Tyler and Y Combinator co-founder Paul Graham sat down at his home in the English countryside to discuss what areas of talent judgment his co-founder and wife Jessica Livingston is better at, whether young founders have gotten rarer, whether he still takes a dim view of solo founders, how to 2x ambition in the developed world, on the minute past which a Y Combinator interviewer is unlikely to change their mind, what YC learned after rejecting companies, how he got over his fear of flying, Florentine history, why almost all good artists are underrated, what’s gone wrong in art, why new homes and neighborhoods are ugly, why he wants to visit the Dark Ages, why he’s optimistic about Britain and San Fransisco, the challenges of regulating AI, whether we’re underinvesting in high-cost interruption activities, walking, soundproofing, fame, and more.
Listen to the full conversation
Read the full transcript
TYLER COWEN: Hello, everyone, and welcome back to Conversations with Tyler. Today I’m here with Paul Graham. Paul, welcome.
PAUL GRAHAM: Thank you.
COWEN: You’ve written several times that your wife, Jessica Livingstone, is a better judge of character than you are. What other areas of talent judgment is she better, or much better, than you?
GRAHAM: Practically everything to do with people. She’s a real expert on people and social protocol — what to wear at events, what to say to someone, [laughs] anything like that.
COWEN: But you have all this data, so why can’t you learn from the data and from her to do as well as she can? What’s the binding constraint here in limiting someone as a judge of human affairs?
GRAHAM: Well, partly, I just don’t have as much natural ability, and partly, I just don’t care as much about these things. The conversations we have are not “Paul, you can’t wear that.” They are like, “Paul, you can’t wear that.” [laughs] I’m like, “Really? Why not? What’s wrong with this?” I just don’t care as much.
GRAHAM: Oh, it’s all the people. The earlier you’re judging start-ups, the more you’re just judging the founders. It’s like location for real estate. The founders are like that, so yes, sure. Which is why the earlier stage you invest, the more you want to have these people who are good judges of people.
COWEN: If she’s better than you at human affairs, comparing you to other people, what exactly, in the smallest number of dimensions, are you better at than the other people?
GRAHAM: What am I better at?
COWEN: Yes. What are you better at? You can’t be so terrible at this, right?
GRAHAM: You mean to do with start-ups?
COWEN: To do with people — how well they will do with their start-ups. What’s the exact nature of your comparative advantage?
GRAHAM: I can tell if people know what they’re talking about when they come and talk about some idea, especially some technical idea. I can tell if they know, if they actually understand [laughs] the idea, or if they just have a reading-the-newspaper level of understanding of the thing.
COWEN: How well can you judge determination?
GRAHAM: Determination — it’s hard to judge by looking at the person. The way you would judge determination — because people act determined; people think they’re supposed to act determined. No one walks into their Y Combinator interview thinking, “Well, I need to seem diffident.” [laughs] Nobody thinks that. They all think they’re supposed to seem tough, which is actually really painful and stressful. It’d be better if they just were themselves because someone trying to act tough — it’s so painful to watch.
But the way you tell determination is not so much from talking to them as from asking them stories about things that have happened to them. That’s where you can see determination.
COWEN: It’s how they tell the story, or it’s what happened?
GRAHAM: No, it’s what they did in the story, right?
COWEN: What they did in the story.
GRAHAM: Something went wrong, and instead of giving up, they persevered.
COWEN: Why are there so few great founders in their 20s today, saying that they are —
GRAHAM: Is that true, though? I’ve heard about that.
COWEN: Ten, fifteen years ago, you had a large crop of people who, obviously, have become massive successes. Today, it’s less clear. Companies seem to be smaller. The important people seem to be older. Sam Altman was important early on, but now he’s very, very, very important. He’s what, 37, 38?
GRAHAM: Something like that.
COWEN: Seems like more of a synthetic set of abilities we need.
GRAHAM: I’m not even sure this is true. It could be that it’s not actually true, and that there’s something else going on. If you look at athletes, for example, athletes learned how to stay good for longer. You have people who are really international quality soccer players at age 34. That didn’t used to happen back in the days of Johan Cruyff, smoking cigarettes between halves. [laughs] He didn’t last till 34. It might have seemed like all the good football players are 34, and it’s just an artifact of them lasting longer.
COWEN: But the 20-year-old stars seem to have vanished. There’s not a next Patrick Collison, who was evidently so.
GRAHAM: Well, Patrick Collison probably didn’t seem to most people to be the big star, and you can prove this, because if he had, they would have invested in early Stripe rounds, and they didn’t. Anybody who claims they knew early on that Patrick Collison was going to be a big star — show me the equity.
GRAHAM: Sure, yes. Some people did, but certainly not everyone. I don’t know if this is true. I’ll investigate this summer when we’re going to go back to YC and talk to some founders, and I’ll talk to the partners, and I’ll see what’s going on, because I had a mental note to check and see if this is actually true.
COWEN: There’s a Paul Graham worldview in your earlier essays, where you want to look for the great hackers, and a lot of the most important companies come from intriguing side projects.
GRAHAM: Oh, yes.
COWEN: But if we want older people who are somewhat synthetic in their abilities, are those other principles still true?
GRAHAM: No, not necessarily. Not as true. Part of the reason you want to find young founders who’ve done stuff from side projects is that it guarantees the idea is not bullshit. Because if young founders sit down and try to think of a start-up idea, it’s more likely to be bullshit because they don’t have any experience of the world.
Older founders can do things like start supersonic aircraft companies, or build something for geriatric care, and actually get it right. Younger founders are likely to get things wrong if they try and do stuff like that. It’s just a heuristic. It’s a heuristic for finding matches between young founders and ideas.
COWEN: Why is venture capital such a small part of capital markets? It’s big in tech, it’s somewhat big in biotech, but most other areas of the economy, you finance with debt or retained earnings or some other method. What determines where VC works and doesn’t work?
GRAHAM: Well, growth. You’ve got to have high growth rates because VCs — it’s so risky investing in start-ups, investing in these early-stage companies that maybe don’t even have any revenues. The only thing that can counterbalance the fact that [laughs] half the companies completely fail is that the ones that don’t fail sometimes have astonishing returns.
COWEN: Can you imagine a future where there’s a lot more tech in terms of the level, but slower rates of growth, and VC quite dwindles because the growth rates aren’t there? Or VC goes to some other area?
