Sam Altman on Building OpenAI & Betting on the Impossible

Key Takeaway

Run a 30-minute “AI-first” audit on one recurring task today. Choose an inbox triage, research, meeting-prep, writing, or data task; give an AI tool the relevant context; ask it to produce a first draft or decision brief; then compare the result with your normal workflow. Do not redesign your entire system at once. Repeated small experiments create the feedback and comfort needed to replace ingrained, low-value computer work.

Episode Overview

David Senra interviews Sam Altman about OpenAI’s origins, the slow human adoption of fast-moving AI, and the discipline required to focus on high-impact bets. Altman connects startup investing, research management, product iteration, AI safety, and personal autonomy, arguing that powerful technology should expand people’s ability to create rather than centralize control.

Full Transcript

Transcript of Sam Altman on Building OpenAI & Betting on the Impossible from David Senra. Auto-generated from episode audio; may contain minor errors.

I just mentioned Toby Lu and the fact that, uh, I recorded with him before. Why do you say you think he's one of the most interesting CEOs right now? One of the things that most caught my attention about Toby is that, from the early days of AI and at every stage of its development, he has been the most visionary CEO. He's there writing the software himself. He's experimenting with it. He sends us extremely detailed feedback on the product offering and the capabilities of the models. Long before anyone else, he was saying, "We're not an NPC company, so we're going to adopt agents, otherwise, you know, we're totally ruined." We're going to build it ourselves.

Every time I talk to him, he's at the forefront of what anyone, CEO or not, is doing. He builds it on his own. He understands that he has a great, deep intuition and is always six to eight months ahead of any other CEO. Do you remember when he wrote this, probably a year and a half ago, maybe in 2024, that letter saying that the first thing you have to do is see if AI can solve your problem? And even then, I don't know, 18 or 24 months ago, people went crazy; They thought it was ridiculous, ridiculous.

That's my point: he's consistently been ahead of the curve, he's been right, he's been involved, and he's very direct, no nonsense. So there's no exaggeration, there's nothing more than: " This is what he can really do now; this is what I think he'll be able to do soon; this is how I'm going to propel the company forward," and he has an extremely deep understanding of where we are. I never thought about that, how beneficial it must be for someone in your position to have someone like that giving you such intense, direct, and clear feedback on the product.

Many people submit comments about the product. He is the only person at the intersection of being the CEO of a large company and offering extremely accurate, detailed, and cutting-edge feedback. Yes, he told me, I don't know if it was in the episode we recorded or if it was later, but he was very, uh, firm. He says, "We'll look back in 2026." And I actually want your opinion on this. It didn't even occur to me to talk to you about this. We will look back to 2026 as the year when all business was at stake .

He said someone was going to build the native AI version of Shopify. And he said, "And it's going to be me." So, at night, he tries to rebuild it. If you were starting from scratch, what would you do with current technology? That was the other thing I was going to say about him: that he does it himself. In other words, he himself is using these tools. He is writing software himself. He is testing the models himself. He himself tries to reinvent his own workflows. Most CEOs, when they reach that level, have teams of people managing other teams trying to implement the idea, please you, and smooth out the rough edges.

I think it's very difficult to have a real perspective if you don't do things yourself. And he does it in such a practical way all night long, as far as I can see. I'm not sure that 2026 will be the year in which all businesses seem to be at stake. I may not entirely agree with him on that point. But I understand the spirit of what he's saying and I understand why it seems like that's happening. Do you think that's even possible? Whether it's in 2026 or 2046.

I mean, obviously not literally all businesses. I think there are things that are very contrary to AI. The better AI gets, the harder it will be to compete with some businesses that have nothing to do with it, because we'll really want these... authentic, non- tech experiences, or we'll care more about ... sports teams or whatever. So, no, not everything, but I think there will be a lot of software businesses that will be very much at stake. So you don't agree with the chronology? What...what...? I disagree with the chronology.

I think it's going to take a little longer . I understand. Can you elaborate on that? I love startups. I believe startups are the coolest thing in the economy, and I've dedicated my career to trying to truly understand them. And I thought that when we got to GPT-4, which was in 2023 I think, very soon after that there would be a lot more disruption and that the software businesses would be at stake immediately, but it didn't turn out to be that way. I think I was wrong about some things , but one of them, in terms of speed, is that the economy simply has a lot of inertia.

People continue doing the same things they've always done. They keep buying from the same...you know, company. They still seem to want to use their tools in the same way. I think this is really positive in many ways and will make this big transition ahead of us smoother and slower, and I'm grateful for that. But I think it means we've all been too ambitious with the deadlines. Even with this incredible technology , I think AI is one of the most amazing technologies that humanity has ever invented.

Society and the economy will adapt more slowly. Yes. It's curious. We were talking before we started recording about all these parallels with history. Obviously, I dedicate myself to reading history. When you were speaking a moment ago, I wasn't even thinking about OpenAI, AI, or Sam Altman. I was thinking of reading Larry Ellison's biography from the 1980s. He said, "Guys, this isn't a software problem." "It's a people problem." We have to convince them; we can install the software, but they don't use it. We need to change their behavior.

The technology is already here. It's as if we now have to adapt humans so that they actually start using technology. My own example of this was after Netflix came out and started shipping DVDs, even before they started streaming. I was surprised that people still went to Blockbuster. I found it incredible. I was just observing this because I passed by a Blockbuster on my way to and from school. And I was amazed that people still did it. And you know, that's an example that has stuck with me about the power of habit and the way people do things; Changing behavior is much more difficult than technology experts believe.

So, if we go back to this fuss about Toby writing that, you know, open letter or that letter to the people within his company, you guys are adopting this faster than anyone else because you're partly inventing it, right? So, is there anything about your own behavior that really surprises you, like: I know there's a better way to do this? I'm even creating the product that could be better, and yet I still can't overcome this habit, this force of custom. 100%. Good. I love that they asked me this before.

I've been waiting for this question. What I find most psychologically contradictory about myself is that I have been using computers in the same way for 20 years. Now I have a magical thing called codecs. You too. And everyone else too. That means I should be using my computer in a completely different way. I shouldn't be clicking here and there, you know, pasting from one messaging app to another. I shouldn't be mindlessly checking my emails , trying to see which one is the least burdensome to open and reply to when I don't want to deal with them.

I shouldn't be keeping a to-do list and doing these mechanical tasks on the computer the same way I've been doing them for so long. And yet, there's something coded in my mind that doing these kinds of things is what it means to work and what it means to be productive. And if you asked me, I would never say that I like doing it that way. In fact, I would say the opposite. And I think he would mean it . But, based on my proven preferences, I have a better way of doing it now.

I can do it faster. I could use Codeex for more of my day-to-day tasks, like finishing up this pile of emails, doing my to-do list, and taking care of all these things. And yet I do it that way . And it doesn't make sense, unless I secretly like it or it makes me feel good. What do you think needs to change for you to truly embrace your own product in a deeper way? I really don't know . In other words, it's happening gradually, and that might be the right answer: these things need to happen little by little, and completely changing ingrained habits and workflows is difficult.

I believe there are better products we can build with this technology that will make it smoother to do so. But right now it seems we 're all torn between these two worlds, where we still have a computer we can use the old-fashioned way and we have Codeex that can use our computer in this amazing way, and we're not sure what to use, when, and for what. And I think this is mostly a product defect. The phase we are in now reminds me of smartphones before the iPhone.

I was a pioneer user. I had a Palm Trio back in, I don't know, 2003 or 2004 or something like that. Have you ever had a sidekick? I never had one, but I thought they were great. I wanted one. And much of the technology already existed. It lacked the multi-touch screen, but above all it lacked the product ideas that made the iPhone the iPhone. And I feel like we 're now in a world where we have all the technological pieces, but we haven't had that iPhone-type moment that completely changed how someone interacts with technology.

