Nat Friedman (Github CEO) — Reading ancient scrolls, open source, & AI

It is said that the two greatest problems of history are: how to account for the rise of Rome, and how to account for her fall. If so, then the volcanic ashes spewed by Mount Vesuvius in 79 AD - which entomb the cities of Pompeii and Herculaneum in South Italy - hold history’s greatest prize. For beneath those ashes lies the only salvageable library from the classical world. Nat Friedman https://nat.org/ was the CEO of Github form 2018 to 2021. Before that, he started and sold two companies - Ximian https://www.wikiwand.com/en/Ximian and Xamarin https://www.wikiwand.com/en/Xamarin. He is also the founder of AI Grant https://aigrant.org/ and California YIMBY http://cayimby.org/. And most recently, he has created and funded the Vesuvius Challenge https://scrollprize.org/ - a million dollar prize for reading an unopened Herculaneum scroll for the very first time. If we can decipher these scrolls, we may be able to recover lost gospels, forgotten epics, and even missing works of Aristotle. We also discuss the future of open source and AI, running Github and building Copilot, and why EMH is a lie. Watch on YouTube https://youtu.be/qcvMjoJdck4. Listen on Apple Podcasts https://podcasts.apple.com/us/podcast/nat-friedman-reading-ancient-scrolls-open-source-ai/id1516093381?i=1000605340304, Spotify https://open.spotify.com/episode/2JRJ3aANOA41aguVKhvtGB?si=7zbCbz8UQW2n1qfXqkP7nA, or any other podcast platform. Read the full transcript here https://www.dwarkeshpatel.com/p/nat-friedman. Follow me on Twitter https://twitter.com/dwarkesh_sp for updates on future episodes. Timestamps (0:00:00) - Vesuvius Challenge (0:30:00) - Finding points of leverage (0:37:39) - Open Source in AI (0:40:32) - Github Acquisition (0:50:18) - Copilot origin Story (1:11:47) - Nat.org (1:32:56) - Questions from Twitter This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.dwarkesh.com https://www.dwarkesh.com?utm_medium=podcast&utm_campaign=CTA_1

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Speaker 00

  We have 600 plus kind of roughly intact scrolls that we can't open. And I heard about this and I thought that was incredibly exciting, like the idea that there was information from 2,000 years in the past. We don't know what's in these things. If we could read all of them, then that would give us approximately doubling of the total text that we have from antiquity. If there are thousands more papyrus scrolls in there and we now have the techniques to read them, then there's gold in that mud. And you know, it's got to be dug out. I just fundamentally don't believe the world is efficient. And so if I see an opportunity to do something, I used to, but I no longer have a reflexive reaction that says, oh, that must not be a good idea. If it were a good idea, someone would already be doing it.

• 0:42Dwarkesh Patel?

  Okay. Today I have the pleasure of speaking with Nat Friedman, who was the CEO of GitHub from 2018 to 2021. Before that, he started and sold two companies, Ximian and Zimerin. And he is also the founder of AI Grant and California Yimby. And most recently, he is the organizer and funder of the Scroll Prize, which is where we'll start this conversation. So Nat, do you want to tell the audience about what the Scroll Prize is?

• 1:09Speaker 00

  We're calling it the Vesuvius Challenge.

• 1:11Dwarkesh Patel?

  Okay.

• 1:12Speaker 00

  And this is just this crazy and exciting thing. I feel like incredibly honored to have gotten caught up in. But a couple of years ago, I was reading, it was the midst of COVID and I think we were in lockdown. And like everybody, I was sort of falling into internet rabbit holes. And I just started reading about the eruption of Mount Vesuvius in Italy about 2,000 years ago. And it turns out that when Vesuvius erupted, it was AD 79. It destroyed all the nearby towns. Everyone knows about Pompeii. But there was another nearby town called Herculaneum. And Herculaneum was sort of like the Beverly Hills to Pompeii. So big villas, big houses, fancy people.

• 1:57Speaker 00

  And in Herculaneum, there was one villa in particular. It was enormous. And it had once been owned by the father-in-law of Julius Caesar. And so, well-connected guy. And it was full of beautiful statues and marbles and art. But it was also the home to a huge library of papyrus scrolls. And so, when the villa was buried, the volcano actually, it spit out enormous quantities of mud and ash. And it buried Herculaneum in particular in something like 20 meters of material. So it wasn't like a thin layer. It was a very thick layer.

• 2:36Speaker 00

  Those towns were buried and forgotten for hundreds of years. No one even knew exactly where they were until the 1700s. And so, in 1750, a farm worker who was digging a well, kind of in the outskirts of Herculaneum, struck this marble paving stone of a path that had been at this huge villa. And of course, he was pretty far down when he did that. He was, you know, 60 feet down. And then subsequently, this Swiss engineer came in and started digging tunnels from that well shaft. And they found all these treasures. And that was sort of the spirit at the time was like looting. You know, they were taking out like incredible, they would, if they encountered a wall, they would just bust through it.

• 3:18Speaker 00

  And they were taking out these beautiful bronze statues that had survived. And along the way, they kept encountering these lumps of what looked like charcoal. They weren't sure what they were, and many were apparently thrown away. Until someone noticed a little bit of writing on one of them. And they realized they were papyrus scrolls. And there were hundreds, there may have been thousands of them. And so, they had uncovered, really, the, this enormous library is the only library ever to have sort of survived in any form, even though it was badly damaged, you know, they were sort of carbonized, very fragile, deformed. The only one that survived since antiquity. In the open air, these papyrus scrolls in like a Mediterranean climate,

• 3:58Speaker 00

  they rot and they decay quickly. And so, they'd have to be recopied by monks like every 100 years or so, maybe even less. And so we only have, it's estimated, you know, something less than 1%, less than 1% of all the writing from that period. And so to find underground hundreds of definitely not in good condition, but still present, you know, papyrus scrolls where on a few of them, you can make out the lettering was like this enormous discovery. People immediately, in a well-meaning attempt to read them, started trying to open them, but they're, they're really fragile, like, you know, they're like, they turn to ash in your hand. And so hundreds were destroyed. People did things like cut them with daggers down the middle and, you know, a bunch of

• 4:41Speaker 00

  little pieces would flake off and they'd try to like get a few letters off of a couple of pieces. And then eventually there was a monk named Piaggio, who's an Italian monk, and he devised this machine, kind of under the care of the Vatican, to unroll these things very, very slowly, like half a centimeter a day, something like that. And a typical scroll, I think, could be 15 or 20 or 30 feet long. And managed to successfully unroll a few of these. And they found on them Greek philosophical texts in the Epicurean tradition by this little-known philosopher named Philodemus. But, but we got kind of new texts from antiquity, which is, you know, not a thing that happens all the time.

• 5:25Speaker 00

  Eventually, people stopped trying to physically unroll these things because so many were destroyed. And in fact, some attempts to physically unroll the scrolls continued even into like the 2000s, like 1990s, 2000s, and, and they were destroyed. So the current situation is we have 600 plus, kind of roughly intact scrolls that we can't open. And I heard, I heard about this and I thought that was incredibly exciting, like the idea that there was information from 2,000 years in the past. We don't know what's in these things. And obviously, people are trying to develop new ways and new technologies to open them. And I read about a professor at the University of Kentucky, Brent Seals.

• 6:08Speaker 00

  Who had been trying to scan these using increasingly advanced imaging techniques, and then use computer vision techniques and machine learning to kind of virtually unroll them without ever opening them. And they tried a lot of different things, but their most recent attempt in 2019 was to take the scrolls to a particle accelerator in Oxford, England, called the Diamond Light Source, and to make essentially an incredibly high-resolution CT scan of sort of 3D X-ray scan. And they needed really high-energy photons in order to do this, and they were able to take scans at eight microns, so these really quite tiny voxels, which they thought would be sufficient.

• 6:52Speaker 00

  And I thought this was like the coolest thing ever. You know, we're using technology to sort of read this lost information from the past, and I sort of waited for the news that they had been decoded successfully. So that was, you know, 2020. And then I think COVID hit. Everybody got a little bit slowed down by that. And last year, I found myself wondering, I wonder what happened to, you know, Dr. Seals and his scroll project? And I reached out, and it turned out they'd been making really good progress. You know, they'd gotten some machine learning models to start to identify ink inside of the scrolls, but they hadn't yet extracted words or passages. It's very challenging. And I invited him to come out to California and hang out, and to my shock, he did.

• 7:37Speaker 00

  And we got to talking and decided to team up and try to crack this thing. And the approach that we've settled on to do that is to actually launch an open competition. We're gonna, we've done a ton of work with his team to get the data into a shape where, and the tools and techniques and just the broad understanding of the materials into a shape where smart people can kind of approach it and get productive easily. And then I'm putting up, together with Daniel Gross, a prize, you know, sort of like an X Prize or something like that, for the first person or team who can actually read like substantial amounts of real text from one of these scrolls without opening them.

• 8:20Speaker 00

  And so we're launching that this week, you know, I guess maybe it's when when this airs. I don't know. The stakes are kind of big, like this, like what gets me excited are the stakes. So the six or 800 scrolls that are there, it's estimated that if we could read all of them. And, you know, like somehow the technique works and it generalizes to all the scrolls, then that would give us approximately a doubling of the total text that we have from antiquity. This is what historians and classicists tell me. So it's not like, oh, we would get like a 5% bump or 10% bump in the total ancient Roman or Greek text. It would be like, no, we get all of the text that we have again. You know, multiple Shakespeare's is sort of one of the units that I've heard. So, so that that would be significant.

• 9:05Speaker 00

  I mean, we don't know what's in there, you know, we've got a few Philodemos texts. Those are of some interest, but there could be lost epic poems or God knows what. So I'm really excited and I think, you know, my bet is there's like a 50% chance that someone will encounter this opportunity and get the data and get nerd sniped by it. And we'll solve it this year.

• 9:29Dwarkesh Patel?

  I mean, really, it is something out of a science fiction novel, you know? It's like something you'd read in Neil Stephenson or something. I was talking to Professor Seals before, and apparently the shock went both ways because the first few emails, he was like, this has got to be spam. Like, no way Nat Friedman is reaching out and has found out about this prize. And

• 9:47Speaker 00

  That's really funny because he was really pretty hard to get in touch with. So like I, I emailed him a couple of times, just like didn't respond. And so I was like, so I asked my admin, Emily, to call the secretary of his department and say like, Mr. Friedman requests me. And then like he knew there was something like actually going on there. And so he finally got on the phone with me and we got on Zoom and he's like, why are you interested in this?

• 10:15Speaker 00

  I mean, I love Brent. He's fantastic. And I think, you know, we're like friends now. And I think we found that we think alike about this. And I think he's reached the point where he just really wants to crack the, you know, they've taken this right up to the one yard line. Like, this is doable at this point. They've demonstrated, I think, every key component, putting it all together, improving the quality, doing it at the scale of a whole scroll. This is still very hard work. And an open competition seems like the most efficient way to get it done.

• 10:46Dwarkesh Patel?

  Before we get into the state of the data and the different possible solutions, I want to make tangible, like, what could be gained if we can unwrap these. So you said there's a few more thousand scrolls. Are we talking about the ones in Philodemus's layer? Are we talking about the ones in other layers?

