Episode 15: Jeremy Howard on Fast.ai - The Network State Podcast

Balaji Srinivasan 0:00
Jeremy, Welcome to Network city podcast. And we’ve been, we’ve been friends or friendly online. I think for a while, you are the founder of fast.ai which is this incredible course that’s online. We both taught large online courses, so we kind of have talked about that. You’re the founder of answer.ai before that, I think you were at Kaggle, right? And you’re Australian, you have an interest in biomedicine, and I think we’re also into, I mean, peace and trade, broadly, internationalism and so on. Give me the spiel. Is that? Danielle, everything is that? Or give me Jeremy on. Jeremy,

Speaker 1 0:34
yeah, no, pretty much. I mean, I’ll say maybe fast. AI, most people know us for the course, because that’s how most people interact with us, but that was only one quarter of it so fast. AI was all about trying to avoid a kind of massive centralization of power and inequality due to what my wife and I saw in 2012 is likely to be a rapid growth of AI. And so we want to similar to open the eyes mission in theory, except we actually were open. Yeah. So we only so through their initial way, yeah. So we basically decided to get AI into the hands as many people as possible, including people with few resources. And so we did a lot of research to figure out how to make AI more accessible, because at that time, only five labs in the world, and yeah, the techniques to actually use AI in practice were not published. They were kind of like little Yeah. So my wife Rachel actually asked earlier, when he was presenting in like, 2012 or something, about some of his work. It’s like, okay, so how did you actually do that bit? What weights did you use? You know what fine tuning with us? He’s like, Oh, we don’t, we don’t publish any of that. That’s our bag of tricks. So we were like, Okay, this is not okay. Like, this is this technology is going to change the world, and it’s requires a bag of tricks that you have to go to Stanford to learn, you know? So we figured out all the tricks and built a lot more tricks of our own. And then, you know, everybody then tried to make it all about money. So then Google eventually started creating TPUs and stuff. Instead of saying, like, oh, you can’t, I remember Jeff Dean saying, there’s no point trying to do stuff with AI unless you’re at Google, because only we have compute, yeah, and we beat them in a global competition to train image net, Kaggle. No, not that fast. Ai, oh, really, I didn’t actually know that. Yeah, yeah, that was a global competition called Don bench, and we competed against Intel. They had, like, a class, Don bench, D, A, W, N, bench, e, n, C, H,

Balaji Srinivasan 2:37
by the I love, I’m friendly with Jeff Dean. I think he’s amazing. And so, yeah, so, so that that’s actually pretty impressive. I mean, I’m sure he was impressed that you’re able to do so much.

Speaker 1 2:46
Oh, yeah, no, he was, he was great about it, you know, they, they published a post, they published a paper, and they credited us. And that’s not hard feelings, you know. But we just wanted, we just, we wanted to say, like, No, you don’t have to be a rich Google person to, you know, how is that successful?

Balaji Srinivasan 3:02
Actually? Maybe you can talk about that because, like, that’s a little surprising to me. Because, you know, obviously deep seek has brought costs down recently, but back then was it, did you like, obviously, Google had massive amounts of clean data and huge compute resources and so on. Why could, how could the student projects be competitive with Google during dawn bench, because

Speaker 1 3:28
these big labs suffer from being over resourced. So in fact, not as bad now, but particularly around that time in the next few years at Google, you were explicitly rewarded for using more compute. Where else we were, like, hey, we don’t have much money. Like, we made no revenue. We had no grants. It was just my wife and I put our own money into fast AI

Balaji Srinivasan 3:51
experience. How are they rewarded

Speaker 1 3:54
for you? So they were basically, if you could use more TPUs, that’s like a good tick on your performance. No, really, yeah, wow. Okay. So, you know, we came along and said, Hey, like, so, for example,

Balaji Srinivasan 4:08
it’s because they wanted people to use the TPU, since they

Speaker 1 4:10
wanted to, like, show off how big their their rig was. Look at our big rig, and these people using our big rig to do these big things. So, for example, in Dawn bench, it was an image recognition competition, be as fast as you can to train a model. And the images were 224, by 224, pixels. And we thought like, Okay, well, 90% of the time the first 90% of training, we’re going to train on 64 by 64 pixel downsized versions. Yeah, makes perfect sense. They look the same. You know, the last 10% we use bigger ones that 4x or 16x delta. Nobody else thought of that. You know, this is one of the major tricks we used. And why would anybody at like an open AI or Google try and do that? Because it’s like, oh, now we’re not using our amazing deep. Use.

Balaji Srinivasan 5:00
Well, it’s interesting because, you know, that’s actually, I’m actually going to put out a little little comic on this, actually, on that, which is, you know, that meme about a secret third thing is, people will say, Oh, you’re not an X or a Y, but a secret third thing. And they’ll say it sarcastically, like, Oh, you must be a Democrat, Republican. You’re not a secret third thing, right? But actually, if you think about like, like a, like, an image zero or one, one pixel is not enough to describe the complexity of an image. You need not just a secret third thing, but a secret fourth and fifth and 1,000th and millionth and so on, right? Pixels. But you know, there is, there is a minimum necessary complexity, right? And it’s interesting, because obviously, if you go all the way down to, like a, you know, if you have the number of pixels, all the way down to just one, you’re not going to get enough, right? So it’s an empirical question, going from 256, to 64 it still works. I don’t know, maybe going to 32 it still works. Maybe going to a Fave icon. It even kind of still works. I don’t know if you did that, if you

Speaker 1 5:59
absolutely did. And I first just did it visually, you know, I just downscaled it, and I looked, and I was like, Can I still see what that is, right? And if I couldn’t see it, then I thought, computer probably won’t be able to

Balaji Srinivasan 6:11
do as well. What was it? Was it like? Was it 16? Was it 32 was kind of 6464

Speaker 1 6:16
Yeah, okay, yeah, at 32 you squint, yeah, it, you can kind of see it’s maybe a dog, but you can’t see what kind of dog it is. I see, instance, interesting.

Balaji Srinivasan 6:27
Yeah, okay, so okay, I want to actually, I love, I love this. So first of all, I want to actually show you something we’ll jump around or whatever. I want to show you something that we have done that I think is a compliment to fast.ai and this also, so I taught a MOOC in 2013 called Startup engineering. I’m a big fan of it. Yeah. Okay, great. So I did that with Vijay Pandey. My colleague is a big fan of Vijay as well. Great. So he’s now at the bio fund. We invested a lot of bio stuff together, so we have that overlap

Speaker 1 6:54
as well. We’re interested. So you and Steve Huffman created those two fantastic courses. I don’t know if you have a look. I don’t know. Steve Hoffman’s ghosts, yeah. So similar thing. They were both like, kind of end to end, like, how to make stuff? Oh, okay, got it. And

Balaji Srinivasan 7:09
that’s a Reddit founder, yeah. He’s my friend. Also, I didn’t actually know, yeah, of course, yeah. So, and neither of them are really available anymore, and they’re, you know, I free web development course by Steve Hoffman, interesting data. We need a we need a modern one. All right, okay, so how about this? Maybe I’ll do a refresher, and we can, we’ll send the fast AI people put it online and something like that. I think that I do think a 2025, version. So actually, you know, let me tell you what I’m planning to do next on this well. So the reason I taught that course very similar, I think, in some ways, to your, you know, kind of kind of thing is, I know there’s a lot of talent on the internet, right? And actually, really, around the world. And you know how, like, the, you know, it comes to the dark matter, and, like, the Hubble telescope, and you can find the dark matter around the globe, or not the globe the in the universe, right? So, like, gravitational lensing, yeah, exactly. That’s right. And so you need, like, a special telescope to see that, right? So, by analogy, just a fun analogy, the mobile telescope, like the phones that billions of people now have, allow us to find, if the Hubble telescope allows us to find the dark matter, the mobile telescope, so to speak, allows us to find the dark talent around the world, right? Basically, people who really have nothing other than their phone and their hunger to learn, right? And we can offer them a course, and that’s like a sky hook and a bootstrap.