We were talking about Toby, about how Toby is building this himself, right? Uh, we have a mutual friend , Josh Kushner. He says there's a big comparison to be made between the way Steve Jobs thought and ran his company and the way he thinks you do it. Obviously, he wasn't the one who wrote the code. He wasn't building the hardware, but he said, "I am patient zero." I'm making products that I want to use myself. And, essentially, everything we saw with Apple was basically what he wanted.

There's a great story in one of the books where they were supposed to have a meeting about, I think, one of the new MacBook laptops, and the team prepares a huge presentation for Steve, and they're very nervous about his imposing presence, and he walks in; They think it will be a one-hour meeting. He comes in, shows them the laptop and says: on, off. Press the button and it turns on. He immediately turns it off and then tries to open the MacBook. There is a delay; He says, "Make this," referring to the MacBook, "like this," and leaves the room.

And that's the whole meaning, it's that simple. There are many examples like that in Apple's history. How do you approach it? How do you improve the product? Do you do it solely based on your own needs? What are your thoughts on this? Most of my effort is now focused on research and computing. I would love to be able to dedicate more time to the product. We have excellent people thinking about the product here, but the most important thing we can do is create smart models and be able to run them efficiently and abundantly for many people.

If we manage to do that well, I think everything else will fall into place. Philosophically, I am very inclined to try to find that difficult, high-impact problem that will continue exponential growth, and for us, that is computing models. I also think those are problems that naturally suit me. Why do they adapt to you naturally? Scale the computing in the way we are doing it. This requires as a model. It's a complex supply chain. There are many interesting associations to define, something I enjoy doing. There are interesting financial challenges about how to finance what is probably already, or at least rapidly becoming, the most expensive infrastructure project in history.

The technological issues necessary to develop computing at this scale, from designing your own chip to the factory supply chain , including rack manufacturers and power systems for these machines. I've always been interested in energy, and all of this comes together . So there are many interesting problems that encompass technology , business, politics, supply chain and logistics, all revolving around building computing at this scale. I used to be a startup investor, and the closest thing I've found in my career to investing in startups is managing a research program.

There are many ways in which they are also really different; The average researcher and the average founder may seem different at first glance for obvious reasons, but there are many similarities in how to find off-the- beaten-path bets, how to decide where to have conviction, how to understand exponential growth, how to manage exceptional talent, and how to identify it further. This is the research building we are in. That's where I feel. Brilliant. Tell me more about the parallels between what you learned investing in startups and doing research.

One very important one is the power law. People talk about this all the time in investing, and it's that you have to reprogram your brain because we don't seem to be naturally adapted to think this way. Where you know your best investment will outperform all your other investments combined. Your second best investment will outperform everything else combined after that. And AI research, at least, works that way too. When we started, people thought it was totally improbable or almost impossible that AGI was possible. What year is it?

2015, 2015. I mean, we got torn apart by all the intellectual giants in the field for saying we were going after AGI. And then, when we started to really focus on the big language models, they tore us apart again, saying that this was completely ridiculous. And we understood, or at least I understood from my experience in startups, and I think others understood it in other ways, this point that high- risk bets are fine as long as you choose those that, if they work, are extremely valuable, and research looks like that .

The type of people who become great researchers are those with an unconventional approach, high energy, and a kind of original approach; "Out of the ordinary" is the word that always comes to mind . Hey, you need to say more about " out of the ordinary". Can you be more specific? Do they have any distinct features? You do n't want to fund a founder who has a slightly different opinion on the same idea that the last thousand people you talked to tried to convince you, and maybe convince themselves, are somehow completely different and doing something totally new.

But mostly it's like trying to fit in with the herd, follow the same path as everyone else and do what you're supposed to do, which is start a startup . And, you know, they've heard Peter Thiel say enough times that there's something you're supposed to do differently, that they try to emulate him, but they do n't mean it. For me, it's very clear when someone simply thinks differently from most people and is willing to defend convictions that are very unpopular; They may be wrong, but if they are right, at least they will be truly correct, unlike someone who only puts a superficial veneer on the same idea that everyone has.

At the end of 2015, when we were starting OpenAI, there were very few AGI efforts in the world. There was DeepMind and one or two others that I can remember. It was a very unconventional thing to do. In that same year, there probably were, and I mention it only because it came to mind, although there are other categories as well. There were probably many thousands of founders creating photo-sharing applications. You know, that probably wasn't such a good thing to do. Today, many people want to start AI labs.

There are a handful of people, you know, two, three, whatever, doing something completely new. Um, that wasn't really possible until the day this got so good. But, well, it still doesn't seem like a good idea, and that's what, as a startup funder, I've always wanted to fund, and what has mostly worked for me. There was something similar for researchers: there were many researchers who pursued the latest thing that had worked, and a small number of researchers who had a strong conviction towards a new idea. And, uh , I think we were and are the best research laboratory for those people.

I remember talking to Deus about this a few months ago, and he thought that, in terms of competing the way you guys do, there are the three big players; that the money and capital required are such that there will not be a fourth, larger player. But he said, "There's like a 10..." I forgot what the number was. I'm just going to make it up. Let's say there's a 10% chance that some lone researcher will approach it in a way that's an angle we've never considered. Completely.

I don't know how to put a number on it, but there is a possibility. I do n't think I'm making up the number, but it was a small percentage. There's definitely a possibility, and I love that. I think that's why things are still exciting. I want to talk to you about the main sponsor of this podcast, RAMP. I've been reading a lot about SpaceX lately. SpaceX is one of the most valuable businesses in the world, and one of the main themes in SpaceX's history is constantly attacking and questioning its costs.

Ramp helps many of the world's most innovative companies do exactly that. The average company that uses Ramp reduces its expenses by 5%. And something SpaceX has demonstrated is that a religious dedication to controlling costs can help increase revenue, because you can take advantage of opportunities you otherwise couldn't. And we also see that in the RAMP data. The average company that uses RAMP also increases its revenue by 16%. So, when you run your business with RAMP and your competitors don't, you have a massive competitive advantage that builds up over time.

Ramp is the only platform designed to make your finance team faster and happier. Many of the best founders and CEOs I know run their businesses with RAMP. I run my business with RAMP, and you should too. Go to ramp.com to learn how they can help your business save time, save money, and increase revenue. That's ramp.com. I found one of my all-time favorite quotes while reading the book "Zero to One". The quote reads: "The most powerful pattern I've noticed is that successful people find value in unexpected places ." And they do this by thinking about business from first principles rather than through formulas.

That's exactly what AppLovin has done with its advertising platform. AppLovin connects you with over a billion new potential customers within mobile games. AppLovin lets you capture undivided attention. AppLovin ads are full-screen videos that are viewed for an average of 35 seconds. It's a retention rate that far surpasses other advertising platforms. And you can get started on AppLovin in a matter of minutes. You set the goal and AppLovin achieves it. There are no complex setups, no experience is needed, and AppLovin scales quickly. They can put your ads in front of more than a billion potential customers.

Other companies have seen immediate results, scaling up to hundreds of thousands of dollars in daily spending and increasing their revenue by millions. So you'll want to get started quickly before all your competitors are on AppLovin. And you can do it by visiting applovin.com. That's applovin.com. Good . So what is it that confuses me? You went from founder to investor and back to founder. But why in 2015? What interested you about artificial intelligence from the beginning to say, "Hey, this is something so lacking in consensus"? People think I'm [ __ ] crazy.

I'm going to do it anyway. Well, I've been interested in AI my whole life. He was a very nerdy kid. As you know, I was the kind of kid who spent Friday nights playing on my computer, watching science fiction and reading science fiction, and I always thought AI would be like the most amazing and crazy thing. I never thought I would actually get to work in it, but I always loved it. I even went to university almost to study it. I worked in the AI ​​lab the summer between my freshman and sophomore years.