GRAHAM: Well, VC has been dwindling in the sense that they have smaller and smaller percentages of companies. If you go back and look at the stories from the 1980s, they would do these rounds where they would get 50 percent of the company. Even when YC started, they would get 30, and now it’s down to 10 in these rounds — God knows what they’re called; they keep renaming the rounds. For a given round size, the amount of equity they get is much smaller than they used to.
COWEN: It at least seems we have a new, dynamic Microsoft that ships products quickly and innovates and, on the surface, appears to act like a start-up again. How can that be true? We all know your famous essay about Microsoft culture. What has happened?
GRAHAM: I did not say that they wouldn’t make money.
COWEN: But they’ve done more than make money. They’ve impressed us with their speed.
GRAHAM: Well, I meant something very specific in that essay. Incidentally, I talk in that essay about how I was talking to a founder, and I was talking about how Microsoft was this threat. You can see he was clearly puzzled, like, how could Microsoft possibly be threatening? I didn’t mention who the founder was, but it was actually Zuck. It was Mark Zuckerberg, [laughs] who was puzzled that Microsoft could be a threat.
Still to this day, if you ask founders, “Are you afraid that Microsoft might do what you’re doing?” None of them are. It’s still not a threat to start-ups. Yes, it makes more money now, but it’s still not a threat like it used to be. It doesn’t matter in the sense of factoring into anyone’s plans for the future. No start-up is thinking, “Well, I better not do that because Microsoft might enter that and destroy me.”
COWEN: In the early years of Sam Altman, what did you see in him other than determination? Because he’s not a technical guy in the sense of —
GRAHAM: He is a technical guy. He was a CS major.
COWEN: But you’re not buying the software he programmed, right? There’s something about what Sam Altman does that —
GRAHAM: Well, now he’s become a manager, but he knows how to program.
COWEN: But that’s not why his ventures have succeeded, right?
GRAHAM: Oh, no, he does a lot more than —
COWEN: There’s some ability to put the pieces together.
GRAHAM: Yes, but he’s not a nontechnical guy. He’s not just some business guy. He’s a technical guy who also is very formidable, and that’s a good combination.
COWEN: There’s a now-famous Sam tweet where he appears to repudiate his earlier advice of finding product-market fit early and then scaling, and saying, “Oh, maybe I was wrong saying that.” What OpenAI has done is not exactly that. Do you agree, disagree?
GRAHAM: Well, maybe OpenAI is a special case because you have to have giant warehouses full of GPUs. You can’t just mess around and throw something out there [laughs] and see if it works. Maybe it requires some amount of advanced planning for something on that scale. I don’t know. I don’t know what he meant by it, and I also don’t know what it’s like inside OpenAI, really. I don’t know either.
COWEN: Do you still take a dim view of solo founders?
GRAHAM: Well, it’s harder. I know I wouldn’t start a start-up alone because it’s just so much weight to bear to do something like that.
COWEN: But it’s very hard to find a partner as good as you are, right? It’s harder and it’s easier. You can just do it.
GRAHAM: Well, people should do some work. When I talk to people who are in their teens or early 20s about starting a start-up, I tell them, “Instead of sitting around thinking of start-up ideas, you should be working with other people on projects. Then, you’ll get a start-up idea out of it that you probably never would have thought of, and you’ll get a co-founder too.” I wish people would do more of that. You can get co-founders — just work with people on projects. You just can’t get co-founders instantly. You’ve got to have some patience.
COWEN: Why is there not more ambition in the developed world? Say we wanted to boost ambition by 2X. What’s the actual constraint? What stands in the way?
GRAHAM: Boy, what a fabulous question. I wish you’d asked me that an hour ago, so I could have had some time to think about it between now and then.
COWEN: [laughs] You’re clearly good at boosting ambition, so you’re pulling on some lever, right? What is it you do?
GRAHAM: Oh, okay. How do I do it? People are, for various reasons — for multiple reasons — they’re afraid to think really big. There are multiple reasons. One, it seems overreaching. Two, it seems like it would be an awful lot of work. [laughs]
As an outside person, I’m like an instructor in some fitness class. I can tell someone who’s already working as hard as they can, “All right, push harder.”
It doesn’t cost me any effort. Surprisingly often, as in the fitness class, they are capable of pushing harder. A lot of my secret is just being the person who doesn’t have to actually do the work that I’m suggesting they do.
COWEN: How much of what you do is reshuffling their networks? There are people with potential. They’re in semi-average networks —
GRAHAM: Wait. That was such an interesting question. We should talk about that some more because that really is an interesting question. Imagine how amazing it would be if all the ambitious people can be more ambitious. That really is an interesting question. There’s got to be more to it than just the fact that I don’t have to do the work.
COWEN: I think a lot of it is reshuffling networks. You need someone who can identify who should be in a better network. You boost the total size of all networking that goes on, and you make sure those people with potential —
GRAHAM: By reshuffling networks, you mean introducing people to one another?
COWEN: Of course.
COWEN: You pull them away from their old peers, who are not good enough for them, and you bring them into new circles, which will raise their sights.
GRAHAM: Eh, maybe. That is true. When you read autobiographies, there’s often an effect when people go to some elite university after growing up in the middle of the countryside somewhere. They suddenly become more excited because there’s a critical mass of like-minded people around. But I don’t think that’s the main thing. I mean, that is a big thing.
COWEN: Think of it as the power of London in the 17th century. The Industrial Revolution happens further north, but the ideas, the science, in or near London, maybe Cambridge.
GRAHAM: And Oxford.
COWEN: Those are the networks people are brought.
GRAHAM: Near London only in the sense that everything in England is near everything else by American standards. It’s not really that London was the center of ideas. There were a lot of smart people there, but things were more spread out. Be careful with English history there.
Back to this idea, though, of how to get people to be more ambitious. It’s not just introducing them to other ambitious people. There is a skill to blowing up ideas, blowing up not in the sense of destroying, like making them bigger. There is a skill to it, to take an idea and say, “Okay, so here’s an idea. How could this be bigger?” There is somewhat of a skill to it.
COWEN: It’s helping people see their ideas are bigger than they thought.
GRAHAM: Yes. Oh, yes. We often do this in YC interviews.
COWEN: People say you’re especially good at that. This is what the other people say.
