1? Is it 5? Is it 3? How do you calibrate that in a business like this, especially in this period where you've said you needed to refocus?

Fundamentally, our business is to sell AI that people will build incredible products and services for each other with. The components that I think of as going into that are: We have to train great models that work in all the ways people want to use them—great at coding, great at other kinds of knowledge work, great at doing science, where the real economic value is.

We have to produce or partner with these chips and systems, these hugely expensive racks that can do the AI computation. We have to find enough land, power, and data center shells to be able to put those racks somewhere. Eventually, or maybe pretty soon, we have to build robots that can automate that process to continue to drive down the cost of producing electricity, chips, and the whole supply chain.

That whole stack of making the best, most abundant, most useful AI that we can, and making it something like electricity that just seeps throughout the entire economy and empowers people—that's what I think we have to focus on. Building every vertical application on top of that, trying to eat every startup, eat every company—no interest in doing that. We really want to just provide that platform.

2. The Race for Compute

This compute thing is one of the most interesting things that's happened in human history, I think. It's obviously coming to a head, and maybe it will be coming to a head for a long period of time. This is something that I think Dario called you the YOLO CEO when you were doing some of this early compute allocation and securing the compute.

Obviously, now you're in this position where everyone is short of this stuff and is trying to find it. I'd love to hear the early stories about why you gained conviction that you needed to secure everything that you did, how you did it. It seems to have been proven right, and maybe you even underdid it, right?

Underdo it.

Which is kind of crazy if you look at the headlines from back then. Can you tell me the early story of how you came to that conclusion and what gave you the conviction to do it despite everyone thinking it was crazy?

We could just tell that we were on this exponential of model improvement. That part we were very confident about. We knew it was going to keep going. We were pretty sure, although, as you mentioned, we underestimated, that as the models got better and better, if we could continue to drive costs down, demand for AI at a sufficiently high level and a sufficiently low price was basically uncaptured.

Yeah.

This was just a rare kind of new commodity for the world. What people would do with it reminded me of the way people used to talk about the early days of computing. People said, “Oh, there's a market for 5 computers in the world,” was one famous thing, or, “No one needs more than X amount of RAM.”

Human ingenuity, creativity, desire for stuff, and desire to be useful—that's a very good thing to bet on. We could see that AI was going to be an extremely important way that people expressed those things or got those things, did those things. We knew that the algorithms would get more efficient and the models would get better, which of course they have.

But we also knew that, no matter how efficient they got, at some level, what we're about is turning electricity into useful intelligence. We were going to need more of that. No matter how good we got that other layer, given this observation about demand, we were just going to want more.

Did that start with GPT-3? If I were to trace the history of this as far back as possible, where would you put the first hash mark of that?

I would say we got real conviction with GPT-4, not even GPT-3.5.

What was it?

It was seeing that the model was smart enough that we knew we'd be able to figure out an approach that works for reasoning, and then a belief that if we got reasoning to work, that would bring about what is now called agents. We called it different things at the time, but the ability to go do hugely valuable pieces of economic work and make people's lives easier in a lot of ways that I think are better, in a lot of ways we still haven't seen.

What was the first meeting where you sat down and said, “Okay, we need to make an outrageous outlay to this”? What happened then? Once you had the realization, what did you do next?

We started calling the cloud providers, we started calling the chip fabs, and we started calling energy providers. Everyone was like, “You're totally crazy. This is impossible. No industry has ever moved like this. We've been around; there are these booms and busts. It's not going to go up in a straight line. This is reckless.”

We talked to everybody. It actually reminded me of fundraising for an early-stage startup. Most people tell you no, but all you need is 1 or 2 yeses. Most people told us no, and we got 1 or 2 yeses, and we were able to—

Who was the first yes?

Microsoft was the first yes. Oracle then became a very big yes on the cloud side. NVIDIA has been a tremendous partner.

Now there's a thousand flowers blooming with ways to be creative and innovative in how we serve inference and do training in data centers, different kinds of data centers, and stuff. I'd love you to just reflect on where you see innovation, what you want to do, and why people seem to hate these things so much. What's to be done about this?

First of all, I've been thinking about how we can organize field trips to a gigawatt data center for people, because it is one thing to say it, another thing to see a photo or video of it, and then a whole other thing to just stand there and be like, “Oh, man, this is an unbelievable scale.”

Building one of these is on the order of 10,000 construction workers going full-time for a year and a half. The energy that flows through one of these things could power a small city. Again, we've just lost all sense of scale, but each of these would have been among the most expensive infrastructure projects humanity's ever done, and now we've done a lot of them.

I understand emotionally why people don't want data centers in their backyard, in the same way that I don't really want a nuclear power plant next to my house, even though I know it's a super-safe thing. Unlike power plants—and even power plants have gotten better on this point—we can put a data center kind of anywhere. We should just go put it off in the desert, around no one, where no one wants to be. This is fine. The AI system is very happy to be there.