And nothing worked. In fact, I remember very well that a teacher told me I could try all these things. All these directions exist. The only thing we know that doesn't work is deep learning . We tried for a long time. You know, it's the most guaranteed way to have a bad career. And I was an impressionable freshman , you know. I assumed that was true . So, you know, I pursued these other things. It was clear to me at that time, this was around 2005, that AI wasn't working.

And it turned out that I accidentally got into startups, but then I fell in love with them. And I wouldn't call it a career detour because it was super useful. Looking back, becoming a startup investor was great. The normal career path in Silicon Valley is, or rather, a common path is to be a founder, then you sort of partially retire and become an investor. We do n't want that. I heard you say that you're going to work on this for the rest of your career. Yes, I hope we're just getting started.

You're going to come on the show several times. I'll take you up on that. We don't need any more founders retiring to invest. No, we have too many investors in this. But what I was going to say is that the fact that I was able to go in the opposite direction. I was an investor first, and then I ran a company. It's quite unusual. Very unusual. And I am very grateful for that. Because you gain an incredible set of learnings and pattern recognition if you actually study and observe companies as an investor, which has been very useful to me running OpenAI, but it's the opposite direction to the normal, so it's a very rare thing and I highly recommend it.

Why is it useful? If you run a business, you've faced a number of similar decisions in the past, such as crucial ones, and you've seen what works and what doesn't. And if you have to make a high-risk strategy change or fire an executive in a very complicated way, you only have your own limited experience from the last 5 or 10 years that you've been doing it. But as an investor, you observe all the crucial moments. So you don't get the day-to-day operational practice of running a company, but you've seen many of those crucial moments all the time.

So the richness of the dataset I got there was incredible. That echoes in your mind when you have to make a decision. Yes. I think, "Oh, this is what happened when something similar happened to this company , or I saw this founder make this mistake or this other one do it very well." We were talking about how you studied the industrial revolution. We were talking about some great biographies that we both read recently. My friend Daniel X says this about me, because I think the benefit of having done this project on my other podcast called Founders for 10 years is, as he says, that you're like a language model trained on the best entrepreneurs in history, but with the temperature high because you're [ __ ] crazy, because I'm strangely passionate about being obsessed with dead entrepreneurs.

But it's useful. It's like, I talk to a founder and they tell me about something they've had to deal with. And I say, "Oh, well, Carnegie did this and Rockefeller did that, maybe you want to try this ." It's like, do you find that you have a great overall idea about all of that, or is it just that for any given scenario, you have what all these people did and how it all comes together? I think it depends on the founder's personality , don't you? So, when I was reading , I realized I've been reading your blog for years.

I think you're a great writer. You are very concise, I find that brevity very appealing. And I love numbered lists. It's strange that we both write in the same way; I feel like when I read your blog I think, "This is the exact conclusion I would reach based on all the reading." And well, there are a handful of principles that can be applied, but it really depends on who the founder is and what they want to do. This is what I'm trying to understand. Let's get back to what we were talking about.

I still think you're taking a big leap, because I know you're one of the best investors of all time in Silicon Valley, from what I hear. You could simply be rich and not have to work because you're lazy. I'm just kidding, by the way. Or maybe not. Hey. No, having done both, I think I can say that it is much, much, much harder to run a company than to invest. Exact. And that's what people should be doing in my opinion. Then you think, "Screw that ." "I'm not going to take the easy way out." " I'm going to do the hardest thing of all." The thing that people think is impossible.

The thing they're going to make fun of me for. Yes. I still need to understand. Okay. So you were interested in it as a child. Why would that appeal to a child? You lived in St. Louis at the time. Yes, I lived in St. Louis. Why would AI appeal to you back then? Well, I think it appealed to any kind of computer nerd. I don't think it's anything unusual for me. It simply seemed impossible. I think most people would say that, of course, it would be the coolest thing in the world, but that it's totally impossible .

I think what was strange about me was thinking: okay, let's try it. But I think everyone would think it's amazing and something worth fighting for. Wait, was that a childhood personality trait that told you that you couldn't do something, as if your initial response was resistance? Not resistance, but rather: Are you sure? Why not? Let's try it. Let's see what happens. Perhaps I can, perhaps we can. I was a very optimistic child. And also, the more impossible something seemed, the more it intrigued me. The idea that we could invent a technology that would allow us to do everything else, that would empower people in a way that no other technology could, always seemed incredibly appealing to me.

It's like: I want that. I want to be able to do everything else. I think another thing that has been a trait of my personality for as long as I can remember is that I find it interesting to give people much more power and much more ability. In a sense, this is the trajectory of technology, and I've certainly always been a technology enthusiast, but AI is the most powerful version I can imagine. What did you think it would allow back then? Like when you were a kid: this seems like cool technology.

I want to do X. I can't do X unless AI is invented. It's always difficult to remember how much of this is what you actually thought at the time versus what you're trying to build now. Yes. How much my current job has influenced my memories about it. As a child, I was certainly very interested in robots; We had a robotics club at my school. And the robots of that era were ridiculous; I even remember that at summer camp we had a little turtle that you could control with a computer on the floor or on the table, and I thought it was the coolest thing in the world.

There was something about seeing physical things move controlled by a computer that I always found amazing. I am now extremely interested in what AI can do to boost scientific discovery. In my adult memory, I think that also seemed great to me when I was a child. But it just feels implausible. And I suppose that's an example of how memories have become nuanced. But now, the fact that we can have AI discover new physics and cure diseases, and what it's already doing with mathematics, I think ...

I think it will be one of the most important areas, even more so than automating other tasks that AI can perform, simply by helping us understand more things. We were talking recently about this book, The Beginning of Infinity. And rereading that book from today's perspective, I think: wow, AI will really help us achieve this, this important task of understanding everything, or as much as we can. I was definitely interested in that Star Trek-style version of great prosperity and abundance, and what AI could do to drive it.

Perhaps the memory of having been interested in science is more real. I always loved science and simply this idea that, because we were intelligent, we could figure out how to understand the world, make predictions, and accomplish things that we couldn't do without this deep understanding. I don't know, that seems inherently amazing. It's interesting how constant what humans want from AI has been throughout time, because some of what you describe is very similar. I just reread Claude Shannon's biography for the second time, and I'd forgotten— because I hadn't read the book for about five years—that he and Alan Turing used to meet every day for coffee when they were both at Bell Labs.

This was back in the 1940s, and they would just talk about AI. They were both obsessed with her, they thought it was inevitable back then and that it would happen about 15 years later. So, like in 1955, when we were going to have computers that did n't exist yet, right? They had the analog versions that were going to be smarter than humans, and they thought that anyone who believed that was n't going to happen was absolutely ridiculous. And they would say, "Well, what would you like the computer to do?" He would reply: "Solving mathematical problems, writing poetry, curing diseases." It's something you hear over and over again.

I've read many things those guys wrote back then and it saddens me greatly that they're not here to see it, because they were right about everything. We are finally at the point where AI is solving novel mathematical problems. He is discovering other things. It's, you know, you can debate how good it is or not, I would say it's not very good, but she 's writing poetry and she gets to what these guys, I think, would have said: "Alright, you got it, that's it, we got it ." Um, and that would have been so cool.

Yes. It's a strange thing where everyone says, "Oh, he'll never do X." Like when I talked to people in the music industry. It's like: "He'll never make good music." And then they say, "Well, do you think I'm going to do a podcast?" I said, "Of course he's going to do it." He's going to do everything we can do, at least I'd say better than even now, better than we can do. It's something very bizarre, it's as if I'll never get over what's happening right now, in the moment I happen to be alive.

