Today I'm sitting down with Sam Altman, the co-founder of OpenAI. A long time ago, you gave a talk at Stanford focused on how to start a startup. What is the biggest thing that has shifted over the past 10 years since you initially did those talks?

Clearly AI, and what is possible to do now with a small team of people, and the speed with which you can do it. And also, not only what's possible, but what you have to do to be competitive given how much the world has shifted, feels totally different. It is amazing to me what a 10-week-old startup now can look like. And also, if you look like a 10-week-old startup from 10 years ago, you sort of are in bad shape.

If you think of a 10-week-old startup in your mind today, what is the best example of a startup that moved extremely quickly?

I don't even know what they're called, but I met what will be a startup that is like two weeks old, something like that, that has an entirely redone office productivity suite. It is all made for a world where AI has got to be able to use documents, presentations, spreadsheets, whatever, as a first-class consumer. I felt like it would have been a year's worth of work for a startup pretty recently.

I would assume that if the barriers to starting startups and the tools out there become more proliferated, you would assume that harder startups are easier to start.

What it means to be a hard startup is changing so incredibly quickly that I don't think I have a perfect mental model for what, over the many years that it takes to build a very successful company, are going to be the really hard and really valuable things.

I've heard a lot of people say, "Well, anything in the physical world is extra valuable right now because software is going to become free." But it won't be that long until robots get really good, and a lot of the expectations of how difficult it is to make rockets or whatever may change quite dramatically.

I love times like this. I think startups have the biggest edge when the ground is shifting the most, and when costs are coming down rapidly and cycle times are coming down rapidly. That's when startups really just have a massive inherent advantage. And that's happening in so many places at once right now that it seems like a great time to be doing startups.

And yet, most startups are like, "I'm going to go build AI agents for enterprise vertical X." Now, that actually will work in a lot of cases. It'll be very competitive, but it probably won't be the most successful, defining startups of this era. But it will work.

But given how much the landscape is in flux, I am surprised there are not more people doing the crazy thing with this completely new set of tools, and also being willing to truly internalize the fact that scaling laws are going to continue, and planning for the things that are not possible or economical this month, but will be possible in two years, four years, whatever. So, it seems like an unbelievable time to be starting startups, and there's very fertile ground. But there's also—and no judgment if this is what people go for—this incredible temptation to just go apply today's agents to the easy wins, and I get it.

One thing that you've said throughout the years is you basically try to, whenever you meet someone new, plot where they are on your mental model of who they are as a person. And then the next time that you meet them, you basically want to figure out how far or how fast they progressed. Did that sort of thing—doing this across maybe thousands of people—impact your ability to say, "Okay, the model is this intelligent today, it's this intelligent three months later. Suddenly you can plot this out, and it makes it easy to believe in the growth"?

I think there's a general thing behind both of those, which is that I developed a great trust in exponentials, whether in people, companies, or models. Subjectively at least, it doesn't feel like watching model progress is the same as watching a founder develop, but I think the underlying belief system about the world is the same.

If I were still advising startup founders, this is the most important thing I would try to get them to wrap their heads around. And it is evidently hard for the same reason that there's free money left in the market for betting on high-growth young founders—the market has still not adapted enough for that. I think the market has also still not adapted enough for the exponential of model progress. It is okay to start working on things now that require smarter or cheaper models.

This idea of constantly shifting chaos, how do you get good at it? How do you get good at operating in a chaotic environment?

I think it's just practice. There are a handful of things that, no matter how much you intellectually understand, take a lot of reps to emotionally be able to handle. Operating in a lot of chaos, trusting that you'll figure it out and it'll be okay, knowing this is not going to be the thing that kills you, and accepting that you don't quite yet know how you're going to solve it but you're going to figure it out—this seems to me to be something that you can only learn by going through it.

And I actually think this is a real weakness of young founders, in that they have not had the career experience of getting to emotional peace with this. So they have a very hard time with it in the early days, and then they eventually learn it, but at great pain, cost, and unforced errors. You definitely will, if you end up in a high-stress, impactful job, learn how to deal with chaos and function very well through it. At least most people will, but I don't think it's teachable. I think it's only learnable.

You had an interesting thing that you said a long time ago, which is basically like the first time that you experience a company-killing event, it feels like the world's falling apart. And then you make it through, and by the tenth time, it's not nearly as bad, and you're like, "Well, I survived the first nine, this one's probably not that bad."

You also had this other idea where you said at some point, I think it was like a year or two ago, you came to the conclusion that shit was always going to be going wrong, and so you had to internalize the enjoyment of experiencing pain or not-great things happening. How did you make that transition?

I was reflecting earlier when you were saying this about how, when I would do office hours at YC, I could always tell who the new founders were because of the way that they reacted to stuff going horribly wrong versus the more experienced founders. I often didn't know how new a company was, but I could tell if it was a founder in the batch versus a founder that had been doing this for a couple of years by their emotional state when they talked about a problem.

I think most people think about the opposite of a bad experience as a good experience, and they would rather have the good experience because it seems like more fun or more pleasant. But if you think about it, the opposite of a bad experience is no experience, and at some point in the not-too-distant future, you'll be in no-experience land. Then you can be grateful for the bad experience too.

Naval Ravikant used to say this thing that I loved, which is if you had a fast-forward button on a remote for your life, your life would be over. And so, the boring parts and the bad parts are much better than no experience. It's all part of the interestingness and the emotional depth and range. I find it fairly easy to be grateful for the bad days.

When you think about hard problems and what rallies people to get excited to go work on them, how do you decide internally at the company which things to go tackle and when? And when do you want to focus on just your core competencies versus expanding the scope?

I think a clear mission and a deep understanding of the problem together do a fairly good job of pointing to what you should do. You won't get it all right. You'll still, at some points, overly expand or not be ambitious enough.

