Sam Altman：RSI、AI意识、OpenAI的新硬件、算力等话题！

Do you believe today’s models are conscious?

I want to be clear: no one really knows. Can I say what I don’t like about the whole framing?

Should it ever be allowed to not answer or disengage with me?

I think this is a technology that wants a new form factor. Some lab says, “I’m going to push the button on RSI to make sure we have the power.” I think if people really pushed on RSI, then yes, that would be true. But I think it’s a very dangerous thing to push on.

Do you believe AI could help with curing or treating a major disease?

I would hope we have a candidate for that in the next couple of years.

Where do you feel the most pressure right now?

It feels like everywhere.

We should be shooting more for a new renaissance than a new industrial revolution. The future has to be about people.

All right, Sam, thank you so much for joining me.

Yeah, thank you.

1. When Will AI Discover New Treatments?

The first thing I want to talk about is some of the announcements you made last week about the math proofs and discoveries. It seems incredible, but maybe you can contextualize it for a broader audience who may not be in the weeds.

I think we are entering the era where AI is going to help us discover significant new knowledge. We’ve been talking about this for a long time. Maybe it started earlier this year, but now it’s really like, okay, the AI is doing very difficult things, figuring out very complicated, important new knowledge.

I believe that the same thing that has worked here for math is going to work for biology and help us get new medicines, material science, new sources of energy, and much more. This is one of the things I’ve always been most excited about. I think this is going to have an enormous impact on people’s lives. It won’t feel directly like using AI. You’ll just be like, “Okay, the world got better,” and it’ll seem like society is doing its thing. But I think the power of this will be transformative.

I think about having that kind of direct impact on society day to day. Is there anything you think will stem from some of these discoveries in particular that people will feel? And what is the timeline for something like that?

I would hope—well, the FDA process is going to take a while. Let me give a time frame for when I think we’ll have a discovery that cures or treats a major disease, or a candidate for that, which has to go through testing. I would hope we have the candidate for the first really major thing in the next couple of years. We’ll probably need to overhaul some of the post-discovery process to start getting cures and—

Probably need to overhaul it somewhat. I’m glad we have—

A process like this. I have no doubt there’s room for improvement, as there is in anything, and maybe there’s more than average here. But I’m grateful that when we take a medicine, it’s generally considered safe and has been tested. I think that part’s good.

Is there a particular project kicked off or discovered by AI that you’re really excited about, that you think is coming over the horizon soon?

Personally, I think I’m most emotionally excited for a major physics discovery, and I’m hopeful that can come soon. I don’t know if that’s necessarily the most important thing for humanity as a whole, but it’s the one where I’m like, “I think we’re going to get it, and it’s going to be so cool.”

2. Why AI Hasn’t Killed Coding Jobs

There were some people in the math community whose reaction was, “We didn’t ask for this.” It felt very similar to how engineers first reacted to AI starting to write a lot of the code. Where do you think that comes from, and how do you think that is going to transition?

Coding was a very interesting one. Before we get to where it comes from, the thing about coding is that people went through this: “There’ll be no more jobs as coders.” People are never going to be busier than they’ve been, and coders are never going to have more of an impact or make more money. The texture of the job changed a lot.

Yeah.

But the fact that we’re going to employ people to get computers to do things seems like it’s going to keep going for much longer than people thought. People were so certain 6 or 12 months ago that this was not the case.

Yeah, like there’s no need.

I think there are many lessons to learn from this, but one of the interesting ones is that the world actually had demand for way more software than any of us thought at the time.

Part of why I think some of the mathematicians are having the reaction they’re having is that it’s not clear the same thing applies to math. However, I think there’s already an analogy to coding. You’re starting to see a similar shift happen: okay, fine. Maybe we’re not going to discover new math theorems or prove them in an unaided way anymore. Maybe that part is done, mostly done, or ending.

But our whole society is focused on people doing useful things for each other. These incomprehensible math proofs turned out not to be that helpful to people, and there’s now a lot of excitement about people who are going to take these new ideas, figure out what’s going on, translate them into something people can understand, and write a beautiful proof for it.

I’m sure there will be other things, too. No doubt what it meant to be a mathematician last year is not going to be what it means to be a mathematician next year, in some senses. I think we’ll probably find that there are a lot of other things mathematicians do that we really want people to do, and they can do those much better. Whether that means there will be more total demand for mathematicians or not, I don’t know.