There is a profound human psychological flaw there. But here, I think, is a more interesting question. Let's just say he does make a great podcast. You know, two AIs having a more interesting conversation than yours and mine. Do you think people will care, or will they prefer the one with real people because we're all obsessed with people? And the fact that they are not real people. Yes, this is more interesting. It's like, oh, these two people that I'm perhaps predisposed to like or dislike are having a conversation, that interests me.

I think for a strict reference, like maybe my other podcast where I just say, "Hey, these are some interesting ideas I read in this book." That could perhaps be interrupted, whatever the case may be. But especially for people who were born before this happened, maybe it's different for your child, you know, but for me it's just that I think humans will always be attracted to humans. I truly believe that. I think there are many other jobs that could face a significant transition , but things that have to do with people, with people connecting, connecting with others and people liking other people, that seems to be more valuable in the post-AI world, not less.

Although perhaps I'm the wrong person to talk about this because, in a way, I deeply desire, even though all my work is digital and broadcast all over the world, you know ... It's just that I deeply desire more , more of an analog life. I like to read physical books. Like when I was talking to Kelly, I told her: I don't want to go on Zoom. Call me or we can talk in person. I just like physical things. I don't like it. I 'm the same.

I don't read digital books. Yes. I don't like Zoom meetings. I like being with people in the real world. I definitely think there's a group of weirdos, and there are probably a lot of them in this city, who, you know, don't like humans and just want to communicate with computers. But I think it's only a tiny percentage of humanity. I think it's a tiny percentage of humanity. That's why I think the world, in general, won't be so different, even with superintelligence. People will still be fundamentally programmed to care about others, want to be close to others, and, you know, interact with them.

And you know, there will be some people who get obsessed with the models and simply think that humans are in the way, or a danger to deal with, but for most, people will be the main target. I think it's very important that, when we find people like this , we point them out and make sure they do n't gain power. I certainly agree with that. Perhaps the two biggest risks that worry me most about AI, which are somewhat in tension, are: one, the loss of control, where AI becomes so powerful that we cannot guarantee the control we want; And the other is that power becomes too centralized, where you have a company, a model, or a person with too much power.

And in both cases, the fundamental point is that I believe it is a very inhumane stance for any of those things to happen. The correct approach is to say: we want people to have profound control over the future. We want people to be deeply empowered. People are the focus of all this. We're not going to sit here, you know, gradually handing over control to an AI model because we don't trust people or because we don't like it. And, you know, it's a very misanthropic thing to say: " Let's put all our trust in this model and let it have all the power and decision-making power over the world." But I think there are people in the world who believe that's the right outcome.

There's another version of this, which is: because we don't trust people, we have to limit who has access to this technology and how they can use it, and all these terrible things could happen and, you know, out of fear of that, we're going to concentrate power in the hands of a few companies. And they're going to, you know, we won't let other people use this, but we'll give them some benefits. As my caricature of this is, I think, there are some people in the AI ​​field who are effectively saying, "We're going to give the world the cure for all diseases and we're going to make everything very cheap in exchange for people giving up their autonomy, their influence over the future, and their power," uh, and also in the name of security and also, as an absolutely rampant inequality.

It's as if some people will have access to enormous amounts of wealth and power, while others will simply receive a "pretty good" amount across the board. And this is like a terrible sales pitch . This is a very inhumane sales pitch that people somehow feel willing to make. Why do you think they feel willing to do that? I think it's like fear and power. I, I think you can...when people talk about the risks of AI, I think there are a lot of people who are so nervous about the magnitude of those risks and feel so shocked by it and feel the need to protect the world from it, um, that they say, "You know, we should trade a lot of freedom for security here because this is unlike other risks we've seen ." But then I think that also ends up being a way of justifying a lot of power-seeking behavior .

Everything I read, like when I was saying what Claude Shannon or Alan Turing said, at least in the books I've read, is more like an optimism that we're going to invent things that will improve our lives and can do things for us. You are absolutely right that if you go back to the era of Claude Shannon and Alan Turing, they were talking about how wonderful AGI would be and all the things it would do. Uh , and when we started, we really had a lot of pressure from the alarmists.

Now, the part of the alarmists that I agree with is that this is a powerful technology and we should lean towards the safety side and act with caution at every level of the technology. The part of the alarmists that I disagree with is that it is an unsolvable problem. If you go back to the beginning of OpenAI, um, I think there would have been two widely accepted opinions. Number one, absolutely not, and certainly not in 10 years were we going to build something very much like IAG, and then, conditioned on whether we did, we certainly weren't going to be able to do it safely.

You know, if you had an AI that was smarter in many ways than many of the smartest people, most of them, doomsayers would say that surely by that point the world would have been destroyed. The alignment issue would have failed, and there were these positions held with great confidence about what would have happened a decade later. We've built something that I think most people would have said at the time would have looked a lot like IAG, and, you know, a lot of good things have happened and those kinds of crazy, negative predictions about the end of the world haven't come true.

So I think that should update people's predictions about the future. We still have higher-risk challenges to solve ahead of us. But our approach—this is another thing I learned from startups—the way you do things is to launch products into the world, get feedback from real customers, see where they fail and where they don't. That's the way to make a good product. That's also how to make a safe product. And we've made far more progress in AI safety than I think most people realized when we started. Because?

Because there are so many people, there are a billion people using your products weekly, and every time we have a new level of model, we launch it to the world. And we see what works, what doesn't, where people need us to relax the barriers because they have good uses for it, where we have misalignment issues, where we have failures in the safety systems. ChatGPT has been around for less than 4 years, and a billion people use it for sensitive and important things. And the fact that we can deliver something that is widely considered safe, although of course it has problems, in such a short timeframe with such powerful technology.

I don't think there's any way we could have done that in an ivory tower , and you know, that's how I believe good, safe, robust, and useful technology and products are built. And I think that's a great, a great lesson from Y Combinator. And it would have seemed to most AI security experts , you know, totally impossible to get to this stage and still have the level of security guarantees we have now. I think it will be more difficult from now on, but I don't think it can be solved by disconnecting from reality.

Why does it get more difficult from here on out? Because the smartest people in the world are almost as smart as the most capable models in the world . And that's going to change right now. Yes. In one direction. I think that's correct. I think the models are so incredibly capable and improve at such a rapid pace that unknown unknowns, perhaps, don't become more difficult relatively, but from an absolute perspective they seem more difficult. And I think we'll have to make a lot of tough decisions about, you know, when to delay development, when to say, "Okay, you know what?

Let's get in touch with reality now" or "Let's wait longer to study it better." Talking to someone recently, something that stuck with me is that the FAA has helped make flying incredibly safe. Flying, on the surface, seems extremely dangerous. And you probably get on a plane without thinking much about it. And this, certainly, you know, airplanes are not that old in the long trajectory of human history. And this was certainly not the case in the early days of aviation. They have extremely solid accident reports, with a very clear vision.

They never try, you know, to ignore something with evasive answers. They want to extract as much information as possible. And, in a sense, I think that with any new technology, an approach like that works very well and is often not appreciated enough. So when we started deploying our models, when we said we were going to launch ChatGPT to the world, we knew the model was imperfect. We know it's amazing, we know it can do these other things, but we also know that the world needs to experience this technology.

We need to learn how to make it safe. And we have to put power in the hands of the people. We cannot simply use this to impose our worldview. We can't use this to sit in a lab and try to think about all the impacts, which won't work anyway because society and models are going to co-evolve. We have to do this all together as a joint product. And then we'll keep a very good record of accidents. We'll study when something goes wrong. We will publish a very clear post-mortem analysis .