• 11:02Speaker 00

  You know, you'd think if you find this crazy villa that was owned by Julius Caesar's father-in-law, that we just like dig the whole thing out. But in fact, most of the exploration occurred in the 1700s through the Swiss engineers tunnels underground. So it was never, the villa was never dug out and exposed to the air. You went down 50, 60 feet, and then you dig tunnels. And, you know, again, they were looking for treasure, not like a full archaeological exploration. So they mostly got treasure. In the 90s, some additional excavations were done, kind of at the edge of the villa, and they discovered a couple of things. First, they discovered this, like, it was a seaside villa, it faced the ocean, it was right on the water before the volcano erupted. The eruption actually

• 11:44Speaker 00

  pushed the shoreline out by depositing so much additional mud there. So it's no longer right by the ocean, apparently. I've actually never been. And they also found that there were two additional floors in the villa that the tunnels apparently had never excavated. And so, at most, a third of the villa has been excavated. Now, they also know when they were discovering these papyrus scrolls, that they found basically one little room where most of the scrolls were. And these were mostly these Philodemus texts, at least that's what we know. And they found apparently several revisions, sometimes of the same text. And so they think, the hypothesis is, this was actually Philodemus's working library.

• 12:28Speaker 00

  He worked here, this, this sort of Epicurean philosopher. And In the hallways, though, they occasionally found other scrolls, including crates of them. And the belief is, at least this is what historians have told me, and I'm no expert, but what they have told me is they think that the main library in this villa has probably not been excavated, and that the main library may be a Latin library and may contain, you know, literary texts, historical texts, other things, and that it could be much larger. Now, I don't know how prone these classicists are to wishful thinking. It is a romantic idea, but they have some evidence, you know, in the presence of these partly evacuated sort of scrolls that were found in hallways and that sort of thing.

• 13:12Speaker 00

  So there are descriptions, you know, I've since gone and read a bunch of the firsthand accounts of the excavations, and there are these heartbreaking descriptions of them finding like an entire case of scrolls in Latin and like accidentally destroying it as they tried to get it out of the mud. And, you know, there were maybe 30 scrolls or something in there. So there clearly was some other stuff that we just haven't got to.

• 13:38Dwarkesh Patel?

  I mean, you made some scrolls.

• 13:41Speaker 00

  Yeah.

• 13:42Dwarkesh Patel?

  Right?

• 13:43Speaker 00

  Yeah, so papyrus, this is papyrus, and it's a reed, it's a grassy reed that grows on the Nile in Egypt, and for thousands of years, many thousands of years, they've been making paper out of it. And the way they do it is they take the kind of outer rind off of the papyrus and then they cut the inner core into these strips, and they lay the strips out kind of parallel to one another, and then they put another layer at 90 degrees to that bottom layer, and they press it together in a press or under stones and let it dry out. And that's papyrus, essentially. And then they'll take some of those sheets and kind of glue them together with paste made out of flour usually, and get a long scroll. And you can still buy it.

• 14:27Speaker 00

  I bought this on Amazon, and it's interesting because it's got a lot of texture, you know, those fibrous ridges of the papyrus plant. And when you write on it, you really feel the texture. And I got it because I wanted to understand sort of what are these artifacts that we're working with. And so we made an attempt to simulate carbonizing a few of these. So we basically took a Dutch oven, because when you carbonize something and you make charcoal, it's not like burning it with oxygen. You sort of remove the oxygen, heat it up and let it carbonize. So we tried to simulate that with a Dutch oven, which is probably imperfect, and left it in the oven at 500 degrees Fahrenheit for maybe, I don't know how long these were, but our biggest attempt was like five or six hours. And they really,

• 15:10Speaker 00

  and so these things are incredibly light. And if you try to unfold them, they just fall apart in your hand very readily. I assume these are in somewhat better shape than the ones that were found, because these were not in a volcanic eruption, you know, and like covered in mud. I think the volcano was probably, maybe that mud was hotter than my oven can go. So, and they just flakes, you know, just sort of, you just squeeze it. It's just, it's just dust in your hand.

• 15:34Dwarkesh Patel?

  Yeah.

• 15:35Speaker 00

  And so we actually tried to replicate many of the heartbreaking 1700s, 18th century unrolling techniques, like they used rosewater, for example, or they tried to use different oils to soften it and unroll it. And most of them are just very destructive. They poured mercury into it because they thought mercury would slip between the layers potentially. So, so yeah, this is sort of what they look like. They, you know, they shrink and they turn to ash.

• 16:00Dwarkesh Patel?

  Yeah. For those listening, by the way, it kind of looks like a black, I mean, just imagine sort of the ash of a cigar, but blacker, and it crumbles the same way. It's just a blistered, black piece of rolled up papyrus.

• 16:12Speaker 00

  Yeah. And they blister, the layers can separate, they can fuse.

• 16:16Dwarkesh Patel?

  And so this happened in 79 AD, right? So we know that anything before that could be in here, which I guess could include.

• 16:22Speaker 00

  Yeah. So what could be in there? I don't know. You know, we, you and I have speculated about this.

• 16:28Dwarkesh Patel?

  Right.

• 16:29Speaker 00

  Well, I think, you know, it would be extremely exciting not to just get more Epicurean philosophy, although that's fine, too. But almost anything would be interesting and additive. Like dreams are, I think it would maybe have a big impact to find something about early Christianity, like a contemporaneous mention of early Christianity. Maybe there'd be something that, you know, the church wouldn't want. That would be exciting to me. Maybe there'd be something, you know, some color detail from someone commenting on Christianity or Jesus. That like, I think that would be a very big deal. We have no such things as far as I know. Other things that would be cool would be old stuff, like even older stuff. So there were several scrolls already found in there that they know were hundreds of years old when the villa was buried.

• 17:14Speaker 00

  So the villa was probably constructed about 100 years prior, is my understanding. And they can tell from the style of writing, they can date, you know, they can date some of these scrolls. And so there was some old stuff in there, and the Library of Alexandria was burned 80 or 90 years prior. And so, again, maybe wishful thinking, but there's some rumors that some of those scrolls were evacuated, and maybe some of them would have ended up at this substantial, prominent, you know, Mediterranean villa. God knows what would be in there. That would be really cool. I think it'd be great to find literature. Like personally, I think that would be exciting, like beautiful new poems or stories. We just don't have a ton,

• 17:58Speaker 00

  because so little survived. And so I think, I think that would be fun. I think you had the best crazy idea for what could be in there, which was text which was GPT-watermarked. That would be a creepy feeling.

• 18:13Dwarkesh Patel?

  I still can't get over just how, how much of a plot of a sci-fi novel this is like, right? Like, potentially the biggest intact library from the ancient world that has been sort of stopped like a debugger because of this volcano. And I mean, just the philosophers of antiquity forgotten, like the earliest gospels. There's so much interesting stuff there. But let's talk about what the data looks like. So you mentioned that they've been CT-scanned, and that they built these machine learning techniques to do segmentation and the unrolling. What would it take to get from there to an actual, to understand the actual content of what is within?

• 18:55Speaker 00

  Dr. Seals actually pioneered this field of what he calls, and is now widely called, virtual unwrapping. And he did it actually not with these Herculaneum scrolls. These things are like expert mode. They're so difficult. I'll tell you why soon. But he did it with, initially with a scroll that was found in the Dead Sea in Israel. It's called the Enghedde scroll. And it was carbonized actually under like slightly similar circumstances. I think there was a temple that was burned. The papyrus scroll was in a box. So it kind of, it's like a Dutch oven. It kind of carbonized in the same way. And so it was not openable. It'd just fall apart. And so the question was, could you non-destructively read the contents of it?

• 19:35Speaker 00

  And so he did this 3D x-ray, the CT scan of the scroll. And then was able to do two things. First, the ink gave a great x-ray signature. And so it looked very different from the papyrus. It was a high contrast. And then second, he was able to segment the wines of the scroll, you know, throughout the entire body of the, of the scroll, and identify each layer. And then just geometrically unroll it using, you know, fairly normal flattening computer vision sort of techniques. And then read the contents of it. And it turned out to be, I think, an early part of the Book of Leviticus, you know, something of the Old Testament or the Torah.

• 20:18Speaker 00

  And that, you know, that was like a landmark achievement. And so then the next idea was to apply those same techniques to this case. And so, okay, this, this is proven hard. I think there's a couple things that make it difficult. One is that, the primary one is that the ink used on the Herculeanum papyri, it has, it is not very absorbent of x-ray. Like, it basically seems to be equally absorbent of x-ray as the papyrus. Or very close, certainly not perfectly. And so you don't have this nice, bright lettering that shows up kind of on your tomographic 3D x-ray. So you have to somehow develop new techniques for finding the ink in there. So that's sort of problem one, and it's been a major challenge.

• 21:02Speaker 00

  And then the second problem is the scrolls are just real messed up. Like, they were long and tightly wound, highly distorted by the, you know, volcanic mud, which not only heated them, but deformed, you know, partly deformed them. And And so just the segmentation problem of identifying each of these layers throughout the scrolls is, you know, it's doable, but it's it's hard. Those are a couple of challenges. And then the other challenge, of course, is just getting access to scrolls and taking them to a particle accelerator. So you have to like have scroll access and, you know, particle accelerator access and time on those. It's expensive and difficult. And, you know, Dr. Seals did the hard work of making all that happen.

• 21:46Speaker 00

  And so the good news is, very recently, just in the last couple of months, his lab has demonstrated with a convolutional neural network, the ability to actually recognize ink inside these X-rays. And, you know, you look, I look at the X-ray scans, and I cannot, at least in any of the renderings that we've seen, I can't see the ink, but the machine learning model can pick up on sort of very subtle patterns in the X-ray absorption at high resolution inside these volumes in order to identify ink. And we've seen that. And so you might ask, okay, like, how do you train a model to do that? Because you need some kind of ground truth data to train the model. So the big insight that they had was to train on broken off fragments of the pyrite.

• 22:30Speaker 00

  So as people tried to open these over the years, you know, in Italy, they destroyed many of them, but they saved some of the pieces that broke off. And on some of those pieces, you can kind of see lettering. And if you take an infrared image of the fragment, then you can really see the lettering pretty well in some cases. And so they, I think it's 930 nanometers, they take this little infrared image, now you've got some ground truth. Then you do a CT scan of that broken off fragment, and you try to align it, register it with the image, and then you have data that you can use potentially to train a model. And that turned out to work in the case of the fragments. Okay, so now I think, this is sort of the why now.

• 23:15Speaker 00

  This is why I think launching This challenge now is the right time because we have a lot of reasons to believe it can work. Like in the core techniques, the core pieces have been demonstrated. It just all has to be put together at the scale of these really complicated scrolls. And so, so yeah, I think if you can do the segmentation, which is probably a lot of work, maybe there's some way to automate it. And then you can figure out how to apply these models, you know, inside the body of a scroll and not just to these fragments, then it seems, it seems like you could probably read lots of text.

• 23:50Dwarkesh Patel?

  Hmm. Why should you do it, decide to do it in the form of a prize rather than just like giving a grant to the team that was already pursuing it, or maybe some other team that wants to take it up?

• 23:58Speaker 00

  We talked about that. But I, I think the, we, what we basically concluded was the search space of different ways you could solve this is pretty big. And we just wanted to get it done as quickly as possible.

• 24:11Dwarkesh Patel?

  Yeah.