But we are very focused on: this is going to massively empower people. It's going to be choppy but wonderful, and it's extremely important to us that power in the world get more decentralized and more spread out. In fact, one of the biggest AI risks I am worried about right now is AI authoritarianism—a small number of people or companies thinking they need to control the world. That'd be very, very bad.

Given this mission, we feel like we need to figure out how to make AI extremely abundant, extremely cheap, and extremely powerful, put it in everyone's hands, and let them use it. As we look at what's in the way of that, there's a whole bunch of new constraints like chips, energy, data centers, robots, and all the pieces that need to come together to build this platform so that we can enable this mission.

Now, there's a bunch of things you can build on top of that platform, like every vertical and every startup. We have no desire to do that. I think a decentralized economy is important and good, and I would like us to just be really great about producing these units of intelligence and making them an incredibly great deal, so people can imbue them in every product and service. That's very instructive to us about what we have to go do.

Interestingly, the key inputs to extraordinarily abundant, high-quality, low-cost intelligence—a lot of those things, like energy and robots—are also things that you really want in a world of abundant intelligence. If ideas are plentiful and there's all this great stuff, we still live in a physical world and we still want stuff to happen. So, we need to be able to make stuff happen in that world.

I have wondered if the fact that energy and robots are going to be so important to staying on this intelligence infrastructure ramp—and also the things that you need immediately after—says anything deep at all, or if it is just the most obvious thing: that to produce anything, intelligence included, you need to manipulate matter. But it is kind of interesting that we need these new areas so much for what we're doing now, and then so much more immediately after.

What do you think your biggest bottleneck is? If you just say, "We want this to continue scaling unabated," what is the biggest bottleneck that you can see?

Transistors and then electrons, in that order.

Let's say you take someone like Jensen, where he's incredibly good at aligning all his suppliers towards his vision for the future. I think you're also extremely good at this. How do you think about not only keeping OpenAI on its timelines, but also keeping the rest of the world on OpenAI timelines as well?

You talk to them a lot. You have to, and you don't just say, "I need you to deliver this turbine or this chip on this date," because people will just say, "Okay, whatever," and they have a bunch of other priorities.

You really have to show them: here is the upcoming model, here's what it enables, here's our research. For our core suppliers, we really show them a lot about what we're doing, why we believe what we believe, and what it's going to enable. The most important thing is to get them to believe in the mission and why they need to prioritize it, but you also try to figure out how to align their incentives with yours as much as you can.

Charlie Munger has this awesome line where he's like, "Every time I've thought that I understood the power of incentives, I've underestimated them." How do you successfully align all the incentives of the different partners, as well as the people inside of OpenAI?

Well, companies are amazing vehicles for this to start. As a kid, I was very fascinated by the Industrial Revolution, and I understood it as a bunch of technologies that happened to come around and scale at around the same time. I would talk to my friends or my parents about whether this technology or that one was the most important invention. My read of it now, from my current lens as someone who has now spent more time thinking about business, is that the important invention was the joint-stock corporation.

Before that was invented, you had these family businesses that ran on trust, where you had to know everyone, and you didn't have the idea of a corporation or stockholders. And then all of a sudden, this new thing was invented. The sovereigns of the world granted this new kind of status, a powerful thing that didn't exist before: we're going to let these new entities emerge and give them much more than the power of a person, though not the power of a state.

Fundamentally, this was about incentive alignment and liability protection, which is another kind of incentive alignment. It allowed them to amass capital, and all of a sudden you could do things beyond what a family business could do. That environment let people raise money to develop technologies of a highly speculative nature, pool capital, figure out ways to get different companies to specialize in different ways and interact with each other, and create serious financial systems and new kinds of instruments. This was really an incredible thing.

The idea of a company has done a huge amount to align incentives of very large groups of people. Now, millions of people can be shareholders of a company. There's a bunch of other things we do, but I don't want to diminish what has been this incredible invention and the ridiculous overperformance of capitalism in human society.

A chart that I think people should look at much more than they do is the fall of extreme poverty over the last hundred years.

Like infant mortality and all the other associated things.

Any of them that you want. If you look back, that's the most zoomed-in view of the last hundred years. But if you look back at economic growth, quality-of-life improvements, and all of these things, and you drew out all of human history and put a line where the company was invented, you would see a really quite interesting change in the shape of that curve after it. Beyond that, I think we have a mission that transcends any amount of economic power anything else could ever do, and that's probably more important.

One of the things that I've thought is the most interesting... Have you ever seen one of those stats where it's like what percentage of CEOs are sociopaths and psychopaths? It's very, very high.

People talk about this.

So I wonder if capitalism is the first system that allowed these people, who are super self-serving, to have their incentives realigned with society's, creating great products where suddenly you help society but you also win.

I definitely know a lot of sociopathic CEOs, but I would say the best ones I know, I would not put in that category. There's a lot of very high-ego CEOs—people with incredibly high opinions of themselves in a way that maybe would have been purely annoying, but now because people can buy their shares, maybe you tolerate their annoyance and antics a little more. So there's definitely some aligning thing there. But I think there are a lot of CEOs of very large companies that I would not put in the sociopath category at all.

What do you think drives those people?

It's different for different ones of them. Definitely seeing how good they can get at the game, or how much they can improve relative to themselves every year, is a huge part of it. It's intellectually quite stimulating.

Mm-hmm.

For all of the other problems, you're playing the most interesting strategic game, and that can be fun.

If you had to think about the way that you operate today versus the way that you would have operated 10 years ago, if placed in the exact same position that you are currently, what are the biggest differences between those two people?

Well, to the previous question, the thing that drives me the most is that this is the most interesting, important thing I can imagine doing, and we are now in the singularity. This is the moment.