GRAHAM: Well, that’s why I’m mulling over what actually goes on, because there is this skill there. The weird thing about YC interviews is, in a sense, they’re a negotiation. In a negotiation, you’re always saying, “Oh, I’m not going to pay a lot for that. It’s terrible. It’s worthless.” Yet in YC interviews, the founders often walk out thinking, “Wow, our idea is a lot better than we thought,” just because of what we do.
You know what we do in YC interviews? We basically start YC, the first 10 minutes of YC is the interview. You see what it’s like to work with people by working with them for 10 minutes, and that’s enough, it turns out.
COWEN: So, you think the 11th minute of an interview has very low value.
GRAHAM: I’ve thought a lot about where the cutoff is. Like, where’s the point? If you made a graph, what’s your probability of changing your mind after minute number N? After minute number one or two, the probability of changing your mind is pretty high. I would say YC interviews could actually be seven minutes instead of ten minutes, but ten minutes is already almost insultingly short, so we kept it at ten. We could have made it seven.
COWEN: I think there’s often a threshold of two, and then another threshold at about seven, and after that, it’s very tough for it to flip.
GRAHAM: Yes. Although that doesn’t mean you’re always right.
COWEN: It could just be, after three hours, you would still be wrong.
GRAHAM: It’s just not going to flip. I didn’t say seven minutes is enough to tell, notice. [laughs] I said seven minutes is the point where you’re probably not going to change your mind.
COWEN: If it’s going somewhat badly, and the person is flipping positive at minute six, what is it that’s happening, both in the interview and in you?
GRAHAM: If it changes from two to seven, uh —
COWEN: Clearly, from zero to one or two, they get over nerves, or they adjust the sound volume. There are plenty of those stories.
GRAHAM: It’s probably that we misunderstood what they’re working on initially.
COWEN: So, great idea, bad at presenting it?
GRAHAM: No, more like they’re near some idea that we’re familiar with, and we just assume they must be doing that idea. And they say, “Oh, no, no, no, we’re not doing that. We’re doing this.” I’m like, “Oh, okay. Thank goodness.” Then, they get extra points because not only they’re not doing the stupid thing, but they understand that the stupid thing is stupid. They get extra credit for what we were subtracting for it in the past.
COWEN: Who falls through the cracks in the YC process, as you’ve experienced it?
GRAHAM: Well, YC — one of the reasons I’m so contemptuous of university admissions is that I am also in the admissions business. I am obsessed. [laughs] We not only measure when we fail, but we’re obsessed with the failure cases. YC has a list of all the companies that we’ve missed, that have applied to YC and we’ve turned down, and they’ve gone on to be successful. We spend a lot of time, like someone thinking about past injuries.
COWEN: What’s the common element, though, now that you think of it?
GRAHAM: When there are common elements, my God, do we act quickly to fix that. I remember early on, there was this company doing email, if you wanted to send mass emails. I forget what their name was. This was back when we were reading the applications. If the first reader gave it a sufficiently low grade, it would never be seen by anybody else.
The first reader was Robert Morris. He was designated wet blanket of YC. He gave this application a C, with the comment “spam company.”
Nobody ever saw it again. They ended up going public. After that, I changed that thing in the software that made it so that no second reader would see it. After that, every application had to be seen by at least two people.
COWEN: How did you get over your fear of flying?
GRAHAM: Oh, really? Did I already talk about this? Do you know the answer to this?
COWEN: There are two or three places in your writing where you mentioned that you had a fear of flying, but you used the past tense, which implies you got over it, but you never told anyone how.
GRAHAM: Well, it’s a bizarre strategy. I learned how to hang glide, which sounds crazy.
COWEN: [laughs] That will do it, though, right?
GRAHAM: If you’re afraid of flying, how could you learn how to hang glide? The answer is, you learn how to hang glide gradually. You start by just running along the flat. If there’s a headwind, maybe you feel a little lift. Then you go 10 feet up the hill, run as fast as you can, and you reach a total altitude above ground level of like a foot. You’re not afraid when you’re a foot above the ground. You go out a little further up the hill until a month later, you’re jumping off a cliff with a hang glider on your back.
After I was good at hang gliding, I took flying lessons. There’s this intermediate point where I was totally comfortable jumping off a cliff with a hang glider on my back, moderately comfortable flying a Cessna 172 — where the instructor had just turned off the engine [laughs] and said, “Okay, land it,” because the glide ratio is actually similar to a hang glider — and still afraid of getting on an airliner.
That shows you how irrational these things are. But when I finally did get on an airliner, my God, it was like a spaceship [laughs] compared to the planes I’d been flying. It was fabulous. It totally worked. My fear of flying was completely cured.
COWEN: What did you learn about start-ups and talent selection from that process?
GRAHAM: Boy, if I learned anything, I haven’t considered it until this moment.
COWEN: But you might have learned it quite well, just not articulated it.
GRAHAM: I don’t know. You ask me these hard questions. I don’t have time to think about them. I don’t think I’ve ever consciously used anything I learned there. It was certainly convenient to be able to —
COWEN: Do you see founders who go through a comparable process with something other than flying, and it’s like, “Oh, I get where you’re at, and here’s what you need to do next.”
GRAHAM: Yes, a lot of technical-type founders hate the idea of doing sales and going and talking to people, and so you can tell them, “Look, just go do this.” You know, after a few months, they’ll be used to it. They may never like it. I would hate doing that myself now, but they can at least do it.
COWEN: Were the Medici good venture capitalists, or do you give greater credit to the Florentine guilds?
GRAHAM: I have no idea. I need to learn more about the —
COWEN: The guilds would run competitions. The Medici would just pick the people they liked. They both have good records in different ways, but obviously, they’re competing models.
GRAHAM: You’re an economist. You’ve read books about this stuff. I don’t know. What do I know about the Medici? [laughs]
COWEN: I think the Medici were overrated, and the guilds were more important, but that’s a debatable view. You could argue it either way.
GRAHAM: They could well be overrated. They were the ones who had all the publicity. I have no idea.
GRAHAM: Man, if you asked how it affected my ideas about painting, I can give you an answer.
COWEN: We’ll get to that, but Florence was all about talent selection because how many people in quite a small area became both great and famous?
GRAHAM: One thing I remember — there was one moment I was sitting in Florence, and I realized I had gone to the wrong place. Because I was studying Florentine history, all the buildings are right around you, and I found myself thinking, “Damn, Florence was New York City [laughs] in 1450.”