We will learn as much as we can. We will not only improve our own technology and products, but we will also try to share those learnings with other people who develop AI. I think that's worked surprisingly well so far . And those were good examples from the history of technology, good examples of startups. Deel is how the best founders turn the world into their talent pool. I've been studying how the best founders in history operate for a decade. And one thing they all have in common is that they understand that recruiting and hiring the best talent is your top priority .

The best recognize other top performers, which is why leading companies like Ramp, Shopify, ElevenLabs , Uber, and DoorDash use Deel. Many of the best founders I know have personally invested in Deel after using their product. And what they discovered is that Deel is the best company in the world building infrastructure for global contracting. Deel will help your company hire, pay, and manage any worker anywhere in the world. This way you can retain top talent anywhere and spend the rest of your time focusing on what you do best: delivering value to your customers.

The founder of ElevenLabs has a great description of the value that Deel can bring to your company. He said : "We created ElevenLabs to break down language and communication barriers." With Deel enabling us to recruit and support exceptional talent anywhere, we can accelerate our innovation and bring more voices, stories, and ideas to every corner of the world. More than 40,000 companies trust Deel. Learn how they can help your business today by visiting deel.com/enra. That's from deel.com/enra. Something that might be puzzling to you because it makes no sense to me is that, you know, essentially what I focus on is only entrepreneurs and entrepreneurship, right?

So all the podcasts I do are for the benefit of entrepreneurs. I'm glad other people are listening, but he's very focused on just trying to serve. Trying to find useful information, whether it's in the biography of a deceased entrepreneur or by talking to someone like you so that other entrepreneurs can benefit from this conversation, right? Or any of the podcasts I do. So, in that sense, it's not that...many entrepreneurs love AI, but what I'm trying to understand, if you can help me reconcile it, is: everyone uses AI.

Everyone hates AI. What on earth is going on there? Well, people always fear, for example, rapid socioeconomic change. We talked about the industrial revolution a moment ago. I love it, I love reading about past technological revolutions too. And people did not have universally warm and pleasant feelings toward the change that occurred throughout the industrial revolution. I think it's probably a good feature of human society that we have a certain built-in inertia . We have some skepticism about rapid change. I think that will probably help society in times of unrest or in times of, you know, localized madness or whatever.

So some of it is probably a good thing, and I think it's a trait of human biology that I believe we should never try too hard to fight . I also think a lot of the people creating AI have been saying, "There's a 25% chance we'll destroy the world, and yes, we're going to go ahead and do it because if we don't, the bad guys will do it first," or it's like, "Wow, this is going to be really terrible, 50% of jobs will disappear next year and we hope everyone is okay," but that sounds really scary.

As a field, we haven't done a very good job of explaining to people what the benefits are and how the negative aspects can be mitigated, and we certainly haven't done a good job even when people have had responses like saying, "Well, there will be a universal basic income or work will be optional, or whatever." There has been very little discussion about how and why it is important for people to have more power and personal freedom in the world, not less. And that matters a lot to most people.

The ability of people to influence their own future and to collectively design where society is going and the autonomy that this entails is very important. And I don't think many people in the AI ​​field, even if they feel it for themselves, spend much time thinking, reflecting, or recognizing how important that is to others. So to return to that characterization of the sales pitch, I don't even know if that's a word, from a moment ago. It is. Like: "Dear peasants, we will bestow upon you these gifts of a cure for cancer, material wealth and some things, and, you know, great entertainment; you stop complaining and we will make all the decisions about the future, just trust us." You know, we'll be little dictators.

It's not good, it's not good. As a lover of entrepreneurs, and something of a student of what has made this incredible economic miracle of the last few centuries work, I truly believe in empowering people to do new things, to pursue what they believe in, and to have the freedom to create businesses, invent technology, and chase ideas, as well as the system that makes all of that happen. There is nothing I believe in more firmly. And I think that even if people don't see themselves as entrepreneurs, even if they never want to start a big company, they understand how important it is.

And when you hear people say implicitly or explicitly that there will be less of that with AI because a small group will have the power, but they will make big decisions and keep everyone safe. I think that's very frightening for them. I also think that, although most people don't want to start giant companies, many people want to start smaller businesses, and that has been difficult. It has been something that required a lot of privilege, luck, and resources, but we are about to see the biggest boom in small business creation we have ever seen.

I think AI is driving that now; For some reason, the industry, including us, hasn't talked about it enough, even though we see all the signs, it's fantastic, and we haven't created enough products to accelerate it, but we'll see much more. It's curious. We talked about Tobi Lütke earlier, at the end of the conversation, and I think that's in the episode. He mentioned something I hadn't thought about. He said, "Oh, yes, you and I are in the same business." He said, "We both try to create more entrepreneurs." He is building infrastructure for entrepreneurs.

I am creating educational and inspiring podcasts for them. And the crazy thing about what you just said is that I don't just think that, of any new industry, the AI ​​industry is doing the worst job I've probably ever seen in my life. And I think part of that is your ability to go out and talk about what you see and educate. There is a great book called "The Intel Trinity" that tells the story of Intel; It's called Trinity after the three main actors: Bob Noyce , Andy Grove, and Gordon Moore, and there's a great story in the book that I never forgot.

They went from inventing, I think, the integrated circuit to the microprocessor and realized that this technology was so important that it would scare off their potential customers. So they went out , they stopped what they were doing, those three people, and they started educating . Yes, potential clients, investors, the whole country. And they said that at one point they were offering more classes than the local community college had in its entire course catalog. They made it that important, a top priority: we are going to go out and educate people about this new technology.

And it's like, why isn't anyone in AI doing that? I mean, there are no excuses. We should be doing more. I think we've tried versions of this. We haven't done it quite right. Well, you're doing it right now. This is one of the main reasons I wanted to talk to you, because I use AI all the time. I find it fascinating, but you have a vision that makes mine seem like that of an ant. There are so many things in your head that I want to extract, you have all this context.

It's like, and you're inventing this amazing technology. I'd love to know how other people are using it. In fact, before reaching other people , I heard you say something interesting that I think is related to what you just said about inventing technology. Technology is advancing rapidly, but adoption should be slow and deliberate. And I heard you on another podcast say, "Hey, I'm considering how much I should let AI see everything on my computer." Do you want to talk about it? With the latest generation of models, I don't want to say they feel smart enough because I think we should always aspire to make them better, but they are pretty clever.

And at this point, I feel more limited by the amount of useful context that AI has about me. I want the AI ​​to know as much as possible to help me. I want him to do things that I can't or don't want to do on my own. For example, I'm not going to read every post in our internal Slack. I'm not going to read every story a customer has to tell about where ChatGPT worked or failed for them. Can't . And then there are other things, you know, I could probably read more research papers than I do.

But, well, it requires a lot of mental energy, and you know. But I would love to have an AI agent that constantly tries to be helpful, that can see and understand more context than I can, or have the time or energy to process on my own, and that can help apply that context to give me good advice when making a decision. So I think we've focused so much on the intelligence of the model that, on the product side, we haven't yet given enough thought to what it means to give a model more context than any person could have and help that person make informed decisions.

I have the impression that we are right on the precipice of being able to see a very different way of working with AI on an axis where people simply cannot reach this level. There are a lot of very intelligent people, but there's no one who can read, you know, tens of thousands of pages of context in a few seconds and actually use that accurately. And this is something that AI can do that will be something very new and an incredible addition. You have read all these biographies.

There are probably times when you vaguely remember something that , if you could recall a specific anecdote from one of them, would really help an entrepreneur at that moment, right when you're talking to him. But maybe you forgot, or maybe you don't remember it exactly. I built my own AI tool. So I use it internally. Do you know what it is? It has only been trained since 2018. I have saved every note and highlight from every book I have used in this database and I consult it for that.