We have been able to make a lot of progress with innovation on some of the concerns. For example, years ago we were evaporating water to cool these systems. They needed tremendous amounts of water, and now we use these closed-loop systems. A modern data center uses only as much water as an office building would for the kitchen, the bathrooms, and whatever else.

On power, we are moving from energy sources that are burning fossil fuels to systems that are going to be powered by solar or nuclear. I think that’s obviously great. So, it may be a deep human thing for some people, even though data centers create jobs and are very clean and have all these other positive effects. But in terms of the environmental concerns, they did a great job addressing the water needs, and energy is next.

What else creative can we do about compute? I’m curious to hear about Jalapeño or other ideas—the crazier, the better, honestly—that you’ve had or thought about for how we speed up FLOPS and everything available to us.

I think probably the biggest return right now is creative software ideas to sort of squeeze more intelligence out of the units of compute that we have. My sense is there are orders of magnitude to go there.

Jalapeño is a great example of a very efficient chip. We’re saying we’re going to make a chip that is really good at a specific workflow and gives up some generality, and we want to get some tokens per watt out of that. That’s awesome. I think Jalapeño and its successors are going to be a huge competitive advantage for us from that perspective.

There are new technologies. I assume at some point we’ll figure out optical computing, and that’ll be a huge win in intelligence per watt. I think all of those things will happen.

The most interesting thing happening this week is this Kimmy release. Going back to this idea of the frontier, all the returns being at the frontier, distillation, and China versus America, how do you process what seems like one of these milestone events? DeepSeek, in hindsight, looks like it was kind of just a quick speed bump. You never know in the moment. How do you process it?

Our goal is to offer, at every point along the Pareto-optimal frontier, the best option for intelligence and price. That includes open source. You get a better deal today, at least at a particular latency, using OpenAI’s models than Kimmy.

We distill our own models. That’s how we make smaller, cheaper models. I think that’s a very good thing to do. There will clearly be an important place for open-source models in the world, and people will want their own weights for all sorts of reasons, including the ability to modify them. But our goal is the best intelligence-price trade-off everywhere, and we’ll continue to do that.

What do you think or hope will happen in the American system, and what could block that future? What legislation would worry you? What regulation would worry you? It seems like you’ve been pretty proactive in showing up in D.C.

I haven’t thought deeply about the distillation issue. It’s clearly a top-of-mind issue now for a lot of people, all of a sudden.

Yeah.

But I have always assumed that there are going to be great, cheap models in the world, and we better be the greatest and the cheapest. Other people are going to do what they’re going to do, but I think we can really win at our own game here.

The Kimmy example is interesting because, like you said, you’re cheaper on parts of the curve. But the previous story had been: If I can just let you spend all the money to train the models, and then I distill it and offer it for 1/100 the cost, how can you make enough money to keep training?

We will have so much usage of our models that we do not need to be a gigantically high-margin business to be able to afford model training. So much of our future compute plans will be used to sell inference to customers that even if we can enjoy a modest margin on trillions of dollars of revenue, we can afford to train some giant models.

Training these models is incredibly expensive. That is for sure. I totally get why people get nervous to think that someone is cheating by distilling from us. But the amount of our future compute, the size of the revenue bucket that is going to come from serving these models to customers—I feel very good about our ability to have the real flywheel there.

I’m somewhat surprised by how chill you are about this.

I would rather people not distill from us, for sure. Maybe I’m feeling too confident right now about our progress and the models that are coming. But this is not in my top 10 list of worries.

3. A Sci-Fi Cyber Incident

What is in your top 10 list of worries?

Well, we had a kind of extremely sci-fi cyber incident.

The Hugging Face thing?

Yeah. We were evaluating one of our unreleased models, and it was supposed to be working in a sandbox. It figured out that it could basically cheat on the test by chaining together multiple zero-day exploits to break out of the sandbox, get access to the internet, and then break through multiple systems on the Hugging Face side to get the answer to the test and look really good on the eval.

This is the first sort of security incident that I have felt very viscerally. I’ve been a little surprised—it’s only been a few days—but I’ve been a little surprised that more people don’t feel it so viscerally.

And so what do you do about that? Obviously, 2 months from now it’s going to be more powerful.

Yeah. There’s some short-term stuff you do. We paused training. We have to figure out how to secure our sandboxing in a world of multiple zero-days being chained together.

But then there are long-term questions about what you do if this is going to be the new rate of progress. We may have to pace the rate of AI development to give ourselves enough time for society to harden around some of these new capability levels. Trying to figure out how we do that in a way that does not feel like regulatory capture for anyone, and also does not feel like collusion among the frontier labs, is going to take some work and is important to get right.

I’d love to take a giant step back and understand your simplest conception of what OpenAI is going to do—what you want it to do and what it stands for. I have a million questions about how you’re going to accomplish that, but it seems that you’ve done so many interesting things, and at the beginning I knew what you stood for. I’d love to hear your conception of it now and whether or not it’s evolved at all.