COWEN: That was 60,000 people, yes.
GRAHAM: That’s why it was good at art. It wasn’t some weird Florentine thing about art or some special Florentine sense of aesthetics. They were just the most progressive city. It’s really too bad that the far left has hijacked that word. It’s such a good word. Maybe we can get it back.
GRAHAM: I remember thinking, “I’ve gone to the wrong place.” I went to the place where the puck used to be over hundreds of years ago. [laughs]
COWEN: You can learn a lot from studying where the puck used to be.
GRAHAM: Oh, not really.
COWEN: Are you sure?
GRAHAM: The Academia was a pretty crappy art school.
COWEN: Being in Florence itself had —
GRAHAM: Oh, I can learn a lot from looking at the works, but I don’t have to go and live there.
COWEN: What is it you learned that was relevant for Y Combinator? Because you’re doing something comparable to what the Florentines did: picking a pretty small area, making it the absolute center for talent selection, a magnet that drew people in, and having a lot of winners. You copied what you were living.
GRAHAM: I had already learned that from Harvard.
COWEN: Maybe you had to learn it twice. [laughs]
GRAHAM: No, no, no. I was already very well aware of this phenomenon, to the extent YC uses anything like that. We were definitely thinking. YC was started within the convex hull of Harvard, [laughs] like the places Harvard spread out through Cambridge, but YC’s original office was within it. We’ve consciously tried to make YC the Harvard of start-ups. No question about that. We had the model right there.
COWEN: Harvard’s very screwed up, as you know. You look at their admissions — how much is dean’s favorites and legacies and affirmative action? It’s not Y Combinator. They’re trying to build a coalition, and you’re just caring about picking winners.
GRAHAM: The thing is, though, all universities have an admissions process that’s corrupt in that way, except possibly Caltech. Caltech might actually do undergrad admissions properly. So, since they’re all messed up, Harvard does still have this draw compared to the others.
COWEN: Does AI make programming even more like painting?
GRAHAM: God, what a question. How do you make these questions? Does AI make programming more like painting? I have no reason to believe that. [laughs] It might be true.
COWEN: You’re piecing things together more, arguably, in this new post-GPT world.
GRAHAM: Well, what I was seeing when I said programming was like painting is that they’re both building something. You’re not building something any more with AI than when you were writing code by hand, so I would guess not.
COWEN: How far is Midjourney from “real art”?
GRAHAM: I don’t know.
COWEN: Is it more decrepit modernism? Is it a fantastic revolution? Or, as art — put aside the start-up angle — how do you view it?
GRAHAM: Well, I see all these AI-generated images, and I don’t know which ones are from Midjourney and which ones aren’t. I can’t say for sure about Midjourney, but I have definitely seen some AI-generated stuff that looks amazing, that looks truly impressive.
COWEN: As art. Not just that it looks impressive. It’s amazing as graphic design, but do you think it’s art in the same way that Rembrandt is art?
GRAHAM: This whole thing about what’s art and what isn’t? I think it’s all a matter of degree. My crap carnation coffee mug is art. It’s just not very good art. [laughs] There’s not some threshold, where above this threshold it’s art. Everything people make is art, just to varying degrees of goodness. I can tell you, some of the things I’ve seen that were AI-generated, I’d be impressed if a person made them. That probably is over your threshold.
COWEN: If you’re good at talent selection, who is an underrated painter and why?
GRAHAM: Ah, wow. [laughs] Boy, there is a topic I think about a lot. There’s a bunch of different reasons people can be underrated. Almost all good artists are underrated.
COWEN: I agree.
GRAHAM: It sounds weird, but if you look at where the money’s spent at auction, it’s almost all fashionable contemporary crap because if you think about how prices in very high-end art are set, they’re auction prices. How many people does it take to generate an auction price? Two. Just two. So, you have boneheaded Russians who want to have a Picasso on their wall so people will think they’re legit, or hedge fund managers’ wives who’ve been told to buy impressive art to hang in their loft so when people come over, they’ll say, “Oh, look, they’ve got a Damien Hirst.”
The way art prices at the very high end are set is almost entirely by deeply bogus people, [laughs] which is great, actually. When I was an artist, I used to be annoyed by this. Now that I buy a lot of art at auction, I’m delighted because it means there’s all this money. You see Andy Warhol’s screen prints selling for $90 million.
COWEN: Yes. Old masters can be, I wouldn’t say cheap, but I would say radically underpriced.
GRAHAM: A couple hundred thousand.
COWEN: Or even less for some good ones.
GRAHAM: Yes, I know because I buy them. [laughs] I used to be annoyed by this, and now I think it’s the most delightful thing in the world because there’s all this loose money sloshing around, and so-called contemporary art is like this sponge that just absorbs all of it. There’s none left. Some of the things I buy, I am the only bidder. I get it for the reserve price. No one else in the world wants it, or even knows that it’s being sold, so I am delighted about this.
The answer to your question, which artists are undervalued? Essentially, all good artists. The very, very, very famous artists, artists famous enough for Saudis to have heard of them — Leonardo, I would say, is probably not undervalued. But except for the artists who are household names — every elementary school student knows their names — they’re all undervalued.
COWEN: If you think that something has gone wrong in the history of art, and you tried to explain that in as few dimensions as possible, what’s your account of what went wrong?
GRAHAM: Oh, I can explain this very briefly. Brand and craft became divorced. It used to be that the best artists were the best craftspeople. Once art started to be reproduced in newspapers and magazines and things like that, you could create a brand that wasn’t based on quality.
COWEN: So, you think it’s mass media causing the divorce between brand and craft?
GRAHAM: It certainly helps.
COWEN: Then talent’s responding accordingly. Fundamentally, what went wrong?
GRAHAM: You invent some shtick, right?
GRAHAM: And then — technically, it’s called a signature style — you paint with this special shtick. If someone can get some ball rolling, some speculative ball rolling, which dealers specialize in, then someone buys the painting with your shtick and hangs it on the wall in their loft in Tribeca. And people come in and say, “Oh, my goodness, that’s a so and so,” which they recognize because they’ve seen this shtick. [laughs]
COWEN: Say, if we have modernism raging in the 1920s, and the ’20s mass media is radio for the most part —
GRAHAM: No, no, no, newspapers were huge. Modernism was well —
COWEN: But not for showing paintings, right? There’s no color in the papers. You had to be —
GRAHAM: Well, the 1920s were good enough to make painting like Cezanne fashionable. It was just about getting going in the 1920s.