Ten years ago, this was a far-off dream at best; it seemed very improbable. Now we're actually in the moment that we used to talk about at the lunch table in a very non-serious way. So, the thing that drives me is I've been waiting for this my whole life, and I think it's going to be incredible, hugely positive, and awesome for the world. I'm excited to get to work on that.

I also think some of the alternative visions painted by other companies are quite terrifying. I'm going to make sure that gets pushed against and is not what happens.

But the main thing that's different than 10 years ago is that we're actually in it. This is real. I think it is both true that it is all one crazy exponential and any one moment is not the tipping point, and also that we are somehow in another one of those decisive periods where the curve can go one way or another, just like it was when we started 10 years ago.

What do you think the biggest impacts are on getting this part of the curve right?

I think there are still major AI alignment issues to solve, as well as safety issues, major economic issues, the future of jobs, and all of that kind of stuff. But I think the fight of the current moment is: are we going to head to a world of AI authoritarianism or liberty?

Are we going to decide that, because of the very real safety issues and economic issues, we want one single model to be the machine god and the company associated with it to control it? Or are we going to say, "You know what, it may be a little messy, but every time humanity has traded off its liberty for safety, it's been a long-term net loss, so we are going to put this in the hands of people and empower them"? We are going to let society express its ideas and use this technology in the way they want, with some necessary guardrails, but with a lot of power and potential. I really care that we do it that way.

When you were talking with Patrick Collison on stage, must have been a couple of months ago, there was this fascinating thing you said, which is basically that you talk or text with 300 to 400 people a day internally.

I text more people than anybody else I know. I don't think this is a good thing; I think this is actually a bad habit.

Why?

Because it takes up most of my time, and I have a very short attention span now.

Interesting. What does that enable you to do that you otherwise couldn't?

A huge amount of context. I mean, I'm like an AI model that's not that smart but has a massive short-term context on whatever is happening right now, and then I forget.

Does that enable you to make decisions that other people wouldn't have, or make company decisions better?

I don't know. It lets me make different decisions than other people would.

What different decisions have you made in the last 12 months that you wouldn't have made if you didn't have so much context?

I don't know if I have a specific one I can share publicly, but there are a lot of things where, because I know about some random research breakthrough impending that is going to affect some customer or supplier, I can make a slightly different decision in the moment. If I hadn't known it that morning, I would have made a different decision, and it would have been worse.

One of the things that I think you are the best at is long-term thinking, and yet it also seems like things are moving so quickly and the world is changing so fast that planning 20 years out and looking backward doesn't even make sense. How do you think about this? Is your time horizon shortening?

I actually don't try to plan backward. I try to have a small number of strongly held convictions about the future that are directions to head toward, but then I try to plan forward from the current state: what can we do now, what can we do this year? Sometimes you do have to plan a few things on a 5- or 10-year horizon, but I try to plan forward, guided by a small number of beliefs about the future.

I think there are a lot of people who have way too many beliefs about the future and a rigid worldview that they try to fit. What happens is they end up chasing—you see space companies turn into AI companies or whatever. Having just a small number of things that you deeply believe about the future and being very flexible on the rest while staying very true to the core is helpful.

One of the people that I know the best, his top company core value is "critical path"—focusing on the critical path and staying focused on the key drivers of whatever the biggest roadblock is. Just unfuck that, go to the next one, and do that again. How do you think about the critical path in your own life when you're trying to make these decisions?

For so long now, I have felt focused on this singular goal of abundant intelligence, and a belief that incredible human prosperity will come from that as long as we don't have a weird power concentration and a new kind of authoritarianism. It has been fairly clear to me at any given time what's on the critical path to get there. I have not been tempted to reconsider the goal or think about other things.

I have been thinking about that a little bit more recently: okay, if we really are close to superintelligence, what's next? But the fact that I am thinking about what the next thing is is the first time in more than a decade that I've had that thought.

And is it the ranch?

Eventually. There might be a few more things on the way, but yes, that is the eventual plan.

If you had to try to guess, what else does Sam want to accomplish before he ranches?

Look, AI superintelligence is going to get built. That's not done, but we are on the glide path to get there.

Abundance, decentralization, wide access to it, and making sure that it ends up with broadly shared prosperity—I would really like to accomplish that. So maybe that's the thing that I've been thinking about as the next step. Technologically, we're going to accomplish our mission. I think a lot of our worldview and a lot of our beliefs about safety and the impact it's going to have—we've gotten those things down too. But very broadly shared prosperity and a real focus on enabling the world, that seems very important.

I've heard you say again and again this idea of getting on planes in marginal situations. I've done this a huge number of times because of you, and so far it's working pretty well.

Good. I think it's very solid advice.

When was the last time that you got on a plane in a marginal situation?

Well, I can't say what for, but very recently, and it worked out.

Congratulations.

Thank you. And I really didn't want to. It was a very inconvenient trip with two overnights. I didn't want to do it—new baby, whole mess—but I did.

Take me into your mental process for deciding to do this world tour. It's hard to rewind three years because now it feels like AI has been here forever, but three years ago we had just launched GPT-4. The world was melting down, everybody was freaking out, and no one knew what to think. I could just feel these storm clouds brewing, with people getting angry, and world leaders asking, "Do we need to take control and shut things down? Is this thing waking up?"

It was a weird time, and because it had all happened so fast, people had not had time to wrap their heads around it or go through the process. So we were getting all of this escalating stuff, and world leaders were asking, "Will you come meet us?" and we were thinking about doing this and that. I felt that if I or someone from the company—but at the time, the company was not very big, so it was probably going to have to be me—did not go show up and talk to people, it was about to go very badly.

I was thinking about doing a bunch of short trips, but I hate long flights and jet lag. So I decided to just get it over with in a short period of time. Brian Chesky advised me to do it; he had done something similar for Airbnb, but he had done like eight cities.