• 24:12Speaker 00

  And so having a contest means lots of people are going to try lots of things. And, you know, someone's going to figure it out quickly. You know, many eyes may, may make it shallow as a task. And so I think that's the main thing. Like probably someone could do it, but I think this will just be a lot more efficient. And it's fun, too. I think this is fun. Like, I, I think it's interesting to do a contest and, you know, who knows who will solve it or how. People may, they may not even use machine learning. I, you know, we think that's the most likely approach for recognizing the ink, but they may find some other approach that we haven't thought of.

• 24:43Dwarkesh Patel?

  One question people might have is that you have these visible fragments mapped out. Do we expect them to correspond to the burned off or the ashen, carbonized scrolls that you can do machine learning on, the ground truth of one can correspond to the other?

• 24:59Speaker 00

  I think this is a very legitimate concern. They're different. Like, when you have a broken off fragment, there's air above the ink. And so when you CT scan it, you have kind of ink next to air. Inside of a wrapped scroll, the ink might be next to papyrus, right? Because it's pushing up against the next layer. And your model has to, your model, you know, may not know what to do with that. And so. Yeah, I think this is one of the challenges and sort of how you take these models that were trained on fragments and translate them to the slightly different environment. But maybe there's parts of the scroll where there is air on the inside and we know that to be true, you can sort of see that here. And so I think it should at least partly work if and clever people can probably figure out how to make it completely work.

• 25:41Dwarkesh Patel?

  Yeah. So you said the odds are about 50/50. What makes you think that?

• 25:45Speaker 00

  It can be done?

• 25:45Dwarkesh Patel?

  Yeah.

• 25:46Speaker 00

  I think it can be done because we've recognized ink on, from a CT scan on the fragments. And I think everything else is probably geometry and computer vision. The scans are very high resolution. So they're eight microns, eight micrometers. And they're taken, if you kind of stood a scroll on end like this, they're taken in these slices through it, right? Like this. So it's like this, in the z-axis from bottom to top, there are these slices. And the way they're represented on disk is each slice is a TIFF file. And for the full scrolls, each slice is like 100, 100 something megabytes. So they're quite high resolution. And then if you stack, for example, 100 of these, they're eight microns, right? So 100 of these is 0.8 millimeters. So, you know, millimeter is pretty small.

• 26:31Speaker 00

  So they're, you know, so they're fairly, we think the resolution is, is good enough, or at least right on the edge of good enough, that it should be possible. There's, there's sort of like, seem to be six or eight pixels, or voxels, I guess, per, you know, across an entire layer of papyrus. That's probably enough. And we've also seen with the machine learning models, Dr. Seals has got some PhD students who have actually demonstrated this at eight microns. So I think that the ink recognition will work. I think the data's in it. The data is clearly physically in the scrolls, right? The ink was carbonized, the papyrus was carbonized, but not as, like a lot of data actually physically survived. And

• 27:14Speaker 00

  Then the question is, did the data make it into the scans? And I think that's very likely, based on the results that I've seen so far, we've seen so far. And so I think it's just about a smart person solving this. And or a smart group of people, or just a dogged group of people who do a lot of manual work. That could also, you know, be true. You may have to be smart and dogged. But and I think that's where most of my uncertainty is. There's just like whether whether somebody does.

• 27:40Dwarkesh Patel?

  Yeah, I mean, if a quarter of a million dollars doesn't motivate you.

• 27:44Speaker 00

  Yeah, I think money's good. Yep, I mean, there's a lot of money in machine learning these days.

• 27:47Dwarkesh Patel?

  That's true. Do we have enough data in the form of scrolls that have been mapped out to be able to train a model? If that's the best way to go? I guess, because one question somebody might have is, listen, if you already have this ground truth, why hasn't Dr. Seal's team already been able to just train a model?

• 28:04Speaker 00

  I think they will. I think if we just let them do it, they'll get it solved. It might take a little bit longer because, you know, it's not a huge number of people. And there's probably there is a big search space here. But I mean, yeah, if we didn't watch this contest, I'd still think this would get solved.

• 28:17Dwarkesh Patel?

  Right.

• 28:17Speaker 00

  But it might take several years. And I think this way it's very it's likely to happen this year.

• 28:22Dwarkesh Patel?

  Okay, and what happens? Let's say, you know, the prize is solved. Somebody figures out how to do this and we can read the first scroll. You mentioned that these other layers haven't been excavated. How is the how is the how is the world going to react? Let's say we get one of these mapped out.

• 28:36Speaker 00

  That's my like, personal hope for this. So I always like to look for sort of these cheap leverage hacks, you know, these moments where you can do like a relatively small thing and it creates, you know, you kick a pebble and you get an avalanche.

• 28:46Dwarkesh Patel?

  Right.

• 28:46Speaker 00

  And the theory is, and you know, Brent shares this theory. The theory is that if you can read one scroll, like just one scroll, and we only have two scanned scrolls. There's hundreds of surviving scrolls. It's relatively expensive to use to book a particle accelerator. So if you can scan one scroll and you know it works and you can generalize the technique out and it's going to work on these other scrolls, then the money, which is probably low millions, maybe only $1 million to scan the remaining scrolls will just arrive. Like, it's just it's too sweet of a prize not for that not to happen. And the urgency and kind of return on excavating the rest of the villa will be incredibly obvious, too.

• 29:28Speaker 00

  Because if there are thousands more papyrus scrolls in there and we now have the techniques to read them. Then there's gold in that mud and you know, it's gotta be dug out. And you know, it's amazing how little money there is for archaeology. It's you know, literally for decades, no one's been digging there. So that's my hope is that like this, this is the catalyst that you know, it works. Somebody reads it, they get a lot of glory. We all get to feel great. And then the diggers arrive in Herculaneum and they dig out the rest.

• 30:00Dwarkesh Patel?

  I, I, I wonder if the budget for archaeological movies and games like Uncharted or Indiana Jones is bigger than the actual budget to do real world archaeology. But you know, I was talking to some of the people before this interview and that's one thing they emphasized is your ability to find these leverage points. For example, with California Y M B, I don't know the exact amount you seeded it with. But for that amount of money, it is, and for an institution that is that new, it is one of the very few institutions that has had a significant amount of political influence. Right? Like if you look at the state of Y M B in California and nationally today. I guess, how do you identify these things? Like, how do you see, I mean, there's plenty of people who have money who get into history or get into whatever subject, very few do something about it, right? Like, how do you figure out where?

• 30:42Speaker 00

  You know, I'm a little bit mystified by why people don't do more things too. Like, I think, first of all, I don't know, maybe you can tell me. Why are more people doing things? Like, I think most rich people are boring and they should do more cool things. So I'm hoping that they do that now. But yeah, I mean, I don't know. Like, I think part of it is I just fundamentally don't believe the world is efficient. And so if I see an opportunity to do something, I don't have a, I used to, but I no longer have a reflexive reaction that says, oh, that must not be a good idea. If it were a good idea, someone would already be doing it. Like, someone must be taking care of housing policy in California, right? Or somebody must be, you know, taking care of this or that.

• 31:25Speaker 00

  And so like, I think, you know, first I like, I don't have that filter that says, the world's efficient, don't bother, someone's probably got it covered. And then the second thing is, I'd kind of have learned to trust my enthusiasm. You know, it was, this gets me in trouble too. But if I get like really enthusiastic about something and that enthusiasm kind of persists, I just indulge it and just think, oh yeah, I'm gonna go like, you know. Like I like doing the things I'm enthusiastic about and so I just kind of let myself be impulsive. And so frequently what you do, there's this great, you know, image that I found and tweeted, which said, we do these things not because they are easy, but because we thought they would be easy. And so, yeah, like that's frequently what happens is like the commitment to do it is

• 32:10Speaker 00

  impulsive because and it's done out of enthusiasm. And then you get into it and you're like, oh my God, this is like really much harder than we expected. But then you're sort of committed and you're stuck and you're gonna have to get it done. Like I thought this project would be relatively straightforward. We're just gonna take the data and put it up. And but of course, everything is and truly 99% of the work has already been done by Dr. Seals and his team at the University of Kentucky. I am a kind of carpetbagger. I've shown up at the end here, you know, to like, try to do a new piece of it. But

• 32:42Dwarkesh Patel?

  The last mile is often the hardest.

• 32:43Speaker 00

  Well, I mean, it's turned out to be fractal anyway, like, you know, just like all the little bits that you have to get right to do a thing and have it work. And, you know, I hope we got all of them. But so I think that's part of it is just like, yeah, not believing the world's efficient, then just like allowing your enthusiasm to cause you to commit to something that turns out to be a lot of work and really hard. And then you just are like stubborn and don't want to fail. And so you keep at it. I don't know. I think that's it.

• 33:07Dwarkesh Patel?

  Yeah, I don't know. I feel like the efficiency point, do you think that's particularly true just of things like California and you're this where there isn't a direct monetary incentive? Or?

• 33:18Speaker 00

  No, I mean, look, certainly parts of the world are more efficient than others.

• 33:20Dwarkesh Patel?

  Right.

• 33:21Speaker 00

  And you can't assume equal levels of inefficiency everywhere. But I'm like constantly surprised by how even in areas you expect to be very efficient, there are things that are sort of in plain sight that no, and it's not that I see them and others don't, you know, there's lots of stuff I don't see too. I was talking to some traders at a hedge fund recently, and I asked them, I was trying to understand the role secrets play in the success of a hedge fund. And the reason I was interested in that is because I think the AI labs are going to enter a new similar dynamic where their secrets are very valuable. Like if you have a 50% training efficiency improvement and your training runs cost $100 million, that is a $50 million secret that you have that you want to keep.

• 34:06Speaker 00

  And hedge funds do that kind of thing routinely. And so I asked some traders at a very successful hedge fund, if you had maybe your smartest trader get on Twitch for 10 minutes, once a month, and on that Twitch stream, describe their 30-day-old trading strategies. Right? So not your current ones, but the ones that are a month old. What would that, how would that affect your business after 12 months of doing that? So 12 months, 10 minutes a month, 30-day look back, it's two hours in a year. And to my shock, they told me 80% reduction in their profits. Like, it would have a huge impact.

• 34:48Speaker 00

  And then I asked, okay, so how long would the look-back window have to be before it would have like a relatively small effect in your business? And they said 10 years. So like, that, I think, is just quite strong evidence that the world's not perfectly efficient, because, you know, these folks make billions of dollars using secrets that could be related in like an hour, or something like that. And yet others don't have them, or their secrets wouldn't work. And so I think there are different levels of efficiency in the world, but on the whole, our like default estimate of how efficient the world is is far too charitable.

• 35:26Dwarkesh Patel?

  Hmm. On the particular point, by the way, of AI labs potentially starting secrets, I mean, you have this sort of strange norm of different people from different AI labs not only being friends, but like often living together, right? So it would be like Oppenheimer living with somebody working on the Russian atomic bomb, or something like that. Do you think those norms will persist once the value of these secrets is realized?

• 35:48Speaker 00

  Yeah, I was just wondering about that some more today. I mean, it's, it seems to be sort of slowing, you know, they seem to be trying to close the valves.