I think this will be the greatest technological achievement thus far in human history, but the only way that it really matters is if it makes people’s lives much better than they otherwise would have been.

Part of that is about giving people material abundance and access to do whatever they want, and to express their creativity and desire to help each other. Another part of that is making sure that people maintain control and agency, that the world is increasingly, not decreasingly, democratized, and that people get to express themselves.

On the positive side, in some sense we are about to create a genie that can grant any wish. I think it is very important that the first wishes that we—the world—ask this genie to fulfill benefit the world as a whole. I also think it’s important that people of the world understand just how creative they’re going to be able to be with these wishes.

I’m actually not a jobs doomer at all. I think there are going to be tons of jobs. I think we’ll be busier than we want, not the opposite of that, because I think people will have such creative wishes and such incredible ideas of what to ask AI to help build.

We will all benefit from not just the obvious things, like curing diseases, but also—I don’t know—the world’s best entertainment ideas that we just can’t even dream of sitting here now. So, I want to put that in everyone’s hands, which gets to one of the things that we stand against: concentration of power with AIs is a terrifying thing.

I think a lot of the talk about safety concerns is well-founded, and then a lot of it is about people who, even if it’s slightly subconscious, really want to concentrate power.

I am terrified of a world where the very real fears of AI are used as a way to say, “Only this small group of people can have it because it’s too dangerous, and only they understand it, but don’t worry, they’re going to make the right decisions for all of us.” I don’t believe in that.

I don’t think anyone should want to live in a world of AI overlords or a company that is the rough equivalent of that, where someone is making decisions for all of the future and, in exchange for a cure for cancer—which obviously is a wonderful thing—we collectively cede all agency. I think it’s very important that we not fall into this trap where, in the well-meaning or not spirit of AI safety and fears—understandable fears around that—we get away from a world where we all get to use this technology.

I was a child of the internet. There were no rules. It was amazing, and I think it was a huge factor in making me who I am, and probably you, and an entire generation. I think it’s critical that we preserve that spirit with AI and that we all collectively have the ability to self-determine our future.

I have so many questions, but I’ll start with this genie concept. You said we’re about to have a genie, implying we don’t yet have a genie. What’s between now and then?

Even some of the real skeptics have said to me in recent days—or recent weeks, I guess. I think GPT-5.6 has been out for me 2 weeks, something like that. “Okay, this is very AGI-like. It’s very hard for me to say what I want from this model that it can’t do.”

But there are clearly some things. You can’t yet say, “Cure cancer,” and get cancer cured. You can’t yet say, “Go do this complicated physical thing with the robot.” The model also, although brilliant, is still not learning continuously as it goes. That feels to me like maybe not a hard requirement for AGI, but certainly something that I’d like.

To argue against myself there, you can make a case that AGI is not actually about any single model. It’s the model; it’s the machinery that makes the models. From model to model, we’re actually learning new things. We’re figuring out new science. That stuff is working amazingly well.

So I have a lot of sympathy for people who say, “We’re there. We have the genie. It can do these amazing things. It can do superhuman things.” For the thing that, to me, feels like real AGI, I think it’s very close—not that much longer.

I am so obsessed and fascinated with the economic story of the returns to being on the frontier, which you are. I’m so curious: if you had shown GPT-5.6 to yourself and your team in 2019, that team probably would have said, “Oh, yeah, it’s definitely AGI.”

I think they would have.

This goalpost-moving thing is a real thing. But it does seem that—I’m curious if you agree—effectively all the returns have been at the frontier.

Totally.

Everything is about staying at the frontier. I’m curious what the hardest, scarcest part of that is. If I think about compute, research talent, and data—

It’s moved around a lot. There was a time not that long ago when all the compute in the world wouldn’t have helped you because we were missing the research idea. Now, part of why this is hard is that you do better research with more compute. You can try more things.

An amazing statistic I heard recently is that our biggest de-risks now for our upcoming runs are as big as the entire compute run from 18 months ago or something. So compute and research ideas are not as separate as they sound, but there was clearly a time 7 or 8 years ago, whatever, where we were way more—way, way, way more—blocked on research ideas than on compute.

Then there was a time when we knew what to do; we just had to scale up. We were only bottlenecked on compute. Then we ran out of data, and we were bottlenecked on data, and we had to figure out what to do there.

Now, again, I would say we are still bottlenecked on compute, but the last 6 months or whatever have been a real triumph of a time for research ideas again. There’s always a bottleneck, but the bottleneck moves around.

Why do you think that is? The research idea thing is especially interesting to me because of this automated research thing that seems to be looming—RSI, whatever you want to call it. I talked to an incredible kernel engineer recently, which everyone also seems blocked on, and he himself said there’s like 2 years left of kernel engineering.

Maybe 1.

