SpaceX IPO, The Erdős Problem, Spotify CEO Joins | Alex Tabarrok, Bill Clerico, Alex Norström, Jordan Schneider, Christina Lee Storm, Erik Bernhardsson

16:34Erik Bernhardsson:Yeah, but anyways, comparing it to anyone can have a pretty wild experience with AI on a ton of different services. Yeah. You could never do that with the metaverse.

16:51Alex Tabarrok:So Lisan Al-Ghaib is contrasting Gary Marcus's sub-stack post with what's happened in the AI industry. Anthropic valuation up 173 % since the start of the year. posting profits in Q2, according to the Wall Street Journal. Open AI evaluation up 67 % since the start of the year. And Open AI general purpose model solves longstanding and well-known Erdos problem. Is it Erdish or Erdos? I believe it's Erdos. Erdos. Erdos problem. The two dots over the O get me on the pronunciation. Without a scaffold. And so there was a lot of questions about how hard is it to solve these problems. But fortunately, we have Tyler Cosgrove, who's going to take us through what actually happened with this solution to this math problem that people are very excited about.

17:43Alex Tabarrok:Noam Brown said, today we are sharing that a general purpose internal OpenAI model achieved a breakthrough on one of the best known combinatorial geometry problems. Less than one year ago, Frontier AI models were at IMO gold level performance. I expect this pace of progress to continue. and Siddhar Ramesh. I don't know if he was joking about this bet, but he says, I have lost my$30 ,000 bet that AI would never solve the planar unit. Yes, I believe that was a joke. That's a narrative violation. He jokes around a lot, but no. But a lot of people were surprised and a lot of people were excited about this.

18:19Alex Tabarrok:So take us through what actually happened.

18:19Jordan Schneider:Okay, yeah. So I can basically go through like a simple explanation of what the problem actually is. Okay. So just for some context, Paul Erdos, kind of this legendary mathematician, throughout the 20th century, He basically proposes, I think the number is like a little over 1 ,200 different little problems. These are the Airdos problems. People talk a lot about these as like goals for AI to solve. And you've heard like over time there's been kind of like small iterative kind of solutions to a lot of these problems. Yeah.

18:45Alex Tabarrok:Sort of like collaborative, a mathematician working alongside AI models. Or an easy one, just getting.

18:51Jordan Schneider:There's like a main kind of like place where all of the solutions go. So sometimes people will find, like AI will find a different paper that wasn't actually put on the website. And then they're like, oh, AI solved it. But it's honestly true. But this is kind of the first time we've really seen kind of a big step change. Like this is actually a new solution. This is using like, you know, kind of novel ideas here.

19:13Alex Tabarrok:It doesn't exist in papers out there already.

19:14Jordan Schneider:So this was problem number 90. So I can kind of read the question and then I can explain what it means. So it's, does every set of n distinct points in the real plane contain at most n to the 1 plus 0 of 1 over log log n many pairs which are 1 apart? Okay, so what does that mean? Basically, we have the real plane, 2D, and we have a bunch of points on it. What is basically the, how many pairs of those will be basically one unit apart? And what's the max number? How do we basically organize those points? such that we have the max number of them. It's not just a grid? No. So you would think that, but I can basically explain why.

19:56Jordan Schneider:So let's kind of like formalize this better. So we have U of N, right? And this is basically the largest number of unit distance pairs among N points in the plane. Okay, so basically, like, we're thinking about, like, how do we solve this? Naively, it's like, okay, what if we just take all the points, we have N points, and we just put them in a line, and unit distance apart, right? So it looks something like this. Yep. Right, so for this example we have four points, but it doesn't matter, it's just n. So one, two, three, four, how many pairs are there? There's three, right?

20:26Alex Tabarrok:Okay, yeah.

20:27Jordan Schneider:So basically this scales with n minus one. Yeah. Right, so you would have a billion points and there's nine, nine, nine, nine, whatever, n minus one. Yeah. Okay, so now if we put it in a grid, what happens, right? So if we have a square grid here, there's nine points here, and then how many pairs are there? I believe there's 12. And basically as this number scales up, it's still linear, so it's 2n. Basically if you do a billion points, it's two billion pairs. Two billion pairs. Yeah. So then basically...

20:54Alex Tabarrok:The pair is specifically that line. Correct. The line represents the pair.

20:57Jordan Schneider:Yes, yes, because it's one unit distance, right? And you specifically don't...

21:01Alex Tabarrok:Diagonal doesn't count because it's not one unit length. Because it's not one unit length.

21:03Jordan Schneider:Yeah, it's not a unit distance length. Got it. So then, okay, what's the next thing we can do? The next kind of configuration is what's called the lattice construction. And so if we can pull up a picture of it, it's this kind of crazy-looking grid that has all these super intricate lines in between. You can see it on the—this is from the OpenAI blog. If we can pull it up here.

21:26Alex Tabarrok:Oh, I think I saw this.

21:28Jordan Schneider:So this is what it looks like. So if you can zoom in on any of these points, you see that it looks like a grid, right? But there's not just kind of pairs at the edges, right? There's like way more.

21:40Alex Tabarrok:Okay.

21:41Jordan Schneider:So this scales at n to the 1 plus 0, 1 over log log n. This is basically the best kind of example that we know works.

21:52Alex Tabarrok:Yes, but not a proof.

21:54Jordan Schneider:So we know we can find this, but is this the upper bound? We don't know. So this is basically the lower bound. So then the question is like, we have the lower bound, which is that this is the best one we found. This is the most number of pairs. And then we theorized that the high bound, upper bound, scales with n to the 4 thirds. And then so Erdos, the original conjecture that he thought that the upper bound is still going to be less than n to the 1 plus O of 1. This means O of 1, it's like as n scales to infinity, right? So O of 1 basically scales to 0. It goes to 0 as n goes to infinity. And then basically OpenAI figured out that this is not true and that there actually are some n's for which this kind of max number of pairs is greater than the original Eirdosh conjecture.

22:55Jordan Schneider:So for infinitely many n, this is not for every single n, so it's not like five points or whatever, but there are infinitely many n's for which this is true.

23:04Alex Tabarrok:Okay.

23:04Jordan Schneider:That it scales with, that it's greater than n to the one plus some constant. Okay, so that was basically the big thing, right? This is like, you know. Huge if you're into math. Decades old problem, right? This is an incredible thing. Terence Tao is like, wow, this is incredible. Yeah, yeah. But yeah, that's basically the overview of the problem. Okay. But yeah, it's very exciting because this is not like a math model. This is just an internal model, general reasoning. You could say it's generally intelligent. Yes, I think you could say that. And then I think it's interesting because from public perception, it seems like this didn't take that many tokens.

23:49Jordan Schneider:This was not millions of dollars of inference time. It was maybe something like hundreds to thousands of dollars of inference compute.

23:57Alex Tabarrok:That's very interesting because we were talking about like Guern's conjecture about like novel ideas coming from just like brute forcing different connections between things and this is more token efficient.

24:08Jordan Schneider:Yeah, this is not just taking some solution to a different Airdish problem and just like spamming it on all 1 ,200 of the problems and oh, one of them works. This is like kind of a new novel idea. Like maybe this solution is, the way that they found this, it's like super complex. You can read the proof. It's like 18 pages long. I don't really know what it means. But a lot of the mathematicians are saying, okay, this actually could be useful to a lot of other problems. This is like a new way to do things. So I think it's very exciting. This is like maybe similar to the alpha fold moment or something where now this is like a real kind of step change in math capabilities.

24:44Alex Tabarrok:Are people going to run out of problems? How many more problems do we have?

24:48Jordan Schneider:I don't know the exact number. I think so there's 1 ,200 in total of just Airdov problems. I believe the number is around 500, 600 of ones that have been solved. But I mean, there's so many of these like famous open math problems. There's like the Millennium Prize, right? A million dollars if you win.

25:05Alex Tabarrok:How much did OpenAI win? If you get a million dollars for the Millennium Prize, how much did they get for cracking this puppy open?

25:11Jordan Schneider:So every single Airdev Prize, if you solve a problem, there is a prize. Yes.

25:17Alex Tabarrok:I'm anchored at a million, so.

25:19Jordan Schneider:Yes. So these are going to be slightly less. I think this one was around$500. I believe there's$25,$100, and$500. This was kind of the big one. Non-dilutive financing.

25:31Alex Tabarrok:Yeah, non-dilutive financing.

25:32Jordan Schneider:Yeah, so it could be a revenue pathway for OpenAI.

25:34Alex Tabarrok:I have to imagine that the inference bill for this was over$500.

25:39Jordan Schneider:I think so, yeah.

25:40Alex Tabarrok:So the mathematician that just works with a coffee still.

25:44Jordan Schneider:So obviously, what does this actually impact? Does this bring us to novel science? How do we make money off this? Maybe not, but it does show that, like, okay, these models are not just, you know, the average of all their training data. They can actually make novel ideas.

25:57Alex Tabarrok:They're outside of distribution, outside of the training data. It's not just knowledge retrieval.

26:01Jordan Schneider:Yeah, I think this is very much contrary to, you know, kind of the Gary Marcus take. These are just, you know, parrots. These are just predicting the next token. These aren't actually intelligent. Yeah, yeah, yeah. Yeah, I think it's very exciting.