Speaker 1 8:28
That’s what fast.ai was about as well. Like, we really reached out to parts of India and Africa and stuff that had nothing. So we had, like, a guy from the Ivory Coast who was like, asking, like, is there some way to get this on CDs, because we don’t have internet here? And yeah, turned out, like one of our biggest markets was in Lagos.

Balaji Srinivasan 8:50
It’s amazing. So actually, I have a fair number of folks in in Nigeria, basically anywhere there’s Anglophones around the world, in India, Nigeria, in the Philippines, right? There’s actually all these Anglophones, meaning just English. I do want to translate into other languages and so on, but I think that’s like, the v1

Speaker 1 9:07
right? Yeah, absolutely, yeah, no. I mean, it was just like, it is all this talent around the world, and it drives me crazy that it’s not being used, you know, they’re like, picking coffee beans or whatever. Yeah, and, and, as you say, like, they’ve got, like, so many of them were saying, like, I’m training a neuro particularly when colab, Google Pay, like, came along, like, I’m training a neural net on my phone, you know, through co lab, you know, can you help me do this or that? And I’m just like, oh, this is great, you know. And so there was a there was a young woman from Bangladesh, one of our first courses, who contacted me, and she was like, Jeremy, you probably don’t even know who I am, but I’m in Bangladesh, and I’m a teenager. And she was like, I want to know if what I’m doing is okay, because I feel shame. Same she said, I don’t know anybody else in my province that does anything with AI. I don’t know any other girls that use computers. Everybody thinks I’m weird. I want you to know. I want to know if you think it’s okay for me to do AI.

Balaji Srinivasan 10:16
Oh, she just needed the social encouragement.

Speaker 1 10:19
And I and I wrote back, and I said, not only is it okay, but like, you know you’re gonna put your province on the map, you know? And you know what? Like, couple of years later, she wrote to me from Google in Silicon Valley. Wow. Thanks to you. I’m now a Google Scholar. They flew me over to San Francisco.

Balaji Srinivasan 10:37
What I like to do is, I like to find these folks, mentor them, train them, stand them up, and now they’re leaders in their own communities. It’s, you know, quote, Teach a man, Teach a man to fish, or teach a man to recognize an image of a fish, right, so to speak, right? Actually, you know, you can use that. That’s a good one liner, you know, because you open with the you open with the bird thing from, from XKCD. So teach a man to recognize an image of a fish or woman, you know,

Speaker 1 10:59
right? You know. The fish specifically you need to know is the Tench. Tench, the Tench. Anybody who’s understands computer vision knows about the tension, because tension is the first image net category. So anybody who’s image, Teach a man to recognize a Tench, yes,

Balaji Srinivasan 11:17
yeah. That’s good. That’s right. Actually, that’s like replaced Lena, yes, exactly, yes. That’s right. Okay, so let’s see, um. Now, why give me the German life story? So, like before? So I know fast. Ai, no, Kaggle. I know answer. Ai, I know the COVID and, you know, masks, what’s uh, like, what’s uh

Speaker 1 11:35
before Kaggle. So yes. So Anthony and I kind of got Kaggle started in Melbourne, in Australia, and then we flew out here. He had this crazy idea that venture capitalists in America would put money into our little startup. And I thought it was crazy. I thought there was no way, but he was right and I was wrong as like, Okay, I’ll come I’ll give it a go. But, you know, is

Balaji Srinivasan 12:03
Kaggle? Have some is it an Australian is just, sort of just like a funny word, made up word, just a

Speaker 1 12:09
made up word, okay, yeah, like Google, Kaggle, yeah, and and, yeah. We spoke to some of your old colleagues. We spoke to mark Andre and it was interesting at that time, Andres and Horowitz hadn’t done anything in machine learning, and in the end, they were very good about it. They passed on our round and they said, Look, we don’t know anything about machine learning. Maybe it’s going to be a big deal, but we don’t have anybody here that can judge that or not. But you know, so we ended up with like Bernard khosler and other folks put the money in. But before that, I had two startups in that I ran out of Australia. One was called fast mail, which became a very popular global email company, and then the other was called optimal decisions, which, if you’re insurance, you would definitely know, and if you are not, you definitely wouldn’t. It. Basically, trans changed how insurance companies price away from using just actuarial methods to using optimization based

Balaji Srinivasan 13:06
methods like convex optimization or something like that. No, yeah,

Speaker 1 13:09
yeah, just, you know, pretty classic optimization. But the key thing was to model elasticity and competitor price, not just risk. Because if all you do is model risk, all you can do is cost plus pricing, which, as you know, is economically very suboptimal. So we made insurance companies a lot more profitable, which I have no pride over. In hindsight, I don’t know why I spent years of my life working on that, but yeah, originally, I don’t know. Like coming out of school, I was a bit lost, to be honest, because, like, I was interested in stuff that nobody else was interested in. So I was interested in, like, spreadsheets and databases and PCs. This is a bit over 30 years ago. I didn’t know any other adults or kids that were interested in any of those things in Australia, yeah, yeah, okay, and there weren’t any university courses you could go to that were about data. So I ended up doing philosophy, but I actually ended up not going to any classes because I happened to get a job at McKinsey and Company, where they really appreciated this odd set of skills I

Balaji Srinivasan 14:21
had. So tell me about so McKinsey is actually interesting to me because there’s the, let me give the negative and the positive view of McKinsey. So the negative view of McKinsey is, oh, you know, you’re hiring overpriced consultants to tell you to fire people and blah, blah, blah, blah, blah, right? And the positive view is it’s something that takes young people and gives them lots of different kinds of business experience, and, you know, lets them actually see the actual numbers of lots of businesses, and actually trains people to make, of course, good slide decks and good presentations, but really to communicate well and understand the gears and nuts and bolts of businesses. Is. And actually, when I’ve hired former McKinsey and Bain and so on, people, they’ve actually done fairly well. They’re very good non technical athletes, like power users or what have you, right? I don’t know. Give me your thoughts on that. Maybe,

Speaker 1 15:13
oh, I mean, you know, I would say this unusual. Sorry to be negative. I mean, it’s just like the Pro and, oh, I love, I love. Like, please. Like, challenge me. Okay, good. If I say something worth challenging, challenge me, because otherwise it’s boring for everybody listening too, and boring for me. Look, I started there when I was 19.

Balaji Srinivasan 15:29
So, oh, really, wow, that’s interesting. Yeah.

Speaker 1 15:33
So I was years younger than everybody else, and for me, it was eye opening, and it was great, because suddenly there were people who cared about what I did. And you’re right. They’re generally non technical people, just one of the reasons why, as a 19 year old, I could be really successful there. You know,

Balaji Srinivasan 15:55
did you feel you leveled up when you were there?

Speaker 1 15:57
Yes and no. It’s funny. You say it’s this kind of polarizing thing. It was polarizing in my life too, right? Because at one level, it’s like, I felt like, Okay, I need to learn business, because I didn’t know any of that stuff and I wanted to create my own companies.

Balaji Srinivasan 16:11
Yeah, you’re very commercial for a professor. Yeah? Professor type, yeah,

Speaker 1 16:15
yeah. Well, I mean, I never went into, I’ve never been a professional academic in my life, right?

Balaji Srinivasan 16:21
But you’ve got, you’ve got the, I think we both have that disposition,

Speaker 1 16:24
yeah, sure. No, absolutely. And so I was trying to learn business, and by being at McKinsey, I did learn a lot about how business worked, but also in a lot of ways, it’s a very conservative organization. Because I was telling my colleagues at the time, hey, this new internet thing. I think it’s going to be big, you know, and they’d be like, I don’t know Jeremy’s computer stuff. It’s, this is pretty nerdy. It’s like, what’s it for? Like, I don’t know exactly, but I feel like, like, very early 90s. I feel like it’s going to impact business. And they’re just like, No, look. Let me explain how business works. You know, business is about relationships and strategy and capital and, you know, and in the end, like they were wrong, you know. And I didn’t have the trust in myself at the time. You didn’t know whether you were wrong or I was sure I was wrong, right? And I just kept trying to figure out why I’m so wrong. I felt really upset with myself for being stupid, that they everybody else can see it. It’s so obvious that they’re just like, look, Jeremy. Let me try to explain it. I just couldn’t get it. So I wish I had, you know, I stayed in consulting for 10 years. Oh, really, well, I should have done it just two because that’s enough. And, like, what I really learned there was a sales. Like, it’s really great for learning sales.