COWEN: Why can’t we build good British country homes anymore?
COWEN: Or do you think we can?
GRAHAM: Well, there’s nothing stopping you —
COWEN: But it doesn’t happen, right?
GRAHAM: — except the planning people. Do you know anything about building houses in England? You just cannot build.
COWEN: It’s impossibly difficult. It’s one of the worst countries, right?
GRAHAM: Well, it’s the reason it looks so nice here.
GRAHAM: If they had America’s zoning rules here, the entire countryside would be plastered with houses. You wouldn’t have any fields left at all because the difference in value between agricultural and building land — it’s like, God, 100X or something like that, and this place is small.
COWEN: Yes, but there are new buildings, say, in Cambridge, many other parts of Britain, and the new buildings are not country homes. Yet everyone, on average, is wealthier. Why the change?
GRAHAM: Wealthier than where?
COWEN: Wealthier than Britain in the old days. Even before the railway, you have large numbers of country homes, and most country homes being built when living standards were pathetically low.
GRAHAM: A country house has got to have a certain amount of land around it. You can’t just have them plastered down a street. It would look wrong. You want a different style of architecture or something like that. There is a place where they build big houses.
You know what it is? It’s like when the Macintosh appeared, and you could have whatever font you wanted, right? Most people have bad tastes. In the old days, you had to have a classical-looking house because that was the only way to build houses. In the Victorian period, actually, was when things went wrong. You could have an Italian villa, or wait, no, it could be a Greek temple, or [laughs] something that looked like it was from the Tudor period. Take your pick and mix them together, Greek temple with Tudor bits. It was all over from that point on, really.
It’s not like they build good houses in America either.
COWEN: No, neighborhoods are worse yet. There are nice individual homes, but is there any truly beautiful neighborhood built after 1950? Where would it be? In any country, put aside Asia?
GRAHAM: No, in Palo Alto, there are neighborhoods of houses built by this guy called Eichler, this developer who hired some of the best midcentury architects. And arguably, those neighborhoods are good, although they messed up the trees.
COWEN: When were those built?
GRAHAM: ’50s and ’60s.
COWEN: Okay, but that’s 70 years ago. Now, I mean, we haven’t done anything in 70 years? What’s your model of that?
GRAHAM: I’m sure someone is building some good houses somewhere. The point is, you don’t really need to, and so it only happens by accident. The developers mostly are thinking, “We need to just turn this land into houses as soon as possible. The buyers don’t have any taste. We don’t need to sweat that, so we’ll just build random houses that look big and have large master bedrooms. People will buy them. We’ll get our capital back and go on and do the next one.”
They don’t need to be good. Houses don’t need to be good. They didn’t need to be beautiful in the old days, either. It was just technology was so constrained, you didn’t have any choice.
COWEN: You get to go back in time. Your health is guaranteed, and you know whichever languages you might need, and you spend six months somewhere safe. Your safety is guaranteed. Where do you choose?
GRAHAM: I often think about this question.
COWEN: You seem like someone who often thinks about this question.
GRAHAM: Yes, sadly, the way I think about it is, I keep trying to escape from the obvious answer, and I don’t manage to, because the obvious answer to that is Athens, right? Which is, that’s where everyone would pick.
COWEN: It’s not what I would pick, but what’s your choice number two, then? I’ll tell you mine, I think.
GRAHAM: Well, there are things where I am obsessed with the mystery of what the hell was going on in Dark Age Europe. I’m deliberately using the term “Dark Age” because they’re trying to outlaw it. [laughs]
COWEN: It’s correct, I think.
GRAHAM: If the term “Dark Ages” hadn’t been invented already, that would be a fabulous invention to describe that period. Yes, sure, new things were getting invented. Some people were doing good things just like always happens. But it was as bad as things have been. There are so few records, nobody knows how things happened.
I would be really interested to know when barbarians infilled the Roman Empire, but there were still big Roman landholders, and they had to give a large percentage of their estates to these new barbarians, who were in a sense running things. They were in the cockpit, but they didn’t know what the buttons did.
COWEN: You want to go to Northumbria or northern France? Or where do you want to be?
GRAHAM: No, no.
COWEN: Dark Ages — you’ve picked the era, now give us the spot.
GRAHAM: Like Provence.
GRAHAM: Provence in 600. What was going on? What was going on in Provence in 600? I would be very interested to see that. Almost like morbid curiosity. I wouldn’t learn as much as I would in going to Athens, [laughs] but I just really want to know what was going on.
COWEN: I’m a big fan of morbid curiosity, by the way. It’s the best kind of curiosity, in many cases.
GRAHAM: I can tell.
COWEN: I would pick the Aztec Empire before the Spanish arrived. I feel I have vague glimpses of what ancient Athens was like. There are still Greek ruins. The Europeans marveled at the cities they saw, and they burnt all the books, and I think I would learn more by somewhere more strange.
GRAHAM: Maybe, maybe. The reason I’m interested in medieval Europe is, it’s where our world came from — the clocks we use, the writing system, all the clothes eventually evolved from that. The reason I’m interested — I long ago realized that the medieval period wasn’t a dip. It wasn’t like there was this high level of civilization, and then it dropped down for a while, and then it rose back up again.
It was more like there was one civilization that was high and went down, and another civilization based in the north that rose up. In a sense, it was the beginning of everything. That’s why I want to know what the hell was going on in 600 or 700.
COWEN: Did Rome have to fall? Or can you imagine a path where Rome has an industrial revolution, and we save ourselves 700 or however many centuries of time? Like you need the fragmentation to get the competition for Britain to become significant?
GRAHAM: Maybe, maybe. I don’t know. Oh, 2,000 years is a long time, or even 1,700 years. Actually, Roman was half decent —
COWEN: Because China never falls.
GRAHAM: — until 200, so, 1,500 years. But 1,500 years — things could have changed a lot.
COWEN: But did they in China? Chinese Empire never collapses. It takes many forms, many dynasties, but it ends up stalling. So maybe the collapse of the Roman Empire was one of the best things that could have happened, for Europe, at least.
GRAHAM: I’m sure people didn’t think so at the time. People were thinking at the time, “It didn’t have to be this bad.” [laughs] I don’t know. Neither of us knows about this kind of thing.