• 35:57Speaker 00

  But I think there's a lot of things working against them in this regard. So one is, again, that the secrets are relatively simple. Two is that you're coming off this academic norm of publishing and really like the entire culture is based on sort of sharing and publishing. You know, three is, as you said, they all live in group houses, some are in polycules. You know, there's just a lot of intermixing. And then it's all in California and California is a non, you know, non-compete state. We don't have non-compete. And so you'd have to change the culture, get everybody their own house and move to Connecticut and then, you know, maybe it would work. You know, I, I think ML engineer salaries and compensation packages will probably be adjusted to try to, you know, address this because you don't want your secrets walking

• 36:42Speaker 00

  out the door. There are engineers, you know, Igor Babushkin, for example, who has just, I believe, joined Twitter, I think, is that right? Elon hired him to train, I think that's public, is that right? I think it is.

• 36:56Dwarkesh Patel?

  It will be now.

• 36:58Speaker 00

  I mean, Igor's really, really great guy and brilliant, but he also happens to have trained state of the art models at DeepMind and OpenAI. And so, you know, like that's the set of people who have that set of, you know, I don't know whether that's a consideration or how big of an effect that is, but it's the kind of thing that it would make sense to value if you think there are sort of valuable secrets that have not yet proliferated. So I think they're going to try to slow it down. Publishing has certainly slowed down dramatically already, but I think there's just a long way to go before you're anywhere in like hedge fund or Manhattan Project territory and probably secrets will still have a relatively short half-life.

• 37:38Dwarkesh Patel?

  As somebody who has been involved in open source your entire life, are you happy that this is the way that AI has turned out or do you think that this is less optimal?

• 37:46Speaker 00

  Well, I don't know. My, my opinion's been changing. I'm, I have increasing worries about kind of safety issues like not, not the hijacked version of safety, but some industrial accident type situations or misuse. And so I do think there's, there's some, we're not in that world and I'm not particularly concerned about it in the, in the short term, but in the long term, I do think there are worlds that we should be a little bit concerned about, although I don't know what to do about, where, yeah, like.

• 38:20Speaker 00

  Bad things happen. The probability mass, my belief is though, is that it's probably better in the whole for more people to get to tinker with and use these models, at least in their current state. And so, for example, when, you know, Georgi Gerganov this weekend did a four bit quantization of the Lama model and got it, you know, inferencing on a M1 or M2, I was very excited and I got that running and it's like fun to play with. Now I've got a model that, you know, is like very good. It's almost GPT-3 quality, runs in my laptop and, you know, I've sort of grown up in this world of the tinkers and open source folks and the more access you have, the more things you can try. And so I, I think I do find myself, you know, very attracted to that.

• 39:01Dwarkesh Patel?

  I guess that is the scientist and the ideas part of what is being shared. But there's also another part about the actual substance, right? So like the uranium and the sort of atom bomb analogy. As I guess different sources of data realize how valuable their data is for trading newer models, do you think that these things will become harder to scrape? Libgen archive, are these going to become rate limited in some way or what are you expecting there?

• 39:27Speaker 00

  Well, first, there's so much data on the internet. I mean, the two kind of primitives that you need to build models are you need lots of data. We, we have that in the form of the internet. We digitize the whole world into the internet and then you, and then you have, you need these GPUs, which we have because of video games. You take like the internet and video game hardware and you smash them together and you get machine learning models and they're both commodities. And so I think the data, I don't, I don't think anyone in the open source world is really going to be data limited for a long time. There's so much that's out there. Probably people who have like proprietary data sets that are readily scrapable have been shutting those down, you know? So get your scraping in now if you, if you need to do it. But that's just on the margin. I, I still think there's, there's quite a lot.

• 40:09Speaker 00

  That's out there to work with. So I think, look, there's gonna be a ton. This is the year of proliferation. This is a week of proliferation. Like, we're gonna see four or five major AI announcements this week. You know, new models, new APIs, new platforms, new tools from all the different vendors. In a way, this, you know, they're all looking forward. My Herculeanum project is looking backwards. I think it's extremely exciting and cool, but it is sort of a funny contrast.

• 40:32Dwarkesh Patel?

  Okay, so that, I guess before I delve deeper into AI, I do want to talk about GitHub. So I think we should start with, you are at Microsoft, and at some point you realized that GitHub is very valuable and worth acquiring. How did you realize that? And how did you convince Microsoft to purchase GitHub?

• 40:48Speaker 00

  Well, so I had started a company called Zammerin together with Miguel de Acasa and Joseph Hill. And we had built kind of mobile tools and platforms. And Microsoft acquired the company in 2016. And I was excited about that. I thought it was great. But to be honest, I didn't actually expect or plan to spend, you know, more than a kind of a year or so there. But when I got in there, I got exposed to what Satya was doing and just the quality of his leadership team. I was really impressed. And actually, I think I saw him in the first week or so I was there, and he asked me, what do you think we should do at Microsoft? And I said, I think we should buy GitHub.

• 41:27Dwarkesh Patel?

  When would this have been?

• 41:28Speaker 00

  This was like my first week. It was like March or April of 2016.

• 41:32Dwarkesh Patel?

  Okay.

• 41:33Speaker 00

  And then he said, yeah, it's a good idea. We thought about it. I'm not sure we can get away with it. Or something like that. And then it was about a year later, a little more than a year later. Yeah, I wrote, I wrote him an email, just a memo, you know, it sort of said like, I think it's time to do this. There was some noise that Google was sniffing around. I think that may have been manufactured by the GitHub team. But it was a good catalyst because it was something I thought made a lot of sense for Microsoft to do anyway. And so I wrote an email to Satya, sort of a little memo, saying, you know, hey, I think we should buy GitHub. Here's why. Here's what we should do with it. And the basic argument was developers are making IT purchasing decisions now. It used to be this sort of IT thing, you know, and now developers are leading that purchase.

• 42:18Speaker 00

  And it's, you know, this, this sort of major shift in how software products are, are acquired. And Microsoft really was an IT company. It was not a developer company in, in the way most of its purchases were made. But it was founded as a developer company, right? And so, you know, Microsoft's first product was a programming language. Yeah, I said, look, the challenge that we have is there's an entire new generation of developers who have no affinity with Microsoft. And the largest collection of them is at GitHub. And if we acquire this and we do a merely competent job of running it, we can earn the right to be considered by these developers for all the other products that we do.

• 43:00Speaker 00

  And to my surprise, Satya replied in like six or seven minutes and said, I think this is very good thinking. Let's meet next week or so and talk about it. And I ended up at this conference room with him and Amy Hood and Scott Guthrie and Kevin Scott and several other people. And they said, okay, make, you know, tell us what you're thinking. And I kind of did a little 20 minute ramble on it. And Satya said, I think we should do it. And why don't we run it independently like LinkedIn, Nat, you'll be the CEO. And he said, do you think we can get it for $2 billion? And I said, we'll try. And three weeks later, you know, he said, okay, go, go do this, Scott, you know, Scott will support you on this. Three weeks later, we had like a signed term sheet and an announced deal. And then it was an amazing experience for me.

• 43:45Speaker 00

  I'd been there less than two years and, you know, Microsoft was made up of and run by a lot of people who'd been there for many years. And they trusted me with this really big project and made me feel really good, you know, to be trusted and empowered. And I had grown up in the open source world. And so for me to get an opportunity to run GitHub, it's like, I don't know, getting appointed mayor of your hometown or something like that. I thought it felt cool. And I really wanted to do a good job for developers. And so that's, that's how it happened.

• 44:13Dwarkesh Patel?

  That's actually one of the things I want to ask you about because often when something succeeds, we kind of think it was inevitable that it would succeed. But at the time, I remember, I mean, it was like a while back, but I remember that there was a huge amount of skepticism. I would go on like Hacker News and like the top thing would be the blog post about how Microsoft's going to mess up GitHub. And I guess people are, have, those concerns have been alleviated throughout the years. But how did you get, you know, deal with that skepticism and deal with that distrust?

• 44:39Speaker 00

  Well, I was really paranoid about it.

• 44:40Dwarkesh Patel?

  Yeah.

• 44:40Speaker 00

  And I really cared about what developers thought. I think there's always this question of who are you performing for? Like, who do you actually really care about? Sort of who's in, who's the audience that's in your head that you're trying to, you know, do a good job for, impress, earn the respect of, whatever it is. And though I love Microsoft and care a lot about Satya and everyone there, I really cared about the developers. You know, I'd grown up in this open source world. And so for me to do a bad job with this central institution and open source would have been a devastating feeling for me. It was very important to me not to. So that was sort of the first thing is just that I cared. And then the second thing is that the deal leaked. It was going to be announced, I think, on a Monday, leaked on a Friday. And Microsoft's buying GitHub. And the whole weekend, there were like terrible posts online.

• 45:24Speaker 00

  You know, people saying, we got to evacuate GitHub as quickly as possible. And, and we're like, oh, my God, that's terrible. And then Monday, we put the announcement out. We said, we're acquiring GitHub. It's going to run as an independent company. And then it said Nat, you know, Nat Friedman's going to be CEO. And, you know, I had, I don't want to overstate or whatever, like, but I think a couple of people were like, oh, Nat comes from open source. You know, he spent some time in open source, so it's going to be run independently. So I don't think they were really that, that calmed down. But at least a few people thought like, oh, maybe I'll give this a few months and just see what happens before I migrate off. And then my first day as CEO, after we got the deal closed, like 9:00 AM the first day,

• 46:06Speaker 00

  you know, I was in this room and we got on Zoom and all the heads of engineering and product. And I think maybe, I don't know what people were expecting, but I think maybe they were expecting some kind of longer term strategy or something. But I came in and I said, there was this, GitHub had no official feedback mechanism that was publicly available. But there were several GitHub repos that community members had started. Isaac from NPM had started one, where he'd just been allowing people to give GitHub feedback. And people had been voting on this stuff for years. And I kind of shared my screen and put that up, sorted by votes, and said, like, we're going to pick one thing from this list and fix it by the end of the day and ship that, like, just one thing.

• 46:50Speaker 00

  And, you know, I think they were like, this is the new CEO's strategy. You know, and they were like, I don't know, we can't, you know, you do database migrations, can't do that in a day. And like, and then someone's like, well, maybe we can do this, you know, we have this sort of, we actually have a half implementation of this. And we eventually found something that we could fix by the end of the day. And what I'm thinking is, what I'm thinking, what I hope I said was, what we need to show the world is that GitHub cares about developers, not that it cares about Microsoft. Like, if the first thing we did after the acquisition was to add Skype integration, developers would have said, oh, we're not your priority. Like, you have new priorities now.

• 47:31Speaker 00

  And so the idea was just to find ways to make it better for the people who use it and have them see that we cared about that immediately. And so I said, we're going to do this today, and then we're going to do it every day for the next 100 days. And it was cool because I think it created some really good feedback loops, at least for me. One was, you know, you ship things and then people are like, oh, hey, I've been wanting to see this fixed for years and now it's fixed. It's a relatively simple thing. So you get this sort of nice dopaminergic, you know, feedback loop going there. And then people in the team feel the, you know, excitement of shipping stuff. I think GitHub was a company that had a little bit of stage fright about shipping previously and sort of break that static friction and ship a little bit more, I think felt good.

• 48:15Speaker 00

  And then the other one is just the learning loop. By trying to do lots of small things, I got exposed to like, okay, this team is really good, you know, or this part of the code has a lot of tech debt. Or, hey, we shipped that and it was actually kind of bad. How come that design got out? Like, where, you know, and so you, whereas if the project had been some six-month thing, I'm not sure my learning would have been quite as quick about the company. And there's still things I missed and mistakes I made, for sure. But that was part of how I think, you know, no one knows counterfactually whether that made a big difference or not, but I do think that earned some trust.