Balaji Srinivasan 17:47
Well, what did you like, I don’t know, what are the top three five things you learn in McKinsey? Like, sales,

Speaker 1 17:53
yeah, so I was and ended it Carney. So I went from there to 80 Carney. What I learned was like, Okay, it’s all about change and influence, right? So it’s not just sales, but it’s a kind of sales. It’s like, you’re trying to sell an idea, or you’re trying to sell a piece of work, whatever. So we were very careful about mapping out the organization, you know? So it’s like, okay, we want to sell this piece of work next, or we want to help our client sell this idea, okay, who’s everybody in the organization, who’s in any way a stakeholder who could have an opinion, who could cause this to succeed, who could cause this to fail? Like, okay, who do we know? Who knows that person and, like, extremely kind of careful and optimized process of creating change through human management and human connections. We brought professional actors in, like, play the role of different types of clients, and we would then interact with them, and then, you know, then talk about what the results were. It was just way more intense human optimization than I’d ever conceived of. I’d always thought of that human side as being like, oh, some people are charismatic, you know, or Oh, some people are just good at convincing people. It’s like, no, they’re their skills. There’s a science, there’s a there’s a logic, there’s it’s like a different kind of logic to programming a computer. But if you want to get an organization to do a thing, you know, you have to know how to map it out and how to react. You know, in

Balaji Srinivasan 19:33
some ways it felt it’s a graph traversal in some ways, yeah,

Speaker 1 19:37
but in some ways it felt cold and kind of calculating and horrible to be like, Oh, this human being. I don’t seeing that as a human being. I’m seeing them as, like, this cog in this machine, and I’m going to use this process. But it totally worked, yes, you know. And so it made me, after a while, I changed my view of it. I was like, You know what? Like getting organizations to do? Things. Is important. It is important. And so if that involves treating people as machine parts, sometimes, because humans are very predictable, yes, you know. And so if you learn how to manage different types of humans and different types of situations, and like, you know, so like, you get the one person to be your kind of inside bowl who’s, like, super champion or whatever, and they’ve recognized that they can use you to advance their career. And then you talk to them specifically about how they can advance their career. And then they tell you who’s going to get in the way. And then you get three more people, and then you use that to put pressure on the fifth person who is well known to, you know, be somebody who likes following rather than leading. And, you know, you structure it out. It’ll play out. And at the end it’s like, okay, it happened. You

Balaji Srinivasan 20:52
know, it’s funny, like the way, you know, do you know Mark cranny at a six and Z? I don’t know if you know him. He’s a very different personality than you, but he also, he’s like a gruff Mormon a few words, but he’s like, a sales genius, actually, right? And very similar, like, the way I think about it, that kind of reconciles all of it is, it’s a nested set of like, Win Win relationships all the way up to the organization level, right? Like, the best kind of sales is when you are genuinely selling them something that will improve their business, or their their product, or something in some way, right? And then it will also improve at a nested level, the career of this person who approves it, and so and so. It’s almost like a, like a venture investment all the way through. And that is actually what I think is the reason that that’ll work is that’s the most consistent kind of thing, where even if you’re flipping them to do it, they will like it in the medium to long run,

Speaker 1 21:39
yeah. And if you’re trying to have a dent on the world, you know, and you’ve got good ideas and develop good things, but you’re unable to influence anybody to buy it or use it, then you’re not going to have a dent on

Balaji Srinivasan 21:53
the world. Like, that’s actually, you know, it’s funny. One of the, I mean, there’s a lot of great things about your course, but one of the best is the domain name fast.ai, right? Like, I learn AI fast, amazing. Okay, that’s what I want, right? So that’s like an example of sort of an inbuilt marketing kind of thing, which is great, right? And I’m sure there was some thought into that, because lots

Speaker 1 22:11
of people could have named it. Oh, yeah, we did a lot of marketing stuff there. We also, as far as I know, we were the first company in the world to do AB tests on our homepage. Oh, is that right? Interesting. I think we were also the first to have all the free email accounts. A little footer would be added to every email message marketing the service like we did a lot of little things like that, things like that, little viral things that today are everywhere.

Balaji Srinivasan 22:36
Yes. So, okay, great. Actually, I want to show you something which is so we took. So let me describe problem and then solution and get your your thoughts right. So you and I have both taught large online courses, right? And the typical thing that happens with large online course is people, it’s a little bit like signing up for for like a workout, right? People, aspirationally want to do it, and then they want to have done it. They want to have done it. Yes, exactly, that’s right, yes. And then they

Speaker 1 23:08
want to be the kind of person that would have done that. There’s

Balaji Srinivasan 23:11
something good out of that, right? But what happens is they sign up for and the problem is allocating the time, or then, if they have the time, the energy, or the discouragement, or what have you, there have been various mechanisms and so on to try to solve that, address that right? There’s like Cohort Based Learning and, you know, and so on. And those things work to an extent. Cohorts are great, yes, so that can work. But let me show you something that we did, which we call a learn a THON.

Speaker 2 23:37
When should you use a random forest? What is the confusion matrix? Dunno. What about collaborative filtering? Dunner, you

Balaji Srinivasan 24:04
When could you use a random forest tabular data, and if you have a lot of like, noisy features, what is the confusion matrix? Like a table of actual answers against like the predicted answers, and then comparing, you know, like how often it gets it right, and then when and how much it gets it wrong? What is collaborative filtering recommendation algorithm by clustering people or items or things by similarity. So basically, well, we’re going to do a virtual, you know, updated version of that, but basically, so the fastest AI. So essentially, literally, we took, because what is it like? About 10 hours, 11 hours of videos, right? Yeah. So over two days, we said, Okay, you really want to do fast AI, okay, sign up, come here, 9am on on Saturday morning and nine to nine. Saturday, nine to nine, Sunday, they watch every single video start to finish. No phones, yeah, right, yeah. And then when it was time to go and type things in, you know, laptops out, do

Speaker 1 24:58
that absolutely. And it drives me. Crazy, because so many people tell me, like, oh, Jeremy, I started your course. I meant to finish. You know, I’ve tried three times I haven’t managed to finish. I always think like, look, yeah, you could just put aside one weekend and just binge it, you know, get it done.

Balaji Srinivasan 25:15
Yes, exactly. And I want to, did I shoot the fellowship video? Okay, hold on. Take a look at this. Okay, global meritocracy is finally here, because we’re awarding $100,000 in funding for the new network school fellowship, and anyone from anywhere can apply. Now you might well ask how well you see we’ve set up shop on an island right off the coast of Singapore, in the new Special Economic Zone, and it has an enlightened immigration policy. That means it’s the perfect place to assemble a global community of tech founders and AI creators. And that’s what we’ve done. We’ve set up housing, food, co work, fitness classes, yoga, fast Wi Fi, office, pots, a state of the art gym, healthy snacks, Starlink, a makerspace, a Content Studio, guest lectures from the most successful founders and investors in the world, Nomad visas and help with everything else you might need. And we have funding too if you’re good. So go and apply for the nervous tool fellowship [email protected] the only connection you need is an internet connection.

Speaker 1 26:16
That’s very inspiring. I want to come great. So also, Malaysia is awesome. So Malaysia, that’s

Balaji Srinivasan 26:23
right? So basically, the combination of Singapore, Malaysia and the new Singapore Johor Special Economic Zone, you know, it was one of the things where there was theory, and then somebody had to put that into practice, right? So the theory is, like, Singapore has a lot of capital, but doesn’t have a lot of land, yeah, Malaysia is actually improving a lot, yeah, but it doesn’t.