COWEN: Looking forward, how optimistic are you about the future of the UK? No real wage growth since 2008, no real productivity growth for as many years. What’s up, and what’s the path out of that?
GRAHAM: I am optimistic because they still have a gear that they haven’t shifted into. I suppose I’ll really have gotten native when I say “we” instead of “they.”
COWEN: [laughs] But you grew up in Pittsburgh, right? Near Pittsburgh.
GRAHAM: Yes, but I’m British by birth.
COWEN: Does that count? Are you really British, or your parents were diplomats here?
GRAHAM: No, no, no. Yes, I’m really British.
COWEN: You’re really British?
GRAHAM: Yes, yes. Just ask HMRC. As far as they’re concerned, I never left. [laughs] HMRC is the British IRS.
Okay, the reason I’m optimistic about Britain — I was just thinking about this this morning — is because people here are not slack. They’re not lazy, and they’re not stupid, and that’s the most important thing. Eventually, non-lazy, non-stupid people will prevail.
I’ve funded a couple of start-ups here. You can see, when you introduce these people to the idea of trying to make something grow really fast and have these really big ambitions, it’s like teaching them a foreign language, but they do learn it. They do learn it. It’s not like English people are somehow genetically inferior to Americans. I think they have all that potential still to go. I’m astonished when I see statistics. I think GDP per capita in the UK is only two-thirds of what it is in America.
COWEN: It’s about the same as Mississippi.
GRAHAM: It’s preposterous, I know. Imagine the potential there. Imagine the potential.
COWEN: You don’t feel that way about Mississippi, necessarily?
GRAHAM: No, I think Mississippi’s probably already up close to its full potential. I don’t know. I’ve never actually been there. I shouldn’t say things like that.
COWEN: Why and how did so many things here end up undercapitalized? The water utilities, the NHS — it seems to be a consistent pattern. That could make us more pessimistic because the flow numbers don’t reflect the fact that capital maintenance is even worse than we had thought.
GRAHAM: I don’t know. You’re an economist. I don’t know what the term capital maintenance means.
COWEN: But you’re a British person, I’m told.
GRAHAM: Undercapitalized — I don’t know what these things mean.
COWEN: They’ve been postponed.
GRAHAM: The NHS is run by the government, so things run by governments are often bad. Although the NHS seems to be pretty good. Even though people attack it, it’s a lot more civilized than the American system.
COWEN: How long it takes an ambulance to arrive if you call one up?
GRAHAM: Well, that’s only recently got bad. That’s just in the last couple of years.
COWEN: That’s what undercapitalization means. You keep on borrowing against the future. You don’t plow resources back in, and then at some point you don’t have anymore.
GRAHAM: Actually, when things get bad enough, they fix things. This place is not run by the kind of yahoos that America is. It may be a small country, but people running things — they’re not just boneheaded political appointees. [laughs] When things are wrong, they notice they’re wrong, and they fix them. This is a very old country. That’s another reason it’s not going to tank. They’ve been through some bad stuff before. There have been ups and downs.
COWEN: Are you an optimist about the city of San Francisco? Not the area, the city.
GRAHAM: Yes, I am.
COWEN: Tell us why.
GRAHAM: I can’t tell you because there are all sorts of things happening behind the scenes to fix the problem.
COWEN: In politics, you mean, or in tech start-ups?
GRAHAM: No, no, no, politics. The problems with San Francisco are entirely due to a small number of terrible politicians. It’s all because Ed Lee died. The mayor, Ed Lee, was a reasonable person. Up till the point where Ed Lee died, San Francisco seemed like a utopia. It was like when Gates left Microsoft, and things rapidly reverted to the mean. Although in San Francisco’s case, way below the mean, and so it’s not that it didn’t take that much to ruin San Francisco. It’s really, if you just replaced about five supervisors, San Francisco would be instantly a fabulously better city.
COWEN: Isn’t it the voters you need to replace? Those people got elected, reelected.
GRAHAM: Well, the reason San Francisco fundamentally is so broken is that the supervisors have so much power, and supervisor elections, you can win by a couple hundred votes. All you need to do is have this hard core of crazy left-wing supporters who will absolutely support you, no matter what, and turn out to vote.
Everybody else is like, “Oh, local election doesn’t matter. I’m not going to bother.” [laughs] It’s a uniquely weird situation that wasn’t really visible. It was always there, but it wasn’t visible until Ed Lee died. Now, we’ve reverted to what that situation produces, which is a disaster.
COWEN: Now, we’re in 2023. Say, two or three years from now, what do you think the regulation of AI will look like?
GRAHAM: Oh, God knows. You keep asking me questions I have no idea about.
COWEN: You have plenty of ideas. You know as much as anyone, I suspect.
GRAHAM: No, that’s not true. That is not true. [laughs] I really have no idea what AI regulation will look like, or even should look like, which is an easier question.
COWEN: Here’s an easier question, yes. There is a broadly tech community in the Bay Area.
GRAHAM: If you said I had to make up the regulations for AI this afternoon, it would be really hard.
COWEN: I agree.
GRAHAM: [laughs] That’s how far away I am from being able to answer that question on the spot. I couldn’t even figure it out in a day.
COWEN: Like when it should be modular, or when there should be a regulation on AI is a thing that itself is intractable.
Making safe AGI is like writing a secure operating system. It's hard to do it, and it's easy to convince yourself you've done it.
Anyone who actually manages to do it, knows that it's hard. If you've written an OS and you think making it secure was easy, that thing ain't secure
— Rob Miles (@robertskmiles) April 29, 2023
GRAHAM: I’ll tell you one meta fact though. There was a guy on Twitter — I think his name was Rob Miles — who said that trying to make safe AI will be like trying to make a secure operating system.
That is absolutely true, and therefore frightening, because the way you make a secure operating system is not by sitting down and thinking at a table with a piece of paper about the principles for making a secure operating system. More like you try and make such principles, and then someone hacks your operating system. [laughs] Then you think, “Oh, okay, sorry.” You patch it, and then they hack that.
Making a secure operating system is like making a fraud-proof tax code. It’s basically a series of patches that were based on successful hacks. Is it Mellon? They say the US tax code is, basically, a series of responses to things Mellon did.
COWEN: Some parts of it, yes.