26:12Erik Bernhardsson:Yeah, cool. Well, thanks for bringing that. Why don't you hit the gong for the researchers?

26:16Alex Tabarrok:Yes. Boom.

26:21Erik Bernhardsson:Thank you for breaking that down. As universities, classes become live podcast settings. Podcasts are becoming university settings.

26:31Alex Tabarrok:This is where we're going. Well, let's take everyone through our lineup today because we have Alex Tabarrok from George Mason University and co-host of the Marginal Revolution podcast with good friend of the show, Tyler Callen, coming on in just a minute. Bill Clerico from Collective Capitals coming on. Alex Nordstrom. Convective. Convective Capital from Spotify. The co-CEO of Spotify is coming on. Then we have Jordan Schneider back from China Talk, going to probably spend a few minutes talking about the underwhelming summit in China and then move on to a bunch of other talks, a bunch of other topics.

27:04Alex Tabarrok:And then we have Christina Lee Storm from Secret Level joining at 1 p.m. Pacific. So the other news, rumors about OpenAI is being close, closing in on the IPO filing. And that pushed out NVIDIA's results, which is normally something I would expect on the front page. But there was too much AI news. NVIDIA results skyrocket on rise of AI agents. We talked a little bit about it yesterday. But the big news is that they're doing an$80 billion share buyback authorization. and Take Him was bullish. People were wondering where he would sit on NVIDIA. He laid out a pretty convincing case. You can go listen to the interview.

27:46Alex Tabarrok:It aired yesterday on TBPN. But without further ado, we have Alex Tabarrok in the waiting room. Let's bring him in to the TBPN UltraDome. Alex, how are you doing? I think we don't have audio. Can we check your microphone? Make sure it's working. We're good. Okay. It was something on our end. How are you doing? I'm still doing great. Fantastic. Perfect. Fantastic. Perfect. It's great to have you on the show. It's been far too long. I'm a huge fan. I've read Marginal Revolution since I studied economics back in 2010 and a huge fan of the Marginal Revolution podcast. I very much enjoyed your debate with Tyler Cowen all about the cost disease or Baumol effect.

28:29Alex Tabarrok:And I I think we should start there and then we can go all over the place in AI and labor and the economy broadly. But do you want to start with an introduction on the Balmol effect? Maybe why it captured your interest and some of the work that you've done around it?

28:48Bill Clerico:Sure. I mean, probably a lot of people have seen this famous chart where you have a bunch of things going up in price and a bunch of things going down in price. Yeah. And the question is, is why? I mean, the things are going up in price. I think we all know is like, you know, health care, education. Right. And then things are going down in price, often manufactured goods like automobiles, quality adjusted or televisions, computers, things like that. And the question is, like, why? Why do we see these big differences? And what Bommel pointed to was that there's a problem with service industries, you know, like education.

29:28Bill Clerico:You know, think about what I do, which is teaching students. And you think about what Pythagoras did. Like Pythagoras, he's got, you know, 10 or 12 students around him. And he, you know, he puts a triangle, you know, in the sand and, you know, draws some math there. It's more or less what I do, right? I mean, thousands of years later, maybe I'm using chalk or maybe I'm using PowerPoint. But basically, you know, it's me and a few students and productivity really has not gone up at all in that in the education industry. So because of that, prices have to go up because you take an industry where productivity is flat and you might say, OK, well, prices are going to be flat.

30:15Bill Clerico:But no, no, because all the other industries are improving in productivity and the education industry, they have to attract labor from those industries which are getting better. So they still have to pay me as much as I would earn in another industry. But my productivity hasn't gone up. So that means prices have to go up. So that's the basic Bommel effect is that industries, especially services where productivity is not going up, prices have to go up.

30:52Alex Tabarrok:Yeah, we have the we have the chart here. And I imagine that there are a few logical pushbacks. One is regulation. Everyone will say that the number of licenses to practice medicine is restricted. It's a taxicab medallion system. It's a regulatory capture situation. Does that not play into this chart as much as people think? Is that to be disregarded or is overregulation still something worth contending with if you want to avoid runaway inflation in or just disproportionate inflation in important services that people have demand for but are maybe paying through the nose for and not happy about?

31:36Right.

31:37Bill Clerico:Yeah, look, I'm a free market guy. You know, I'm anti-regulation, anti-bureaucracy, all that kind of stuff. But you have to understand with these trends, we're talking about increases in prices, which have happened over 100 years. And, you know, medical care, people were complaining about medical care going up in price in like 1920, 1930. This is before, you know, Medicaid, Medicare, before a lot of government involvement. Education has been going up in price. So I think it's deeper than just regulation. And let me give you just one other example is think about car repair. OK, you know, or cobblers.

32:25Bill Clerico:When was the last time you took your shoes to a cobbler? Yeah. You know, my mother will say, oh, the new generation, you just don't care about repairing things, you know. And I'd say, no, no, mom, I love you. But look, it's that the cost of repair has just gone up so much compared to just the cost of buying a new pair of shoes.

32:47Erik Bernhardsson:Yeah. How would you like to spend$70 to fix your$60 pair of shoes?

32:54Bill Clerico:Exactly. Exactly. So, you know, I had a my car had some I bashed it in the parking lot, you know, And it was like, not a serious, you know, just a surface injury, but it was like a third the price of the car, the value of the car, just to repair the side.

33:15Alex Tabarrok:And that's pretty typical. Yeah. So, I mean, a lot of people in the AI world are saying that AI will do to services what technology has done previously to manufacture, to goods, to manufactured products. does that mean that the the bommel effect goes away is it more pronounced in the industries that are ai resistant and you see some sort of runaway inflation in the things that can't be automated because when i think about teaching a college course i i do i do see the the sort of maybe linear scaling that you're discussing but we have microphones now we have amphitheaters you can teach more students there's online resources there's ways to delegate and manage teams of homework reviewers and and dis like sort of disaggregate the work and so you would imagine that technology would be a lever on services in some ways but i'm wondering how you think it will

34:15Bill Clerico:change in the age of ai yeah so the big question is robots right if you can replace labor with capital, then a lot of the Bommel effect goes away, which would be great. I mean, that would be great. Basically, you know, anything that improves productivity is good. But I think people also get a little bit too upset about the Bommel effect because really why another way of putting it is why are services getting more expensive? It's because manufactured goods are getting cheaper, right? It's because what we have to pay for the services in terms of giving up other goods has gone up in price. But still, we're richer than ever before, which is why people keep buying more education and more healthcare, right?

35:14Bill Clerico:So that's another reason why I don't buy entirely the regulation story. Because if it was sort of regulation, and the price of healthcare was going up, people would buy less of it. Yeah. But actually, they're buying more of it. Yeah. So it's the fact that, you know, our productivity is going up, we can afford more of it. That's really why the price of medical care is going up is because the price of computers and other things is going down.

35:42Alex Tabarrok:Yeah. How are you processing the latest and greatest in behavioral economics around maybe this concept of the vibe session, this idea that, yes, we are richer than we have been in history. And yet, I mean, right now, consumer confidence is very low. There seems to be a lot of dissatisfaction with the progress that the economy is making. How much of that is grounded in real economic data versus psychological factors?

36:11Bill Clerico:I'm amazed at the amount of psychological factors. It's something I've changed my mind on, how big the psychological factors can be. If you just look around the world today, no other country has done as well from globalization as the United States. I mean, it was us which kept open the sea lanes and globalized the world to our benefit. And we're the richest country in the history of the world, the richest at any point in our history. And yet somehow we're upset about free trade and globalization. And again, no other country has done as well at assimilating immigrants and doing well with immigrants than the United States.

36:57Bill Clerico:And yet we're upset about immigrants. So I think it's very disappointing. I hope we get over the bad vibes because the U.S. is a lot to be proud of and a lot to feel good about.

37:13Alex Tabarrok:Yeah. Is one of those psychological factors, you know, the average American's perception of debt to GDP? We seem to anchor debt, like nominal values of debt to GDP, which is sort of an income stream. and we get very irritated or anxious when debt sort of touches GDP and relative values. How are you thinking about the level of indebtedness that is appropriate for a modern economy to sustain itself?

37:48Bill Clerico:I'm not happy about the debt. You know, we still don't know. The U.S. has a big choice. Do they want higher taxes or do they want less spending? And the U.S. voter just keeps saying, how about neither?

38:09Bill Clerico:Sorry, this is not working, right? You know, add your cake and eat it too. No, you cannot do that. So the American public has just not decided which way it will go. So that is what is driving the malaise. I don't know. I'm not I'm not sure. I don't know whether I don't know whether it's cell phones. I don't know whether it's Instagram, you know, but it does seem that there is an anger, a grievance culture in the United States. At first, you know, I thought it was just on the left. Right. You know, when we had everyone's complaining, oh, African-Americans are treated so poorly. The women are treated so poorly.