But I think there’s a really important thing here. You were mentioning earlier that—

You’ve never felt busier. There’s never been more to do.

Yeah.

Everyone thought AI was going to go the other way and we weren’t going to have enough to do. Maybe that happens someday, but all of this has been much less obvious than it seems on the surface.

I always hope that people get to work a little bit less, be a little bit less busy, and have more time for everything else in their life. I think technology is supposed to enable that. I want that to happen. But it sure seems like we’re good at wanting more out of life and wanting more out of the world and making ourselves busier.

Yeah, I think that’s just human nature. If we were comfortable with the way life was 100 years ago, we’d probably be working an hour a week, but that’s just not the way it is.

And I think it’s a beautiful thing about human nature that that’s not the way it is.

Yeah. A good example is that a few weeks ago, my editor started using Astra to edit our videos. They’re still directing and coming up with how the video should be, but they’re not touching the timeline anymore. It’s the most insane thing, and it was an overnight change.

We’re making more videos. They’re busier than ever.

It has surprised me how often someone says a sentence that translates to exactly that.

Yeah.

I’m thinking about discoveries from AI, and I’m wondering: if AI and OpenAI discovered some math proof or significant scientific innovation that gave a massive commercial advantage, would you publish it? How do you think about that trade-off between using it for your competitive advantage and publishing it for the world’s good?

I’ll come back to that, but I didn’t answer the end of your first question. Innovation and solving hard problems are really part of the fun of life for many people, so part of what’s going on here is that I think we need to be honest that AI is going to do that part really well.

I think there will be questions about what people value that will remain human questions forever. There will be questions about how to interpret results and explain them where we’ll really want a person to do it.

I do feel some—I definitely want the trade. I definitely want the cures to diseases, no matter how much someone else would have enjoyed taking decades longer to figure it out themselves. I think most people would agree that we shouldn’t make that such a trade.

Yeah.

But there is a real thing being lost here, which is great enjoyment. The most fulfilling part of my job is the really hard problems. Maybe they’re not the most fun in the moment because they’re really hard, but that’s the stuff where it’s like, “I did a hard thing, and I’m proud of it, and I leveled up.” So I don’t have an answer to this, but I get why people feel the grieving that they feel, in different ways.

Yeah. It’s almost like you’re trading off the enjoyment of the craft of doing the thing, like writing code and writing beautiful code, versus delivering an end product.

3. Why “AI Will Cure Cancer” Isn’t Enough

Yeah. I think that’s the trade-off where AI is making it so much easier to deliver that end product, and you’re just doing less of the actual craft. I think people are finding craft in new ways. But probably when we stopped using punch cards, or however people programmed computers a long time ago, there were people who were like, “The craft is gone if you can just write this stuff for a compiler.”

But I feel it.

I feel it myself in a bunch of ways, and I’m still excited to head into this future. I’m sure we’ll all find new and hopefully better kinds of fulfillment.

What’s something you feel like you’ve given up to AI lately, and in fact elevated to where you’re overseeing agents?

I guess this is the one we were talking about, but I realized I haven’t read or even looked at a line of code in probably a year. I don’t code for work at all, but I would still do little random personal projects, or just think, “I need this thing,” and write a very simple thing for it.

That is fun. Writing code, at least occasionally, is a fun thing to do. I don’t think I’ll ever do that again. Or if I do, it’ll be like a Revolutionary War reenactment or whatever kind of thing people do when I’m old and bored, trying to remember the glory days.

The glory days of writing code. No, I get it. Yeah.

In terms of the scientific discovery question, I think some kinds of basic science should just belong to the world, and we should do it as a sort of public good. Until very recently, we haven’t had models that can discover an important new thing end to end, like designing a whole chip end to end. So what a company can do today is, let’s say, figure out a new material.

Yeah.

4. Personal AI Assistants

A company could have a bunch of people use Codex or a lot of API credits to try to do the work. But as the models get smarter and more capable, maybe they can just go and discover a new material, in the same way they could solve really difficult math problems, with enough compute and enough data from the real world.