Speaker 1 26:41
Malaysia’s got a good education system. It’s a strong,

Balaji Srinivasan 26:45
yeah, very underrated, and it’s improving a lot, and you can basically live a pretty good life there, I think, super good. And it’s right next door, right? Yeah. So Malaysia has land and has less. Can literally drive there, you literally drive there, I literally drive back and forth all the time, right? In fact, we’re just, like, 30 minutes from Singapore. Basically, just, literally, you know, just go over the bridge, pop. You can see, you can see Singapore directly from from it, right? So, and we’ll have probably have a ferry or something back and forth that’ll give down to, like, 15 minutes, yeah? So I want, like, these autonomous boat kind of things, right? So why not? Yeah. So those knock on road, let’s get, let’s get that, right? So this is something, what you’re seeing in a video, is something I’ve wanted to do for more than 10 years, right? And you just have to build all the, you know, overnight thing 10 years in the making. So certainly, anybody who’s like doing fast AI, who’s taking the deep learning courses, we’re looking for the kinds of people who’ve completed your course, and yeah, we can fund them and help them build things, and in particular, to think about. So let me explain kind of the motivation behind what we’re doing with network school, right? So a it’s very hard, obviously now to get student visas, skilled worker visas, into the US. It’s, I mean, even like people who are tourist visas, like they’re getting strip searched, or crazy things happen. You saw there’s actually some Australian or what have you, like, some terrible thing happened to them or that,

Speaker 1 27:57
right? And you think every, almost every country now has some story examples of people, citizens of their country, that have been screwed around,

Balaji Srinivasan 28:05
tourist visas, student visas, skilled worker visas, like

Speaker 1 28:09
the US and in Southeast Asia, these countries are now competing for that talent with their digital visas, with their startup visas.

Balaji Srinivasan 28:17
Exactly, it’s so smart. This is exactly, that’s right. And this is the thing I was

Speaker 1 28:21
like, I want Australia to get on there. Australia to get on that boat too. You know, we’ve had this global talent visa in Australia, which is pretty good. So, yeah, I have it. Everybody

Balaji Srinivasan 28:29
needs to do this. You know, the country’s offering digital nomad visas, right? So there’s this weird thing where the US is taking itself out of the global economy, yeah, just as everybody

Speaker 1 28:38
else, everybody else is diving in, exactly, and all of America’s big value creators are tech,

Balaji Srinivasan 28:46
that’s right, exactly. And they’re globally mobile, because there’s no silicon in Silicon Valley. No, we’re not like

Speaker 1 28:50
mining. So our team NS AI is fully distributed in Turkey, Japan, Australia, Ireland, if

Balaji Srinivasan 29:01
you ever want to co locate them, we can host them in every school for a week or a month or something like this. And one of the things you want to do is, like, co location for remote teams.

Speaker 1 29:09
That’s a nice idea because, like, we’ve got together for the first time ever in person here in Singapore. Oh, great. And we’re all like, Oh, it’s so nice to spend a week together. Eric Ries and I at answer AI, we did something a bit unusual. We decided to only have one policy. And our only policy, at answer AI, is to only have one policy.

Balaji Srinivasan 29:33
Okay, what is that policy? The

Unknown Speaker 29:35
policy is to only have one policy. Oh,

Balaji Srinivasan 29:39
it’s very meta. Is like, one of those recursive kind of things. Go ahead,

Speaker 1 29:42
I’m done that we only have one policy, and it’s to only have one policy. So you can’t have no policies, because that’s a policy,

Balaji Srinivasan 29:50
okay, okay,

Speaker 1 29:51
so we have no policies other than the policy that we’re only going to have one policy, I see, okay, got it and why? Well, policies. They’re like, ideologies, they’re like, they’re these fixed things which say like, oh, you can turn your brain off now, because, because we’ve decided, X, you know, in this situation, this is how you’re meant to behave. Like, I am deeply skeptical of ideologies and policies and all of these cognitive shortcuts that basically say like, oh, I believe in this thing, because that’s what my ideology says, you know,

Balaji Srinivasan 30:26
yes. So let me give an analogy or a way of thinking about this that I have from the network state book, which is, um, you know, like programming paradigms, you can have imperative programming, functional programming, declarative programming, and so on and so forth, right? And for certain problem domains, you know, certain style, it just makes it very easy and concise to solve that problem domain, right? But then you also want, like, a multi paradigm language, like, like Python, or something like with Haskell, you know, you can just do everything as f of g, of H, of x, and you can actually get far with that. But it’s sometimes nice to do things in an imperative style, or what have you, right? And, and so that’s how I think about political paradigms, right? Like, and every analysis. You know, I’m not a big UFC guy, but like, Ultimate Fighting Championship is some people are using grappling, some boxing, some Muay Thai, and it’s situational. As to do I solve this with a kick or a punch, right? Do I Solve this as functional or imperative? And I think, like Lee Kuan Yew was someone who was like that, where he understood many different political schools of thought, and then he just, like applied the right technique that was sort of self consistent in that school of thought for that situation, right? And so that’s like the beyond ideology thing, which is, you’re aware of a lot of these different things, and you situationally figure out which one is appropriate, and then you use that because

Speaker 1 31:50
Andrew, you know, you’re constantly curious and interested. And you know, what you care about is doing a good job, you know, rather than being consistent with other members of your tribe, most humans are mainly interested in being consistent with other members of their tribe. That’s right, number one driving force.

Balaji Srinivasan 32:11
And the thing about that is, there’s, there’s a meta rationality to that. I think it’s kind of like general, like evolutionary game theory, right? So, like, you can imagine you have two populations of people who are conformists and dissidents, so to speak, right? And the dissidents are constantly exploring, and they’re taking a high risk, and sometimes they fall off a cliff, and sometimes they have reward, and the tribe follows them, right? And the conformists are just, you know, they’re like, this is risk capital, and this is just, you know, stay home, money, or what have you, so to speak, right? So you can make an argument for a portfolio strategy as to why you want a small number of dissidents who are sometimes wrong when they’re wrong, or contrarians, or whatever you want to call it entrepreneurs, right? And then most people should actually, like go the tribe, so they don’t run off a cliff, but they could actually find, you know, a better, better pasture, or something over here, that’s, that’s one way of thinking about the respective balance. Go ahead. Yeah. I

Speaker 1 33:04
mean, I’m kind of curious about this, because, like globally, somehow every jurisdiction has settled on the same education system, and the education system teaches children to be conformist. Yes, if you, if you you know the test tests whether you can feed back the things you were taught in the way that you were taught them, you will get rewarded if you do what you’re told and like, I’m kind of curious about how much of this thing we see in the world is because every single child, basically in The Western world, at least, has learned this same behaviors. Have you

Balaji Srinivasan 33:43
heard the concept of the Prussian educational system? Yeah. Okay, do you know what preceded that? No. Okay, so there’s this great book we can put it on screen called, called the craft apprentice. Okay? And one of my macro kind of theories of the world is that history is running in reverse. And I can show you a bunch of graphs on that, or what have you, but literally, like a U curve, where, in many ways, our future is more like our past, like, more like, let’s say the 1850s and then eventually the 1750s than the 1950s like, there’s a lot of U curves which have their minimum or maximum in 1950 and I can, I can show you some graphs on that. And so one premise of that is, like, prior to the Prussian educational system, which was, which is what we currently know, K through 12 and so on. That was all set up. It was inspired by Bismarck after German unification to have all the children get basically the same software in their heads. It’s like, you know how with Windows, you have, like, the default install that comes off the factory. And then you have like, you know, Windows premium, ultimate, maybe for college graduates. And then you have the service packs from, you know, mainstream media. That’s how I kind of think about right, right? And, and there’s a reason for that, because then everybody kind of has the same references. They they salute the flag, and, you know, they’ve just got the same basic install, and they can interoperate, right? There’s, there’s a rationale for that. It’s how you it’s a softer part of constructing a nation. In fact, arguably, that’s even after. As important as, quote, the hardware part, right, which is like the physical territory and the people and so on. But before that, there was a different system, which was all based on apprenticeship. And they would start working from an early age, and they would just learn practical skills very, very early on. Or they’d be like Jebediah and Abigail would have 12 kids, and they’d all be working on the farm, and they’d be like mini industrial robots, so to speak, picking fruit or something like that. You know, mending fences very, very early on. So the entire concept of extended adolescence wasn’t there, the concept of being on your parents health insurance till 26 or whatever, it wasn’t there. And now, the reason that that stuff got introduced in part, is because, I think, in the in the late 1800s with the advent of like industrialization in factories, these kids were no longer under the supervision of their parents. Or if people the parents knew they were under the supervision of factory owners who would push them too hard, right? Like these were like the child labor factories, you know, and so and so forth. And that was a dis alignment between, like, the interest of the factory owner and the kids. That’s when the child labor laws were passed and so on. I mean,