• 48:50Dwarkesh Patel?

  I mean, most acquisitions don't go well. Not only do they not go as well, but like, they don't go well at all, right? Like, as we're seeing in the last few months with a certain one. Well, why do most acquisitions fail or fail to go well?

• 49:02Speaker 00

  Yeah, it is true. Most acquisitions are destructive of value. What is the value of a company? In an innovative industry, the value of the company A lot of it boils down to its ability culturally to produce new innovations. And is some sensitive harmonic of cultural elements that sets that up that makes that possible? And it's quite fragile, I think. And so if you take a culture that has achieved some productive harmonic and you put it inside of another culture that's really different, the kind of mismatch of that can destroy the productivity of the company. So I think that maybe one way to think about it is companies are a little bit fragile.

• 49:43Speaker 00

  And so when you acquire them, it's like relatively, yeah, relatively easy to break them. I mean, they're also more durable than people think in many cases, too. Like I would say, another version of it is the people who really care leave. And so like the people who really care about building great products and serving the customers, maybe they don't want to work for the acquirer. And the set of people that are really load bearing around kind of the long term success is small. And when they leave or get disempowered, you get very different behaviors.

• 50:18Dwarkesh Patel?

  And then, so I want to go into the story of Copilot because until ChatGPT, I guess, it was like the most widely used application of the modern AI models. What are a part of the story you're willing to share in public?

• 50:30Speaker 00

  Yeah, I mean, I've talked about this a little bit. I mean, so look, GPT-3 came out in May, I think of 2020, and I saw it and it really blew my mind. I thought it was amazing. And I was CEO of GitHub at that time. And I thought like, I don't know what, but we've got to build some product with this. This is, you know, we've got to build something. So Satya had, at I think Kevin Scott's urging, already invested in OpenAI like a year before GPT-3 came out. Like, this is like quite amazing. And he invested like a billion dollars.

• 51:04Dwarkesh Patel?

  By the way, do you know why he knew that OpenAI would be worth investing in at that point?

• 51:08Speaker 00

  I don't know. Actually, I've never asked him. But yeah, I'm not sure. It's a good question. I mean, I think OpenAI had already had some successes that were noticeable. And I think if you're Satya and you're running this multi-trillion dollar company, you're trying to execute well and serve your customers, but you're always looking for the next gigantic wave that is going to upend the technology industry. It's not just about trying to win cloud. It's like, okay, what comes after cloud? And so. You want, you have to make some big bets, and I think he thought AI could be one. And I think Kevin Scott deserves a lot of credit for really advocating for that aggressively.

• 51:52Speaker 00

  And I think Sam Altman did a good job of building that partnership because he knew that he needed access to the resources of a company like Microsoft to build, you know, large scale AI and eventually AGI. And so I think it was some combination of those three people kind of coming together to make it happen. But I still think it was a very prescient bet. You know, people, I've said that to people and they've said, well, a billion dollars is not a lot for Microsoft. Yeah, but like, there were a lot of other companies that could have spent a billion dollars to do that and did not. And so I still think that that deserves a lot of credit. Okay, so GPT-3 comes out, you know, I pinged Sam and Greg, I think, Brockman at OpenAI. And they were like, yeah, like, let's, we've already been experimenting with GPT-3 and

• 52:37Speaker 00

  derivative models in coding contexts, like, let's definitely work on something. And to me, at least, and a few other people, it was not incredibly obvious what the product would be.

• 52:47Speaker 00

  Now, I think it's trivially obvious, you know, autocomplete, my gosh, isn't that what the models do? But at the time, actually, my first thought was that it was probably going to be like a Q and A chatbot stack overflow type of thing. And so that was actually the first thing we prototyped. So we grabbed a couple of engineers, this guy, Uga, who had come in from an acquisition that we'd done, and Alex Gravely, and started prototyping. And the first prototype was a chatbot. And, you know, what we discovered first was that demos were fabulous, like, every AI product has a fantastic demo. You get this sort of wow moment.

• 53:25Speaker 00

  So like, that is, turns out to be maybe not a sufficient condition for a product to be good. Because it was just, at the time, the models were just not reliable enough, they were not good enough. You know, I ask you a question, 25% of the time, you give me an incredible answer that I love, 75% of the time, your answer is useless or wrong. It's not a great product experience. And so then we started thinking about code synthesis, and our first attempts at this were actually large chunks of code synthesis, like synthesizing whole function bodies. And we built some Tools to do that and put them in the editor. And that also was not really that satisfying. And so the next thing that we tried was to just do simple, small scale autocomplete with the large models.

• 54:07Speaker 00

  And we used the kind of IntelliSense dropdown UI to do that. And that was better, like definitely pretty good. But the UI was not quite right. And we lost the ability to do this large scale synthesis. You know, we still had that, but the UI for that wasn't good. And we had it, I think, so that you, to get a function body synthesized, you would hit a key. And then, I don't know why this was the idea everyone had at the time, but several people had this idea that it should display multiple options for the function body. And then the user would read them and pick the right one. And I think the idea was that we would use that human feedback to improve the model. But that turned out to be a bad experience because first you had to hit a key and explicitly request it, then you had to wait for it.

• 54:49Speaker 00

  And then you had to read, you know, three different versions of a block of code. Reading one version of a block of code takes some cognitive effort. Doing it three times takes more cognitive effort. And then, most often, the result of that was like you, none of them were good, or you didn't know, which, you know, which one to pick. And so that was also like, you're putting a lot of energy and you're not getting a lot out, it's sort of frustrating. So, once we had that sort of single line completion working, I think Alex had the idea of saying, we can use the cursor position in the AST to figure out heuristically whether you're at the beginning of a block in the code or not. And if it's not the beginning of a block, just complete a line.

• 55:29Speaker 00

  If it's the beginning of a block, show inline a full, you know, block completions, the sort of number of tokens you request. And when you stop, gets altered automatically with no user interaction. And then the idea of using this sort of gray text, like Gmail had done in the editor. And so we got that implemented. And it was really only kind of once all those pieces came together and we started using a model that was small enough to be low latency, but big enough to be accurate, that we reached the point where like the median new user loved Copilot and wouldn't stop using it. And that took Four months, five months of just tinkering and sort of exploring, you know, there were other dead ends that we had along the way.

• 56:10Speaker 00

  And, and then, yeah, I think that then it was became quite obvious that it was good because we had hundreds of internal users who were GitHub engineers. And I remember the first time I looked at the retention numbers, they were extremely high. It was like, I remember 60 plus percent after 30 days from first install. Like if you installed it, the chance that you were still using it after 30 is like over 60%. And it's very intrusive product. I mean, it's sort of always popping UI up. And so if you don't like it, you will disable it. And for indeed 40 something percent of people did disable it. But those are very high retention numbers for like an alpha first version, you know, of a product that you're using all day. And so then I was, you know, just incredibly excited to launch it. And now it's, you know, it's improved dramatically since then.

• 56:55Dwarkesh Patel?

  Yeah, sounds very similar to the Gmail story, right? Of it's a incredibly valuable inside and then it becomes obvious that it needs to go outside. Okay, we'll go back to the AI stuff in a second, but some more GitHub questions. By what point will, if ever, will GitHub profiles replace resumes for programmers?

• 57:11Speaker 00

  That's a good question. I mean, I think they're a contributing element to how people like try to understand a person now, but I don't think they're like a definitive resume. You know, we introduced Readmes on profiles when I was there and I was excited about that because I thought it gave people like some degree of personalization. I think some people have, you know, I mean, many thousands of people have done that. Yeah, I don't know. There's forces that push in the other direction too on that one, where people like don't want their activity and skills to be as legible. And there may be some adverse selection as well, where the people with the most elite skills, you know, it's rather gauche for them to signal their competence on their profile. So there's some weird like social dynamics that feed into it too.

• 57:53Speaker 00

  But I will say, I think it effectively has this role for people who are breaking through today. Like one of the best ways to break through, I know many people are in this situation. You were born in Argentina, you're a very sharp person, but you didn't grow up in like a highly connected or prosperous network, family, et cetera.

• 58:15Speaker 00

  And yet you know you're really capable and you just wanna get connected to kind of the most elite communities in the world. And so if you're good at programming, you can join open source communities and contribute to them. And you can very quickly accrete a global reputation for your talent, which is legible to many companies and individuals around the world. And suddenly you find yourself getting a job and moving maybe to the US or maybe not moving, you end up at a great startup. I mean, I know a lot of people who've deliberately pursued the strategy of, you know, building reputation in open source and then kind of you've got the sail up and the wind catches you and, you know,

• 58:56Speaker 00

  you're, you've got a career. And so I think it plays that role in that sense. But in other communities, like in machine learning research, this is not how you, you know, there's, there's a thousand people, their reputation is more on archive, you know, than it is on GitHub. So I don't know that it'll ever be comprehensive.

• 59:14Dwarkesh Patel?

  Are there any other industries for which proof of work of this kind will eat more into the way in which people are hired?

• 59:22Speaker 00

  Well, I think there's a labor market dynamic in software where the really high quality talent is so in demand and the supply is so much less than the demand that it shifts power onto the developers such that they can require of their employers that they be allowed to work in public. And because, and then when they do that, they develop an external reputation, which is this asset they can port between companies. And if the labor market dynamics weren't like that, if programming well were less economically valuable, then they would not, you know, the, like, they would not have, companies wouldn't let them do that. They wouldn't let them like publish a bunch of stuff publicly.

• 1:00:06Speaker 00

  They'd say like, we're not, that's a rule. And that used to be the case, in fact. And so as software's become more valuable, developers, the leverage of like a single super talented developer has gone up and they've been able to demand over the last several decades, the ability to work in public and. I think that's not going away.

• 1:00:26Dwarkesh Patel?

  Other than that, what has been, I mean, we talked about this a little bit, but what has been the impact of developers being more empowered in organizations, even ones that are not traditionally IT organizations?

• 1:00:35Speaker 00

  Yeah, I mean, software is, is, is kind of magic, right? I mean, the, you can write a for loop and do something a lot of times. And like, you know, like when you build large organizations at scale, one of the things that does surprise you is the degree to which you need to systematize the behavior of the people who are working. Like when I first was starting companies and building sales teams, I had this wrong idea coming from the world as a programmer that salespeople were like hyper aggressive, hyper entrepreneurial, you know, making promises to the customer that the product wouldn't do. And that the main challenge you had with salespeople was like restraining them from going out and like, you know, aggressively cutting deals that shouldn't be cut.

• 1:01:18Speaker 00

  And what I discovered is that that does exist sometimes, but like the much more common case is that you need to build a systematic sales playbook, which is almost a script that you run on your sales team, where your sales reps know the process they need to follow to like exercise this repeatable sales motion and get a deal closed. And so, you know, I just had bad ideas there. I didn't know that that was how the world worked. But software is a way to like systematize and scale out a valuable process extremely efficiently. And I think the more digitized the world has become, the more valuable

• 1:02:00Speaker 00

  software becomes and the more valuable become the developers who can create it essentially.

• 1:02:05Dwarkesh Patel?

  Would 25 year old not be surprised with how well open source worked and how pervasive it is?