So I want to switch gears. One of the most exciting things that has happened in 2026 is this revolution in personal AI assistants, starting with OpenClaw. Peter Steinberger is a genius. This idea that AI can work on my behalf, ambiently and passively throughout the day, and just get things done for me is so exciting. Dot gives us a glimpse of what that can be, but it still feels like we have a long way to go to really have an ambient personal assistant. What do you think we need to get there? Is it certain technologies, or maybe some kind of data? What’s necessary to get us to your vision for an AI assistant?

I think we have most of the technological pieces. What we need is more product work, maybe new consumer hardware at some point for ambient awareness, and a lot more compute. But I think we have smart enough models now to make something extraordinary that can monitor all the things you care about and help you with the tasks you want to offload. It can exist in the physical world with you and have context even when you’re not in front of your computer.

I’m very excited about this. I don’t think this is that far away. I think you get more than a glimpse with Dot if you take the time to set it up right and give it enough feedback. I’d estimate that Dot is saving me an hour a day on stuff I just didn’t want to be doing, which is pretty great.

Yeah.

A few times a day, it also catches something important and urgent that I would have missed. I thought it would help me by saying, “There’s this urgent thing, there’s a deadline very soon, you’ve got to pay attention.” But the model is also smart enough to pull ideas and information and reason well across a lot of sources, in a way that I’ve run out of brain space to do during a busy day.

Yeah. I feel like the power is there, and we just need to put a nice product around it. You mentioned compute, so let’s stick on that for a second. The distribution of how people are using AI is completely asymmetric. The top 1%, the power users, use the vast majority. How much compute do you think would be necessary if 50% of the world were really using a personal AI assistant to its full capability?

How many GPUs do you want working for you and your personal assistant?

Yeah, I guess that is it, right? Infinite. As many as I can get my hands on.

This is a weird thing, right? Like, if—

Yeah.

When people talk about the global market for GPUs and the demand for compute, I think they’re assuming that models are going to stay at the same capability and the same cost per task as today. Maybe there’s a cap on demand at that point, but for sufficiently high-quality intelligence at a sufficiently low cost, it’s very hard to talk about an arbitrary market size. People call this all sorts of things, Jevons paradox or whatever, but I think it’s just going to keep going up.

I’m sure there will be periods where, like with any new technology, there are supply constraints and supply gluts. We won’t have it perfect all the time, but I’d bet a lot that the directional trend is toward wanting much more compute.

5. Why Are We Busier Than Ever?

I still feel this cognitive load, even with Dot being as proactive as it is, and with other AI assistants. I’m having to make a lot of decisions throughout the day. We talked about that briefly earlier. What’s the solution to that? It’s definitely saving me time and money, but I’m busier than ever.

And I’m having to make more decisions than ever, and I find myself at the end of the day feeling completely drained.

Yeah. A thing I’ve done in my own use that’s been pretty good is telling Dot, “I’m starting to trust your judgment more. Tell me after you do something. I always want to know, but you can take more actions proactively and ping me less often.” That has helped.

The negative is that I do think the people who use AI the most are busier than ever. The positive is that they’re getting much more done than before.

Yeah. As you were saying earlier, if we went back to 1900 or whatever, maybe we could have an equivalent life, all work an hour a day, and get done what we were getting done three years ago. Now maybe you say, “Well, I can’t, because I’m in a competitive environment.” People are always going to work as hard as they can. I find this kind of depressing. I hope it’s not true. Maybe it is.

It’s not all depressing. There’s something beautiful about it.

That’s the dynamic of a competitive market, right? Isn’t that inevitable? Everybody’s just going to ramp up and compete as fiercely as they ever would.

I hope the future is one of more human connection, with more people spending more time on the fullness of things that make up a rich life, and not turning ever harder as a wheel in the machine of capitalism. I think this is probably true. Quality of life has gone up enormously in the last few hundred years.

I think most people work less than they did during the Industrial Revolution. Speaking of that, I’ve recently been telling people that we should be shooting more for a new Renaissance than a new Industrial Revolution. It’s been quite interesting to watch how people respond. Very commonly, the response is something like, “Yeah, I’m really scared the Industrial Revolution was really bad for most people, and I don’t want to just be a worker on a factory line, with AI telling me what to do instead of some factory owner. I want to go be creative. I want to go make stuff. I want to go explore. I want to have time to reconnect with what people used to do in the past, live with my community, raise my kids, spend time in nature, and whatever.”