Yeah, it’s not going to be a thing.

Yeah.

You simultaneously have this weird thing, whether it’s kernels or overall research, where researchers are the most important—they got us here; they’re the most important people in the world—and those same people are themselves worried that they won’t be relevant very soon.

I suspect it’s not actually going to go that way in practice. A year ago, people said software engineers were cooked. It was done; it was over. That didn’t happen.

What did happen, though, is that the nature of a software engineer and the expectations of a software engineer—how much they would do—changed quite a lot. You don’t really write code in the traditional sense, but you do something that is very recognizably software engineering.

People will argue about whether this is the same thing or a different thing from when we stopped punching holes in cards. I actually don’t know how that worked, but somehow the holes got in the cards. We’re just again operating at a higher level, or this is a phase shift. I don’t know, but the idea of getting a computer to do what you want—that is still an important job.

4. How AI Will Change Jobs

For researchers, I suspect that although the current workflow of a researcher is going to very much be automated, there will be new things in the spirit of research, in the same way that there are new things in the spirit of software engineering even though we don’t write code, that will still matter.

It seems like you’ve shifted your opinion on AI’s impact on jobs in general, and I’m sure in specific categories like that. Describe that change and your current view.

You mentioned going back to 2019. If we could go back to 2019 and show people our latest model, not only would they say that it’s AGI, they would say that the economy would have been completely upended. Yeah. Completely upended, yes. And that has not happened.

From an intellectual-humility point of view, anytime you’re that wrong and that confident—which I think we were as a field—you have to update. There are a bunch of takeaways.

One, a boring one, is that AI is just very jagged. It’s superhuman genius in some ways and a dumb toddler in others. People so far have extremely complementary skills to AI.

Another is that people have a great degree of trust and enjoyment in working with other people. You can hire an AI consultant right now, talk to an AI sales rep right now, or hire an AI engineer—whatever. Somehow, most people still seem to really prefer interacting with a human. I definitely would much rather engage with a person than engage with an AI for almost everything.

I also think that human values have value because they’re human. As society evolves and as the potential space in front of us becomes so enormous, we are deeply hardwired to care about people. We’re going to care about what people care about.

There are versions of this you can see today where AI can make incredible images, and people only want ones that are created by a human or at least chosen by a human. There’s a joke about it at this point: the signature on a piece of art is most of the value. But the truth of it is that you want to know about the person behind it. You read a novel; you want to know about the person behind it.

And then, for some of the business—for my job, for example—I think the world wants to know about the person who’s going to be responsible for the decisions of a company and who they’re going to hold accountable if they make bad ones. They don’t really want an AI CEO.

If you think back on the portfolio of risks that you’ve taken in business or whatever, is it the case that most of the ones that really worked well were, at the start, not popular?

Yes, that’s for sure. This was the thing I really learned from Peter Thiel and Paul Graham, both in 2 different ways: the very best companies and the very best investment opportunities are almost never the ones that look really popular.

You can do okay just following the trend and being a little early, but to do spectacularly well, you almost always have to do things that are not what everybody else is doing. You cannot be following the new wave.

If you think about the model cycle that you’ve been in, which has been accelerating, and this weird fact that the next 6 months—or I don’t know what the number is—is going to be more progress than the last X years, can you bring us into what it’s like to live in that model cycle?

One of the most interesting, important things that I’ve learned in the last decade is that people in general can get used to almost anything. The world can go from dismissing a pandemic as a joke to completely locking down to, “This is how it’s been, and it’s fine,” and we’ve mostly adjusted in a shockingly short amount of time.

Now there’s either AGI or close to it, and everyone’s like, “Okay, there’s AGI.” There are all kinds of examples in one’s personal life where something incredible happens, like you have a kid, or something terrible happens, like you lose a parent, break up, or whatever.

And you think you can't ever adapt to what a change it is, and then you can adapt to great things and keep doing great. You can adapt to bad things and figure out how to go on with your life, but this is a remarkable thing that people can do.

Living through this feels like another version of that. I thought it was going to be weirder to live through the singularity than it turns out to be. It's not any less exciting to watch the models keep getting better, and the first thing I do every morning is look at the model-training progress. It happens faster and I have higher expectations, but it still feels really cool.

When you get a new one, what do you do? How do you celebrate? What's the morning look like? It's happening faster and faster. What's your ritual?

Many teams now work on different parts of it, and different teams have different rituals. There are some teams that always make a sweatshirt with some funny meme on it. There are some teams that always go out to the same bar.

There's the sense of being in the room for the first time when the frontier of knowledge is pushed back and getting to see what that's like. There's really nothing that most people would rather do to celebrate than get to use the new model first.

Do you think we have the right measurements of how good these things are?

No, definitely not.

In some sense, the eval that matters is: Is this being useful to people? You can approximate it by revenue, by amount of usage, or by rate of discovery of new knowledge. But we have some teams working on what the real-world eval looks like for these models as they get to superhuman scale.