And you were like, "Fuck it, we ball. Let's just do 30"?

We did 28 countries in 35 days. I lived on an airplane. It was very strange.

Did you have a bed?

I was comfortable, yes, I had a bed.

I fly coach, so it's a little bit different.

This is better than that—a lot better than that. But travel still sucks. You can make it as comfortable as you want, and you still miss your own bed, your own time zone, your office, and whatever.

One thing that I've learned is no matter how many people have cures for jet lag, they're all BS. It just sucks; your brain doesn't like it.

Because I went to so many places, my biggest single time zone change was only four hours the whole trip. Mostly, it was a one-hour-at-a-time hop, so I wasn't that jetlagged. I was just exhausted.

It was cool to get to see the world so quickly because you really do get a sense of subtle cultural differences as you go from place to place. But by the end of it, the last few days, I started having this weird, half-asleep, half-awake feeling where I envisioned being in my childhood bed.

Interesting.

Which I had never experienced before or since. But I would sort of vaguely wake up and think I was in my childhood bed in my childhood room. I was like, "Okay, this is some deep subconscious message that it's time to go home."

Your subconscious fighting back or something.

Something like that.

Interesting. When you're in zombie mode or goblin mode, whatever you prefer, and you're trying to function, how do you change? What is different about you, and how do you decide to keep on going?

I definitely got to a point near the end where I was counting down the days until I could be home. Each day was like 14 hours of non-stop interaction, and I'm not an extroverted person. I was meeting hundreds of people a day in some cases.

Tensions lowered by the end of it, but at the beginning, the world was very nervous. Once things calmed down, it helped that I knew when I was going to get home and that I wouldn't have to do it again for a while. I've done smaller trips since—I probably do something like that once or twice a year.

But one thing I learned is that I would much rather put a bunch of international travel together in a 7- to 10-day chunk than do a bunch of one-day trips spread out. So I always try to do it that way now.

I think a lot of people overestimate the risk involved in taking most actions. There is a lot of risk if you're just buying a bunch of call options on Robinhood, but there's probably a lot less risk in getting on planes and stuff. What have been the best examples of times where you did something that seemed really risky to other people, but you knew or believed wasn't?

I guess the obvious example would be that I was always extremely high-conviction about buying a lot of compute.

Have you ever seen the meme, "Everything's high risk if you're a bitch"?

No, but that's a good meme. [laughter]

I spend a lot of time trying to talk to people about why they think a specific decision is high-risk or low-risk. I have found that just getting people to speak about it out loud will often help them get through their intellectual blocks in either direction, because people are usually wrong one way or the other—not always just on the emotional part of it, but sometimes on the logical side too.

If you had to think through one of these meetings where someone is firmly against a decision that you want to make, what is the most effective way to help them see what you see?

I don't think I'm very good at this, to be honest. I think the right answer is to spend a lot of time really trying to explain it and get people there. I actually think I used to be better at that. As life has gotten so busy, I'm more prone to getting upset or frustrated and saying, "This is what we're going to do." I don't think it's a positive trait, and I would like to get back to more of the "let's really talk it through" approach.

You said that you used to be much harder to work with back around 2008, 2009, 2010 than you are today.

I don't know if people would describe me as fun to work with. I think they'd say "interesting to work with," "effective," "ambition-raising," or "very mission-focused." But day-to-day, really fun to work with? I don't think I'm a particularly fun person in general.

I feel like if you are having a good time, it tends to be infectious. And if you are working on things that you want to work on, you're having a good time. Maybe I'm wrong.

You could go ask a bunch of people I work with. I don't think you would get a resounding "fun to work with." Maybe not that language. I think people would say stuff like: creative solutions to problems, new insights, a very high level of ambition, and getting them to do things they didn't think they could do. That kind of thing.

What does it typically look like when you go into a meeting with someone and realize they are not ambitious enough, and you need to help them become more ambitious?

The version of this that came to mind was the first office hours I would have with new YC founders. You take the average person who has worked in corporate America, at a big tech company for a few years, and their level of ambition, scaled thinking, and self-belief is catastrophically, horribly low.

You realize that this person has never not had a boss in their life. They have never not had a parent, a teacher, or a manager telling them what to do or what they were allowed to do. They also got punished for being too creative, wanting too much, or thinking too big and being too ambitious. Most countries or cultures have some sort of phrase for being too ambitious.

Tall poppy syndrome.

That's one. You realize how deep this is in people, and that they heard it from their parents, teachers, and friends from when they were quite little. Getting people to not think that way takes real time.

What's the most effective path to get someone from not believing in themselves to believing in themselves?

I think small, repeated wins. You do something you didn't think you could do, or you try something that felt too ambitious or had too high of an expectation, and it works. Then you say, "Okay, let's try it again."

Do you do anything like Steve Jobs, where if someone was worried, he would stare into their eyes and get them to believe?

No, nothing like that. I don't think so. I'm not like an inspirational speaker; I don't do motivational speeches or hard-charging rallies. None of that.

What's your favorite strategy game?

I have not played a game other than my poker night with friends in so long. I wish I had. I guess I'll have to say poker, but I miss great board games.

What was your favorite board game?

There's no one board game where I'd say, "I want to play that again and again." I'm not someone who thinks Settlers of Catan is the best board game of all time. But they're all fun because they're new, and you have to figure out a new set of challenges, rules, and everything else.

In the business world, you build your company differently than almost anyone else is able or willing to build companies. What allows you to do that?

I actually think every big company is pretty different; I don't think this is a unique thing. If you make a list of the 20 biggest tech companies, they're all shockingly different from each other in terms of their strategy, how they operate, and their culture. We are quite different from everybody else, but that's not a unique phenomenon.