GRAHAM: Secure operating systems are like that. I’m worried about AI because you’re not going to be able to figure out — whatever the regulations are, they’ll be wrong, I’ll tell you that. They’ll be overregulated in some ways, and miss and just have huge holes in others.
COWEN: There’s a Bay Area tech community. At least in the not-too-distant past, they agreed about many things, but very recently, it seems, on AI, there’s quite a divergence of views. People who are very, very worried: “Oh, it’s going to kill us all in the world.” People who say, “Oh, there are problems, but this is going to be great.”
In as few dimensions as possible, what accounts for that difference in perspective from people with broadly similar backgrounds, and who used to agree on many things? Is it temperament? Is it genes? Is it like a snake bit when —
GRAHAM: It’s probably which aspect of the problem they choose to focus on. You could focus on either one. If you focus on how it could be good, there are all sorts of exciting things to discover, and you discover lots of genuine ways it could be good. If you focus on how it could be bad, it’s true there, too. So it could be either.
I manage to keep both thoughts in my head simultaneously. I simultaneously think there will be all kinds of good things and all kinds of bad things. They will be unimaginably good and unimaginably bad. [laughs] It sounds like that produces oscillation, doesn’t it?
GRAHAM: That’s worrying. That in itself is worrying.
COWEN: Why hasn’t Lisp been more successful? Or do you think it has?
But why doesn’t everybody use Clojure or some other dialect of Lisp? Because the notation is frightening. There’s an initial hump with the notation, and if you give people initial hump — you must know about this. There must be names for this in economics.
If you put some sort of obstacle, like a container, right in front of people’s front door, they’ll go off to the left, and then they won’t go right back in front of the container and resume their original path. No, they’ll take this other path that goes miles out of their way, just because of that one block in front of their front door. The syntax, the reverse Polish notation, puts people off.
COWEN: Is AI-generated programming going to vindicate you on Lisp over time, or cousins of Lisp?
GRAHAM: No, no.
COWEN: Because the AI doesn’t care about the notation, right?
GRAHAM: It does.
COWEN: It does?
GRAHAM: Because it’s trained based on the amount of code that’s out there.
COWEN: But you could train it on something more like Lisp if you wanted to.
GRAHAM: You can’t. You have to train on actual examples people have written.
COWEN: But you have AI write some code in a perpetual motion machine, train other AIs on that code, and converge to something better, better, better. No?
GRAHAM: There was some research paper recently, where they trained an AI on the output of AIs, and it converges on crap. Maybe there’ll be some solution because it is a very rapidly evolving field, but I think you have to have a large corpus, initially, of examples written in the language.
COWEN: Other than hackers and hacking, what other human activities are what you have called high-cost interruption? That is, if you’re interrupted, you lose your train of thought, have to start all over again.
GRAHAM: Oh, math, I think must be like that. I think anything.
COWEN: Painting — yes, or no?
GRAHAM: Not as much in my experience, not as much. You don’t have to think ahead as much in painting. It’s annoying to be interrupted, no matter what, but it doesn’t absolutely destroy you, like it does in the middle of writing a program or something. You don’t build a big mental model of something in your head when you’re painting. That’s what it is. It’s when you’ve got this giant house of cards in your head. That’s when you get destroyed by an interruption.
COWEN: Do you think kids today spend too much or too little time learning high-cost interruption activities?
GRAHAM: Do they spend too much or too little time working —
COWEN: Like, are we underinvesting in high-cost interruption activities?
GRAHAM: Well, if you think about what it’s like in schools, kids are constantly being interrupted, and they always have been.
COWEN: Of course.
GRAHAM: So, whatever the answer is, it’s not going to be kids these days. It will be a statement about school for centuries. Kids have such short attention spans. The one thing they can’t do is things that are big and long. I think if you said, “Okay, you have five hours to sit in a quiet room and build something,” I don’t think they’d be up to it, anyway.
COWEN: Our last segment is what I call the Paul Graham production function, which is how you got to be Paul Graham. Would you major in philosophy again at Cornell, if you were doing it all over?
GRAHAM: No, I would not major in philosophy.
COWEN: Why not? Didn’t it allow you to think at a very general level?
GRAHAM: Well, it’s hard to say. It’s hard to say. It’s good to be able to take ideas and flip them around like a Rubik’s Cube, and take them apart, and notice the two parts are the same shape or something like that, but I don’t think I actually learned that in philosophy classes. I think I would’ve learned that in classes about anything hard.
I mistakenly thought that you could just go and learn the most abstract truths. It sounds great to a high school student. “Why do I have to learn all the specific crap? I’ll just learn the most general truths.” Needless to say, that’s one of those things that sounds too good to be true, and it is, because if you go and look in philosophy classes . . .
I remember when Bill Clinton was saying, “It depends on what the meaning of ‘is’ is.” I’m like, “Hey, that’s what I majored in, [laughs] what the meaning of ‘is’ is, literally.” I think it was pretty much a waste.
COWEN: Which kinds of ideas come more naturally to you while you’re walking?
GRAHAM: Which kinds?
COWEN: Yes. Maybe not painting ideas, but some kind of ideas, because you’ve written about how using walking to learn ideas is a good thing, but which kinds of ideas? There’s cross-sectional variation, right?
GRAHAM: Yes. Well, ideas about whatever you’re thinking about. I remember, until a few years ago, I was working very intensely on programming. I had the problem I was working on loaded into my head, and whenever I was doing something without any interruptions, I would start to think about that, so I think it’s good for whatever you happen to be thinking about. Mathematicians, apparently, walk a lot.
COWEN: I find it best for learning from what the other person knows, not so good from my own ideas. It’s better —
GRAHAM: What, when you’re walking with someone else?
COWEN: With someone else, and I’m talking with them, and I learn from them better. I don’t find that I’m very generative when I’m walking.
GRAHAM: Walking makes all kinds of thinking better. I’ve seen images — like MRI images or something like that — of brain activity. I don’t know how they do MRI images, [laughs] but some kind of images of brain activity. Your brain is definitely more active when you’re walking. Classic YC office hours were to walk down the block and talk as you walk, which also has the side benefit that you are side-by-side, and not looking the other person in the face, which I think may be better. It’s certainly better than —
COWEN: It’s less threatening. It’s like confession in the church. You don’t see the priest. Or you’re on a therapist’s couch — probably you’re not looking right at them, and vice versa.
GRAHAM: Or you’re driving your kid back to boarding school. You’re taking kids —
COWEN: And they’ll say things they would not otherwise have told you.