Speaker 1 36:06
that took a long time. It took a long time. It’s like, what was it like 6070, years, Britain was the first in the world to introduce child labor laws. But yes, much still took much longer than it should

Balaji Srinivasan 36:16
have. That’s right, this old Dickensian kind of era, or what have you, right? So then, so now there’s a good to that at first, but then that’s what actually led to the modern era of adolescence. And you know, I’m having fun as a kid for a long period of time, and now we have this extremely extended adolescence and training period where some people are like students as doctors, all the way up into their 30s before they start their career. And they’re almost middle age before this, you know. And I think that the the corrective to that is, because everything good, you can always overdo it, right? And so you can go from quote, you know, like being opposing to child labor, to not allowing people to even work until they’re in their 30s, as a doctor, for example, or right? So I think the opposite of that, the thesis, antithesis, synthesis is when the kid is at home and they’re under the supervision of their parent, but they’re able to start earning online by doing development, software development and so on. Even 1012 years ago, I had a bunch of kids. Some of my best students at Stanford 1012 years ago were were kids who had actually earn their first dollar doing online programming in their teens, right? And it’s not even so much about the amount of money. It is that it’s that the market is greater. This is how I think with Have you seen the grade inflation graphs? Yeah. So, like you put that on screen, but basically, kind of crazy. Everybody gets a 4.0 basically, students are the customers, so they’re basically buying a job. And so how do you how do you deal with that? And my answer is, the market is a greater right? So now you have kids. They’re doing software. They can’t hurt themselves, like in a factory. They’re under supervision because they’re working remote at home, but they’re also like, apprenticing, right? I think we network school, we also want to make that happen where now they’re in a friendly environment along a bunch of other adults. They can run around and roam and so on. And then they can level up. They can be next to an electrical engineer, next to a mechanical engineer, as they’re building robots and stuff like that, and just help them with small things, right? And they start to see what the like, what adults are doing. And it’s not just being, you know, sitting at a desk the whole day, right? So let me pause there. That’s kind of how I’m thinking about part of the future education. Maybe you have some thoughts. Maybe you

Speaker 1 38:23
have some thoughts. I have a lot of thoughts. Yeah. So, I mean, I I know a lot of kids who are in that kind of interesting group who are basically ready to go to university when they’re like, 11 or 12, and adults all try to stop them. Oh, interesting. It’s like, we don’t. For some reason, the vast majority of adults I deal with don’t want children to learn when they’re ready to learn. They have to learn at the speed which they’re expected to learn. They want a speed limit. Yeah, yes. And they assume any kid that’s keen to learn more, it must be the parents fault that they’re pushing them kids. Kids are not allowed to have curiosity and drive and passion, but actually, not every kid learns everything at the same speed. Yeah. So I’m very interested in, like, how do we help this that talent at the much younger age, not because I want to, like, make them more productive or whatever, but just because I know so many of these kids are deeply unhappy when they’re artificially held back, and I want to, all you know, help them all have the opportunity to to have that excitement of feeling like they’re achieving their potential, that they’re that they’re just really happy with the things they’re building. So I’ve got a kid, you know, she’s nine, and she’s we let her basically have whatever opportunities she wants. You know, when she chooses her curriculum, and she chooses what she does, and she’s happy for us to provide her some guidance as well, you know, but we don’t. Don’t force her to do anything. And, yeah, she’s got this great cohort of friends all around the world now who learn in this way, and are all doing it at their own speed. Obviously, with AI, there’s a lot of opportunities to help more and more of these kinds of kids develop as they’re ready, you know, and get a much more customized, personalized, dynamic education experience, one that’s not focused on conformity or authority. You know, sometimes my daughter comes back, she’s like, she does lots and lots of extracurricular things, you know, one of them is trampolining. She comes back from trampolining, sometimes she’d be like, Oh, I got a gold star for good behavior. Isn’t that great? And I always say, like, I don’t know. I’m not sure I want you to have great behavior. You know? Why do you think that’s so important to have great behavior?

Balaji Srinivasan 41:02
Well, well, of course, it depends, obviously, like a layer of dissidents and so on, on top of a fundamentally pro social attitude is good, but if people are, like, anti social and they’re littering or they’re, you know, yelling in the street, that’s

Speaker 1 41:14
exactly it’s, it’s, it’s not necessarily, you know, being the best behaved kid in the class, and getting the gold star that week is not necessarily the great thing, and it’s not something, not something I want her to be proud of, right? You know, yeah, she’s incredibly pro social, she’s incredibly kind, she’s incredibly generous, but that doesn’t mean she has to do everything she’s told as soon as she’s told

Balaji Srinivasan 41:36
to do it. That’s right? And this is, it’s funny, says, because

Speaker 1 41:38
basically, particularly for a girl, like, like, like, girls are particularly taught to like, fit in and do what they’re told. And I don’t want her to be somebody in society who just fits in and does what she’s told.

Balaji Srinivasan 41:52
I think, I think this concept of like, the balance and so on where it’s like, you know, as you said, they’re pro social, and they’re kind, but they also don’t obey every single command and so so

Speaker 1 42:04
yeah, I tend to focus on empathy with my daughter, which maybe ends up in a similar place, yeah, just like, particularly for younger kids, empathy doesn’t necessarily come as easily. So I have to kind of say, like, Okay, you thought that was funny. Now, can you try to imagine what that person’s situation was. Do you think they would have found it funny? You know, if you were them in that situation, has anything similar happened to you before and eventually? So I’m just like, oh, wow, did I just do that thing to them that another person did to me that made me sad. Like, Oh, wow. I feel so sad. I didn’t want to make upset that person.

Balaji Srinivasan 42:39
It’s funny because, you know, sometimes you can get to like, just like with religions, you can often get to a similar behavior pattern by different kinds of religion, different so I had a recent tweet a little bit viral on actually, that exact topic of empathy. And essentially what I said is, because I was, I was talking to conservatives, and I was saying, Look, empathy is actually a useful concept, even for a completely cold blooded capitalist, right? Why? Because you have to understand other guys point of view and their win, win, right? And a lot of the like, especially in today’s America, they’ve gotten themselves in the in the mental state where I think everybody’s exploiting them, everybody’s ripping them off, right? And that, like, Australia is an enemy and Canada’s an enemy and Vietnam is an enemy, and whatever, right? And it’s like, you know, lots of people are just neutral, right? They’re just business partners, or they’re just, like, living their lives. And you don’t have to, like, fight, and you can’t fight the entire world, and you also have some understanding of, okay, what’s their win? And how can we get to a win? Win, often a win, win is more profitable for both parties involved, and so on and so forth, right? So

Speaker 1 43:48
you can and actually, altruism is programmed into us, like this is something we’ve discovered, like, evolutionarily, it’s been programmed into all of us to not be altruistic is to fight against your basic instincts, and that’s really dangerous, because when you fight against things that evolution has programmed you to do, you’re creating a new unstable equilibrium. So why has that happened? Well, presumably, there were plenty of groups that had no altruism in their villages, you know, just genetically. They didn’t have that as part of their DNA. Didn’t cooperate in the died out. They died out, you know. And so we, we as a species, you know, we’re not perfect, right? But you don’t want to underestimate the power of what we’re born with. You know, when we’re born. You know, altruism is not weakness. Altruism is is strength. These are the people that survived. And if you want to fight against that, then you’re fighting against a basic survival instinct. Also it’s, it’s nigh on impossible to design and organize such a complex system. They they arise. Over a very long period of time to create these marvelously stable equilibria, you know, and this is what kind of terrifies me at the moment, is there are so many opportunities to destabilize that equilibrium right now, you know, with with technology and the connectivity we have, and historically, each time you get a previously stable equilibrium is damaged, sometimes ending up with, you know, hundreds of years of societal misery. And so I always just like, I’m definitely very keen to see change and growth, but I want people to understand the power of where we’re at, and know how hard it was to get there, and to know enough history to know that, you know, creating, you know, destabilizing an equilibrium creates a power vacuum. And there are certain people who are extremely motivated and good at taking advantage of power vacuums, and the pair, the people you definitely don’t want in power you know? Yeah, well, I don’t know. Like, somehow Singapore did an amazing job, like, the one country in the world that, like, I think they just got lucky with Lee Kuan Yew. Do you know what I mean? They ended up with a guy who’s kind of incorruptible. He doesn’t have a huge chip on his shoulder. He just cares about outcomes. Most places around the world, in that situation, end up with, you know, basically a, you know, deeply insecure chip on their shoulder, power hungry person.