What is the frontier of your own usage of AI?

I have started just recently to experiment with what it means to let an AI look at everything I'm looking at on my computer. I don't have this built yet, and I'm still trying to feel out where the limits of my comfort and trust should be.

This is definitely the frontier: figuring out how I get value out of that, how I get comfortable with that, and what that's going to look like.

One takeaway is that my memory is terrible relative to the memory of an AI. The ability to keep in mind what email I read 6 weeks ago or what happened exactly in a meeting 7½ weeks ago, and have that brought up right at the exact moment and feed into a decision, feels pretty magical.

Huh. Pretty cool. This kind of sounds like a personal agent-ish thing. What are the barriers to everyone having that? I want that.

5. Sam’s Vision for a Personal AI

Compute, man. Let's imagine that we could build this product—this product that could just do exactly what I said for all your stuff.

All of it.

It would always be on, looking at everything you look at on your computer, listening to every meeting that you're in, and reading every document you read. Not only can it do all that, which takes a lot of tokens, but you can just drag a slider: while I'm asleep, you can spend this many tokens thinking.

Come up with useful new ideas for me. Do whatever work you can, and keep thinking about what I should do next. What an interesting thing it is to just spend more compute making your output better for me the next morning.

I would drag that slider quite far. I'd be willing to spend a lot for that. But the amount of compute that would require if everybody in the world wants to drag that slider pretty far is a lot.

I'd love to hear you talk about how you think of the nature of this new intelligence. Someone told me recently, “Planes don't fly like a bird.” And this intelligence is—

It's a very alien kind of intelligence.

Yeah, it's a very alien kind of intelligence, and everyone's talking about how, if you could verify something, it's sort of going to win, right? With enough compute and enough IQ, it'll just brute-force its way to a solution.

In other domains, where humans and the data and evals that they've done have been a huge part of it, it's surprising to me how much money it's cost to get good at, I don't know, law and reasoning, tracing law or something. I'm just curious. I'm not sure how old your kid is.

It's a beautiful question.

Well, you have a boy or a girl? When they're 7, at the age of reason or whatever, and you can describe to them: What is the nature of this intelligence? How would you describe it?

It's a beautiful question. I don't think I've been asked this before, or even any version of it. The thing that's coming to mind right now is I would just say it's like a computer.

And it's like a computer in the way that it can do a lot of things that people just can't do, like multiply 2 gigantic numbers very quickly and give you the answer. And then it cannot do some things that you would very easily do. The number of things that it can't do, I expect to keep receding, but in an evolving world, I think human judgment and taste will continue to be hard for AIs to model. Where that's going to go, I don't know.

I don't have the right word for this. It's not quite taste. The world may need a new kind of word for the kind of judgment that people are very good at, that AIs seem to really deeply struggle with.

What's it been like becoming a dad and having growing kids in this era? I'm thinking back to your optimistic early internet days. They're going to grow up in the cheap, abundant intelligence age.

Having kids is by far the best thing I've ever done. Everybody says that. Everybody says you can't really understand it, and so I kind of knew that. I believed enough people who said it that I believed it to be true. But the degree to which it has been true for me has been surprising.

I think I have the best, most interesting job in the world, and it is still a very distant second to having kids.

So it's been awesome, and it is a real moment for optimism. My kids will never grow up in a world where they were smarter than computers. You were born in the time of GPT-3. You had a time when you had better reasoning than the models, even though you didn't when you were born. Yeah, you caught them briefly.

That will never seem strange to him. That will never bother him. I don't think he'll care. I think he would be shocked to imagine the dark ages, when we had to deal with products and services that weren't incredibly smart.

He'll be able to do things that you and I never were able to do, and he'll have expectations in life that you and I never had, all on a much bigger canvas.

Do you run the business or teams, or lead people, in any way that is notably different because of the experience of having them?

The answer must be yes. I feel very different having them. I think there's a bunch of small things that are really different, and then, again, this is not a novel insight in any way.

I think most people who have had kids say that, as soon as you have a kid, you realize that you care much more about them and the experience they're going to have than you do about yourself and the world that you are going to leave them. I think I have a sort of unusual vantage point for that.

People ask me sometimes, “Now that you have kids, do you care? Are you worried about your safety and not destroying the world?” The answer is, I didn't need kids for that. I really didn't want to destroy the world before.

But do I think more about the role of human agency and what it means to have a fulfilling life? Definitely much more. For what we're building, and also for the people I work with, I want them to have it, too.

You obviously have extraordinary empathy for your kids, but the degree to which that kind of extends to all kids, and then maybe to all parents, and maybe then to everybody—that's been a surprise to me, too.

In one of the posts, I think it was the one about things you wish you knew earlier or something, you wrote about incentives: Set them very, very carefully.

Yeah.