38:54Bill Clerico:The poor are treated so poorly. You know, and then with Trump, we've just changed our set of grievances, you know. And so now, oh, it's it's it's the the the foreigners who are ripping us off. You know, crime is terrible. And of course, it's not. You know, all of these things. We've just changed our set of grievances. without actually focusing on... Is that not... How much does the fact that, you know,

39:26Erik Bernhardsson:so many people are downwardly mobile? I mean, I was born in the 90s, and I remember as probably an early teenager hearing, like, you know, statistically, this is the first generation where you're more likely to do less well economically than your parents did. And I feel like part of the challenge with that is that America's culture is so progress oriented, right? Like the American dream is just centered around, uh, doing, uh, doing, doing more, doing better. You know, if your, if your father was a, was a cobbler, like you own a shoe store, that, that kind of thing. And, and so now when you have, you know, still this like massive wealth and, and as you said, you know, the richest country in history, uh, uh, there's this constant comparison to, uh, to the past.

40:18Erik Bernhardsson:And I think frustration from that, which is making like, you know, huge swaths of the country frustrated and you want to, you want to blame it. You want to blame it on, on, like you said, foreigners or, or free trade, et cetera. But, uh, but it's just like this sort of latent frustration.

40:40Bill Clerico:Yeah, there's definitely frustration. I agree with that. Compared to the past, we are doing better. Yeah. There's no question of that. Now, to be sure, there are some key areas like housing, right? Housing is much more expensive, you know, that it should be. You know, that's a zoning problem. That's a choice people have been making, I think a bad choice. I think we ought to do something about it. But even with housing, you know, houses are so much better today. They're larger. Kids have their own rooms, you know. Air conditioning, parking, enclosed parking structures. Yeah. A large fraction of the housing stock in 1960 didn't have indoor toilet, didn't have indoor plumbing.

41:31Bill Clerico:That was still pretty common not to have indoor plumbing. So, so I don't think it's compared, I think if people are saying that we are worse off in the past, I think that's incorrect. Maybe what's going on is that inequality has gone up somewhat, not as much as people often think, but it has gone up somewhat. And of course, we have more access to seeing inequality. You know, Instagram and stuff like that, which is sort of a fake, you know. The Lamborghini might be rented.

42:11Alex Tabarrok:No, I completely agree.

42:13Erik Bernhardsson:Yeah, the other factor is that the forms of entertainment that are constantly being chosen today, like doom scrolling, right? When you maybe had the hippie generation, it was like, hey, we don't have jobs. Let's go to Yosemite. Just live there. And like what that would do to your sense of well-being versus like, I don't have a job. I'm going to sit on my phone all day long. Or, you know, last week.

42:42Alex Tabarrok:Everybody really hitchhikes across America when they're in between gigs anymore.

42:46Erik Bernhardsson:What happened?

42:47Alex Tabarrok:Yeah.

42:47Erik Bernhardsson:Yeah. But like just I don't have anything going on. I don't have opportunity. I'm going to go outside. I touch grass. is like going to have a wildly different impact on your psyche.

42:57Alex Tabarrok:I know a super successful corporate lawyer today that like was in the middle of a career transition, spent a year like surfing in Southern, in South America. And like, that is unheard of by today's standards. Like that just doesn't happen. Sorry. Did you have a, yeah. Yeah. I have also been thinking about like, if there was a, if there was an, if there was an alien that showed up and had a quadrillion dollars, Like if you include that alien in the genie coefficient, it skyrockets it. But the existence of that alien doesn't affect your perception until you're made aware of the diamond spaceship that the alien comes to town with and starts flexing on you on Instagram.

43:35Alex Tabarrok:But the mere existence doesn't change your economic well-being or anything real. But it does change your psychology once you see them flexing on you. I'm interested in the housing thing, though. I want to go back to a Bommel effect. because housing seems to potentially fall in the middle. Like there's manufactured goods that go into building a house. There are also services that go into a house. Like how do we apply the thinking of the Bommel effect and what you've learned to what would happen to housing? Like, is there a is there a second step? Everyone, when they talk about housing, will say, let's fix the zoning rules.

44:11Alex Tabarrok:Let's make it easier to build. But is that enough?

44:15Bill Clerico:Yeah, I mean, look, with housing, it's not the construction of housing, which has gotten so much more, you know, extensive. It's almost 100 % the land. The cost of the land.

44:29Alex Tabarrok:Yeah.

44:29Bill Clerico:So, and you look, I mean, you just go to, you know, San Francisco, San Jose, which ought to be right some glorious metropolis you know of the future with you know sky high buildings and people traveling around and you know uh fantastic uh it should look like a chinese city where

44:54Alex Tabarrok:there's led walls and yeah we have that on salesforce tower but outside of that there aren't these massive skyscrapers and it just it's a land of strip malls yeah you know

45:04Bill Clerico:Silicon Valley is a land of strip malls. It really is. And strip malls are just extremely, extremely expensive. They're just lying on this very valuable land and their people are not allowed to build. I'm hopeful of something like California Forever, where they're trying to get permission to start a new city. I'm hopeful for things of that nature. But none of this, this is all policy. All of this is under our control in some sense. This is not like we haven't been hit by a tornado and we're all poor and have no housing. It's because we've said, no, you cannot build. Yeah.

45:49Alex Tabarrok:I've been thinking about the push into cities, which makes so much sense in the sense that the economic opportunity is in the cities. And so we've seen, you know, successful college graduates leave their hometowns, go to San Francisco, Los Angeles, New York, Miami, Austin. These these mega hubs that have been very expensive if they aren't building. But I was I was really optimistic that between remote work, self-driving cars like faster highways, like we would get just an extension. Because I think if you look through history, the 45 minute commute is sort of a sweet spot before the horse and carriage.

46:27Alex Tabarrok:People would live a couple miles because that's where they could walk in 45 minutes. We got the horse carriage. Once we went to 60 miles an hour, it became, okay, you live 30 miles away from your workplace. But I was hoping that we would get another leg up on that. Maybe it's coming, but we certainly haven't seen it yet. Have you been surprised by any of the fallout or lack thereof of the COVID era, the shift to remote work, any changes, technology or otherwise, in just the housing and labor markets?

46:58Bill Clerico:Yeah, I agree with you. I've been a little bit surprised that we haven't seen another city really take off. I mean, we used to build new cities, right? You know, like, you know, Chicago, you know, not that long ago was a city of 50 ,000 people. Yeah. And Trump briefly talked about freedom cities for a while, which I thought was a good idea, but it's sort of gone away. OK. It does tell you that there's something very strong about these so-called agglomeration, the technical term in economics, agglomeration effects. Yeah. That like people just get more productive when they are near other productive people.

47:44Bill Clerico:And you really need this big push to try and get this into the new city. I mean, Miami briefly, you know, seemed to be jumping ahead, but that trend seemed to have gone away. Las Vegas, you know, tried it with Zappos, but that, you know, modestly, only took off modestly. I am somewhat surprised that we can't plant our flag and say the new city is going to be here and have a lot of companies all agree to move in at once.

48:15Alex Tabarrok:Yeah, I've heard it referred to as like the rainforest theory that why San Francisco is so resilient, because you have so many different participants and the VCs can go to Miami, but they're just a phone call away while so many other key pieces of the economy are still chugging along. And so San Francisco clearly made it through a trough and is on a major, major upswing. Jordy?

48:39Erik Bernhardsson:What data are you most obsessed with following to try to understand the current moment? We see AI in the GDP data primarily through CapEx right now. But what kind of productivity data are you looking at? We've been very excited about what we've seen from Stripe. They have this incorporation product and they're seeing companies, you know, more companies formed, growing revenue faster. That's very exciting. But it's also like a certain type of person finds himself incorporating, you know, their business with with Stripe and is not perfectly a reflection of the economy more broadly. So what are you looking at to try to understand the impact and try to ignore maybe the headlines from CEOs that say, oh, well, we laid off this 20 percent of people because of AI, because, you know, as we know, oftentimes it's marketing.

49:40Bill Clerico:Yeah, it's very interesting. There's a lot of theories, you know, about is this going to be a job apocalypse or something like that. Not much data. And, of course, all of the data we have so far is that AI is increasing the number of jobs, not decreasing the number of jobs. What I'm most excited about and most interested in seeing is the effect of AI on medical care. so you know we just saw yesterday that uh ai had oh you just had it on your on your show uh you know had proved a new mathematical uh a theorem or a counter counter proof uh so ai is making uh these inroads into the highest levels of mathematics if we could do that for drug discovery You know, if we could have, you know, a 5 % reduction in cancer mortality, that would be worth trillions.

50:43Bill Clerico:You know, that would be worth trillions. So the opportunities there for AI to make tremendous leaps in human welfare by improving medical care, health care, I think are really exciting and well within the realm of possibility. You know, one new drug like solving an air dose problem, that would be incredible.

51:11Alex Tabarrok:And what does that mean in the labor market? I'm just I'm thinking back to like there was a time when there was sort of only one track for doctors. He was just like a generic doctor. Now there's much more specialization. There's a dermatologist, a podiatrist, all sorts of different doctors. Is this like there's a there's a fracturing and further employment creation from the administration distribution advisory around new treatments as it rolls out? because you could imagine, okay, there's a new drug. That's great. That helps everyone live longer. But I'm unclear on how it interfaces with the labor market.

51:51Bill Clerico:Yeah, I mean, I think the trend is, you know, increases in the division of labor, which Adam Smith talked about from the pin factory. Yeah. And when you apply that idea of the pin factory, you know, somebody shapes the pin, somebody puts it on, you mess it, you get pieces in your product. When you apply that to the knowledge economy, then it's exactly as you said. You no longer have a physician. You have a podiatrist, and you have an optometrist, and you have an ear, nose, and throat specialist, and so forth. And yeah, I think that will continue, and they will all be using AI for sure, right? But yeah, they're going to get more specialized, and the tasks which physicians do will differ, will change.