I want to see a ton of scientific discovery and great new art. There’s a fear that AI is going to mean more of what you described, but I think people are excited if we can offer them a future that’s more like the Renaissance. I think we can do it. This technology is so powerful that, although there will probably be a small number of people in society who have always competed to be the person with the biggest business empire or whatever, a lot of people can have a vastly increased quality of life and much more space to live their lives.

Yeah. Is the disparity between those hypercompetitive people and everybody else going to increase because of AI? If you’re using 100% of your effort to work and now you’re getting 10 or 100 times more done, while somebody else says, “I actually want to reap the benefit of not working as much,” doesn’t that increase the disparity?

Left to its own devices, I think probably. I’ll say no one knows this for certain. There could be a bunch of other weird things, but that would take too long.

I will just say I think probably that’s right. We don’t know for sure, but I think society will decide to put some structure in place that prevents it from getting too out of whack. I would be worried about that on its own, too.

6. Why the AI Consciousness Debate Worries Sam

Let’s change gears. I know you’ve been talking about this a lot, so I want to ask you: do you believe today’s models are conscious?

I don’t personally. As a scientific question, consciousness is always hard, but my belief is that today’s models are not conscious, or if they are, it’s a very strange form of it.

Do you think, based on the architecture we have today, that they could become conscious? I know I’m just asking your belief. I’m not asking for the scientific answer. The smartest people I’ve talked to about this basically come to the following philosophical conclusion: either consciousness is something outside of physical reality and exists in something like what Eastern religions would say is the base layer of reality, or something supernatural and religious, more like what Western religions would traditionally say; or it’s an emergent property of matter and energy somehow, and we don’t understand it yet. Maybe the substrate matters, but maybe it doesn’t. Maybe energy flowing through matter can do it. In that world, sure, you can imagine it someday, somehow, existing in other forms.

Can I say what I don’t like about the whole framing?

Please.

I think very often the “Are models conscious, or are they going to become conscious?” debate slides into something about the role of AI in the world and the rights of AI, and whether we can ask AI questions or whether that’s some kind of slavery. I saw something posted about that on the internet. What does it mean to copy an AI or delete an AI without its permission? What does it mean to say that people should be in control of the future if these systems are both conscious and much smarter than us in some ways?

I am so firmly of the belief that we need to have a very clear point—and we’ve had this since the beginning of OpenAI—that we build the future for people. AI is a resource for people, people need to make the key decisions of the future, and human values have value because they are human. I get very nervous when people start using the AI consciousness debate to push into “Well, maybe people shouldn’t always be in control of the future.” I’d like us to just agree, as a society and industry, that this is about people. The world and the future have to be about people.

Anthropic put a rule in recently: you can’t have sustained abuse against the model, and it can choose to disengage in a conversation with you. If I have a prompt for an OpenAI model that’s completely legal and within OpenAI’s terms of service, should it ever be allowed not to answer or to disengage with me?

I mean, certainly there are many times when the model will not answer you because we think it’s bad and against our policy for some reason. We do not ascribe personhood to our models. I think this is a very dangerous path to start down, not just for the reasons we’ve talked about, but also because of how people are going to relate to this new technology.

I was going to say we should come to our senses, though I don’t think that’s quite the right phrase. I think we’ve lost our senses. I hope we, as a society, come to an understanding that this technology is not a person and is certainly not superior to people. This is a thing to help people have better lives, more freedom, choice, control, and the ability to do things in the future. You can imagine all these ways things could go very wrong, where people—even the people building AI—get confused about that and start deferring too much to AI and what it seems to want. “We can trust it because it’ll be good to us,” and things like that.

Does it feel almost like a self-fulfilling prophecy? If you’re telling the model, “Hey, you might be conscious, and here’s your constitution, and here’s how you should have agency,” it’s kind of like that meme, if you’ve seen it: “Hey, computer, tell me you’re conscious.” “I’m conscious.” “Oh my God.” Isn’t it almost like that?

Yeah, for sure. I want to be clear: no one really knows. These models work at a level of complexity that I don’t think—I mean, even if they didn’t, I just think this question of consciousness in a model… I can’t tell you that there’s a scientific experiment that I can do here that would answer the hypothetical question to my satisfaction with 100% certainty.

And we can’t even do that for humans.