What do you think your superpower is?

I think we did a really good job of maintaining a principled conviction on something that no one else believed, even though it was the obvious thing one should believe, and then putting together all of the pieces and the talent around that to make it happen.

What allowed you to have that conviction when other people didn't?

I have thought about this a lot, and I do not know. It seemed so clearly obvious to all of us that I was more worried we were drinking our own Kool-Aid than that everybody else was wrong.

But with the benefit of hindsight, we were clearly right. I don't understand the deep mental block and lack of conviction from everybody else. It's very strange. I have spent a lot of time trying to understand this.

Did you find any explanation in your search for why, or no?

There's all the obvious stuff about groupthink, and the fact that big transformations don't come along that often, and exponential curves don't stay powerful for this long, this often. But between 2016 and 2018, we had to get really lucky, and it required a lot of belief. By 2019, and certainly by 2020, we shouldn't have really gotten to exist—Google should have just run away with it.

Yeah. I think Jeff Bezos called it a business miracle that they built AWS and then didn't have any serious competition for seven years. I feel like OpenAI is another business miracle. There's clearly something here about why big companies get sclerotic and set in their ways, though it helps them in a lot of ways too. But this is wonderful. I think it's great that big companies don't stay the only dominant forces forever—that would be really bad. Still, there's clearly a phenomenon where these business miracles happen when they shouldn't. Is that right, or should they?

Well, I think it's great for the world that they do. But to go back and explain in 2019 why OpenAI was allowed by the giants, with their huge amounts of capital and talent, to do what we've done... it's hard to explain.

Why did Microsoft give you the money?

That, I think, is easier to explain. Google had DeepMind and Microsoft did not have an AI bet, and big companies feel like they should have something in play in all of the major areas of technology.

Was it because they thought it was going to be big, but underestimated how big it would be?

I think it's been great for them. I don't know if we've added $1 trillion or $2 trillion of market cap to Microsoft, but I think it's been a lot. They have gotten incredible technology and cloud growth, and will continue to. So, I'm sure it's bigger than they thought it was going to be, but they made a good bet, have done super well with it, and we are very happy about that.

There have been different versions of OpenAI. The first one was the research lab. The next was the product company, and you've said that you bolted the product company onto the research lab, which is the exact opposite of what most companies do. Now you're basically going to be a massive infrastructure company, and almost all the value from OpenAI is going to come from this low-margin infrastructure.

You also said that transitioning into the next phase of this business is not necessarily aligned with how your brain is naturally designed to operate. What do you have to change, and what is that next phase of the business?

I won't answer this because it gets into something we're not quite ready to talk about. But I am very excited for the next phase, and I think I have figured out how to align something that I am very good at and very passionate about with what will be the next 10x or 100x of our growth.

How did the transition go from Sam the research lab head to Sam the product company head? What were the biggest shifts for you?

The whole thing was just extremely different. It has felt like two almost completely unrelated jobs. It's not to say I did one and now I do the other, because I am still responsible for the research lab. But they are separate things, and running the research lab did not prepare me for running the product company in any way.

Did the YC experience prepare you more?

Yes, because I had watched a lot of people scale big companies.

Was it fun to finally get in the driver's seat?

No. [laughter] The fifth day after ChatGPT launched was the day we crossed a million users. I had watched it go up the first day, come down in the afternoon, go up the second day to a higher peak, come down, and go up the third day to a higher peak.

Each of those days, the researchers were like, "Oh, that was just some crazy flash-in-the-pan PR thing. This is over, and it's not going to continue." But I had seen enough at YC to know that when something grows like that completely organically, it has the signature of something massive. On the fifth day, it crossed a million users, and when I came home, it really hit me that we were about to become a very big company very fast.

I had watched what other founders went through, and when I came home, Ollie said, "I saw that you crossed a million users, congratulations." I told him, "You have no idea how bad this is. You have no idea what's about to happen. It's not just bad for me, it's bad for you too. We have this nice, quiet, wonderful life, and it's about to be shot through a cannon."

And that is what happened. It was a very abrupt transition. I knew it was coming, and I had watched other people go through it, so I knew it was not a pleasant experience. But I was like, "Well, we're being shot out of the cannon. Here we go."

Were you subconsciously preparing yourself for years, knowing that this was eventually going to happen once the intelligence got good enough to support a magical product?

Intellectually, I knew it was going to happen at some point, though I had no idea when, and I didn't know *that* would be the moment. But I had successfully deluded myself into acting like it wasn't going to happen. Running the research lab was the coolest, most fun, most amazing job and lifestyle setup I could possibly have imagined.

How so?

It wasn't stressful at all, and it was intellectually incredibly satisfying. It was the smartest group of people doing probably the most important work of the last century—maybe longer. Having a front-row seat to that was a once-in-a-generation moment. It was unbelievable.

Are you guys experiencing that same kind of growth again? I keep seeing Tibo post on Twitter that there's another reset, and another reset. Are you seeing that same ChatGPT-esque experience again?

Totally. I think there have been two giant form factors and associated growth waves so far: the chatbots, and then the coding agents. The coding agents are going totally nuts.

There will be a third one soon, which is this idea of persistent agents—chiefs of staff, co-workers, colleagues, whatever we call them. I think that will come pretty soon. So we're going to go through the third of these waves very fast.

A long time ago, apparently in August of last year, I saw you tweet something, and I thought to myself, "This is not going to help the reputation." Why did you do that?

Let's see. Why? [laughter] First of all, I love that scene from the movie where the Death Star comes out of hyperspace, there's dramatic music, and it's just a great scene. I thought it was funny. I was scrolling Twitter late at night—I don't really remember—and it amused me at the time. It was not a great tweet, to be honest.

I actually thought it was great. I thought it was funny.