52:44Bill Clerico:But so far, you know, I'm not terribly worried about the job market per se. You know, look, this is a problem of people are worried about, oh, the AI is going to do all the jobs, right? Like, this means we're going to be fabulously wealthy. And, you know, even without any jobs, just being fabulously wealthy, we'll figure things out. This is the sort of problem you want to have. Again, this is not like a tornado or hurricane, a tsunami which destroys wealth. This is a tsunami which creates wealth. And yes, it could be a tsunami in the sense that it's going to be very dramatic. OK, but it's going to be very dramatic, like, you know, Santa Claus coming and leaving us goods, you know, under the under the Christmas tree.

53:39Bill Clerico:So that's drama that we can we can handle. It won't be it won't be without problem. OK, but problems where the pie gets bigger are problems that we can solve. You know, it's problems when the pie gets smaller, when we are forced into a zero sum society of one person versus another. That's when society breaks down, not when the pie is getting bigger. Like we'll figure out ways to make sure everyone gets a decent slice. If the pie is getting so much bigger, we can solve the problem of dividing it up with everybody being happy. I'm not I'm much less worried about that.

54:22Erik Bernhardsson:How do you think about value capture with this technology wave versus historical technologies? You know, if you get really good at inventing engines, you can sell a lot of those engines and hopefully have a nice margin. And maybe other people copy the engine and also make similar engines and benefit from that. But we're at a moment right now where frontier intelligence is like very widely available, right? There's certain internal models that aren't with the public yet. And if you have one of those Erdos, you know, moments in medicine, it could just be an off the shelf model that helps make a breakthrough and you pay for the tokens or you pay your subscription.

55:08Erik Bernhardsson:But then right now, the labs are not really set up well to capture that value at all. It could be, you know, some it could be a big company that captures the value. It could be another startup. There's small businesses. And so how are you thinking about value capture versus like public benefit of this technology cycle and how it would differ to prior technologies, let's say like the Internet and telecom and all the way back to railroads and the steam engine, et cetera?

55:36Bill Clerico:Yeah, it's a very interesting technology because this will sound odd, but it doesn't seem that hard.

55:48Bill Clerico:The fundamentals is linear algebra. It's very surprising. I don't think anyone predicted this. And it's true, of course, that the frontier models are ahead. Open AI and Anthropic have the best models. But they're like six months ahead, you know, of open source models. And for most of what you want the models to do, you don't even need the frontier models. And, you know, what today is a frontier model, like tomorrow, you know, it'll be much cheaper second rate model. Right. Like people are worried. Oh, you know, some people have access to, you know, 5.5 and other people are still working with 5.4.

56:40Bill Clerico:Well, the big point is that even 5.4, you know, the free model is 100 times better than 3.0, which was also incredible. So the models seem to be getting more powerful and cheaper at a faster rate than any other technology that I have ever seen. So I think the gains will be fairly widespread, if not at first, then soon afterwards. I mean, some people are obviously open AI and anthropic. the people who got in early and the programmers there, they're going to be fabulously wealthy. No question about that. But the technology itself, so much of it is open source. So much of it is really quite accessible.

57:41Bill Clerico:There's no magic there, as far as I can see. It was just you had a few good ideas, and then those ideas just turned out to be incredibly powerful. So I'm expecting to see really the technology being quite widespread. Yeah.

57:58Erik Bernhardsson:I know we're out of time, but one last question. How do you think about part of the reason I think we feel this insane acceleration with this technology shift, maybe more so than certainly mobile, but looking back to the Internet and everyone in this moment wants to figure out, okay, are we in 98? Are we in 99? Are we in early 2000? Trying to figure out the moment that we're in. And it feels like because we have the internet today, there's this compression in the technology cycle because ideas get distributed faster, products get distributed faster. If there's a breakthrough, it's instantly everywhere.

58:40Erik Bernhardsson:There's instantly hundreds or thousands of companies working on, you know, improving it, furthering it. And so, um, I find it hard to try to think about where, where we are because it felt like in Q4 of last year, we actually did have a correction. Like we were joking, uh, we were joking like, great, the bubble popped because like there was over like maybe too much excitement around chat bots and there was somewhat of a correction. And then we got agents that really worked. And then it feels like in some ways we're We're in a new cycle now. And so I'm curious if you have any sort of frameworks or thinking around that, like compression in progress that we're getting because we're building on top of all these other technology cycles for this new one.

59:26Bill Clerico:Absolutely. And of course, the AIs themselves are starting to improve the AIs. Right. So some people worry that precisely because of that, we'll get a sort of a foom scenario under which one day everything is fine. And the next day you have a God in the laboratory. Right. I'm not too worried about that, but it's not it's not insane. It's not insane. Yeah. So far, my view is that I trust the technologists when they say that the technology is going to keep getting better quite rapidly. Where I think the technologists are not quite right is that it's going to take much longer than they think for this to start affecting jobs and the economy writ large and things of that nature.

1:00:34Bill Clerico:That'll be slower. So the economists generally are on the slow side in terms of it's going to take time to adapt to this technology. I mean, we saw like with electricity, for example, you know, electricity was another, you know, incredible technology. But it took time. You know, it took time to adapt to that, even when the frontier, you know, was very far ahead. But for that to work its way in the economy and for people to figure out how they're going to change their production structures, that took time. And I think it will take time here as well. But as far as I can see, the technology is going to keep getting better, which it does give one pause.

1:01:24Bill Clerico:And I will say never in my life have I felt that the window of what is possible is as large as it is today, both on the possibilities for superintelligence, huge gains, the boomers, and also on the doomer side. I don't discount those entirely. My view is more in the middle. But those two sides, they're not insane. so we do have to do a lot of thinking.

1:01:56Alex Tabarrok:Yeah. Well, thank you so much for taking the time to come chat with us. We'd love to have you back on the show soon. It was long overdue. But yeah, thanks for the time, Alex. Rest of your week. Have a great weekend and hopefully we'll talk to you soon. Talk soon. All right. GMU economists, always so much fun to talk to. Well, our next guest is Bill from Convective Capital here with a new fund. Get that gong ready. Let's bring Will from Convective Capital into the TVP. Welcome to the show. How are you doing? Doing great. Thanks for having me, guys. Kick us off with the news. What happened today?

1:02:35Jordan Schneider:Yeah, so we raised a new fund,$85 million. There it is. Thank you so much. It's up 2x from our last fund, and we're focused on disaster resilience. The thesis of the fund is that the world's getting warmer, our infrastructure's getting older, and that's literally a recipe for disaster. And so as disasters rise and volatility rises in the world, there's got to be private markets that can respond and that can build solutions and services and technology to stop that. And so we back founders that are building those things.

1:03:09Erik Bernhardsson:Give us an overview of the FIRE. Give us a FIRE market map. What's exciting in FIRE? It's top of mind right now. There's a fire burning in LA slash, I think, Ventura County. And so my house has been very smoky around my house. And yeah, I want to know. This feels like a particularly brutal industry because there's these insane spikes of interest and then people just kind of forget about fire.

1:03:38Alex Tabarrok:Are you thinking watch duty for-profit conversion?

1:03:40Erik Bernhardsson:Yes. It's got to happen. Yes. Lead a series A in watch duty, please.

1:03:45Jordan Schneider:Lever it up. Yeah, John, I watch it. He's a good friend. And we started, you know, I started my company around the same time he started that. And he will never take that for profit. He's a dyed in the wool. It's such a good app. You know, mission first. And it's actually become the standard, not just for people that care about their homes, but also, you know, people in the fire service use it. He's an amazing example of how technology can address this problem. So I guess to get to get to your question around market map, you know, we think about a lot of like, who is the end buyer? Because that's actually what's kept people out of this space.

1:04:15Jordan Schneider:Historically, you know, the, you know, VCs typically don't get excited about companies that sell to utilities or insurance companies or government agencies. And, you know, our thesis is that that is in the middle of a really big change that, you know, if you see like PG &E went bankrupt a couple years ago, the insurance companies have had to leave California in some of these really large markets. And that's changing behavior. You know, you lose$70 billion of market capitalization and you respond and you do something about it, you act differently. And so, you know, we think a lot about the market in terms of who's going to buy these technologies.

1:04:47Jordan Schneider:And we've had a lot of success investing in startups that sell to utilities like Overstory, which, you know, uses satellite imagery to help utilities trim trees around power lines. We just invested in Fund2 in a company called Voltair, which does autonomous drone power line inspections. so you know utilities cause about 11 percent of fire ignitions but about 60 percent of the acres burned and so if you can actually just help reduce utility ignitions that's like a huge leverage

1:05:13Alex Tabarrok:point interesting interesting interesting what what what's going on on uh disaster prevention in like more of the consumer prosumer space i uh i saw a house for sale over in malibu and the it was uh by some actor and he had installed like fire shutters and like all this different equipment. He was like living off grid and this was like his getaway. And it feels like I saw after the LA fires, like the autonomous water sprinkler that would be bolted to the top of the house. I think a lot of homeowners in Los Angeles at least were like, I want one of those. And they would have clicked the button to buy it if they'd seen an ad that day.