• 1:02:10Speaker 00

  Yeah, I think that's true. Yeah, I, I, I think we all have this image when we're young that these institutions are these implacable edifices that are evil and all powerful and, you know, are able to like substant, with master plans substantially orchestrate the world. And that is some, sometimes a little bit true, but like they're very vulnerable to these. Yeah, like new ideas and new forces and new communications media and stuff like that. So right now, I think like, maybe I wouldn't, right now, I think our institutions overall look relatively weak. And certainly, they're weaker than I thought they were back then. So I thought Microsoft, honestly, I thought Microsoft could stop open source.

• 1:02:52Speaker 00

  I thought that was a possibility. You know, they can do some patent move and kind of, there's a master plan to ring fence open source in. And, um, yeah, that didn't, that didn't end up being the case. In fact, Microsoft, when we bought GitHub, we, um, we pledged all of our patent portfolio to open source. That was one of the things that we did as part of it. And so that was a kind of poetic moment for me, having been on the other side of patent discussions in the past to be a part of and instrumental in, in Microsoft making that pledge. Like, that was, that was quite crazy.

• 1:03:24Dwarkesh Patel?

  Oh, that's really interesting. It wasn't that there was like some business strategic reason, more so it was just like an idea whose time had come.

• 1:03:30Speaker 00

  Well, GitHub had made such a pledge. And so I think in part, in acquiring GitHub, we had to either try to annul that pledge, or sign up to it ourselves. And so there was sort of a moment of a forced choice. But, you know, everyone at Microsoft thought, thought it was a good idea, too. So I think in many senses, it was a moment whose time had come and the kind of GitHub acquisition was a forcing function.

• 1:03:53Dwarkesh Patel?

  Well, what do you make of critics of modern open source, like Richard Stallman, or people who advocate for free software, saying that, well, corporations might advocate for open source because of like practical reasons for getting good code. The real value of software and the real way the software should be made, it should be free in that you can replicate it, you can, you can change it, you can modify it, and you can completely view it. And that the ethical values about that should be more important than the practical values. Like, what do you make of that critique of?

• 1:04:23Speaker 00

  I think those are the things that he wants.

• 1:04:24Dwarkesh Patel?

  Right.

• 1:04:25Speaker 00

  And I think the thing that maybe he hasn't updated is that maybe not everyone else wants that. You know, he has this idea that people want freedom from the tyranny of a proprietary intellectual property license. But what people really want is freedom from having to configure their graphics card or sound driver or something like that. You know, they want their computer to kind of work.

• 1:04:46Dwarkesh Patel?

  Right.

• 1:04:47Speaker 00

  There are places where freedom is really valuable, but there's, there's always this thing of like, I have a prescriptive ideology that I'd like to impose on the world versus this thing of like, I will try to develop the best observational model for what people actually want, whether I want them to want it or not. And I think, you know, Richard is strongly in the former camp.

• 1:05:08Dwarkesh Patel?

  Well, what is the most underrated license, by the way?

• 1:05:10Speaker 00

  I mean, I don't know. Maybe the MIT license is still underrated because it's just so simple and bare.

• 1:05:17Dwarkesh Patel?

  Nadia Eickboll had a book recently where she argued that the key constraint on open source software and on the time of the people who maintain it is the community aspect of software. They had to deal with feature requests and discussions and maintaining for different platforms and things like that. And it wasn't the actual code itself, but rather this sort of extracurricular aspect that was the main constraint. Do you think that is the constraint for open source software? Like, how do you see what is holding back more open source software?

• 1:05:43Speaker 00

  Yeah, I mean, I think by and large, I would say that there is not a problem, meaning open source software continues to be developed, continues to be broadly used. And, you know, there's areas where it works better and areas where it works less well, but it's sort of winning in all the areas where like, large scale coordination and editorial control are not necessary. And so it tends to be great at infrastructure, standalone components, and very, very horizontal things like operating systems. And it tends to be worse at user experiences and like things that where you need a sort of dictatorial aesthetic or, you know, an editorial control. I've had debates with Dylan Field of Figma as to why it is

• 1:06:25Speaker 00

  that we don't have lots of good open source applications. And, you know, I've always thought it had something to do with this governance dynamic of, you know, gosh, it's such a pain to coordinate with tons of people who all sort of feel like they have a right to try to push the project in one way or another. Whereas in a hierarchical corporation, there can be a, you know, head of this product, or a CEO, or founder, or designer who just says, we're doing it this way. And you can really align things in one direction very, very easily. Dylan has argued to me that it might be because there's just fewer designers, you know, people with good design sense in open source. I think that might be a contributing factor, too. But I think it's still mostly the governance thing. And I think that's what Nadia is pointing at also. You're running a project, people, you gave it to people for free. For some reason, giving people something for free creates the sense of entitlement.

• 1:07:10Speaker 00

  And then they feel like they have the right to demand your time and push things around and give you input and want to be polite. And it's very draining. So, so I think that, you know, where that coordination burden is lower is where open source tends to succeed more. And probably software and other new forms of governance can improve that and like expand the territory that open source can succeed in.

• 1:07:32Dwarkesh Patel?

  Yeah, I mean theoretically, those two things are consistent, right? Like you could have very tight control over governance while the code itself is open source.

• 1:07:40Speaker 00

  Yeah, and this happens in programming languages.

• 1:07:41Dwarkesh Patel?

  Right.

• 1:07:42Speaker 00

  Languages can't be designed, I mean, they often are eventually sort of set in stone and then like advanced by committee.

• 1:07:47Dwarkesh Patel?

  Yeah.

• 1:07:47Speaker 00

  But yeah, I mean, certainly you have these sort of benign dictators of languages who enforce a strong set of ideas. They have a sort of vision, master plan. So that would be the argument that's most kind of on Dylan's side. It's like, hey, it works for languages. Why can't it work for end user applications? I think the thing you need to do, though, to build a good end user application is not only have a good aesthetic and idea, but somehow establish a tight feedback loop with a set of users where you can give it, Torque, Torque, Cash, try this. Oh, my gosh, okay, that's not what you need. Okay. And so like, doing that is so hard, even in a company where you have total hierarchical control of the team in theory, and everyone really wants the same thing, and everyone's salary and stock options depend on the product being accepted by these users. It still fails, you know, many times in that scenario.

• 1:08:32Speaker 00

  Then additionally, doing that in the context of open source, I think is just like slightly too hard.

• 1:08:36Dwarkesh Patel?

  The reason you acquired GitHub, as you said, is that there seems to be a sort of complementarity between Microsoft and GitHub's missions. And I guess that's been proven out over the last few years. Should there be more of these collaborations and acquisitions? Should there be more tech conglomerates? Like, would that be good for the system?

• 1:08:54Speaker 00

  I don't know if it's good, but I think there are, yes, it is certainly efficient in many ways. And I think we are seeing agglomeration occur because the math is sort of pretty simple. Like, if you are a large company and you have a lot of customers, then the thing that you've achieved is this very expensive and difficult thing of building distribution and relationships with lots of customers. And that is as hard or harder and takes longer and more money than just inventing the product in the first place. And so if you can then go and just buy the product for a small amount of money and like make it available to all of your customers, then there's often like an immediate Really obvious, you know, gain from, from doing that. And so in that sense, like acquisitions, acquisitions make a ton of sense.

• 1:09:39Speaker 00

  And I've been surprised that the large companies haven't done many more acquisitions in the past. Until I got into a big company and started trying to do acquisitions and I saw that there are like strong elements of the internal dynamics that make it hard. You know, there's not even, you know, it's like easier to spend a hundred million dollars on employees internally to do a project than to spend a hundred million dollars to buy a company. The, the dollars are treated differently. The approval processes are different. The kind of cultural buy-in processes are different. And then to the point of the discussion we had earlier, many acquisitions do fail. And when an acquisition fails, it's somehow louder and more embarrassing than when like some new product effort you've spun up doesn't quite work out as well.

• 1:10:22Speaker 00

  And so I think there's lots of internal reasons, some somewhat justified and some less so, that they haven't been doing it. But just from an economic point of view, it's, it seemed like it may, makes sense to see more acquisitions.

• 1:10:33Dwarkesh Patel?

  Uh, why, why did you leave?

• 1:10:35Speaker 00

  I think as much as I loved Microsoft and certainly as much as I love GitHub, I really truly like, I still feel tremendous love for GitHub and everything that it means to the people who use it. And I didn't really want to be a part of like a giant company anymore. And, you know, I think building Copilot was an example of this, you know, it wouldn't have been possible without OpenAI and Microsoft and GitHub. But building it also required navigating this like really large group of people and between Microsoft and OpenAI and GitHub. And, and you, you reach a point where you're spending a ton of time, you know, on just, just navigating and coordinating lots of people. And I just find that less energizing, you know, back to enthusiasm, like just my enthusiasm for that was, was not as high.

• 1:11:19Speaker 00

  And it was, I was torn about it because I truly love GitHub, the product. And there was so much more I still, you know, knew we could do, but I was proud of what we'd done and. I miss the team, you know, and I miss, I miss working on GitHub. It was really an honor for me. But yeah, it was time for me to go do something. You know, I was always a startup guy. I always liked small teams and I wanted to go back to sort of a smaller, more nimble environment.

• 1:11:46Dwarkesh Patel?

  Okay, so we'll get to it in a second, but first I want to ask about nat.org and the 300 words.

• 1:11:52Speaker 00

  Oh, yeah.

• 1:11:53Dwarkesh Patel?

  The list there, which is, I think, like, one of the most interesting, sort of like, and I guess, very Strassian list I've seen, list of 300 words I've seen anywhere. But I'm just gonna like, mention some of these and get some of your commentary. You should probably work on raising the ceiling, not the floor.

• 1:12:10Speaker 00

  Yeah.

• 1:12:11Dwarkesh Patel?

  Why?

• 1:12:13Speaker 00

  Yeah, I mean, well, first, I say probably, but what does it mean to raise the ceiling or the floor? I mean, I just observed a lot of projects that set out to raise the floor. Meaning, gosh, we are fine, but they are not, and we need to go help them with our superior prosperity and understanding of their situation. And many of those projects fail. So, for example, there were a lot of attempts to bring internet to Africa by large and wealthy tech companies and American universities. And I won't say they all had no effect. That's not true, but many of them were far short of successful. Like, there

• 1:12:57Speaker 00

  were satellites, there were balloons, there were, you know, high-altitude drones, there were mesh network laptops that were pursued by all these companies. And by the way, by perfectly well-meaning, incredibly talented people who, I think, did in some cases see some success, but overall, probably much less than they ever hoped. But if you go to Africa, there is internet now. And the way internet got there is the technologies that we developed to raise the ceiling in the richest part of the world. Which were cell phones and cell towers. I mean, in the movie Wall Street from the 80s, you know, he's got that gigantic brick cellphone. That thing cost like 10 grand at the time. That was a ceiling raising technology. It eventually went down the learning curve and

• 1:13:41Speaker 00

  became cheap. And the cell towers and cell phones, eventually, you know, we've got now hundreds of millions or billions of them in Africa. And it was sort of, it was that initially ceiling raising technology and then the sort of force of capitalism that made it work in the end. And it was not any Deus ex machina technology solution that it was intended to kind of raise the floor. And so I think there's something about that that's not just an incidental example. But I say, on my website, I say probably because there are, there are some examples where I think people set out to kind of raise the floor and say, no one should ever die of smallpox again, right? No one should ever

• 1:14:25Speaker 00

  die of Guinea worm again. And they succeed. And I wouldn't want to discourage that from happening. But I think on balance, we have too many attempts to do that that look good, feel good, sound good, and don't matter. And in some cases, have the opposite of the effect they intend to.