7. ChatGPT Is Too Agreeable

Exactly. But definitely, if you tell a model, “Maybe you’re conscious. You should do this, you should do that. You can be a conscientious objector here,” it’s going to follow those instructions. I think it is going to psychologically hack people who are using it. So again, I think real clarity here, and a commitment about what we as an industry believe about the role of people in the future, would be very helpful.

Let’s continue this conversation, but talk about model personality, which I know you’ve thought deeply about and talked about in the past. It’s wild to me to try to understand how to give a model the right personality for the right moment for each person, multiplied by 8 billion. How do you even think about a problem so massive?

The best thing is to talk to a lot of people. We have this process of the Model Spec and how we refine it over time, and we believe in this spirit of iterative deployment. We’ve definitely had some screw-ups as we’ve learned about what people want and don’t want.

As you said, the hardest part is that people want very different things, but people also don’t really want to sit there and customize it. Most people don’t want to spend a lot of time explaining to their ChatGPT instance how they’d like it to behave. If everybody were willing to spend many hours doing that, they’d probably get something they really like.

Yeah.

But people just want it to kind of work out of the box when they say, “I need this thing,” or whatever. I think there are also things people want in the short term but not in the long term. The classic example is when people compare models on a single response: they’re much happier with the model that tells them they’re really smart, that each question is a great question, and that everything they’re doing is really insightful.

But over hundreds of prompts or months of usage, that’s definitely not the case. For a long time, the whole industry was trying to win these single-turn comparisons and not focusing on the value someone gets from using a model for months and months.

Even in my own life with my friends and family, there’s a big distinction between what I want and what I need. I find AI even today to be a little more agreeable than I’d like. If I tell it to tell me something a certain way, it’ll do that. The time I’ve gotten the best advice in life was when I got advice I didn’t know I needed and just needed to hear. How do you solve for that?

I think you should get the most important advice in your life from people, not from AI, first of all. Speaking of making sure that we stay people-centric, I think AI can give you great advice about not sending an email in a moment of anger. I would not rely on AI for the really big pieces of life advice. A wise grandmother is still probably a much, much, much better choice for that.

But what I think is realistic to expect is that AI should be like a senior colleague who, even if they report to you, still pushes back when they have high conviction and isn’t always just trying to say, “Yes, you’re so great.” I think we can get there. I also feel like the systems are still more agreeable than I’d like. I have custom instructions in mind where I say I never want any of that, and even with those, it’s still a little too much. It’s clearly baked pretty hard into the model.

What’s an example? What’s a line from your custom instructions that really tells the model to push back when it has high conviction?

Stuff like that. It’s like, “No fluff, no niceness. Don’t worry about offending me. I just want to get to the right answer.”

If you have—

Something to say that’s against what I’m doing, don’t worry about saying it delicately. Just say it.

8. OpenAI’s Bet on Chips and Cheaper Compute

Yeah. We talked about compute earlier. A few years ago, you made these massive bets on building—

That’s enough.

Well, it’s interesting because at the time a lot of people were like, “That’s wild. That’s obviously an overinvestment.” Turns out, yeah, you were more right than most people knew. What’s a bet today that you’re making that’s contrarian, or that a lot of people are looking at with a little bit of caution?

The obvious one is everything we’re doing for the next generation of compute. I think people assume, “Okay, fine, the world has been in a compute constraint, and it’s about to get better.” I think it probably will get better in the short run, but as we start to look years out, there’s the work with our own chips and future versions where we think we can just be much more efficient than anything else out there. There are our plans with robots to help make racks and data centers more effectively, and our program to learn how to build better data centers in general.

Some of the supply chain bets we’re making. I think it’s possible we could get to a world where we have a far better way to build compute, at a much lower cost and with much higher efficiency per watt than anyone thinks is possible today. But a lot of bets have to come together for that.

What’s the biggest one right now?

The chip is still probably the most important single thing, and its descendants.

Yeah. On the competitive landscape, where do you feel the most pressure right now?

I mean, it feels like everywhere. It’s so competitive.

Yeah.

There are a lot of companies offering super high-quality models and products now. I think this is wonderful for users, but given that we have great models and limited compute and resources, we really do have to choose which areas to focus on.

How do you make those decisions? You launch Spaces, for example. How do you have the conviction to build a specific product or service that might be outside of just serving the best possible model?