1:05:50Alex Tabarrok:But then a year goes by and nothing happens in the next fire season. And they think, oh, maybe not. But it does feel like there's some fertile ground in the consumer space. But is that more challenging than people might think it is in reality?

1:06:03Jordan Schneider:I think you're totally right. Consumer demand goes in spite. It's a very seasonal, volatile business. I just saw today Watch Duty is the number two downloaded app on the App Store today. That probably was not the case a couple of weeks ago. So, it's certainly, that's the nature of the beast here. In terms of home hardening and things you can do to protect your home, I think that also kind of goes in waves to your point. But to me, the real unlock is going to be when insurance companies create incentives for people to actually install these things around their home. So we're investors in a company called Stand.

1:06:32Jordan Schneider:They help model homes. They use computational fluid dynamics to simulate wildfires moving through the property. They come up with a list of recommendations for the homeowner of what they should do with shutters and windows and remove certain vegetation. They remodel it and then they can actually provide discounts on insurance. I think that's going to be the real unlock that drives, you know, at scale consumer behavior.

1:06:53Erik Bernhardsson:Yeah, it has to be a real time like you're renewing your home insurance and they're like, if you do spend, you know, in California, it could be like, spend$10 ,000 on this new system. And if you do that, we'll give you a$10 ,000 reduction this year and, you know, further discounts in the future so that you end up, you know, effectively, you know, saving real money.

1:07:15Jordan Schneider:Yeah, exactly. I think historically insurance companies have not done that, but that's the, that's the key to protecting, you know, homes in this kind of new era. And it's got to happen. The California Fair Plan, which is like the state-backed insurer of last resort, just announced that they're going to raise rates 30 % this year after the LA fires. And so if we don't actually reduce the probability of homes burning, housing and insurance is just going to become unaffordable.

1:07:41Alex Tabarrok:How are you thinking about selling to the government, maybe in California or elsewhere? Anderle has this interesting story where they built a firefighting tank. They were trying to sell it to California firefighters, but there was pushback around job displacement, even though it was ideally a new capability that would actually have support staff and not really take anyone's job because no firefighter can sit in the middle of a blaze like this particular firefighting tank could, but it was still became a political issue and ultimately did not become a real product. Is there movement there? How are companies positioning themselves as additive?

1:08:17Alex Tabarrok:When I think about drone review of power lines, there's probably someone that went up there earlier. Cost savings is good. Doing more with less is great. But also, there's always that pushback around job displacement.

1:08:30Jordan Schneider:Yeah, I think there's kind of two issues at play here. First is like, is there actually job displacement? I think the reality is no. I mean, it's just, we are so under-resourced relative to the scale of these disasters. They're happening three times more frequently with huge severity. Cal Fire is the largest and best-resourced firefighting agency in the world, bar none, like that's not going to change. And I think it's really about how do we get leverage out of those investments. I do think, though, you're pointing out a cultural issue, though. And that's something that we've really worked at trying to bridge.

1:09:00Jordan Schneider:You know, I think it can be really harmful if like a bunch of guys in Palo Alto sipping lattes, like walk out to the fire line and try to tell people how to do their jobs. You know, there's just this huge disconnect. And so one of the things that we built is this conference called the Red Sky Summit, where we actually get 600 fire chiefs and other emergency managers together every year in San Francisco. And we kind of create like an off the record venue for them to talk to people that are building stuff and technology. And it kind of creates this great sharing back and forth where, you know, you can show the value of this technology.

1:09:29Jordan Schneider:You can have these two way conversations. It sort of changes the tenor. And that's been a real unlock. We've seen a lot of buying behavior come out of that event. I think things like watch duty that show how impactful technology can be, the firefighters see that. And, you know, I think the tide and the cultural tide is really changing. And so it's an exciting time to be building in this space. Yeah.

1:09:49Alex Tabarrok:Is there any relevance? Like we talked to some other sector specific funds or thematic funds. And the classic example is like in CPG, they can be a harder business sometimes. But there's a number of like clear acquirers for mid-scale companies like a Unilever or Coca-Cola will take out a lot of these companies. at a unicorn valuation. And so it sort of changes the underwriting. It's worked for a lot of funds in that category. Is there a 800 pound gorilla in this category that's maybe not being disrupted, but maybe can be a partner at some point in time? Or is the thinking like everything is IPO, it needs to be a standalone business or bust, or is anything different financially about these businesses that you see?

1:10:37Jordan Schneider:Yeah, that's a good question. I don't look at wildfire as like a market in and of itself. It's this sort of dynamic that touches these huge markets like energy, insurance, housing, real estate, forestry, government emergency response. And in each one of those categories, there are really large companies. There are contractors in the utility space or providers in all of those markets. So I think there's certainly exit paths. But the economics here are just like immense. You know, Bloomberg just published a report, disasters cost the U.S. economy a trillion dollars a year. That's like on par with what we pay for interest on the national debt or defense.

1:11:16Jordan Schneider:And so it's like, you know, there are really big public companies that will be built, you know, solving that because that cost falls on these really large deep pocketed institutions. And, you know, we think that there's just very big businesses to be built.

1:11:29Erik Bernhardsson:yeah uh is there opportunities for wearables uh for firefighters i was uh once the the the sandy fire started here in la i was uh i was showing uh my son some videos on like wild uh just like wildland firefighters and i was shocked that a lot of them just like weren't wearing gear they were wearing don't wear something like a piece of fabric over their face seems like there's probably opportunities there and potentially a decent size market. Yeah.

1:12:02Jordan Schneider:I mean, it's, it's a travesty what we equip our firefighters with and send them out there. I mean, it's like, you know, the, the health impacts of wild and fire smoke are just terrible. I mean, and so, and there's really not a practical way currently to help filter that air for these wild and firefighters. So they're out there, you know, really with no masks, sleeping in smoke days at a time, carrying packs, it's 110, 120 degrees, you know, they're working hard with chainsaws and axes. I mean, it is, it's a really dangerous, really grueling job. And I think, you know, part of the original thesis for convective was I was actually volunteering with my local fire department up in Mendocino County.

1:12:38Jordan Schneider:And, you know, I just was watching this and I'm like, I can't believe we're doing this with, you know, trucks from the 1970s and, you know, axes and like, that's the state of the art. And so there's a huge opportunity there. And I think, you know, it's the, and you can look at, you know, wild and fire as a market, and it's a certain size, but it's also a path into all types of field service jobs and all types of worker safety. And so we've looked at a lot of companies there. We haven't made any investments yet, but that's a category we're really interested in.

1:13:05Alex Tabarrok:We jump straight into the discussion of the fund, the strategy. Can you take us back a little bit to what you were doing before? Why start this fund? When you started it? Sort of the prehistory of the fund.

1:13:18Jordan Schneider:Yeah, so I was a founder. I started a company called WePay back in 2008. It was an early fintech company, and we built that up over about 12 years and sold it to J.P. Morgan.

1:13:28Alex Tabarrok:Where was that headquartered? Overnight success. Yeah, I wish.

1:13:35Jordan Schneider:It was headquartered in Palo Alto.

1:13:38Alex Tabarrok:Palo Alto, okay. Yeah, I just remember running into it in Boston, I think in like 2012 or something. That was around the time. I actually started it in Boston. Okay, that's right. Yeah, we started it in Boston.

1:13:49Jordan Schneider:That's right, that's right.

1:13:50Alex Tabarrok:Yeah, okay, got it.

1:13:51Jordan Schneider:I started in Boston. We were funded by Y Combinator, moved to the West Coast, built that up for about 12 years. And then we sold to J.P. Morgan. I left J.P. Morgan. And I think one of the things that really suited us well at WePay was we were early to like the financial technology market. People called it like banking technology at the time. And a lot of the same skepticism that I hear around utilities and government. We heard, you know, we're trying to raise money from Boston VCs in 2008. And, you know, they were wrong. I mean, the fintech market was like the most exciting sector of technology over the next 10 years.

1:14:23Jordan Schneider:And, you know, so as I was leaving JP Morgan, I wanted to work on something that was, you know, a market that was early and that we could be sort of one of a kind, that we could be a market leader in. And that also had an important mission where we were helping people and doing good for the world. And this sort of lined up all of those things. My wife and I own a ranch up in Mendocino County, a couple hours north of San Francisco, and a fire almost burned under the property. and all the dots started to connect and here I am.

1:14:50Alex Tabarrok:Yeah, that's amazing. It's sort of like the VC version of build something you use yourself, like the YC ethos, make something people want. Well, thank you so much for coming on the show and breaking it down for us. Congratulations. Yeah, I'm really glad you're doing this.

1:15:03Erik Bernhardsson:Yeah. Thank you. We need more. Thanks, guys.

1:15:05Alex Tabarrok:Love what you're doing. Have a great rest of your day. Great to meet you. We'll talk to you soon. Cheers, Bill. Goodbye. Up next, we have the co-CEO of Spotify joining Alex Nordstrom. On a huge day. Calling in from Spotify Investor Day. Welcome to the show. How are you doing? What's going on?

1:15:21Alex Norström:Thank you. Hey, John and Jordy. It's good to be here.