Yeah, I thought it was funny too. [laughter]

I do feel like this is more and more what the world feels like.

Yes. Yes, that's a good meme.

The way somebody explained the marketing challenge for OpenAI, and the industry in general, is: Number one, we are close to creating a genie that can grant any wish. Number two, we are going to make sure that our first wishes broadly benefit humanity, and that we get the world to a place where many more people get to have wishes. And number three, the space of what you can wish for is incredibly big and creative, and it'll be exciting to figure that out with real human values and preferences.

But I think there's a fourth thing too, which is exactly that: you start making these wishes, the computer grants them, and then you're like, "I didn't think that was actually going to work. What now?" It's a weird feeling.

Yeah. It's like when people spent a hundred years trying to prove the Jacobian conjecture, and then someone just asked Claude, and it solved it within a few days.

I think there are going to be plenty of great jobs in the future, and most jobs will adapt more than it seems they should. But math is a very important example of something for us to study closely right now, because it may not go that way—in fact, it probably won't.

Even as the systems get better, the people I know are not less busy; everyone is busier, just doing more stuff. Except for the people spending a trillion tokens. I loved seeing a post where someone said, "I spent an ungodly number of tokens," and someone replied, "Why aren't you more successful then?"

Technology has long promised that people will work less and have more leisure, and it has moved in that direction in terms of higher quality of life. But we never get the promise of the four-hour workweek at a mass scale, and I don't expect AI to change that.

Clearly, society is not working for a lot of people, and more productivity gains that directly benefit individuals would be a great thing. I suspect that will happen. But a thing I really admire about people is that our expectations always go up. We always want more, and we think of new things to create and want.

It's a relative game. People are very focused on how they are doing relative to others. The competition that drives the economy, and the wonderful desire to be of service and create, will keep going. We're all going to be much busier than we thought we would be in a post-superintelligence world, and we'll still complain about it, but secretly we'll be happy.

How do you think the status games change as less of our direct input is correlated with value creation?

I don't think it will feel like our direct input is gone. The things we value, I suspect, will remain very human.

Like cooking for people. You're not creating massive economic value, but it shows love.

You are creating value, in that those are the experiences we cherish. Cooking for people is a great example. I suspect people cooking and eating together will remain important long after robots can do a great job of cooking food.

Do you think a good heuristic would be looking at the longest-surviving works? For example, the Bible is likely to still be impactful a thousand years from now because it has been impactful for the past 2,000 years. If you had to extrapolate what humans will still care about, is it the most primal, earliest things? We love adventure, we love a good quest, we love cooking.

Betting against evolutionary biology is usually a bad bet. So yes, I would assume those things continue.

When was the last time that you realized you weren't being ambitious enough?

I definitely badly undershot on our compute investments.

Could you have known that going in with the right mental model?

Yes, but I got psyched out by the financial markets or something. That was clearly a mistake.

How do you correct that in the future?

Hopefully, I'll learn from it. I'll make some new mistakes, but I won't make that one again.

On the compute side, this is the biggest infrastructure project of all time, or is about to be. Are you going to try to vertically integrate under OpenAI's hood everything from power generation to token generation, or are you going to rely on partners?

We're not literally going to try to do it all inside of OpenAI, obviously, but we will try to do a better job of managing a well-functioning supply chain than we have so far. We have brought chip design and model design together, which has been good. I don't think we need to bring electron production under our roof, because that's more of a commodity.

However, when you think about the whole supply chain required to produce intelligence, there is relatively too much focus on algorithms that create better algorithms, and not enough focus on data centers that can create more data centers.

In a world of robots and a truly automated supply chain, you can totally imagine using a data center's thinking power to drive a fleet of robots to build more copies of the data center. That is probably a very wonderful use of compute, relative to other applications, if we're right about what additional compute will eventually unlock.

How much of your time is spent thinking about how to successfully design that fully robotic, end-to-end future where intelligence can just produce more intelligence?

Almost all of my thinking there goes into execution. The high-level idea is very obvious and easy to state: here are the pieces that need to come together. But getting that whole machinery and all those companies working together—there's no glory in just thinking the big thoughts, it's a lot of grinding.

When you think of execution, what does that typically mean for you?

There's no typical day. To the degree that there is, you just do whatever step is required to solve the immediate problem and figure out how to make it happen.

But figuring out how to finance a new fab build-out is very different from figuring out how to get a great chip design team together, get them working with the research team, and actually run the supply chain. Each of those requires a completely different approach.

One thing I know Brian Chesky does is he shamelessly finds whoever the absolute expert is in any given subject, and asks them whatever questions he has to rapidly come up to speed. What is your process when you need to become good at something new in a very short period of time?

Definitely finding the experts and talking to them, and reading as much as I can. I don't have novel strategies here, but I've always been really grateful and pleasantly surprised by how willing experts are to help if you just ask. It's a very nice trait of humanity.

One thing I've heard you say is you should always ask for what you want, because you won't always get it, but sometimes you do, and when you do, amazing things can happen. When was the last time you asked for something that seemed absolutely insane to someone else, and you got it?

Actually, it was this recent trip I took on a plane that I still can't talk about.

Okay, go back one more that you can talk about. [laughter]

I mean, in some sense, Codex was an example of this. It's not the most recent one, but it is instructive. We were way behind on coding, and it seemed like a crazy kamikaze mission to try to beat others with a coding app.

The consensus is usually that this kind of thing never works—you just move on to the next thing, because trying to win when someone else already has massive momentum in a product category is a fool's errand. But we decided it was really important, we asked a team to do it, and they performed a truly unbelievable, very rare feat in the history of business. Now, it is the model that most of the best coders I know use. That felt like an impossible thing, but if we hadn't asked the team to take on an extremely hard mission, it wouldn't have happened.