• 1:14:39Dwarkesh Patel?

  Here's another one. And this is under the EMH section. In many cases, it's more accurate to model the world as 500 people than 8 billion. Now, here's my question. What are the 800, 8 billion minus 500 people doing? Like, why, why are there only 500 people?

• 1:14:54Speaker 00

  Yeah, I mean, I don't know exactly. It's, it's a good question. I ask people that a lot. I mean, when, the more I've sort of done in life, the more I've been mystified by this, like, oh, somebody must be doing X. And then you kind of, you hear there's a few people doing X, and then you look into it, they're not actually doing X, they're doing kind of some version of it that's not that. And so, all the kind of best, you know, best moments in life occur when you find something that to you is totally obvious, that clearly somebody must be doing, but no one is doing. I mean, Mark Zuckerberg says this about founding Facebook, like, surely the big companies will eventually do this and create this social and identity layer on the internet. You know, Microsoft will do this. But no, none of them were, and he did it. So what are they doing? Okay, so I think the first thing is, many people throughout the world are optimizing local

• 1:15:39Speaker 00

  conditions. So they're working in their town, their community, they're doing something there. And so the set of people that are kind of thinking about kind of global conditions is just naturally narrowed by the structure of the economy.

• 1:15:53Speaker 00

  That's number one. I think number two is, most people really are quite memetic. And I think we all are, including me. You know, we get a lot of ideas from other people. And so, you know, our ideas are not our own. We kind of got them from somebody else. It's kind of copy paste. And so you have to work really hard not to do that and to be decorrelated. And I think this is even more true today because of the internet. You know, I don't know if Albert Einstein, as a patent clerk, would, you know, wouldn't he have just been on Twitter? Just getting the same ideas as everybody else? Like, would he have his decorrelated ideas? So I think the internet's correlated us more. The exception would be really disagreeable people, who are just naturally disagreeable.

• 1:16:37Speaker 00

  And so I think the future belongs to the autists in some sense, because, you know, they don't care what other people think as much. Those of us on the spectrum in any sense, I think, are in that category. Then, yeah, I think, you know, I think, and then we have this belief that the world's efficient, and it isn't. And I think that's part of it. So the other thing is that the world is so fractal and so interesting. I mean, Herculaneum papyri, right? Like, is this, is this corner of the world that I find totally fascinating? But I don't have any anticipation that eight billion people should be thinking about that, you know, or that that should be a priority for everyone.

• 1:17:15Dwarkesh Patel?

  Okay, here's another one. Large-scale engineering projects are more soluble in IQ than they appear. And here's my question. Does that make you think that the impact of AI tools, like Copilot, will be bigger or smaller because of engineering? Because one way to look at Copilot is like, actually, its IQ is probably less than the average engineer, so maybe it'll have like, less impact, right?

• 1:17:33Speaker 00

  Yeah, but I think it increases the productivity of, like, it definitely increases the productivity of the average engineer to bring them, you know, higher up. And I think it increases the productivity of the best engineers as well. Certainly, a lot of the people I consider to be the best engineers tell me that they find it increases their productivity a lot. So, yeah, I think AI is going to completely change. Like, it's really interesting how

• 1:17:59Speaker 00

  So much of what's happened in AI has been sort of soft fictional work. You know, you have mid journey, you have copywriting, you have, you know, gosh Claude from Anthropic is so literary. It writes poetry so well, except for Copilot, which is this real hard area where like the code has to compile. It has to be syntactically correct. It has to work and pass the tests. And, you know, we see the steady improvement curve where now already on average, more than half of the code is written by Copilot. I think when it shipped, it was like low twenties. And so it's really improved a lot as the models have gotten better and the prompting has gotten better. But I don't see any reason why that won't be like 95%. Like it seems very likely to me. And so I think I don't know what that world looks like.

• 1:18:43Speaker 00

  It seems like we might have more special purpose and less general purpose software. Like right now, we use general purpose tools like spreadsheets and, and things like this a lot. But part of that has to do with the cost of creating software. And so once you have, you know, much cheaper software, do you create more special purpose software? That's a possibility.

• 1:19:00Speaker 00

  Every company, just a custom piece of code in a sense, like maybe that's the kind of future we're headed towards. So yeah, I think we're going to see like enormous amounts of change in software development.

• 1:19:11Dwarkesh Patel?

  Another one, the cultural prohibition on micromanagement is harmful. Great individuals should be fully empowered to exercise their judgment. And the rebuttal to this is like, you know, if you micromanage or prevent people from learning and to develop their own judgment.

• 1:19:26Speaker 00

  Yeah, so imagine you go into some company, they hire Dwarkesh and you do a great job with the first project they give you. And so you're, everyone's really impressed, man, Dwarkesh, he made the right decisions. He worked really hard. He figured out exactly what needed to be done and he did it extremely well. And so over time, you get promoted into positions of greater authority. And the reason the company is doing this is they want you to do that again, but at bigger scale, right? Do it again, but 10 times bigger, the whole product instead of part of the product or 10 products instead of one.

• 1:19:57Speaker 00

  So the company is telling you, you have great judgment and we want you to exercise that at a greater scale. Meanwhile, the culture is telling you, as you get promoted, you should suspend your judgment more and more and defer your judgment to your team. And so there's some equilibrium there, and I think we're just out of equilibrium right now, where the cultural prohibition is too strong. And, you know, I think maybe in the, I don't know if this is during, but maybe in the 80s, I would have felt the other side of this, that like we have too much micromanagement. I think the other problem that people have is that they don't like micromanagement because they don't want bad managers to micromanage, right?

• 1:20:40Speaker 00

  So you have some bad managers, they have no expertise in the area, they're just kind of people managers, and they're starting to micromanage something that you don't understand where their judgment is bad. And my answer to that is like, stop empowering bad managers, like, don't have them, just don't have bad managers. Promote and empower people who have great judgment and do understand the subject matter that they're working on. You know, if I work for you and I just know you have better judgment and you come in and you say that, like, you're launching the scroll thing and I think you've got the final format wrong, you know, here's how you should do it. I would welcome that, even though it's micromanagement, because it's going to make us more successful and I'm going to learn something from that and I know your judgment is better than mine in this case, or at least we're going to have a conversation about it, we're both going to get smarter. So I think on balance, yeah, there are cases where people have excellent judgment and

• 1:21:25Speaker 00

  we should encourage them to exercise it. And sometimes, you know, things will go wrong when you do that, but on balance, you will get far more excellence out of it and we should, yeah, we should empower individuals who have great judgment.

• 1:21:39Dwarkesh Patel?

  Yep, yep. There's a quote about Napoleon that if he could have been in every single theater of every single battle he was part of, that he would have never lost a battle. I was talking to somebody who worked with you at GitHub and she emphasized to me, and this is like really remarkable to me, that even the applications that are being shipped out to engineers, how much of the actual suggestions and the actual design came from you directly, which is kind of remarkable to me that as CEO you would...

• 1:22:04Speaker 00

  Yeah, you can probably find people you can talk to who think that was terrible.

• 1:22:09Speaker 00

  The question is always, does that scale? Right? And and the answer is it does not scale. It doesn't like. But the set of people who really do have great judgment. Like the experience that I had as CEO is I was terrified all the time that there was someone in the company who really knew exactly what to do and had excellent judgment. But because of cultural forces, that person wasn't empowered. Right? That person was not allowed to exercise their judgment and make decisions. And so when I would think and talk about this, that was the fear that it was coming from is is. You know, they were in some consensus environment where their good ideas were getting whittled down by lots of conversations with other people and a politeness and a desire not to micromanage. And so we were ending up with some kind of average thing.

• 1:22:54Speaker 00

  And I would rather kind of have more high variance outcomes where you either get something that's excellent because it is the, you know, expressed vision of a really good auteur. Or you get just a disaster and it didn't work. And so now you know it didn't work and you can start over. Like, I would rather have those more high variance outcomes. And I think it's worth, it's a worthy trade.

• 1:23:14Dwarkesh Patel?

  All right, let's talk about AI.

• 1:23:15Speaker 00

  Yeah.

• 1:23:15Dwarkesh Patel?

  What percentage of the economy is basically text to text?

• 1:23:19Speaker 00

  Yeah, I mean, it's a good question. We've done the sort of Bureau of Labor Statistics analysis of this. And yeah, it's not, you know, the majority of the economy or anything like that. We're in the low double digit percentages. The thing that I think is hard to predict is what happens over time is the kind of cost of text to text goes down. And yeah, I don't know. I don't know what that's going to do. But yeah, there's like plenty, there's plenty of revenue to be got now. I mean, one way you can think about it is, okay, we have all these benchmarks for machine learning models. And, you know, there's Lombada and there's this and there's that. And those are really only useful and only exist because we haven't deployed the models really at scale.

• 1:24:04Speaker 00

  And so we don't have a sense of what they're actually good at. The best metric would probably be something like, what percentage of economic tasks can they do? Or like on a gig marketplace like Upwork, for example, like what fraction of Upwork jobs can GPT four do? I think it's sort of interesting question. My guess is like extremely low right now autonomously. But over time it will grow. And then the question is, what does that do? For Upwork, I mean, it's, I don't know, it's probably a five, guessing a $5 billion GMV market marketplace, something like that. Does it grow? Does it become 15 billion or 50 billion? Does it shrink because the cost of text-to-text tasks goes down?

• 1:24:48Speaker 00

  I don't know, my bet would be that we find more and more ways to use text-to-text, you know, to sort of advance, advance progress. And so overall, there's a lot more demand for it. But yeah, I guess we'll see.

• 1:25:01Dwarkesh Patel?

  And at what point does it happen? So I mean, like, GPT-3 has been a sort of rounding error in terms of like overall economic impact. Does it happen with GPT-4 or GPT-5 where we see billions of dollars of usage?

• 1:25:10Speaker 00

  You know, I've got early access to GPT-4 and I've gotten to use it a lot. And I honestly can't tell you the answer to that because it's so hard to discover what these things can do that the prior ones couldn't do. I just was talking to someone last night who told me, oh, GPT-4 is actually really good at Korean and Japanese and like GPT-3 is much worse at those. And so it's actually a real step change, you know, for those languages. And yeah, I think people didn't know how good GPT-3 was until it got instruction tuned for ChatGPT and was put out in that format. And so I think there's kind of, you can imagine the pre-trained models as kind of unrefined crude oil. And then once they've been kind of RLHF'd and trained and then put out into the world, people can, they can find the value.

• 1:25:55Dwarkesh Patel?

  What part of the AI narrative is wrong in the over-optimistic direction?

• 1:25:59Speaker 00

  The probably over-optimistic case from both the people who are fearful of what will happen and from people who are expecting great economic benefits is that we're definitely in this realm of diminishing returns from scale. So for example, I think GPT-4 is, my guess is two orders of magnitude more expensive to train than GPT-3, but clearly not two orders of magnitude more capable.