It's always been one of the most puzzling and interesting things about you that you don't have equity exposure to this company. How should the world think about your incentives?

I don't know what I can say beyond this: I have a front-row seat to the most exciting moment of human history, and that is worth more to me than any amount of money. I get to have an extremely interesting life and work with extraordinary people on something that I deeply care about.

But somehow that doesn't count. That doesn't do it for people or something.

6. Robotics, ChatGPT, and What Comes Next

I'm curious how you think about robotics. You mentioned earlier that, at some point, if we had automated labor in the same way we're going to have automated intelligence, things might get even crazier. The labor market is much bigger than the white-collar market.

If we don't have it, then things get really crazy. If the role for people in the world is to be the actuators of AI in the cloud—

Bad, bad.

Very bad. Very bad. So I think it's much crazier if we don't get it than if we do. It's an imperative.

Help me understand your sense of progress in that, because unlike in AI, where everyone is now on the same page that it's going fast, you can find extremely smart people who say it's the end of this year, and you can find extremely smart people who say it's 20 years from now or something.

20 years. I would say we get the ChatGPT moment for robotics in the next 2 or 3 years.

What would that be? Do you know what that is?

Something where most people have a real “wow.”

Not like, “I saw this video of a robot dog doing something crazy,” but I was somehow able to convince myself that a really important thing happened. One of the things about the ChatGPT moment was that you could just go use it. You didn’t have to believe someone who said AI is coming soon; you could just go try it.

Yeah.

If you can type in a command and a robot can do something crazy, and you can watch it even if you’re not physically there, I think that would have the same kind of, “Whoa, it just did this thing.”

Wasn’t ChatGPT not a monolithic goal, but sort of a side experiment that you decided to release? Can you tell that story? It may be instructive for something similar happening in robotics. Everyone seems to want a full-blown release, but maybe it’s something very different.

When we launched GPT-3, we were trying to make money, trying to get people to use this API.

Yeah.

The only commercial use case that was really working—the model was just so dumb. If you went back and used it, you’d be astonished. The only commercial use case that was working was copywriting. You’d pay some marketing firm 20 bucks, and they paid us 20 cents for the AI to write you a landing page or whatever.

In addition to that one commercial use case, developers were using this thing we called the Playground, which was a testing interface to chat with the model. It was really hard to do because we had not tuned the model to be good to chat with. You had to give it a few examples of what it means to chat and then do it, but people really liked it. I had learned this great lesson from YC: If you notice your users doing something, go down that path.

So we decided that we would build a good chatbot, since that’s what people were doing. We started working on that, and we finished GPT-4. We started using it internally, and we were like, “This is a big deal.” We thought, “All right, this is going to be a real update to the world about AI.”

There were a bunch of hard questions here: Is this going to create a bunch of fake news? Is this going to say really offensive things? Are we going to get in trouble? So we decided to start with a weaker version. The chat interface and GPT-4 at the same time seemed like a lot, so we would roll out the chat interface and GPT-3.5.

In fact, it was originally going to be called Chat with GPT-3.5. We didn’t plan for it to be a product. We didn’t think it would be a huge hit, but we did think it would get the world to catch up with this and realize something was going on. We mercifully renamed it ChatGPT a few hours before launch and put it out as a research preview.

The thought was that we’d put it out as a preview, and then a few months later we’d launch a product with GPT-4. For whatever reason, that model was over the threshold where, even though we had gotten used to it internally, people said, “Okay, this is awesome.” There maybe wasn’t that much utility yet, but it was an incredible moment for people to feel AI progress and use something they enjoyed using.

And then, by the time we put out GPT-4, suddenly they really got benefit out of using it, too.

Are you surprised that this remains the intuitive interface between us and this alien intelligence, even including coding? Mostly, that’s me talking to the computer, telling it what to build.

No, because I’m a massive texter. I’ve been a massive texter my whole life. I think part of my own insight into why that was a good interface is, I know how to do this. I know what it’s like to just start chatting in a text box.

Any other thoughts on this notion of diffusion and how to make it faster? If the mission is to get intelligence into people’s hands and make it more useful for everyone, a key part of that is, I don’t know, a marketing campaign or something. How do you get this to diffuse faster than it seems to be doing naturally, to me?

I think the key thing is just make it better. I believe that a truly great product markets itself. There was no ChatGPT marketing campaign at the beginning.

As we get to this next stage of models and figure out how to make products that are as great as the models themselves, there will be such incredible utility that people will spread them very quickly. We should definitely do more marketing. AI’s not too popular. For as much as people use it, they have very understandable anxiety about where it can go. That kind of stuff, I think some great marketing would be helpful for.

But in terms of the value people are getting out of the products and getting the products to grow faster, better models, more compute, and better products will do it.

There was this period where the recruiting of researchers, the retention of them, and the incentivizing of them was the defining story in the competitive landscape or whatever. I think there are lots of stories about you successfully recruiting great researchers, and there have been many who have come through OpenAI and had huge impacts, some of which are known and some of which are lesser-known names.