Why did you make that decision? Why didn't you just give up?

Because coding felt like one of those few highly strategic areas that was going to happen very quickly, and it is so important to recursive self-improvement—to say nothing of its economic value—that it didn't feel like something we could give up on.

At YC, your mission in its most distilled version was trying to increase the amount of innovation in the world. My hunch is that if you thought about why you decided to work on OpenAI instead of YC, you could make the argument that you could have a bigger impact on increasing global innovation through OpenAI.

It wasn't that intellectualized at the time. It was just that my whole life I wanted to work on AI, and I knew it would be the most important thing I could ever touch. It was my passion, and I just wanted to do it.

What were those first few weeks like? I know there was this retreat where people got together, and it wasn't even meant to be a company at the time.

The first few weeks, we were working out of Greg Brockman's apartment. There were maybe 10 or 12 of us. It had been all of this work to get it going, and then we showed up on the first day, January 4th or something like that.

We were like, "So here we are. What are we going to do? Maybe we should get a whiteboard, start talking about ideas, or write papers." I had this momentary feeling of, "Oh shit, what have we done?" It was a crazy moment. It took a couple of years to really get our groove and figure out what we were going to do; beyond knowing we wanted to build AI, it was very unclear.

When you're in the stumbling-through-the-woods phase of this type of Manhattan Project, how do you most efficiently stumble through the trees to rapidly figure out what not to work on?

If something is not working and you run out of ideas, you can kill it—that's the easy part. The really hard thing is when something is working super well, deciding when to kill other good things to make the best thing work even better.

There have been several critical moments in OpenAI's history where one thing started to really work, and we killed other promising projects to double down on the best thing. When GPT-3 started to work, we shut down robotics—which we were really excited about—to focus on it. And when coding agents started to work recently, we shut down projects like Sora and the browser to focus on coding. That is very difficult to do.

Let's say you invest a significant amount of resources into something like Sora. How do you decide whether that is going to become a core OpenAI pillar?

Oh, Sora would have been super successful. It was just that it was more important to put our compute and energy into coding agents.

Take me into that meeting where you say, "We've spent a year-plus on this, invested massive amounts of money, compute, and resources, built a lot of momentum, and people love it, but we're going to pull the plug." How do you make that call?

It's not a single meeting; it's a gradual realization that there is a more important use of this compute, these people, and this product direction, and that we have to make a very painful decision to get there.

How do you realign people once you have to kill their project to move them to another one?

People understand the mission, the stakes, and the need to reorient. So even if they are unhappy in the moment—though sometimes they are very understanding and agree it's the right choice for the mission—they get why we're doing it. The continuing focus toward superintelligence is a powerful aligning force.

Do you think there are projects currently that are going to get killed?

I assume some will become harder to pursue.

There's this amazing line from John Collison where he said basically everything in the world, if you just look around, is so difficult to make happen. Even getting a park bench built is incredibly difficult, and so you can think of the world as a universe of passion projects.

If you are in a position like Steve Jobs, where you are creating something that a billion people might interact with every single day, you have to think deeply about how to design something people are going to love and not be unhappy about. How do you think about design?

I feel super lucky to get to work with Jony Ive on designing beautiful things. I have learned so much from him about how he really studies a problem before trying to get to a solution—not even letting himself think too much about the solution until he truly understands the problem.

I think this is a key that I didn't fully appreciate before. Great design is way more about understanding the problem than a sudden flash of insight. If you rush toward a solution or lock yourself into a specific concept too early, you won't do as good of a job.

I think the iPhone is the greatest piece of technology humanity has collectively made so far. It is an incredible thing, but I don't love my relationship with it anymore. I turned off my notifications, so I actually like it much better now.

Yeah, I keep myself on Do Not Disturb.

That wasn't even enough for me. I turned off notifications for everything except a very small number of things, and I never have them on. I can just look at it when I want to.

Even messaging apps?

Even messaging apps. That was a big life upgrade. And I deleted TikTok because it was just too powerful.

You were addicted to TikTok?

Here's a crazy thing that happened. When we were building Sora, I made myself get addicted to TikTok because I wanted to learn about it. I had never really used it before—people would send me a video occasionally, but I never got sucked into the feed. I wanted to make sure we didn't build something that would just absorb people's time without offering real value.

And I loved it; I thought TikTok was great. But then I thought I could control it: "I'll only use it for 10 minutes to wind down before bed." And then it became an hour one night, and then some Saturday afternoon I was on the couch for three hours. I realized this was not what the iPhone was supposed to be about. I was enjoying it in the moment like a drug, but I could tell it was bad for me.

I briefly got it back under control, down to 5 or 10 minutes a night, but then I decided to delete it. The iPhone is amazing, yet I didn't feel I had enough self-control to keep that app, nor did I feel notifications on messaging apps were a net good for me.

We will make an incredible, beautiful, and helpful set of devices. I am also sure some people will misuse them and it will make their lives worse in ways we can't imagine, and we'll adapt. But this is the nature of powerful technology.

When you're inventing a new device, what is that process of exploration through the problem space? Like, if you're Jony...

I won't answer on his behalf, but he has talked about this process before. When he's exploring a concept, like a car, he will study the history of motorsport, the different typefaces used in the cabin, the materials and why they were chosen, and the different sounds engines have made over time and why some appealed. He writes literal books of his detailed explorations of these unbelievably small components.

Did you guys basically take those books and try to use them as training data to make a Jony Ive model?

No, but that would be a great thing to try. There is definitely a part of his process that I don't fully understand. I understand the studying and the refinement of a concept into something great, but there is a middle step—the inspiration that moves from understanding the problem deeply to the truly novel idea—that seems to happen all at once, and I don't understand that.

When you're thinking about designing something, what goes through your mind?