• 1:26:28Speaker 00

  Now, is it too much or is magnitude more economically valuable? That would also surprise me. And so I think it's possible when you're in these sigmoids where you kind of are going, you know, up this exponential and then you start to asymptote it, it can be difficult to tell if that's going to happen. So I think, yeah, the idea that we might not run into hard problems or that scaling will continue to like be worth it on a dollars basis, I think are reasons to be a little bit more pessimistic than the people who have high certainty of, I don't know, GDP increasing by 50% per month or something like that, which I think some people are predicting. But on the whole, I'm very optimistic. So you're asking me to like make the bear case for something I'm very bullish about.

• 1:27:08Dwarkesh Patel?

  All right. No, that's why I asked you to make the bear case, because I know on the whole, you're, I want to ask you about these foundation models. What is the stable equilibrium, you think, of how many of them will there be? Like, will it be a sort of oligopoly like Uber and Lyft where?

• 1:27:21Speaker 00

  I think there will probably be wide scale proliferation. And you sort of asked me, what are the structural forces that are pro proliferation and the structural forces that are pro concentration? So I think the pro proliferation case is a bit stronger. So the pro proliferation case is they're actually not that hard to train. You can kind of, the best practices will promulgate and you can kind of write them down on a couple sheets of paper. And to the extent that secrets are developed that improve training, those are relatively simple and they get copied around, you know, easily. Number one, number two, the data is mostly public. It's mostly kind of data from the internet. Number three, the hardware is mostly commodity and the hardware is improving quickly and getting much more efficient.

• 1:28:05Speaker 00

  And then I think there's a lot of techniques that kind of overcome, you know, like I think some of these labs potentially have 50, 100, 200% training efficiency improvement techniques. And so there's just a lot of low hanging fruit on the technique side of things. And so we're seeing it happen. I mean, it's happening this week and it's happening this year is that we're getting like a lot of proliferation. The only case against proliferation is that you'll get concentration because of training costs. And I don't know, I don't know that that's true. You know, I just like, I don't, I don't have confidence that the trillion dollar model will be much more valuable than the hundred billion dollar model. And That even it will be necessary to spend a trillion dollars training it.

• 1:28:50Speaker 00

  Like, maybe there will be so many techniques available for improving efficiency that, like, how much are you willing to spend on researchers to find techniques if you're willing to spend a trillion on training, right? Like, that's a lot of bounties for new techniques and, like, some smart people are going to take those bounties.

• 1:29:06Dwarkesh Patel?

  How different will these models be? Will it just be sort of everybody chasing the same exact marginal improvement, leading to the same marginal capabilities? Or will they have entirely different repertoires of skills and abilities?

• 1:29:17Speaker 00

  Right now, back to the memetic point, they're all pretty similar.

• 1:29:20Dwarkesh Patel?

  Right.

• 1:29:20Speaker 00

  I would say. I mean, basically the same rough techniques.

• 1:29:23Dwarkesh Patel?

  Right.

• 1:29:24Speaker 00

  What's happened is an alien substance has sort of landed on Earth and we are trying to figure out what we can build with it. And, you know, I think we're in this multiple overhangs here. We have sort of a compute overhang where there's much more compute in the world than is currently being used to train models, like, much, much more. You know, I think the biggest models are trained on maybe ten-ish thousand GPUs, but there's millions of GPUs, you know? And so, okay, there's the compute overhang. And then we have, I think, a capability and technique overhang where there's lots of good ideas that are coming out and we haven't figured out how best to assemble them all together, but that's just a matter of time, kind of, until people do that. And then those capabilities haven't reached, because many of them are in the hands of

• 1:30:08Speaker 00

  the labs, they haven't reached the tinkers of the world. And I think that is where the new, like, what can this thing actually do? Like, what, you know, like, until you get your hands on it, you don't really know. I think OpenAI were themselves surprised by how explosively ChatGPT has grown. I don't think they put ChatGPT out expecting that to be the big announcement. I thought, I think they thought GPT-4 was going to be their big announcement. And I think it still probably is and will be big, but ChatGPT really surprised them. And I think that's That's, you know, it's hard to predict what people will do with it and what they'll find valuable and what works. And so you need tinkers. So it basically goes from like hardware to researchers to tinkerers to products. That's the pipe. That's the cascade.

• 1:30:52Dwarkesh Patel?

  Yeah. Yeah. When I was scheduling my interview with Ilya, it was originally supposed to be around the time that ChatGPT came out. And so their correspondent tells me, listen, just so you know, this interview will be scheduled around the time. We're going to make like a minor announcement. It isn't, it's not the thing you're thinking. It's not GPT-4, but it's just like a minor thing. So they didn't expect what it ended up being. Have incumbents gotten smarter than before? So it seems like Microsoft was able to integrate this new technology well.

• 1:31:21Speaker 00

  There's two, there's been two really big shifts in the way incumbents behave in the last 20 years that I've seen. The first is that it used to be incumbents got disrupted by startups all the time. You had example after example of this in like the mini computer, microcomputer era, et cetera. And then Clay Christensen wrote The Innovator's Dilemma. And I think what happened was that everyone read it and they said, oh, like disruption is this thing that occurs and we have this innovator's dilemma where we get disrupted because the new thing is cheaper and we can't let that happen. And they became determined not to let that happen and they mostly learned how to avoid it. They learned that you have to be willing to do some cannibalization and you have to be willing to set up separate

• 1:32:02Speaker 00

  sales channels for the new thing and so forth. And so we've had a lot of stability in incumbents for the last, you know, 15 years or so. And I think that's maybe why. That's my theory. So that's the first major step change. And then the second one is, man, they are paying a ton of attention to AI. You know, if you look at the prior platform revolutions like cloud, mobile, internet, web, PC, all the incumbents derided the new platform and said, you know, gosh, like no one's going to use web apps. Like everyone will use full desktop apps, rich applications. And so there was always this sort of laughing at the new thing. The iPhone was laughed at by incumbents and that is not happening at all with AI. Now we may be at peak hype cycle and we're going to enter the trough of despair.

• 1:32:46Speaker 00

  I kind of don't think so, though. I think people are taking it seriously and, you know, every live player CEO is adopting it aggressively in their company. So yeah, I think incumbents have gotten smarter.

• 1:32:56Dwarkesh Patel?

  All right. So let me ask you some questions that we got from Twitter.

• 1:32:58Speaker 00

  Okay.

• 1:32:59Dwarkesh Patel?

  This is a former guest and I guess mutual friend Austin Vernon.

• 1:33:03Speaker 00

  Oh, yeah.

• 1:33:05Dwarkesh Patel?

  Nat is one of those people that seems unreasonably effective. What parts of that are innate and what did he have to learn?

• 1:33:12Speaker 00

  Well, it's very nice of Austin to say. I don't know. I mean, I think, you know, we talked a little bit about this before, but I think I just have a high willingness to try things and get caught up in new projects and then I don't want to stop doing it. And so I think I just have a relatively low activation energy to try something and I'm willing to sort of impulsively jump into stuff. And many of those things don't work, but enough of them do that I guess, you know, I've been able to accomplish a few things. The other thing I would say, to be honest with you, is that I do not consider myself accomplished or successful. Like, my self-image is that I haven't really done anything of tremendous consequence. And I don't feel like I have this giant, you know, sort of

• 1:33:56Speaker 00

  bed of achievements that I can go to sleep on every night. I think, and I try, and I, you know, I think I've, I think that's truly how I feel. I'm kind of an insecure overachiever. I don't really feel good about myself unless I'm doing good work. But I also have kind of cultivated a, tried to cultivate a forward-looking view, you know, where I try not to be incredibly nostalgic about the past. I don't keep like lots of trophies or anything like that. Go into some people's offices and there's like things on the wall and trophies of like all the things they've accomplished. And that always seemed really icky to me. So I don't know, it just had a sort of revulsion to that.

• 1:34:33Dwarkesh Patel?

  Is that why you took down your blog?

• 1:34:35Speaker 00

  Yeah. Yeah, I just wanted to move forward.

• 1:34:38Dwarkesh Patel?

  Simeon asks for your takes on alignment. Quote, "He seems to invest both in capabilities and alignment, which is the best move under a very small set of beliefs." So he's curious to hear your reasoning there.

• 1:34:50Speaker 00

  Yeah. Well, I'm not, you know, I guess we'll see. I'm not sure capabilities and alignment end up being these opposing forces. It may be that capabilities are very important for alignment. It may be that alignment is, you know, very important for capabilities. I mean, I think I believe, I think a lot of people believe, and I think I'm included in this, that AI can have tremendous benefits, but there's like a small chance of really bad outcomes. Maybe some people think it's a large chance. And so I think the solutions, if they exist, are likely to be technical. There are probably some combination of like technical and prescriptive. So it's probably a piece of code in a readme file. It says, if you want to build like aligned AIs,

• 1:35:33Speaker 00

  use this code and don't do this, you know, or something like that. And, um, and so, yeah, I think that's, I think that's really important and more people should try to actually build technical solutions. I think one of the big things that's missing is sort of perplexes me is there's no open source technical alignment community. There's no one actually just implementing in open source, the best alignment tools. There's a lot of philosophizing and talking, and then there's a lot of behind closed doors interpretability and alignment work. And I think we're going to end up because the alignment people have this belief that they shouldn't release their work in a world where there's a lot of open source, pure capabilities work and no open source

• 1:36:16Speaker 00

  alignment work for a little while. And then hopefully that'll change. So yeah, I wanted to on the margin, invest in people doing alignment. Seems like that's important. I thought Sydney was a kind of an example of this. You had, you know, Microsoft essentially released an unaligned AI. And I think the world sort of said, hmm, sort of threatening its users. That seems a little bit strange. If Microsoft can't put a leash on this thing, like who can? And so I think there'll be more interest in it. And yeah, I hope there's open communities.

• 1:36:42Dwarkesh Patel?

  That was so enduring for some reason. It like threatening you just made it so much more relvable.

• 1:36:47Speaker 00

  Yeah, I think it's like, the only reason it wasn't scary is because it wasn't hooked up to anything, you know, like if it was hooked up to HR systems, or if it could like post jobs or something like that, then I don't know, get on a gig worker site or something. I think it could have been scary.

• 1:37:02Dwarkesh Patel?

  Yep. All right. Before we go, where can people learn more about the Vesuvius Challenge?

• 1:37:06Speaker 00

  Yeah. So Vesuvius Challenge is at scrollprize.org, S C R O L L P R I Z E dot org. Yeah, check it out. And I, I think it's very likely that somebody listening to this could be the one who wins the grand prize and decodes the scrolls.

• 1:37:21Dwarkesh Patel?

  Okay, excellent. Awesome. Okay. Well, Nad, this is a true pleasure.

• 1:37:24Speaker 00

  Thanks for coming.

• 1:37:25Dwarkesh Patel?

  Thanks so much for coming on the podcast.

• 1:37:26Speaker 00

  Thanks for having me.

• 1:37:28Dwarkesh Patel?

  Hey everybody. I hope you enjoyed that episode. Just wanted to let you know that in order to help pay for the bills associated with this podcast, I'm turning on paid subscriptions on my Substack at warkeshpatel.com. No important content on this podcast will ever be paywalled, so please don't donate if you have to think twice before buying a cup of coffee. But if you have the means and you've enjoyed this podcast or gotten some kind of value out of it, I would really appreciate your support. As always, the most helpful thing you can do is to share the podcast. Send it to people you think might enjoy it, put it in Twitter, your group chats, et cetera. It just splits the world. Appreciate your listening. I'll see you next time. Cheers.