Balaji Srinivasan 46:33
Is funny about Lee Kuan Yew, which I think is very underappreciated, is he like he could argue his case in English. I think this is the most underappreciated aspect of Lee Kuan Yew, because he would argue his case in English. He could argue on the global stage right other people understood at least his point of view. He could make it cogently. He could do it in short form. He could do it in long form, sound bites and then extempo, you know, long speeches, extemporaneously or in policy papers, and he made sure that Singapore won the argument. And if you win the argument, then you often don’t have to fight, right? Because there’s like, that swing vote in the middle who’s like, you know what? He has a point here. We should do it his way, and so on and so forth, right? And I feel that, for example, there’s other other folks in East Asia who delivered comparable economic results to LKY, right? For example, in South Korea or in Taiwan or what have you. But they couldn’t make their argument in English, right? That’s a really exceptional aspect of they could speak in Korean, they could speak in Chinese, but, like they couldn’t, they couldn’t make their case on a global stage, right? And, and I think that’s very underrated, and it’s something I think about a lot because so let me, let me actually slightly counter argue with you on the power vacuum thing, which is there, right? I think that we are about to enter a period where the the future is China versus the internet. Should I elaborate on what I mean by that, China versus the internet. China versus the internet, right? So the 20th Century was sort of a symmetric thing, you know, almost like basketball, like the final four plays, and it then ends up as US versus USSR, everybody slugs it out, right? Sean McMeekin has this book called Stalin’s War, where he kind of makes a point that World War One and world war two can be seen almost as like a 30 Years War, like, years war, like an extended bar brawl with people like smashing chairs over each other’s heads all around the world, right? And then it kind of lands up as the US versus USSR, right? With Japan and Germany eliminated, and and, and other powers too, US, UK, France, blah, blah, right? I think this century is going to be different where it’s not a symmetric thing, but asymmetric. Like China and the internet are, I think the balancing things and China’s obvious. I think the internet is not obvious. What I mean, but China’s obvious. China, if you take the quote, American empire, I think China inherits the manufacturing and the money and the military, or not all the money, but the manufacturing, the military, and really the might of it globally, like blood, the alliances and so on. The world is, after this tariff thing, recentralizing around China, totally right? Quickly, interesting

Speaker 1 49:11
to see how eminent that is, but that’s, it’s something very deep happening there. Yeah.

Balaji Srinivasan 49:16
So, so I think what’s going to happen, and it’s not just economically,

Speaker 1 49:19
also culturally. You know, America’s cultural power has been enormous. It has been, that’s right. So now in Australia, I’m seeing people being like, oh, America’s kind of cringe. Now it’s

Balaji Srinivasan 49:30
cringe. Now, that’s right. But I think that the other air that the less visible but as important air, is the internet, which it has the people, the values and the language, okay? And the reason I say that is the only thing that has economic scale comparable to China is actually the internet like so that’s that. Why? Why am I into crypto? I’m into crypto because everybody in the internet is equal, meaning you’re peer to peer. You can send packets back and forth. You’re the same. Property rights, you have same contract law, right? You have the same monetary policy. And so whatever you were born into, you can opt in to a system of law that is superior to the one that you were born into. And it’s like emigrating to at least half of what a government is, right? It’s not the land, it’s not the physical territory. Yet I’ll come to that, but it’s at least the property rights, the and you have to have some sacrifice. You have to buy some of the coin, or whatever. You start to start interacting with this. Now you have, like, a system of law that’s often superior to the one that you inherited, whether it was in Nigeria or is in, you know, Lebanon or something like that. These places have destroyed currencies. They don’t guard property rights. Now you can finally save because, you know, the blockchain protects your savings, right? So I think that the internet has half of what we want, which is it has a system of government, and with all these blockchains, multiple systems of government, and actually compare it, one of the ways I think about it is, you know, early America, it didn’t actually think of itself as America at first. They were British colonists, right there. They’re, you know, like the Virginia colony, Massachusetts colony, and they had a land and they had a people, but they didn’t have a government, right? Because the government was in London, and took a while for them to develop a sense of national consciousness and realize, oh, that’s actually not our government. Our government is here, right? So they had land, people government. They became America, right? I think the internet is evolving in the opposite way it has the people, and actually as a government, in the form of the blockchain. I blockchain, but doesn’t yet have land. I think that’s the

Speaker 1 51:26
next step, and hopefully it won’t be versus unfortunately, Xi Jinping has moved into a power vacuum in China. Prior to that, actually, China was much more of a democracy than people realized. Talk about this. Well, I think a lot of people don’t understand how the political situation in China worked. So there was a lot of voting, but unlike most western democracies, the voting was entirely within the party. Party. Yep. Now people might think, Oh, that’s not very big.

Balaji Srinivasan 51:59
It’s actually 100 million people. Chinese guy was very big. And then you

Speaker 1 52:02
go and it’s not like of my so I spent a lot of time in China, and with a lot of really great people in China, young people, and the vast majority of the best of the people, most what they wanted to do was to get into the party. So the not commenting on whether this is good or bad, but it ends up with that kind of a democracy of you know, the hardest working, most intellectually capable people

Balaji Srinivasan 52:29
can I make a provocative comment? So there’s a book called The party decides. Point of that book was the American unit. Party decides who’s actually running on the Democrat and Republican side. For many years, there people have said a choice Don X, choice, Don echo or whatever, right? And so there’s a similarity to that where there were, quote, smoke filled rooms where the candidate was determined. And certainly with a recent Democrat primary, it was something where, basically the party determined who was running, and so on and so forth. Then there’s a whole disaster, the whole Biden comma thing. So there’s more similarity to the American system for many years, where there was essentially a unit party that decided, like, who the candidates were. Then some would argue, and now I’d say, in a sense, we’ve had true democracy burst forth, but that some people conceptualize this democracy. Let me pause

Speaker 1 53:14
there, yeah. So, yeah. So that’s another whole connection I’ll leave aside for a moment, which is that actually, yeah, there’s, there’s actually a lot more conspiracies in the world than people realize. There’s a lot of smoke filled rooms. I’ve been in plenty of them. Yes, I think I just wanted to mention is the thing that was missing in what you said, is the is the key power for me, the key issue for me, which is the presence of positive feedback loops. Now, when I say positive feedback loop. I don’t mean good feedback loop. I mean a feedback loop which goes back and causes more of itself. So power and wealth, like, like viral reproduction, something like that. But like, power and wealth are naturally positive feedback loops. Getting more power puts you in a position to be able to get more power. Getting more wealth puts you in a position to get more wealth, and then you’ve got the cross correlation. Getting more power helps you get more wealth. Getting more wealth helps you get more power. I talked to them earlier about the importance of a stable equilibrium. How can you get a stable equilibrium in a situation where somebody getting ahead can let them get more ahead, right? There’s a the compounding interest. There’s a huge tension here, right? And this is where democracy and capitalism and the market economy come into a huge, a huge problem, right? Which is if, if you allow those positive feedback loops to happen, then you end up with people who have incredible riches and incredible power because they’re on the right side