I'm not a designer in any way, and I try to be pretty good about realizing what I'm not good at, so I wouldn't try to have a strong opinion there.

What are you the worst at on the business-building side?

Maybe product.

Really?

Maybe.

Okay. How do you find great product people if you're bad at product?

This is a thing that I've never quite agreed with people on. There's a business meme that you can only hire people in things that you deeply understand to know if they're great. I think this is just clearly not true. I don't understand design either. I know Jony's great at design. There's lots of things I can point to. When you talk to him for 30 minutes, it's very obvious.

Which things do you try to actively get better at, and then which things do you just try to outsource and understand that you're just never going to be one of the best?

I am a big believer in you should try to get better at your strengths, and the whole obsession with "I'm going to get better at the things that I'm just not good at at all and never going to be good at" is a huge trap. Be super good at your strengths.

Okay. What are your biggest strengths if you had to force rank?

I hate this question. I don't hate this from a false modesty perspective. I think it's very hard to say anything insightful about your own strengths.

You're clearly good at really rallying people around some objective.

Sure, but I don't think I could teach someone who isn't good at that how to be good at that. I think the things that come super naturally to someone are very hard to—I don't know, at least in my own case, I don't deeply understand why I'm good at that or what to do about it. Although I think we talked about earlier, like this category of things that you can learn but you can't teach.

Mhm.

If you want to get good at something like that, asking someone to explain it to you, in my experience, never works. Really studying them and sitting with them in meetings and just trying to observe it and learn it yourself, that does work.

When there has been something that I do want to get better at, I have tried to just be around someone who's great at it and really deeply study it. But I don't think they would have been able to explain it to me. I don't think they could have taught it to me by talking to me about it.

I think that's totally right. If you look at the way that the best people in the world learn video games—if you're playing like CS:GO, you play some CS:GO so you have the basic game dynamics, and then you just go watch a pro and see how they interact on the map at different points in the game and stuff.

When you think about organization pace, there's ambition and all these other things, but ideally, over time, you want your organization—you obviously don't want it to slow down, but ideally, it even moves faster. How do you bake that in?

I think it's like 90% the people you put in leadership roles. There's other things you can do. People have all these different sort of operating rhythms and how they try to run the company, and this management technique and that one. But I think it mostly just comes down to the people. Definitely, something I think about with everybody in a role like that is: are they a fast mover or a slow mover?

How do you measure that without actually having worked with them in the past?

Well, ideally, you have worked with them in the past. I think most of the time, executives at a company should be promoted internally, not hired externally.

But when you do need to hire someone externally, you spend a lot of time talking to them, spend a lot of time reference checking, and you try to work together in some kind of casual way.

Slightly different note. What was the most painful thing that happened in the last 12 months?

Honestly, having kids and working really hard at the same time is brutal. It's just extremely painful. I think I'm a very present dad, relatively. Like, I don't do anything but work and hang out with my family really at this point, but I still feel like I'm missing so much of this one-time thing. It's very painful.

When you think of someone like Masa, for example, what about his brain and the way that he sees the world is so different?

I think he will end up being one of the wild successes of this decade. Incredible person, dear friend, incredible person, massive conviction and belief, and not afraid at all of big numbers and scale, and it's great. There are not many Masa-like people in the world, and we should be very grateful for them.

What made him—I don't know if you've talked with him, but—

I mean, as far as I can tell, he was always like this.

Do you also have the same sort of just, like, no-ceiling concept of scale?

Not—I think Masa is an N-of-1 character.

Do you kind of view OpenAI as a little bit like the golden goose?

In what way?

Spawning a bunch of very valuable things over time and just building a core competency of increasing intelligence in the world.

Not really how I think of it. I get what you're saying, but it doesn't feel like we're just spitting out golden eggs. It feels like we're building this kind of compounding thing.

I think one of my favorite mental models that I've taken from you is this idea of real trends versus fake trends. And—

Oh, yeah, this is an important thing.

Whenever people talked about bubbles or anything else, I was always kind of confused. I studied Warren Buffett and Charlie Munger a bunch, and they love to talk about bubbles and things being overvalued. I was trying to decipher how do you figure out whether or not something is real or fake, and whether or not it's going to persist and continue, or if it's just going to revert to the mean.

And what did you figure out?

Well, it was, I think, your mental model of basically, if there's a small group of people... What is a real trend versus a fake trend? A fake trend is, you know, the VR situation where there's a lot of hype, but then someone buys the thing, they don't really love it, they don't really start to design their life around the experience of using it, and then it sits on the shelf.

Whereas, like for me with ChatGPT, it's pretty much every day. Sometimes I need to work a lot so I spend like three hours, or some days it's like almost nothing, but it's persistent in my life. How do you kind of figure out now whether or not something is a real trend or a fake trend inside of the company and just generally?

Same principle. Like, is there a real, deep, enduring—

How did you come to that philosophy or mental framework?

Well, I watched a lot of startups. There are many amazing things about working at YC, but just the amount of data you get—as long as you're willing to spend the time trying to analyze it and make sense of it, you can really figure out a lot of things.

Are there any big mental frameworks that you used to have that you think are wrong in today's environment and will continue to get more wrong over time as intelligence explodes?

I mean, I kind of think a huge amount of them are—they weren't necessarily wrong at the time, although I'm sure I was wrong about a bunch of things too, but a huge amount of them are wrong now.

I think a startup of today still looks mostly like a startup of 10 years ago because that's what the received wisdom says you're supposed to do. And sure, it's different in some ways. People are like, "Well, I'm going to hire fewer people and spend more money on tokens for Codex," or whatever. But it should probably look very different.

And I have met a few people who are trying to do a startup in a completely different way, but most people just use—it means like use more Codex, and that doesn't seem like enough.