1:15:23Alex Tabarrok:Great to have you. Fantastic. First time on the show, why don't you kick us off with the news today? What's new in Spotify's world?

1:15:34Alex Norström:We just had a fantastic Investor Day 2026. We talked about what we've delivered in the past four years, which put us in a really great position in terms of user growth and financial metrics. We also laid out our four big ideas for the future. And then we also talked a little bit about how we can further monetize Spotify and grow all the sort of financial metrics and goals of the company. I want to send my greetings from Gustav Sutterstrom, who was also on your show a while ago. Yeah, we talked to him

1:16:05Alex Tabarrok:in South by Southwest. It was a great conversation. We're very excited to have you here. Um, where should we start? I'm, I'm, I'm, I want to jump into big ideas.

1:16:15Erik Bernhardsson:Yeah. I mean, there's so many different things.

1:16:17Alex Tabarrok:Uh, maybe, uh, we should start with the reserve ticket access because I saw that organically promoted to me and it seems so logical. Uh, and I'm fascinated by the rollout, the strategy, and also, um, like sort of, why did it take so long? Has this been in the chamber for a long time. What has been the plan? What has been the go-to-market with this particular product?

1:16:40Alex Norström:Great question. So I've been with Spotify for, I think it's now almost 16 years. And this is by far one of the most, one of the most lovely improvements to Spotify Premium that we've done, I guess since the founding really. So you're right. You know, it's long overdue. It's a great one. And the reason why it's so great, it's really that it solves a couple of different problems. One is, you know, who hasn't been sitting there in front of a website trying to get tickets for a concert? It's very hard. And if you get it, it's going to cost you a lot, right? And the second thing is, even if you get tickets, it may not be to the artist that you actually listen to and love.

1:17:17Alex Norström:So this solves that. So what we're launching is basically we're giving you, we're holding tickets for you for a certain time window for you to just go and pay and collect with our partners. So it really solves the problem in a unique way because we're matching the tickets and the concerts and the participating artists and tours with the users on our platform that are actually the true fans of this. And we look at it from many standpoints, like not just the number of streams that you're doing, so you can't just grind yourself there. It's also catalog engagement and so on. So it's terrific. This is a fantastic improvement in Spotify Premium.

1:17:53Erik Bernhardsson:Yeah, I mean, I have a new problem, which will be ticket scalpers setting up infinite Spotify accounts. Paying the monthly subscription fee for years. Just to try to run an algorithm. No, but I think it's absolutely brilliant. Solves a problem, you know, even from the artists. Like, being in this position where you're trying to make your ticket prices affordable for your fans, but due to market dynamics, no matter where you price it, they'll go way up. and then the only beneficiary in that transaction is the person that's just buying the tickets to resell and everyone else loses. You're totally right.

1:18:30Alex Norström:I mean, like many things that we do at Spotify, we look at what organically happens on the platform and what we do. And then when we see that there is some sort of signal or trend that actually just come by way of organic behavior on the platform, then we double down on it. So we've done many events, you know, with artists and groups and bands and so on over the years. And, you know, on a yearly basis, probably 150 to 200 events. And the magic really happens when we look at who to invite to those events. And if we can really strike the balance of, you know, getting enough people there that are really sort of on a very, very high engagement level with that artist, Again, not only just listening to the artist, but doing it maybe on a daily basis, having listened to the whole catalog and do it for at least an hour per day or something like that.

1:19:21Alex Norström:It really gives the artist also a magical experience of facing your truest fans. And this is something that uniquely we can do for artists.

1:19:29Alex Tabarrok:Yeah. AI is obviously at the top of everyone's mind. The large taste model is the latest technology from Spotify in the AI world. but I feel like you have to have been embedding every song into some sort of machine learning model for probably over a decade. Can you take me through a little bit of the history of machine learning or AI at Spotify? And then what has actually changed? Like, where are you seeing a break in the graph? Like, where are you seeing, okay, we're experiencing some sort of discontinuity in the progress we were making? Yeah.

1:20:07Alex Norström:So you're right. It was actually just about a decade ago when Daniel and Gustav and I started investing into AI then, it was called machine learning. And it was very basic, the algo's then. It was almost like, okay, similar to what you do on Amazon, John listened to this. He's similar to Jordi. He also listened to this other thing. Therefore, he should listen to that too. So lookalike algorithm, if you will. Collaborative filtering, probably based on text,

1:20:35Alex Tabarrok:not actually looking at the waveform even.

1:20:37Alex Norström:Exactly like that. Yeah. And so now many, many years later, having continued to invest in that and just the data, not just the data, but the learnings we have and the feedback loops we've had over the years, it's been compounding. So, and now, you know, you're asking me what changed lately. Well, obviously what changed is that, you know, we've got access to, you know, large language models, like general intelligence that we can use to reason over the data that we have. But what it comes down to really is the unique data at scale here, right? So, you know, we logged probably, you know, between, I think it's three and four trillion events that happen per day on Spotify.

1:21:12Alex Norström:So we really get billions of signals that are relevant to feed it, to feed the algorithm with. And then the second thing that happened is, it's obviously because we started to sort of, you know, understand English. Algorithms started to understand English. I mean, we sort of started importing a lot of knowledge from outside, from the outside world onto Spotify.

1:21:30Alex Tabarrok:Yeah.

1:21:31Alex Norström:And also the third thing that's important is...

1:21:34Alex Tabarrok:has gotten so much better at detecting lyrics. Like I used to have to go in Google or search a lyric and then go to Spotify, and now I can just type it right in and I get it. I'm like, this is an amazing experience. I'm interested in your build versus buy versus fine tune and open source model. What's working? What's interesting? Music is, you know, obviously you're benefiting from transformer architectures from all the open source work, But I imagine that a lot of the tasks that you're doing aren't exactly okay, just fine-tune like a text-based open source model. You might need to go a layer deeper.

1:22:13Alex Tabarrok:You might need to partner with someone. How have you been thinking about the way to deploy AI?

1:22:18Alex Norström:Yeah, we certainly believe in general intelligence. And, you know, we've been riding that cost curve that's going to come down, collapse by 1 ,000x over the last four or five years or so. So we believe in this industry commoditizing more and more. So we are going to buy the best reasoning capabilities that are out there. And then we're going to use that on top of what we call a large taste model, which we are uniquely able to using proprietary data to actually build out. Much thanks to the stuff that I was talking to. The other side of the equation here that is important, I think, is also that we get a lot of feedback and data from artists as well via our Spotify for Artists application, which feeds us with data that's also proprietary and unique, which is great.

1:22:58Alex Norström:So you ask about build and buy and so on. I think it's a great question. You know, if we see opportunities to actually build something to help, you know, bootstrap or add a capability, we will do it. And I'll draw two examples that are quite good. I think three years ago we bought Synantic, which really helped us create AIDJ.

1:23:18Alex Tabarrok:Okay.

1:23:19Alex Norström:DJX. Yeah, it really accelerated the build out of AIDJ. And now AIDJ is all over the world and we have it in different languages and people are using it at scale. So that was a great, I think, transaction that we made. Another one that is also, it's more recent, actually. It was just before the end of last year. We bought Who Sampled, which provided us not with so much capability, but a unique set of data. So I don't know if you've used SongDNA lately. It's been killing it. I think 50 to 60 million people are using it already. You can listen to a song and then you can check which were the samples in that song.

1:23:55Alex Tabarrok:Oh, interesting.

1:23:55Alex Norström:And then you can also follow the threads down. Go back through the lineage. What's the original? Yeah, you can listen to the original track and then hear how they sort of flip that into the new one. And you can see who's in there, the studio technicians, the artists, the composers, and so on. And a lot of this we couldn't have done or it would have taken us much longer time without two samples.

1:24:15Alex Tabarrok:Yeah. Structurally, how is AI changing the work that Spotify employees are doing? Matthew Prince from Cloudflare was talking about this concept of you have builders and sellers, but just the raw measurement, that's less important as a discipline. And so he's pushing more of his staff to build new things and then go and sell it. Do you have a framework for thinking about how your employees should be using AI in the future?

1:24:44Alex Norström:Yeah. I mean, we, you know, a couple of years ago, I think we all you went through that the chat GPT moment and just before the holidays of last year we had a similar moment around coding tools and we all sort of got enamored with it and started adopting it but at Spotify the engineering and development teams have just you know done a crazy good job adopting it so we now have I think close to 99 % adoption across Spotify and this goes you know beyond the engineering team but also into the marketing team that are using it not just for design but also for better gaining insight into what type of compelling stories to tell on top of the story.

1:25:24Alex Norström:So we do have very many different use cases. And, you know, people are crazy about just creating prototypes now all around Spotify, just test things in a way faster way. So what you get basically is a lot of like productivity gains. And I know for a fact, one of my dear colleagues, the chief architect of Spotify NGN, he was just at Anthropic and he was doing a presentation for their staff. And I hear from Anthropic that we are actually one of the leading developers out there in adopting AI and using it to gain productivity improvements.

1:25:59Alex Tabarrok:Yeah. Have you thought about the, I don't know if there is a trade-off, you might be doing both, but there's a world where you have some sort of engineering pod and there's an opportunity to have more members of that team pushing code, shifting to individual contributor roles. but of course they're managing agents, so it's also a managerial role, versus embedding more technically-minded folks across different organizations. So you have more cross-functional teams. A marketing team gets an engineer who can oversee agents. How are you thinking about just spreading different AI tool use across different organizations that might not previously have been in a position to build a piece of software?