Balaji Srinivasan 54:46
of that feedback loop. Yes, okay,

Speaker 1 54:49
it’s a natural disequilibrium, and it’s not compatible with actual market forces or with democracy. Because you’re now in a situation where you can, like you can buy the media, you know you can, or nowadays, like the social networks or whatever you can, you know, stick all the odds in your favor. And that is not again, that’s not a resilient state to be in. So somehow, many societies in the world have managed to create sophisticated, complex equilibria that have avoided this for decades, you know, but it’s not the natural state of things. The natural state of things is for there to be, you know, one incredibly wealthy and powerful person that you know is there because of the power of

Balaji Srinivasan 55:40
positive feedback. Okay, so let me, let me disagree with that in two ways. And then you counter argument. Counter argument. The first is, there’s a saying like shirt sleeves to shirt sleeves in three generations, right? Which is to say that, like this guy, he starts a factory, his son inherits it, and his discipline, grandson puts a fortune up his nose and, you know, does drugs and, you know, basically spends down the whole thing, right? And this is like the resource curse concept, where when people get too wealthy or too powerful, these get extremely lazy. They forget cause and effect, especially if they’re two or three generations out, they don’t even know what hard work resulted in that fortune in the first place. And they just blow the whole thing up. And that’s actually what’s happening with the US right now. Like in many ways, I think the people are currently running the US government are not founders. They’re heirs. They’ve inherited the system that, like better people, set up decades and decades ago. They don’t even understand how it works. It’s like a factory they’ve inherited. And they don’t understand how it produces widgets, or how it maintains global order, global peace, and they just think I’m big and powerful, and they don’t understand why it

Speaker 1 56:45
exists. I think that’s true, but it doesn’t matter, because the what the data shows is that over multiple hundreds of years periods, the wealthy families stay the wealthy families and and at like, the highest levels of power you know, you like, if you look at the history of the, you know, English royal family, whatever, or of Chinese emperors, like, they stay there for hundreds of years, you know, and they create, they create feudal systems underneath themselves, which are critical for establishing loyalty and all that. That’s the more natural state of things that things fall into, unless you can maintain that equilibrium.

Balaji Srinivasan 57:29
Okay, so I’m on a counterarguance from an argument that I think is interesting, at least, maybe to you know, maybe you’ll disagree. So if you have an heir, or you, let’s say you have a like a Genghis Khan, right? They have two like they have a child, they’ve got half their DNA, then another child, they’ve got a fourth, then another child, they’ve got an eighth, right? And most of the time, people don’t have an exponentially increasing number of children. So that means that that fortune, for example, would or whatever it is. It’s very hard to pass a fortune down many generations, number one and number two is that person almost doesn’t even exist anymore because their genes are being split up, diluted, like Does the person even in what sense is somebody who’s only 116

Speaker 1 58:10
part of the same family, right? I think you’re dramatically though, over emphasizing the importance of genes over context

Balaji Srinivasan 58:20
so, like, they’re four generations down. How is it they’ve got a bunch of descendants, right? The vast majority of their descendants must like, what does it even mean to say a family across four or five generations that family doesn’t like?

Speaker 1 58:34
Arguably, not how power is transferred, right? So power is transferred by picking an air, and they have an air, and they have an air, and then as soon as there’s like a lack of a clear air, then you get 100 years of war, and then somebody wins, and now they have another, you know, air, air, air, like they, that’s the thing They they generate this system of hierarchical loyalty and and they do like you can see historically, sure that people do maintain it,

Balaji Srinivasan 59:06
but I had made two points. First is, most of their heirs are not inheriting that fortune, so the majority of the family or the descendants or whatever are not right, because it would be divided. The second is even this fourth or fifth generation guy is now, like 1/32 Genghis Khan, or, or what have you. And so they may just not have the the zeal or the energy of the original Genghis, right? Say, lose, and then there’s a new guy who, who takes over, right? So basically, what I’m saying is it’s almost like there’s a there’s a huge tax, like a 50% tax every generation, that makes it very hard to keep concentrating the same stuff in the same because the same people don’t even exist three or four even there’s some In brief,

Speaker 1 59:48
you’ve got the premise wrong. Premise is that what better there is the genes. And what I’m saying is no Balaji. What matters is the power of the positive feedback loop. Power gets begets power. Yeah, it doesn’t matter if my I’m five generations away from Genghis Khan. What matters is I’m the king of England or I am the king of France, right? But you know, like, if, like you saw what happened in China, hundreds of years of terrible emperors, opium addicts, destroying the country. They still maintain the power, right? And the country went from like during the Tang Dynasty, the, you know, the vast majority of GDP in the world was in China. Cultural Center was was in China. Scientific center was in China. And then, through power concentration, the civilization died, you know, sure. So we don’t want that to happen,

Balaji Srinivasan 1:00:46
I guess so. Let me agree with you on that, and I do think that it needs to be alternatives and so on and so forth. I’ll just make one other point, which is, if that person is only 1/32 or 1/64 Genghis Khan, then there were 31 or 63 other people or families that rose so so like the mobility is actually there. If there’s it’s if it’s a sufficiently exogamous society, then all these folks did rise to become rulers, because their bloodlines actually did get up there. So basically, what I’m essentially, what I’m not, what I’m agreeing with you, is the title got passed down, but the family doesn’t even exist beyond 564, whatever number of generations, right? The family just gets diluted out. Does that make any sense?

Speaker 1 1:01:31
Yeah, but that’s what I’m saying. It doesn’t matter, right? What matters is that you the positive feedback loop created a power and wealth concentration that was maintained for hundreds of years, and most people in the country suffered, right? And that’s the thing that we want to avoid, and it’s incredibly difficult to avoid, because that’s the natural state of things. It’s positive feedback loops,

Balaji Srinivasan 1:01:57
I guess. I guess maybe this is an empirical question, and we can look at different trajectories, but I think it is difficult to maintain that power and wealth concentration without zeal, and that zeal, if it’s not there, what people get fat and happy a few generations out, like we’ve seen that. I mean maybe, maybe we’re just thinking of different kinds of examples, right? And for example, in tech, it’s almost entirely quote new money, right? And what I find is that people who’ve inherited fortunes are just lethargic, right? They don’t have that energy. So we are seeing this internet disruption, right, this dark talent that’s hungrier. I would always invest in that. I would always back that, because it’s hungrier and it wants it, right? Whereas, so I’m always seeing, seeing anti compounding. I’m only seeing, I guess, yeah, no,

Speaker 1 1:02:43
and I agree with all that. But I’m trying to get you to think about the end state, okay, like, like, I, maybe I agree with everything you’re saying, right? But what I’m trying to say is, okay, consider the positive feedback loop here, right? You’ve with AI now you’ve got the ability to create more power, you know, and and more wealth, and we’re more connected, like we could literally end up with a global dictator, and we could literally end up with a permanent underclass representing 99.99%

Balaji Srinivasan 1:03:19
of the world. So let’s talk about how we prevent that, right? Because I, because this is something I do think about, right? So my view is, and you may may or disagree with this or not, or is that we got people got more left than they expected. Now they’re getting more right than they expect, more Maga, and then they’re gonna get more China than they expected. Like, basically, I think what’s gonna happen is China’s rolling up a lot of alliances, like the EU is doing deals with China, all its historical rivals in Southeast Asia are now just all folding in. So the whole global economy is recentralizing around China and America has not just become isolationist. They’ve isolated itself from the world and the most punishing, they’ve sort of self imposed the most punishing sanctions of all time on themselves, like a rogue state North Korea, Iran would face this kind of embargo, but it was, like, self imposed because they think it’s going to make them strong. It’s really kind of crazy stuff, Maga Maoism or whatever, right? So, as a consequence, I think a lot of power gets centralized in China. And

Speaker 1 1:04:14
along with that, interestingly, you’re seeing a huge this kind of cultural isolationism happening in America, yes, also, like, quite difficult to undo, potentially extremely difficult, because they don’t end up like Japan pre the myji restoration. No, they thought they’re powerful. They thought they’re strong, but actually they separate themselves in society and