GRAHAM: Or talk at all. [laughs]
COWEN: What are the best environments for learning while walking? Urban, British countryside, or —
GRAHAM: Oh, my God.
COWEN: How do you optimize this dose?
GRAHAM: I actually have —
COWEN: I think Britain is wonderful for learning while walking because it’s never too hot. You heat up while you’re walking.
GRAHAM: It doesn’t pour with rain on you like it did today? I was drenched today.
COWEN: It ends in four minutes. You can bring an umbrella. You run to the gazebo, but there is a gazebo.
GRAHAM: There was no gazebo where I was this morning.
I think it’s good to always walk in the same place. You don’t want to see things that distract you. If you’re trying to have ideas, you’re not going to get ideas from things you see. Probably not. Not relevant ones. You want to walk in the same place, and it should be something where there’re no distractions, so I would think the countryside. Where I go walk is on this preserved medieval common — to an American, it would look like a large park — and it’s the perfect thing. I just always take the same route. There’s not much on it. [laughs] I see grass and trees. That’s about it.
COWEN: What’s the most important thing you’ve learned about soundproofing?
GRAHAM: Boy, there’s a good question. I’ve learned a few things about soundproofing. Can I only say one thing?
COWEN: One thing is fine. The most important.
GRAHAM: No, no. You said, “What’s the most important?” Can I say some more?
COWEN: Oh, no. You can say more than one.
GRAHAM: Okay. Well, one is that sound comes through holes. It doesn’t come through your walls; it comes through your windows, probably, in most places. If you fix the holes, you fix the noise problem.
The other thing I’ve learned is basically, the solution — it’s either multiple layers, in the case of windows, or simply mass. You make some big, thick door, and make it have hinges that make it sink down. That’s what recording studios have. When the door opens, it rises up a little bit, and when it closes, it goes shoomp right down onto the floor.
So, great big, heavy doors, multiple-paned windows. And, weirdly enough, I’ve managed to soundproof some places so effectively that I’ve noticed this phenomenon you only notice with soundproofing — all kinds of things make annoying noises you never noticed before.
COWEN: That’s right. It’s a war of attrition of sorts.
GRAHAM: Yes, exactly. Soundproofing is worth it, though. Quiet is really good, at least for me.
COWEN: There’s an optimal level of fame. Do you feel you have too much or too little?
GRAHAM: What is the optimal level of fame? I suppose it’s when you can get resources you need or something like that, or if there’s someone you need to talk to, they’ll talk to you. I can now talk to most people I need to talk to. If I want to talk to somebody, I can find somebody who will introduce me, so that must be enough.
COWEN: Are you past the optimal?
GRAHAM: I don’t know. That’s the thing. This sounds very arrogant, but I realized this with Y Combinator. I realized that Y Combinator had become famous long after it had become famous. As far as I was concerned, I was just doing what we’d always been doing. Every six months, we would get all these applications. We’d have to find the needle in the haystack, [laughs] get all these start-ups, help them grow, find investors for them, and then it would start again.
It didn’t seem like YC was any different. It was the same building, the same people. We would get more start-ups. Meanwhile, YC was starting to be considered as this giant gatekeeper for Silicon Valley or something like that.
COWEN: But Jessica knew it was different, right?
GRAHAM: You’re not aware. Famous people don’t know how famous they are, unless they’re experts on it, like movie stars or something like that. They’re always basically taken by surprise. We were especially taken by surprise because the thing is, the companies we funded would grow until they had thousands of employees, but YC itself didn’t grow.
YC’s market-wise — the value of the portfolio grew with these giant companies, but we didn’t see it. We were still just a few people doing the same thing we’d always been doing, so how could we be famous? I discovered that was one of the biggest mistakes I made with YC. I didn’t realize how many people were watching us. I thought we could just keep doing what we were doing, and nothing really mattered.
COWEN: Why was that a mistake, per se? Maybe it was better being oblivious.
GRAHAM: No, because when — basically, anybody outside Silicon Valley who wants to blame Silicon Valley for something, well, who do they blame? They’ve never heard of the people who are actually powerful in Silicon Valley. They only know a handful of people who have consumer brands, me among them. Basically, the world sucks because of tech, and tech sucks because of Paul Graham, [laughs] because they’ve never heard of any of the other people.
I don’t seem to get quite so much of that anymore. I don’t know. I’m glad about that. With YC, definitely, I didn’t know how prominent YC was becoming, and how many people would be out to get us as a result.
COWEN: Very last question. In my view, a life properly lived is learn, learn, learn all the time.
GRAHAM: That’s what Charlie Munger said, right?
COWEN: Yes. Now, what have you recently been learning about, other than soundproofing?
GRAHAM: Well, Tintoretto. [laughs]
COWEN: What are you learning?
GRAHAM: Well, Vasari had a very low opinion of him.
COWEN: Vasari is unreliable on most things, right?
GRAHAM: I don’t know. I don’t know.
COWEN: He way overrated his patrons, the Medici.
GRAHAM: Yes. You have a thing about the Medici, clearly. He said that Tintoretto was too independent-minded, that Tintoretto was a mad genius, and that he would’ve been better if he had constrained his creativity and stayed within the limits of proper art. You know what I mean? A very Florentine sort of idea. I think Tintoretto would have looked down on Vasari as a minor-league artist, but that was interesting. That was interesting to learn that’s how at least some of Tintoretto’s contemporaries viewed him, and Vasari in particular.
COWEN: And a Y Combinator co-founder is not going to buy that argument, is he?
GRAHAM: Which Y Combinator founder?
COWEN: Of Vasari’s, that he was too radical, and too off on his own.
GRAHAM: Oh, I thought you meant Jessica was and agreed with me about Tintoretto.
COWEN: Well, there are two of you. Neither of you would agree with Vasari.
GRAHAM: I don’t know. I’m now going to look. I never thought about this, but I was just looking at some Tintorettos. I was just in Venice, looking at the Scuola Grande di San Marco, where all those Tintorettos are, and they were so dirty. It was hard to tell [laughs] what the paintings actually looked like, [laughs] but I’m going to go look. I’m going to go look and see if they seem freakish.
COWEN: Paul Graham, thank you very much.
GRAHAM: [laughs] Thank you. Boy, that was so many hard questions.
Photo credit: Dave Thomas