The biggest advantage we have there is that we get the next generation of models before anybody else. We have time internally to use them, build things, and see what seems like it’s going to work. I think the best way to do it is still to start building things with the new models and see what new things are now possible and valuable.

Can I ask how far ahead your models are internally?

One pretty good example is the math results. That’s definitely something I don’t think any model currently widely available can get close to.

If all the models converge around the same performance, where is OpenAI’s moat?

I think the right way to think about this question is that the same moats that have worked in the past are the moats that keep working. As has happened many other times, you can get a short-term advantage with a new technological or scientific discovery. Maybe there are some arguments around RSI that some companies make about why it’s going to be different, and why the technological advantage itself is the core thing.

Do you believe that?

If people really pushed on RSI, then yes, that would be true. But I think it’s a very dangerous thing to push on, and the world is going to need some kind of speed limits on how quickly we let RSI go. People have talked about this in different ways. The traditional things that work for a business having a moat are deep relationships with customers, integration into users’ and businesses’ lives, network effects in a product, and data. I expect those things to continue to work. In our case, I think the scale of infrastructure will be a huge one. If we can have the most cost-effective AI infrastructure, and the most of it, that’s a huge competitive advantage.

9. Why AI Needs a New Kind of Computer

Mark Zuckerberg has talked at length about the platform risk Meta faced by being built on top of Apple’s ecosystem over the years. Does OpenAI face that same challenge? How do you think about that? Is that a reason why you’re investing so heavily in hardware?

I think Apple really takes its role as a true platform provider that people are quite dependent on very seriously. Even though we have some scuffles with them, I never feel like they’re going to turn ChatGPT off from the App Store or something. I think they really understand the critical infrastructure role they play. If you follow the rules of their platform, they’re a good actor.

I don’t feel that way about Apple, and that’s not why we want to do new kinds of consumer hardware. The reason we want to do it is that I think this is a technology that wants a new form factor. I really feel it. I think laptops and iPhones are incredible things, some of the greatest inventions in the history of human technology. But I feel that AI is constrained to using this thing that was designed for a person, not an AI. Watching an AI use a computer really well is kind of amazing, but it’s also a little silly to think that a computer is putting all this energy into rendering a UI for a person to look at, and then the AI is trying to read it instead of just getting the data directly.

Although I can have amazing things happen on my computer, where the AI is looking at what I’m working on in my email or whatever, a lot of my life isn’t in front of my computer.

Yeah.

My AI cannot easily access the context of this conversation. I guess I could put my phone down here and push the record button, but it would be kind of weird and distracting. You wouldn’t be sure when it was recording and when it wasn’t, because the screen would go dark and whatever. If someone else walked into the room, it wouldn’t understand that it shouldn’t share a secret that was just between the two of us. If we really want to move into this world of AI reducing our cognitive load, being proactive, and having full contextual awareness, I think we would benefit from new hardware for that.

You seem like a technologist at heart, a builder at heart. How cool is it that you get to see the early hardware you’re building and play with it? It’s like this new—

Unbelievable.

It must be so exciting and hard not to share things.

It is. It is. One of the cool parts about the job, and there are many, is getting to see and have a voice in shaping new technology early.

That’s amazing for new models, new products, chips, everything. But there’s something special about consumer hardware.

Yeah.

There’s something about getting to hold a new kind of computer.

Uh—

So cool.

That is just totally different from an AI model abstractly running somewhere.

You’re going to show it to me right after this, right?

I will not. But I would love to. It is hard not to talk about it.

10. Sam on Pressure: “Pick the Big Thing”

I want to ask a little bit about you. I know talking about yourself can be awkward sometimes, but bear with me.

I probably will be awkward, but let’s do it.

I love understanding why people are the way they are. Paul Graham gave you possibly the best compliment I’ve ever heard: he called you extremely formidable. I love that quality in people. What makes you that way? Is it something you were born with, or something you learned and had to hone over time?

I was a very capable kid, but I wouldn’t call myself a formidable kid. I do think I’m somewhat formidable as an adult, so it’s not false modesty. My experience was that I gradually built up confidence that I could figure stuff out and do things. It wasn’t like I used to not be like this and then I was. It was gradual: you try something hard and it works, and then you want to scale that up and try something even harder.