Balaji Srinivasan 1:04:36
become weak. That’s a good outcome. I actually, I think it’s quite that’s a good outcome. Yeah, fair enough. I think, I mean, because that’s actually something where they give up the Empire, but they’re just like, you know, a country, or

Speaker 1 1:04:47
would stay isolationist, except they’ve got nuclear weapons. Well, that’s, that’s a

Balaji Srinivasan 1:04:51
problem. The thing is, I think, you know, there’s a lot of people who will say, like, actually, both on the left and the right, will say, we need to, you know, over. Public non empire, or we need to shut down, you know? And the problem is that, first of all, maybe you’ll agree with these things. I’ll give a view, and then maybe shoot at it, right? I think the first thing at least, that I start with is American Empire is real, and it was spectacular, in the sense of, arguably, for all its faults, one of the greatest of all time. Absolutely, it did have capitalism, democracy, world peace, in many ways, then lost its way, especially recently. And now you’ve got a very common kind of thing where folks on the left think, Oh, the US is bombing lots of countries. It should stop doing that. Folks on the right think the US is being exploited by all these foreigners abroad. It’s being cheated. We’ve de industrialized. We need to stop all that. Bring all those jobs. Okay, fine. So this group thinks the US is harming the world. This group thinks the world is harming the US. Both them think they want to shut down the Empire, bring the troops home, you know, and so on. Okay? I remember

Speaker 1 1:05:51
also, like, during that heyday of the 50s, you know, the American top marginal tax rate was like 80% like 90% Yeah, yeah, that’s like, they’re working very hard to avoid this positive feedback. I loop, I mentioned, you know, redistributing the wealthiest So,

Balaji Srinivasan 1:06:06
okay, so on that point, just to talk about that, the at that time, though, power was completely centralized in the US government, right? So you almost have, like, a toothpaste tube squeezing where, like, if you avoid centralization on one axis, you often get it in another kind of thing, right? So, um,

Speaker 1 1:06:24
because, because the people who want power will find ways to get it. Yeah, you can have

Balaji Srinivasan 1:06:28
total centralization of government power, or you can have totalization of corporate power, or maybe military power and so, or you can have checks and balances. And where I think the world is going to go is a billion person, Chinese super state. And then eventually, like, 1000 million person network states like and then I think India is going to be in the middle. I think there’s other countries are going to be in the middle, and so on and so forth. But that’s, that’s where I think things go by, like 2040, or so, right? And and so hopefully that gives, I’m not saying they’re all a million person networks. Some might be bigger, some might be smaller, but I but I do think that we’ll have a lot of choice of jurisdictions.

Unknown Speaker 1:07:04
I mean, that would be nice. That’s at

Balaji Srinivasan 1:07:07
least a hope, perhaps. Yeah, good.

Speaker 1 1:07:09
I just got to say, keep thinking about the positive feedback problem, because I think it still has it, you know, it feels, you know, Rosy to the level of being late. Well, that’s that seems not in line with power

Balaji Srinivasan 1:07:24
dynamics, I guess. I guess my biggest argument against that is arbitrage, because it’s very difficult to get. Or let me give a game theoretic argument, right, which is going back to your sales example, right? If you have two people, you have four possible outcomes in a Win, Lose thing you can have win, win, win, lose, lose, win, lose, lose, right. If you have three people, you have two thirds. So eight possible outcomes, win, win, win, win, win, lose, right. And you have k people, you have two to the k possible outcomes where you know n of them can win and n minus k can lose, and so on for any value of n and k, okay, so this is how I think about, like managing a startup, right? With a startup, if you have 100 people, what you don’t want is political behavior where some subset of them loses and the other subset wins. You want to have a single thing which aligns everybody, and that’s like equity, and that’s like the exit so they all know, if I work together, we all get the maximum payoff when it’s all win, win, win, win, across the board, right? However, there’s limits to how large you can make that, right. You might make that 100 people. Might make that 1000 you might make it even a million people, like cryptocurrencies, of getting it to 10s or hundreds of millions of people, right? But I don’t think you can get to everybody. And the reason you can’t get to everybody is at some point there is an incentive to break away, to dis line, is what I call network defect, right? And so that is a counterweight to kind of, I think what you’re seeing about infinite compounding, it’s actually,

Speaker 1 1:08:50
if you’re allowed to go ahead, if you’re allowed to, like, I mean, like, yeah, it’s like, oh, you know, the people in Wessex could have left, or whatever. It’s like, no, they’re in a feudal state, and they would have got killed, and there’s violence and, right? And, like, if you add AI in the mix, then you can have, like, absolute global surveillance and power and total control, right? So now, okay, so here’s it’s fine. In theory, you could go and do something else. In practice, if you even talk about it, you get shot in the face, yeah.

Balaji Srinivasan 1:09:16
So the right, so the practical way, where I do agree with you is the Chinese drone Armada, right? Because they can manufacture huge numbers of robots, and those robots are they’re no longer like human beings who can defect, right? Because they can’t defect all these concepts I’ve been talking about with the game the principal agent problem goes away and just one guy pushing a button, and it’s like a machine that just enacts our action around the world, right? That is definitely something which changes these dynamics. That is actually something where you could have centralization and power for a long time, and that is actually something we should think of as the most important thing to build counterweights to going three to five.

Speaker 1 1:09:58
So I think your network states idea can hit the. That too. So

Balaji Srinivasan 1:10:01
fast.ai. You’ve got this practical deep learning for coders, part one, part two.

Speaker 1 1:10:06
We’ve done a new course called How to solve it with code, and we built a whole new platform for it, which we basically beta tested it. We opened up sign ups for 24 hours, kind of reasonably quietly, 1000 people signed up within 24 hours. So then we closed it, we did that, and the reactions we got were amazing, like we’ve had hundreds of people come back and say, this changed my life. I’ve got a new job. Well, it’s not open for everybody, but it’s it’s solve it.fast.ai. So we’re trying to figure out how to now make the most of this. Because we’ve come up we’ve created something clearly extraordinary, basically the fundamental idea. I don’t know how familiar with the polio book, but it’s basically

Balaji Srinivasan 1:10:53
like it’s a bag of tricks for solving math problems. Yeah, but it’s a bag of tricks.

Speaker 1 1:10:57
It’s actually a fundamental idea, which is that to do things iteratively, step by step, and when you apply that idea to coding, and then you bring AI into the mix as well, you can, we’ve kind of come up with this way of solving problems with code and AI, where you’re constantly in control of the AI. You never get into that situation where the AI is kind of controlling you, yeah? So, yeah. So we, like, I say we, this was from months ago. We haven’t let anybody use it for months because we’ve been running it and testing it, yeah. So it’s a bit of a long story, but basically it’s a whole different way of thinking about problem solving, which is the exact opposite of the whole

Balaji Srinivasan 1:11:39
vibe coding kind of, it’s like, let’s think step by step for humans. Yeah, let’s

Speaker 1 1:11:44
think. Let’s think step by step for human plus AI together. The AI sees all of your thinking. You see, the AI is thinking. You write code. The AI write code. You’re constantly focused on learning and iteratively improving. You know, vibe coding, it’s just like one shot thing where you don’t learn anything, you get up more more technical debt. So it’s actually, it’s interesting, like my co founder, Eric Ries, has this Lean Startup approach, which it turns out, is really similar to the Polya approach. Again, it’s like highly iterative learning based. So we’re hoping that through this solve it course, that we’re going to eventually build something like your startup engineering. Oh, great, but, but using this solve it approach and with, with the help of AI, to allow and then to create, like, 1000 new startups from that course, and then work with investors to give each of them, you know, a start financially and maybe hopefully build the next generation of founders.

Balaji Srinivasan 1:12:45
Amazing. And I think, you know, would be good. I want to actually talk about the network school fellowship with your fast AI folks. I think a lot of them could benefit from applying. So, okay, awesome. Thank you very much.

Unknown Speaker 1:12:57
Jeremy, great.