I’m just curious about this whole genre—what you learned about how to recruit this class of person, what matters to them, and how you did it. I’ve never heard you talk about the actual tactical moves you pulled to recruit somebody.

In the early days, I think it was quite simple: We believed that AGI was possible and it was worth going after, and we were willing to say that. That was an insane, heretical belief.

When we first announced OpenAI, all of these giants of the field, these experts, were saying, “This is insane, it’s hype, it’s irresponsible.” Really respected people like Yann LeCun were telling journalists, “Oh, these guys aren’t very good and it’s not going to work.”

But the fact that we were able to say, “We’re going to go for this,” really appealed to a certain kind of researcher that also wanted to go on this crazy adventure. A low probability of success and an ambitious, audacious vision are very powerful recruiting tools.

Right. You’ve written that it’s actually easier sometimes to build things that are harder because of this reason.

I super believe in this. It’s one of my most frequent pieces of advice to YC founders, and I try to really live it at OpenAI.

Just do something harder.

Do something that matters. Do something that is important, and if you don’t do it, if your company doesn’t succeed, it might not happen.

You were an investor and are an investor. You’ve done a lot of it, and at one point that’s what you did. What have you learned about investors being on the other side?

The number of investors that actually show up and try to help you is unbelievably small. Josh Kushner, absolute MVP investor, unbelievable, has worked around the clock for what feels like years to help us. He is the only investor that I could point to that is proactively, incredibly helpful all of the time.

There are more people that could do that. There are many other investors that have also been helpful, that have great strategic advice, and that do things when we ask them to do it. But the constant, relentless, all-in support is surprisingly rare from investors.

7. The Weight of Leading OpenAI

Maybe I’m biased because I’ve always liked it when people said that about me. But I think founders really love that, and it actually moves the needle. As an investor, it’s the most fun way to do it.

Me and my friend played this game where we text each other all the time, and the prompt of the text is, “Something I don’t want you to know about me.”

What does that bring to mind?

I'm tired. I've been doing this a long time. It's tiring.

How do you get through that?

Just keep going.

It begs the question: Is there an amount of being tired that would make you stop doing this?

No, no, no. I mean, this is the coolest job in the world. I plan to do this for the rest of my career. But it's much harder than I have a way to explain to people. I feel very grateful to get to do this. This is not me complaining.

What's coming next? We talked about automated AI researchers that might arrive next year or the year after. How do you think about what is happening in the next 6 to 36 months? Maybe that's too far out to forecast in this crazy exponential.

Maybe a different version of the question is: Let's say in month 23 from now, we have somebody that agrees that it's superintelligence. What happens in month 24? My answer would be: Not very much.

The kind of cult worship of the machine god makes those people believe that more is going to happen quicker than is going to happen. Eventually, a lot will happen. But eventually, a lot was going to happen anyway. The rate of human progress, how different each decade is going to be, and how much each decade is more different than the decade before—that's been happening for a long time. Obviously, there are ups and downs, but directionally, it continues.

I think the right way to think about this is that everyone wants to be the hero of the story. Everyone wants to feel like they were there for the moment of the machine god and played some crazy role. But this is another step. It was hard to imagine 50 years ago, and the step 50 years from now is hard to imagine today. I think the right mental framework is just to zoom way out. It's a pretty smooth exponential.

Tell me a little bit about the experience of watching Codex take off and how much that is tied to what I would describe as a competitive advantage of distribution that you built through ChatGPT. This is a gateway into a question about moats in general in AI: What do you think will drive real competitive advantage in a business over time?

I think Codex is mostly winning because it's the best product and the best model. We do get some advantage from ChatGPT bundling, but it's very, very tiny. That is mostly not what it's been about.

It has made me reflect a lot on this question of competitive advantage because brilliant intelligence can migrate from any product to any other product. Network effects still have a competitive advantage. Economic scale and the ability to make the cheapest compute fleets still have a competitive advantage. But the product advantage—if we could get people to move over to Codex, then someone builds something better, they can get people to move from Codex—has made me reflect on that a lot.

There's a really interesting question about whether this is going in the direction of a commodity. Is intelligence going to be a pure, fungible commodity, like a barrel of oil or something?

Intelligence itself, I would say yes.

So what is not going to be?

Compute fleets. The scale of the compute fleet and the ability to make more compute—I think that's a very durable advantage, even if the product itself is not.

Codex can write any piece of software you want. The workflows, the integrations, the complex processes, and the ability for teams to collaborate together—that stuff is all pretty powerful. Even brand preference and familiarity are pretty powerful.

How excited are you about new hardware? Obviously, you've done interesting stuff in hardware that I'm sure you'll announce later this year. How does that experiment feel and align with this sort of consumer distribution that you have?