1:26:43Alex Tabarrok:Maybe they're in the market to buy a couple of SaaS products for whatever their vertical is, but now they have the opportunity to actually build. Yeah.

1:26:51Alex Norström:Well, you're seeing actually happening organically at Spotify is that as adoption goes up, it's typically you're by your own and you're looking at what's out there, which MCPs to connect to and how that can help your work. And then you're basically sort of copying some other people's skills around at the company and you're just sort of experiencing how it is to use AI to do your work, your particular field. but what now is more happening is that we all of our slack groups and in our communities we see people sharing sharing you know use cases with each other and the next step really is that for instance someone a product developer and an engineer you know sits down with a marketer starts to understand hey okay we the three of us can actually build something of a you know a holistic experience of not just new product and feature but also tell a compelling story about it you know in a much much more sort of amplified and leveraged way which i think is

1:27:46Erik Bernhardsson:fantastic when that happens uh we got to talk about fan-made covers and remixes the new deal with umg yeah super super excited about this personally and uh of course it's been you got a

1:27:58Alex Tabarrok:song you want to cover you're gonna do that you're finally gonna do the gunna cover you were planning

1:28:02Erik Bernhardsson:on what's your favorite track right now is it a great track i've been uh no i think i i think uh I was expecting Drake to kind of grow up by this point in his career and maybe evolve a bit more. I feel like I grew up maybe faster than Drake. He's still obviously an incredible, incredible superstar. But, yeah, John was talking about Gunna. I'm a big Gunna fan. Oh, yeah. That's great. So I'll be making some Gunna covers and remixes. But, yeah, talk about this deal. It's certainly controversial, but this feels like one of those things that people will have one reaction and then they'll have the experience, which I'm assuming will be incredibly magical.

1:28:49Erik Bernhardsson:And we've had Mikey from Suno on the show. He's a friend of ours. People fundamentally, they say one thing on the internet, but when they use the products, they love them. And so I think it's super exciting. So, yeah, talk about this deal. what were the factors that were most important in getting this right so that fans win, artists win, and everyone involved wins?

1:29:13Alex Norström:Yeah, yeah. I'm certainly very, very pleased with it. I mean, it's a landmark deal, and it's a landmark deal from the vantage point of we're enabling something that is for the first time a legal product for users and fans to actually create these remixes and covers. And what's more is that it's really the first time in a controlled and licensed medium that artists get to partake in the AI economy, really. So that is like the core of it. And then you think a little bit about what it is that we're doing here. Well, you know, people love, to your point, like people love remixes and covers and they've always done that, but there hasn't been any scaled way to really sort of tap into this opportunity.

1:29:54Alex Norström:And what we're doing right now with our product is that we're unlocking this market. So where we think actually in the end, And it's going to be very additive to both the music industry and ourselves, including artists and songwriters. So it's very cool. We're launching it as a paid add-on, which means that if you have Spotify Premium, you're eligible to actually buy into this add-on. And when you have the add-on, you can actually go ahead and create remixes and covers. And what's cool with Spotify is that we're very experienced when it comes to free and premium and how to convert people and lead people up the ladder.

1:30:29Alex Norström:and so what we're going to do is always to give you a certain limited usage on premium so you can actually try it out, play with it create habits around it and then commit to buying add-on but the important point here is that you can think about it as creation is paid for and consumption is included so basically the output that you create around the Gana the Gana remix that you're going to create I'll be able to listen to it, everyone will be able to listen to it

1:30:58Erik Bernhardsson:Yeah, is there, so I can imagine a world, you know, maybe six months from now where the number one track globally on Spotify is a fan-made remix. Like, you know, it may take more time, but it doesn't feel impossible. Do you think this could potentially create a new category of actual artists on Spotify? would the person that creates a remix ever be able to get, you know, some elements of, of rev share? Or is this like entirely going to be more like, you know, more like a live DJ where, you know, you're putting, you're mixing and mashing and putting things together and, and maybe you can build up your brand, but you're not monetizing, you're not getting royalties or any, any revenue share.

1:31:48Yeah.

1:31:49Alex Norström:I can't speak to the specifics of it, but I can tell you how I think about it. You know, when I started at Spotify together with Gustav 16 years ago, we had roughly, I think it was 2 million tracks in the catalog. And I've lost count, but I think it's in the order of 200 million tracks now, you know, 15, 16 years later. So the catalog being big and growing is a good thing. And the problem you're talking about, a real opportunity is for us to do recommendations well. And that's always been at the heart of Spotify. If you ask someone, why do you love Spotify. Well, most people say, Hey, it's because they seem to know me.

1:32:25Alex Norström:So we're able to sort of help you discover by way of personalization, uh, help you kind of find your tracks again, that you like and so on. Uh, so really this is what you're talking to is, is really a personalization problem, right? So it, when, when this product creates more catalog, our job is to make sure that, you know, the best song gets to the best place.

1:32:44Alex Tabarrok:Last question, uh, talk about the app icon redesign. I thought it was really cool. I loved it. But I'm interested in the actual, I mean, I want to know how you process it internally.

1:32:58Erik Bernhardsson:Did you ever imagine that a disco ball could get people? People were really talking about it.

1:33:01Alex Tabarrok:But my assumption was that, you know, it was the Spotify icon is on my home screen and I click it when I want to listen to music, but it jumped out to me. And I feel like that would show up in the metrics of jarring people awake, but I'm interested to know, like, what actually happened? Are you going to be changing the icon randomly every five years? Is it got to be every 20 years? Like, what's the, what is the learning from doing something sort of bold? Right.

1:33:31Alex Norström:So about 10 days ago, we decided to change our, the Spotify app icon or Spotify logo to one that is more of like a glittery disco ball version of the Spotify logo. And, you know, to your point, this just sparked a massive conversation around the internet on basically every major social platform. And, you know, what a lovely and wonderful piece of culture to have a conversation around the change of our logo icon. Right. So not only users, you know, have been sounding off, but also other big brands. Yeah. Chopping in. I made their version. suggested that they would make a version like that. Kit Kat join in on the phone.

1:34:15Alex Norström:So yeah. So when you have, you know, hundreds of millions of people that passionate about culture and art and, and, and Spotify, this is what happens, right? So this sparks like a massive conversation that I think is, is, is lively and fun. And, and to be honest with you, the reason why this is happening, you know, having reflected on it a few days later now, it's pretty intense when it happens because there are two sides, right? And then there's also all of this like byproduct, you know, the internet coined a new design term around this called discomorphism. Discomorphism. Oh, yeah. I love it.

1:34:49Alex Tabarrok:And people have been asking for an answer to flat design. Lots of people have been lamenting how boring things have gotten. You spice it up and then people, I didn't mean that, which is a very funny way to process it. But sorry, continue.

1:35:01Alex Norström:I mean, I just did the reason why this is happening after having sort of thought about it for a few days and discussed it internally. It really is because we are truly at the intersection of the humanities and technology. And I think we're in a good place when it comes to that. We're at scale. You know, when hundreds of millions of people, potentially even billions, are talking about you, then you did something interesting.

1:35:23Alex Tabarrok:I love it. I love it. Yeah, a really, really well-executed stunt. And, yeah, just got everyone talking and reminding. And it's hard to break through with something like a 20-year anniversary with a milestone. like but it's important to do something like that so congratulations thank you we did it mostly for

1:35:40Alex Norström:ourselves in the beginning but now it turns out that you know music should be fun building a

1:35:45Alex Tabarrok:company around music should be fun like all of this like it's i i know people have their think pieces and their deep dives on uh contrast ratios and all sorts of things but like at the end of the day it's fun for a company like spotify to do something fun with an app icon for a few days

1:35:59Erik Bernhardsson:i love it also i put back on those people i thought i thought it looked exactly like a disco ball would look if you were in a dark room. That's right. Listening to music and you look up.

1:36:08Alex Tabarrok:Yeah. So I thought it was pure to the disco ball brand. I agree. I agree. That's good.

1:36:14Erik Bernhardsson:Thank you. Keep it on that.

1:36:17Alex Tabarrok:Anyway, thank you so much for coming on the show. Great to hang, Alex. Great to hang. Thank you. Congratulations on the progress.

1:36:21Erik Bernhardsson:Yeah, congrats to the team on all the new milestones.

1:36:23Alex Tabarrok:We will talk to you soon. Have a great rest of your day. Thank you so much. See you soon. Bye. And if you haven't been following, the stock is up 13 % today. Spotify is a$103 billion company. Not too shabby. Our next guest, Jordan Schneider from China Talk. You know him. You love him. He's back on the show for the third, fourth, fifth, sixth time. Who knows how many times it's been. We lost count, but it's been too long, and we're happy to have you here. How are you doing?

1:36:49Jordan Schneider:I'm good. I got to be honest, was also pretty confused by the Spotify. You were confused by it. I thought, well, yeah.

1:36:56Erik Bernhardsson:Why a big$100 billion company can't have fun in this country? Yeah. What's wrong?