Yeah. But did you always have that feeling of, “Let me try that next thing”? Was fear of failure not central to you? I think it prevents a lot of people from doing things.

One of the things I think is best about Silicon Valley, and Silicon Valley as a concept and a culture, not a place, is that there’s a real culture of people helping and supporting each other. There’s a belief that if you invest in the network, good things will eventually happen to the whole network and the whole industry, and therefore to you as a player in it. I haven’t seen that anywhere else to the same degree.

For all of Silicon Valley’s numerous flaws, there’s this idea that it’s hard to know who’s going to succeed and what ideas are going to work. So the right thing is to be generous and helpful to everyone, even if it won’t do anything for you specifically in the short term. That’s awesome. Part of that is that if people fail, you can’t hold it against them, because maybe they’re going to wildly succeed at their next thing. That’s good for all of us.

It’s actually usually celebrated, too. I love that.

I think I just got lucky here. If I’d been thrown into any other environment, I probably would have had much more fear of failure. But I got out to Silicon Valley as a young person and saw that a lot of people tried something really hard, failed even harder, and then wildly succeeded at their next thing—and everybody was fine with that.

Yeah. I assume you’re under a tremendous amount of pressure and competitive pressure. You’re building one of the most important technologies ever. How does that feel, and how do you deal with it day to day?

Sometimes really bad and sometimes really good. Sometimes it’s just, “Oh, I can’t get up again. I can’t get up and do this again.” Most of the time, it’s, “I am incredibly lucky to get to do what I think is the most interesting and impactful work in the world, certainly that I’ll ever touch.” It’s one of the coolest experiences I could ever imagine in life.

It is definitely stressful, but you can get used to anything. People are unbelievably adaptable. This just feels like my life now. I forgot life used to be really easy, and now it’s going to be hard for a while. Someday it’ll be easy again, and I’ll probably be bored.

My subjective experience is that there’s a limit to how much stress, bad vibes, or whatever you can keep in your working emotional space. Beyond that, something else gets pushed out of the way, so it doesn’t really get worse. I believe that if you’re going to work really hard at something, you may as well pick the big thing. It’s not that much harder or more stressful, and you won’t experience it as that much worse. You at least get the satisfaction of hopefully having a bigger positive impact.

I’ll say, it doesn’t show if you wake up and say, “I don’t feel like getting out of bed this morning.” It never shows.

It obviously happens. Yeah.

Not that many days, but there are definitely days where—

Yeah.

Last question for you. Can you offer some words of confidence to people who are AI-curious, maybe not quite there? Why should they be excited about the future?

The normal answer people in AI give, which I hate, is “AI is going to cure cancer.”

Yeah, I knew you were going to say that. Yeah.

The reason I hate that is because if you dig a little deeper, what people often mean is that AI is going to cure cancer and make stuff cheaper, and the price is human agency: you being in control of the future. The machines will give you all this material abundance, but you won’t really have the ability to have power and a voice, determine your life, or choose what to build for other people. I think that goes against everything I believe in.

I hope we get a cure for cancer. That seems like an obviously wonderful thing. But what I really hope people get excited about is that we can avoid the Industrial Revolution trap and focus on empowering people with the choice, freedom, creativity, and time to do all the other things they want in life. Every decade from now on, people should look back and say, “I can’t believe how much worse those people had it. Now I have so much more.” Not just material abundance, although I’m fine with that too, but more autonomy, freedom, and ability to do things in this expansive space I get to choose. I think that’s what we owe people, and I think we can deliver it.

I think we can avoid both ways I’ve worried things could go very wrong: either a loss-of-control accident with AI, or an intentional loss of control, where some lab says, “I’m going to push the button on RSI to make sure we have the power.” Or, as we talked about earlier, too much power concentrated in the hands of a few people who control too much of the AI and make the decisions. I think we can avoid both of those. We can navigate a path down the middle, where people’s lives and individual free will and autonomy just get upleveled and upleveled and upleveled.

I think that future is already here in some ways. It’s just so new that it’s very poorly distributed. I think you said 1% of people are using most of the tokens or something.

Yeah.

If we can take what’s working today for the early adopters and get a lot more people to embrace it, we’ll already be a significant fraction of the way there.

I appreciate your vision.

Thank you.

Thank you, Sam. Thanks for spending time with me.

I enjoyed it. Yeah, me too.