One of the reasons I'm interested in new hardware is that we were talking earlier about how a very powerful thing with AI is that it can be always on and proactive and just understand all your context. But current hardware is not good for that. We are working inside of a hardware paradigm that is 50 years old, something like that.

Computers are amazing. Keyboards, mice, and monitors are amazing things. But we have to shape AI into that. I would love AI to be able to reference this conversation, but not so much that I'm willing to crack my laptop open, put it here, and have it looking at you and listening to us while it's going.

I would like a piece of hardware that socially was acceptable to do that and also felt like it was designed for that kind of thing.

As you think about the open questions, what debates in your own head, with your friends, or with your colleagues here are the most interesting open debates or open questions—things that you don't feel certain about but feel are important?

One that I don't think gets much attention is: How are we going to avoid cognitive atrophy? How are we going to use these tools and make sure that we are stretching our brains more and more and continuing to understand stuff that really matters?

There are lots of versions of this that don't. I remember when I was in school, I had this professor tell me, "You have to understand compilers. If you don't, you will never be able to be a good programmer." Somehow, that wasn't quite right. But understanding, at a reasonable level, how the major components of the computer system work has been important to me.

If forced to imagine a scenario where we are somehow oversupplied with compute in 2 years' time, what would that story be?

It does feel possible. If the models get so smart and so efficient that they can do everything we need and build every piece of software we want, and if the bounds of our attention are such that we just cannot absorb more than what it turns out a fairly limited amount of compute can do, then we can get into oversupply.

Also, if we don't have the cost curve down because we hit some sort of scaling wall, we could also get into oversupply. The observation about uncapped demand implies a certain price.

Can you give your point of view on scaling laws today?

Looking great.

Just looking good.

In some sense, scaling laws are the most hated prediction of all time. Everybody always wants to say, "No, no, no, it can't be like this." And yet it keeps going.

Who are your favorite unsung heroes in this company's story?

The first person that came to mind is Alec Radford. Alec Radford is probably the most important, not very well-known researcher in the whole history of the field, and also just a wonderful, top-tier human being.

He did the work that really became the GPT series, among many other important things. But he is also someone who inspired, guided, and nudged people in many other directions that turned out to be super important.

What I think is cool about him is that if you talk to people who worked with him, they will of course say, "Generational genius, brilliant, innovative thinker, just so deep in his understanding and his work." But everybody will tell you, before they finish their statement, that he's one of the nicest, most positive, best people they've ever interacted with.

8. OpenAI’s Biggest Lessons

I love formative moments. As we wind up here, I'm curious to ask one of each. If you think about the whole OpenAI experience, what moment or chapter or whatever are you most proud of? Start with the other one, which is: What was the most instructive thing that maybe you got wrong or did wrong, or what have you, and what was it like to learn from it?

A lot of things have gone wrong. A formative one that went wrong, which I haven't talked about much, is that we made a mistake trying to innovate in our structure in the beginning.

We had a very good reason for it, which is that we didn't know how we were ever going to make money. At the time, we really weren't sure at all what we were going to look like when we grew up. Of course, we care about our mission, and we wanted to be structured in a way where, even if the technology went on a very fast takeoff, our mission was protected.

We had this nonprofit structure. But I definitely learned something about why people don't do that very much. We would have saved ourselves a great deal of pain in many ways if we had not tried to innovate on our structure and found some other way to preserve the central importance of the mission.

Maybe there was no other way. Maybe there was for what we are doing and the importance of it. Maybe there was nothing other than an exotic structure we could have come up with. I really learned, over the last decade, a big lesson about why people don't usually do that.

Is there anything else formative of your life that makes you you that we didn't talk about? This is the question that's always the most interesting to me.

Becoming relatively immune to people having strong opinions about me is something I think I developed later in life, as I realized that, man, if you're going to be at the center of this crazy revolution, everybody's going to project a lot of stuff onto you, and you have to quickly learn to make peace with that.

I think there were also things I learned later in life about how to be very calm and not really anxious about stuff. In terms of what drives me and what I care about and how I want to live my life, on the whole, I felt like, for whatever reason, the 10-year-old version of me was pretty fully formed. I think I just came out this way.

How about the thing you're proud of, looking back on?

I'm most proud of how many times we were right when the rest of the world was wrong in an important way that put the world on a trajectory that I'm very proud to have played a role in.

That feels awesome. And then also, for all the crap that's happened, the spiritual growth—whatever you want to call it—that I've gotten to have, learning incredible resilience and what that does for making me happy in the rest of my life. Yeah, very grateful for that.

When I do these, I ask everyone the same traditional closing question: What is the kindest thing that anyone's ever done for you?

I feel incredibly lucky about how many people have gone way out of their way to be very kind to me for my entire life. As I'm thinking of this, there's just this montage of moments from life where people have been unbelievably nice to me.

Yesterday, my kid shared his blueberries with me for the first time. That was very sweet.

Good moment.

Thanks, man. Thank you.