And then, when the work you do came out, I added it; So I've trained her with that, in addition to every note, every highlight, and all the transcripts of my founders episodes. I use this thing every day to make each of the episodes. So, I'm working on Claude, called Shannon, right? The Claude Shannon episode came out, I don't know, two or three weeks ago, whenever that was. And I ask questions about everything, I was like, "Hey, what did Bob Noyce say about this? " or "What did Rockefeller do about this?" and I made the episode.

I read the book. I took note. I don't remember, because it was about seven years ago. It's incredible. This is what I mean. I was like: this is [ __ ] incredible. That's great. Let's talk about how you think about running the company, shall we? So you're dedicating your time. You said your main focus is getting more computing power and then research, right? Alright. You want the models to be the best in the world, but what do you think about creating your own products? So, you created Codex, right?

I don't even know the damn product lines, where does all the revenue come from? Actually, I think we should be more of a platform company than a product company. Okay, of course we 'll build products, but how many do you have? Let's back up a bit. How many products do you have now? We just merged ChatGPT and Codex. So we used to have ChatGPT, Codex, and the API. You know, and Codex, unfortunately named, but it wasn't just for coding. He could do any kind of job, which confused people.

That confuses me. Yes that's fine. Like many other people. I think what most people want is a single interface for their own personal or company AGI that can help them with whatever they need, and then the ability, via an API, to build anything on top of it. And that's the platform we should offer the world. We will sell excellent AI at every point on the cost curve, on the cost-performance curve; we will be the best. If you want a very high-end AI for scientific discoveries, that's great.

Do you want an inexpensive AI to do a lot of work that may not require genius-level intelligence? We 've got that covered too. And thinking of this as a new utility or raw material, whatever you want to call it, people want to use a lot of AI that is low cost, fast, works well, has context, and is fluid. We've got you covered. And then there's the unique product, which is: "I need to ask the AI ​​something." Eventually, perhaps it will be: "AI should proactively offer me things ." But you'll have this interface, which started as a chatbot and now also has programming agents , and I think at some point it will feel like a more persistent agent of this AI running wherever you need it.

But that's all. I don't think we should build all product categories. I don't think we should try to compete with all of our customers. I don't think we should try to cover the entire economy. I believe we should offer this platform and get 100 million new businesses and 8 billion people to use it in all sorts of ways. So, a direct interface to the product, an API for people to use however they want. These eventually converge more and more as well. And then it all depends on what people do with it, build on it, or anything else.

What mistakes did you make that made you have to learn that? I feel like you've had to kill some good ideas and sacrifice going after the big ones with all your intensity and focus. Yes, I believe that killing good ideas, sacrificing good ideas to go after big ones, is the hardest lesson for any entrepreneur or business. It's terrible to have to discard good ideas. And no matter how much you think you will, people, everyone, perhaps because of the nature of those who choose to be entrepreneurs, seem to be terrible at it.

I'm terrible at this. I know I 'm bad at this. But last year, for example, we eliminated Sora, which was a good, fun, and great product, but it consumed a lot of computing power and wasn't as important as Codex, where we allocated those resources. We removed our web browser called Atlas. Which, again, I think was a great product and the best browser, but it wasn't as important to us as focusing that talent elsewhere. In a world of limited computing, limited people, and limited resources, we thought a lot and said: general intelligence for intellectual work and eventually for science is the most important thing we can do.

Anything that contributes to generating that intelligence—building our own chip, our own data centers , writing good infrastructure software, and certainly training models—is really important. But then, let's offer this AI as a service and have people use it for intellectual pursuits, for work, for scientific discoveries, and to be more productive in their personal lives. Let's have a flexible general platform and not do many other things. Anyone who does complicated work—and you must be at the top of the list of people who today need someone to help them organize their ideas, right?

It is extremely beneficial. You see it in each and every biographies. You see in the story that you need someone to talk to. In fact, there's a funny story about how extreme this can get. Charlie Munger has something called the orangutan theory. Have you ever heard of this? Where it said that a relatively intelligent human could sit down with an orangutan, tell it all their problems, tell it everything on their mind, and then, you know, the orangutan obviously doesn't say anything about anything else. The human leaves and the human feels better.

Just the idea of being forced to put your thoughts into some kind of structure. Now, obviously with a very intelligent partner , Munger played this role for Buffett. Buffett is one of the smartest people who have ever lived. The best investor of all time. Even so , he needed to organize his thoughts in front of someone else. You're going through...I can't imagine...you're having an almost unique life experience, especially for someone as young as you. So I 'm curious, who fills that role in your life? Who do you turn to who can at least remotely empathize with what the hell you're dealing with on a daily basis?

There are three categories here. One, many of the researchers who have been here forever; We've been through all of this together and developed this shared set of language, intuition, standards, or whatever you want to call it, which I haven't been able to replicate with anyone outside the company. When it comes to the shape of what's happening, what might happen next, and where technology is likely to go, in terms of questions about business and the world, for a long time in my career, Paul Graham and Peter Thiel have been two of the people I've learned the most from in many different phases of my career, and they remain the two people I turn to if I really have a very obscure problem that I 'm stuck on.

And after searching extensively, I haven't found anyone else who has the same ability to think in a super non-linear way. If what language models do is predict what word comes next, they are two of the people from whom I can least predict what word they will say next. And that's a super valuable skill. You go in with a "oh man, I feel really stuck and I've thought about all these options" and someone might tell you "I don't think any of those options are good." Here's something that now seems totally obvious and correct that you didn't think of, a completely different perspective that you haven't heard anywhere else .

Is this more of a prompt for your own thinking rather than explicit advice like "do X", for example? It's often like: here's something specific. Oh really? Yes. So, what's an example you could share of someone like Peter? Peter is very fascinating to me. It's very fascinating. And that's precisely who I thought of. The reason I thought of this question just now is because you said we had to kill good ideas for other great ones. We are eliminating Atlas. We are computing. We need to focus, focus, focus; It's something that becomes very obvious when you hear him talk about the importance of focus.

And if you have something that works, making it work better and continuing down that path; taking an hour of that to explore something else is too costly. You should delve deeper into what's already working. There is a lot of value in the extremes. After launching ChatGPT, it was kind of strange because people did n't really know what to use it for and it was growing super fast, but it felt very unstable or almost like low-value growth. It was like people were using him just because they were interested in talking to him and seeing what they could do.

So there were a lot of people in the company who were saying, " Uh, well, we need to look for something else ." "This is not a sustainable value." And I remember talking to him about this list of five or six other things we could focus on instead. This was like 2 months after ChatGPT or something like that. And he said: it's an obvious mistake to do anything about it beyond the fact that it's growing, which is rare and great. It wasn't growing as fast back then as it started to later.

He says the power of this is the power of the Google text box . It's like a text box where you can type anything and it does the right thing. And the fact that it doesn't align with the current Silicon Valley wisdom of, you know, that you have to have feeds, have a network effect, and have, because we didn't have any of that and that's why everyone was worried , you know, you have to have a...you know , the way people are going to build more.

This was before we had any memory. People are going to build more context. Are people going to get trapped? Are people going to have all that? It's like, people have been chasing Google's business model for 20 years and this is the first thing that's come up and, you know, clearly the empty text box worked for Google, so why not double down on that? It's growing, it's very flexible, and, you know, it has all the signs, except that it doesn't fit with the current wisdom of Silicon Valley, and I was like, okay.

So we focused all our attention on CHP and it was great. He's a simple genius, judging by what he just said. Sometimes there is more complexity, but that was a very important example of simple genius. What about any advice from Paul Graham or any guidance or direction that he has encouraged you to take? When you just said that, this is like a meme for many YC founders, where you would go to see him during his consultation hours and he would say, "You know what you should do?" and he waved his hand, his finger, like this.

Do you know what you should do? You know what you should do. And sometimes what came after that was great. Sometimes what came after that was terrible. But the important thing was that there was a kind of, there is a creativity and an open panorama and just, you know, let's try a lot of things. We talked about the spirit of iterative deployment and how, you know, just like with startups, he really pushed the entrepreneurial ecosystem towards this world of "you have to release a version one, even if it's embarrassingly early, and it doesn't matter if it could be much better; you'll improve it much more thanks to customer feedback." I don't even think I asked him before releasing Hatch: "Hey, do you think we should release this?" But I knew what she would say.

I knew it was still early. I knew it was still embarrassing. And I knew the right thing to do was to take it out and put it in front of the people . So, wait. Your mental model of Paul Graham is so complete that you don't even have this case. That's the point where you would say there is certainty. Let me tell you something funny. Just before he died, a few months before he died, I went to Charlie Munger's house and had dinner with him. And I asked him, " How often do you talk to Buffett?" He replied, "Never." I said, " What?" He said, "We talked every day for hours and hours." "Buffett can simply pretend to pick up the phone to call me and he already knows what I'm going to say." Obviously, that comes after 65 years of working closely together, but I found it hilarious.

That's hilarious. It's a really funny story. No, there are many times when I couldn't predict what he's going to say, and that's why I think he's valuable. But in terms of launching a product when you're embarrassed, I know what you're going to say about that . Like that, uh that's been it, I won't say YC's most valuable tactical advice , but it's been among the best . Looking back at all my data points from YC founders over the years, I'm struck by how much the ability to move quickly and be iterative correlates with success.

Okay, you've mentioned YC too many times in this conversation. I need to explore this because we've talked about it before. It's like, listen, I'm not a journalist. I am an enthusiast. I don't have a list of questions. It's like I have a world-class founder in front of me. I want to know what the hell is going on in this person's mind and I want to selfishly extract information for myself. So why? I'm surprised by how much you mention it in conversations, how going through YC and then running YC, being affiliated with them, has clearly impacted your life.

Can you explain this? There is a band that wasn't very successful. They didn't sell that many albums. I mean, but they influenced all the musicians who came after. I think it's literally called "The Band". Rick Rubin told this story. I think it could be Velvet Underground , but you get the point I'm trying to make, whether it's fair to talk about YC this way, because YC, measured by traditional metrics and created market capitalization or whatever, would be one of the few most valuable tech companies. But the degree to which YC totally influenced everything that has happened in the last 20 years in the tech industry, startups, entrepreneurship, you name it, I think is only half understood.

OpenAI is an example of that. Not only because of how we have launched our products to the world, but also because of the philosophy of how we run our research laboratory. I think if you talk to many of the people running this generation of big tech companies, they would tell you similar stories even if they didn't go through YC . But what's going on there? Is it an operating system that gives you YC because they tell you, "do these five things " or something like that, or is it more of a philosophy for building companies?

This is the confusing part for me as an outsider. I think there are two main things. I mean , there's some operating system stuff about what to do, but I think it's mostly the philosophy of how to run companies. The idea of iterative deployment, putting technicians in charge, betting on young people with a lot of energy and ambition but perhaps with less experience at all levels, and also the related change in the entire technological ecosystem prior to YC. So if we turned back the clock to 2004 and projected technology to 2016, but without changing anything else about the shape of the startup ecosystem, what it meant to be an entrepreneur, how capital flowed, who got to run companies, all those things.

I don't think OpenAI would have been possible. I think the changes that YC induced throughout the ecosystem, more power for founders, young technical founders with the ability to raise a lot of capital, the ability to work on ambitious things without a very proven resume. I don't think OpenAI would have been possible. So this is a big change. Is all of this related to the fact that you believe there was a benefit in your transition from founder to long-term investor and then back to founder? There are many benefits too, although I wouldn't say I really went from founder to investor and then back to founder, because the first time as a founder didn't go very well.

It's like a company; You learn lessons from failure, but I think you learn much more from success. Oh, wait, we're not going to move on to another topic yet. You need to say more about that. There's one... I can't believe it, I think it's from some great Russian novel. I'm so embarrassed not to know. He begins by saying that all unhappy families are unhappy in their own way. All happy families are the same. Yes. That explains why it came to mind . Um, but I think this is very true.

When I analyze the lessons from my failures, I learn something generic about determination and what not to do. But most things don't work. So there are many reasons why things don't work out. And I think it's more difficult to determine the correct cause. And when something has actually worked for me, understanding which parts of Y Combinator or OpenAI were successful, trying to apply those lessons has been much more helpful to me than trying to apply the negative lessons of what didn't work. And well, of course, one should learn as much as possible from each piece of data .

So learn from failures and learn from successes. But in my own experience, when I've tried to apply those lessons, the ones I learned from success were very good and I should have applied them more. And the lessons from failures were fairly generic and I already knew them, or they simply got in the way of other things. And I think this is generally true for many people. Yes. But isn't it true that we already know what we should do or avoid, and that what's needed is the reminder, the constant reminder ?

The best description I've heard of my other podcast, Founders, is that it's a church for entrepreneurs. If you think about it, I used to publish it on Sundays and I should start doing that again. But in reality, I'm just counting the same type of personality that has appeared throughout history. It's just that now this person builds ships and this other person builds technology, and they live in different times. It's the same personality. Yes, definitely. I'm a little obsessed with this idea of things lasting for a long period of time.

And as you know, companies, the best companies, can last a long time, but not as long as cities, and cities don't last as long as religions. And I ask myself: of all the things made by man, what has lasted the longest? I'd say I can't think of anything else, unless it's religion. So I started studying. I grew up with a fundamentalist Christian mother , so I was forced to go to church my whole life. And I simply began to analyze what all the major religions of the world have in common.

It's like, oh, we have a shared knowledge base , usually some kind of book, right? We meet with like- minded believers at regular intervals. And it's not like I go to church on Sunday. It's like, okay, we talked about Jesus last week, but let's talk about this other guy. It's like, no, we keep going back to these same books and the same stories over and over again. So, I read your blog and you even said this, something about when YC ended. It's like you're repeating the same thing.

You tell them all the time and then they leave the church, you know, to use this analogy, and then they stop doing the same things . It's like they're not even lessons. It's like a constant reminder that this is important. I completely agree with that . But I think it's better to be reminded of things like talk more to your users, you know, launch products earlier, get more feedback, uh, maintain a higher standard for who you recruit and who you hire, and do it faster. But it's the positive aspects that I think are good.

You have one of the best tweets I've ever said to my phone. You say something like, you know, skip the conferences, the [ __ ] dinners, everything else. Simply create the product or sell and sell the product. If you're not creating it, you're not selling it. That's really all you have to do. And I think it's like it returns to that simple genius. So, this is the other part that I find most fascinating, because someone asked me, uh, yesterday, they say: " What is your...?" Usually, there's some kind of historical equivalent for every founder I know, that I can say, "Oh, that guy is like Vanderbilt , that guy is like Rockefeller, Lewig, or any of these people," and I was like, "What's your historical equivalent for Sam?" I was like, there isn't one , I can't think of any because I do n't know him well enough, I don't understand how he thinks yet.

What do you think now? Well, this is the first of, hopefully, eight conversations. So I'll tell you in conversation 7, but this is very strange. I just spoke with Doug Leone and he talked about a guy he hired, who is the founder of New Bank, and he was a damn associate at VC and then he goes and founds one of the most successful companies. I'm like, I've never heard of that. Doug, are you ? And he dedicated his life to this. He says, no, that's the only one.

So, again, it's very strange. Most people go from founder, unfortunately sell their business, and then become an investor, instead of running the business until they die, which is my preferred method. What I'm curious about is when you said that I learned more from successes, right? Well, those are the successes because you were exposed to, what?, 10,000 different companies in that decade or decade and a half that you did this and obviously you saw maybe half a dozen or a dozen that were the best in the world.

So, do you also see their successes as instructive? No. Okay. Absolutely, and I think people do , I mean, you're an amazing student, but there are a lot of pretty good students of entrepreneurship, and people, I think, often try to find those lessons from the things that really worked. And as you said, it's more or less the same thing over and over again, done in different industries, but you have to remember it a lot, and it's not glamorous. I had no idea when we started this conversation.

I think I understood a little better when I started this. I told you before we started that this is just for my own notification, but even people who influenced you, like Peter Taylor saying, "Don't be a fool," would obviously say, "Don't be a fool, this works, why are you doing anything other than what works?" So there must be examples where you say, " Hey, I've given this advice to other founders a million times," and then you're surprised, "Damn, I 'm not even applying my own advice right now." Completely.

Yes, I will give many examples of that. I think it 's also instructive, what, what was new...? Why didn't you have the council? And what was really different about OpenAI compared to everything I had seen before is that it took four and a half years from when we started the company until we launched our first product. The opposite of YC's advice, right? Yes. Okay. Yes. And while there were all these ways in which managing a research team was similar to selecting and advising founders, learning what we did—to the extent that we learned it, because I don't think we did it perfectly—about how to manage this part of the world where you don't have the external signal from customers and you're just trying to...you know, do what would normally be the catastrophic startup advice of not, uh, not launching a product for four and a half years.

That was very difficult, and we tested all these things about how we replaced the signal of whether customers actually like the product with whether our research actually works. One of the things that actually worked was that during the Dota 2 days when we were trying to use real-world gameplay to win this video game, to beat people at a video game, we put a sort of leaderboard in place and people could just see how the different ideas worked out and that was, you know, objective and real and people wanted to climb up it.

But we had to test all these things to, uh, basically simulate end users, and that was a totally interesting new problem for which I had no reference pattern. How did you manage to overcome that? What was your reasoning? So, how did you do this? We asked a lot of people who had been in large research labs in the past, and it had been, you know, OpenAI started as a time when everyone in Silicon Valley, as their vanity project, including myself, wanted to start a research lab.

And there were all these books about the heyday of Bell Labs or Xerox PARC that were very popular. Everyone was talking about it. There was a huge amount of discussion. In fact , I even see one of the books over there about, uh, about Bell Labs. But there weren't many people who had, in their living memory, how to actually do it. So we talked a lot with Alan Kay. We talked to a handful of other people and got some advice from them about, you know, what made a really good research lab, and some of it was very good.

Some of that didn't translate so well to the present moment. Well, you also didn't have this gigantic monopolistic money-printing machine like Bell Labs that spun off independently. Polaroid did much more research when they essentially had a monopoly on their photography. I just read the biography of the founder of Honda, you know? The guy created the most successful motor vehicle of all time. The Honda Cub has been sold continuously for about 60 years, millions of damn vehicles. And in his entire approach, he came to the same conclusion as Bell Labs, that he thought research and development really had to be separate.

He went, separated from the company, and had separate ownership, just like Bell Labs did. We didn't have that. No, they didn't have it . When I look back on those early days, I mainly feel like I was trying to raise money and failing. That's kind of my dominant memory of the early days of OpenAI. A lot of effort, very frustrating. I wish we had had some kind of money machine like that. I remember it as one of my clearest memories of all of OpenAI. announced as we were leaving 2015.

But the first day was right after New Year's in 2016 and about 11 or 12 of us showed up at Greg Brockman's apartment , you know, around 9:30 on a Monday or Tuesday morning, something like that . Let's say it's January 4th, everyone is there, and it had been a great effort, and everyone enters with a lot of excitement. It feels like the first day of school, whatever. And then, very quickly, people look around the room and it's like, well, what do we do now? Someone says, "Okay, we should get a whiteboard." Greg, you know, has someone go and get a whiteboard.

The blackboard arrives. Um, we looked around again. You know, what are we supposed to do now? And you simply feel the energy in the room plummet. And none of us knows what to do. Since there wasn't one, it wasn't like creating a product startup. It wasn't like, "Let's build this product." "We're going to talk to the customers." It's like, okay, we said we want to create AGI. Perhaps we should write some articles. Okay, let's write some articles. Perhaps we should think of some ideas. Okay, let's think of some ideas.

You know, everyone has their "I have no idea what I'm doing" moments. That was one of mine. As I have, you know, we just launched this. None of us have any idea what we're going to do. So we did what we knew how to do and eventually discovered that many things didn't work. Eventually we discovered a kind of rhythm for making and then evaluating research bets, and it was all far from perfect, obviously. Uh, but we found a gradient along which we could progress and figured out how to get, you know, the resources that very intelligent people needed and how to make sure we didn't get completely lost in nature .

And over a number of years, mostly through chaotic stumbles, we eventually made most of the great discoveries. You know, what started as an unsupervised feeling on paper ended up becoming GPT1 and then GPT whatever ; The work on loss of scale that gave us confidence, not only in the computation, but in understanding how to scale our models, simply came together, and many other things as well. Um, through this process, we learned things like the idea of leaderboards did work. We also learned the incredible power of external demonstrations, such as for an eminent person whom the researchers really wanted to impress.

And then we learned a lot of things that didn't work, like fake deadlines. That must be very disconcerting to live through, to go through that experience. You're in an apartment with 12 people , you don't even have a whiteboard, you don't know what to do. Fast forward a decade and you have a billion people using it. A very strange experience. Do you keep a diary? When my son, my first son, was born, I would come home at the end of the day, rock him to sleep, and just talk to him or whatever.

So, I just needed to invent things to talk about. So I would tell him about my day, what we were struggling with, what was worrying me, and what was going on. It was fun for me, and I thought it was interesting and that someday it would be interesting for him to have this. So I started writing to him every Sunday. I was writing her a letter. He would just talk for the sake of talking and then write it down. I only made like eight of them or something.

How many children do you have? Two. Okay. Bezos has this great quote about building Amazon. He says, " We're trying to do things that we can talk to our grandchildren about and be proud of, right?" "And those things are difficult ." The fact that you were writing to your son. Oh, man. Keep writing the letters. And if you don't , this is quick only because I 've read enough books on this subject most of the time. Guess what? Founders don't write autobiographies when they're 40. They write them when they are 70 and look back and wish they could do it again.

And much has been lost in the sands of time. Everyone repeats this. They say, "I wish I had kept a diary." So even if you don't , you have enough resources. What I would do: write, have a book written, even if it's only for internal purposes. Have you ever read The Little Kingdom by Michael Moritz? I never read it. Oh, you have to do it. It's the story of Apple's first six years, written by Michael Moritz. Isn't it crazy that he wrote that book? He is a phenomenal writer.

Writer. He ends up being one of the best venture capitalists of all time, I guess. But the point is how that book ends. Steve hasn't even been fired from Apple yet. So you understand what really happened. You're going to want this. You may not want it now, but you'll definitely want it when you're 60 or 70 years old. What was so interesting was the mindset of writing for your child. You really can't hide behind anything. It's like I really care what my children think of me.

So, if something happened and you didn't feel good about it, it's best to do things differently next week. It was an extremely interesting mindset . Maybe I'll find a way to do it again. Oh, maybe you will. Okay, Sam, thanks for taking the time. This was incredible, man. I appreciate it. I hope you enjoyed this episode. Please remember to subscribe wherever you listen and leave a review. And be sure to listen to my other podcast, Founders, which has been around for almost a decade. I've obsessively read over 400 biographies of the greatest entrepreneurs in history, looking for ideas you can use in your work.

Most of the guests you hear on this show first found me through Founders.