1:37:01Jordan Schneider:I'm just saying it took me longer to click it. I thought I had like some weird iOS update, you know, Tahoe, whatever. Oh, yeah.

1:37:09Alex Tabarrok:Sure, sure, sure. I just think that like the fact that you processed it, even if it slowed you down, it put you in the fast thinking, slow thinking. You were in slow thinking mode. You're not addicted to your phone enough.

1:37:19Erik Bernhardsson:You're not addicted to your phone enough. Are they still rocking the – are they still rocking it?

1:37:23Alex Tabarrok:It's still a disco ball. I like it. I think they should just keep this thing. It stands out so much from every other app on my phone right now. I love it. anyway how are you doing give us maybe let's start with the post-mortem on last week how is the china summit and then we can just go into everything else so boring man right i knew i knew you were going to say that i was like we're going to talk about everything but china because it was

1:37:46Jordan Schneider:so boring but they let me down i mean look i think there is a clear um period of stalemate which we've gotten into where the U S has some leverage over China, which has been something that really since like 2017, Trump won Obama, Trump won Biden. And then the start of Trump truth thought that they could kind of keep turning the temperature up and squeezing on tariffs and export controls and investment restrictions. And then all of a sudden China, April, 2025 punches back with rare earths. And all of a sudden, they realize they run the experiment and it works that actually the PRC has real leverage over the US and is able to kind of meaningfully constrain what American presidents want to do when it comes to more coercive actions on China's technological and economic expansion.

1:38:48Jordan Schneider:So both sides have kind of realized that and what we got is a you know uh what was my headline prestige on the cheap um where uh the u.s decide where um you know you have the whole cabinet uh flying over you have every fancy ceo flying flying over just being awed and wowed by um you know their 36 hour vacation to beijing yeah like tourism yeah and look there's an interesting debate to be had about whether the like giving them the face and giving them the atmospherics of that is actually like kind of in some level costless and maybe this like makes them feel less aggrieved less likely to start world's war three i mean this is the argument with putin right it's just like we kept disrespecting him and then he decided no, like I'm going to show you just how tough I am.

1:39:45Jordan Schneider:So I'm not kind of on face opposed to it, but you know, it was, it was kind of just more or nothing.

1:39:52Alex Tabarrok:Yeah. I mean, you mentioned every fancy American CEO was there, but were you surprised or were you expecting, uh, or did it break your expectations that, uh, no Sam, no Dario, no Demis, no Sundar, uh, a sort of lack of AI representation. Yes, you have Elon, and yes, you have Jensen. Very important figures in the AI, American AI build-out, but no true lab leaders from my rough estimate, if I define it a certain way. And it felt like, okay, maybe this discussion should be about AI, and they're not even bringing the people to let them get a word in edgewise. Not even a single line of conversation will happen.

1:40:34Alex Tabarrok:Was that as expected or how do you process that? Yeah, I mean, it's interesting because like this is phrased, this is framed as like a trade delegation.

1:40:46Jordan Schneider:Sure. Right. And so almost all of those companies like Meta excluded. I mean, I guess they had a person there, but Zuck wasn't there.

1:40:53Alex Tabarrok:Yeah.

1:40:54Jordan Schneider:Yeah. But, but that person is like a Trump one person. It's kind of just like keeping with the vibes. And I guess Meta builds glasses, builds glasses in China.

1:41:02Alex Tabarrok:um yeah and they and they acquire ai agent companies manage they they're a buyer of chinese companies or formerly chinese companies that have moved to singapore which they're now not allowed

1:41:13Jordan Schneider:to do so yeah they had some lobbying to do i guess but like the you know the asks i think are really inchoate um look the u.s government we just had news today that trump um 86ed a uh executive order because he doesn't know what he wants to do about this sort of thing about AI safety. And if like you can't even figure out what you want from a domestic regulatory perspective, like are you really ready to like put all the cards on the table and have this big, you know, grand summit with the Chinese to make sure that bioweapons don't kill us all in 2029? I don't think so. I mean, it'll happen at some point, but it's just a little premature.

1:41:56Jordan Schneider:I think maybe not from the technological perspective, but from a sort of policy perspective that you do expect much of a serious discussion.

1:42:06Alex Tabarrok:Yeah. Are you Sebastian Maliby had an op ed where he was sort of gesturing towards like the possibility of more collaboration on AI between America and China? You're you have your hands in your face or your face in your hands. So I assume you are skeptical that that will take place anytime soon. But I mean, also the AI 2027 folks, very good at forecasting, have predicted a lot of what's happened. Say China wakes up in 2026.

1:42:35Jordan Schneider:Yeah, I mean, it's not impossible. I think the sort of look, if the American political system has not yet woken up to an understanding that you need a different regulatory strategy in order to keep humanity safe. I think it is unrealistic to expect a strategic adversary who is, you know, 6, 12, 18 months behind where you are on the sort of technology development trajectory to come to the same realization. Now, you know, it'll happen at some point. I think this stuff is really powerful. And the the sort of logic that America doesn't want, you know, bioweapons to be developed by Al Qaeda or ISIS or Hezbollah, just in the same way that China wouldn't want it to be done by, you know, the Falun Gong or Tibetan separatists or, you know, any any.

1:43:40Jordan Schneider:group that they see as like a non-state actor that would want to kind of take down the party applies, right? But China, I think, is deeply, the Chinese leadership is deeply hardware-pilled. I think they're deeply skeptical of these sorts of arguments on face because they're kind of sci-fi. And the sort of journey, the analogy or the journey that the U.S. and Soviet Union went on over the course of the Cold War to start to have real kind of discussions about strategic stability and arms control happened after you had the Berlin crisis and after you had the Cuban Missile Crisis. So like, look, Anthropic putting out a paper saying Claude Mythos is really scary is one thing, but it is different at a fundamental level than like completely shaking a society and like having the entire world realize that you were on the verge of apocalypse.

1:44:39Jordan Schneider:So I hope we don't need that, but I am not particularly optimistic that much will come of this before it really accelerates.

1:44:51Erik Bernhardsson:How much have you read into the theories that maybe China would fund data center pushback in the United States? I don't know that they would need to, but if they were, if they felt that was important enough, They would signal something about how they think about the technology.

1:45:09Jordan Schneider:I think this is cope. Look like, yeah, the Russians won the election for Trump. Okay. I mean, you know, it is. I think Ben Thompson had the right take on this a few days ago. This can be solved with money. like if if you really um put it uh you know put put the choice to a town saying okay we will pay everyone within a 20 mile radius ten thousand dollars a year um people will start moving

1:45:44Erik Bernhardsson:ugly thing people will start frantically trying to live next door or like look make the data

1:45:50Jordan Schneider:centers beautiful build a park on top of them like give me some playground that is the one that i'm

1:45:56Alex Tabarrok:the most skeptical of in the American society. I feel like they will be companies will be paying dividends and mailing checks far beyond long before anything looks beautiful in this country. We build massive like monolithic structures all the time. It's a lot of cement over here. I don't know.

1:46:16Erik Bernhardsson:That's sad. Turn the GPU racks into free slots.

1:46:19Alex Tabarrok:Every time I see one of these like beautiful industrial structures, it's always in the Nordics. It's never in America. sometimes it's in China. They build a train through a mountain or something. So you have

1:46:28Erik Bernhardsson:the data center. You've got all the different racks of GPUs and you just put a slot machine attached to each one of them. Maybe that's it. You can walk through and you get like three or four.

1:46:38Alex Tabarrok:I wonder what's the... You know those cell phone towers that look like trees? And I don't think they look any better, really. But I guess they do sort of blend in. I wonder what the economic mechanism was for going that direction. Because when I'm driving to LA, I see normal cell phone towers and I see tree, like fake trees, cell phone towers. And there had to be a reason, like it has extra costs to make it look like a tree. Like, is that just a choice? Is this like a nonprofit that's funding this? Like who's, who's lobbying for this? Is there a law? I don't know. We got to, can we please get the TBPN

1:47:11Jordan Schneider:segment? Yeah. We need to dig into this.

1:47:14Alex Tabarrok:The tree phone poll. Come for the questions, not the answers i'm just riffing about things i don't know and would like to know every time i

1:47:23Erik Bernhardsson:drive by and they get me i just start slapping wait really and i'm like you got me big telecom you got i thought it was a tree from the distance from the distance yeah every time well the bad

1:47:35Jordan Schneider:ones are like when they're five times as tall as all the other trees around just like and like the branches only start like four fifths of the way up yeah it's just yeah yeah yeah yeah it's

1:47:47Alex Tabarrok:It's very half-hearted in its attempt at disguising itself. Ridiculous. On other China issues, did you see there are two sort of dueling predictions? Maybe I'm making them dueling predictions. Chamath Palaiopitea says Taiwan, not going to be a factor geopolitically in 18 months. Dan Wong has talked about how technologists over-represent the chip story in the Taiwan story. And in fact, the discussion over Taiwan goes back much further and is grounded in more of like capitalist versus communist decision making and alliances. Where do you stand on the on Taiwan today versus in two years?

1:48:29Jordan Schneider:I have to acknowledge Chamath's existence. I refuse to do that. I think that Taiwan is a place that has mattered for a while and will continue to matter for a very long time to come. I mean, first, only on the chips thing, the idea, like, I don't know, ask any analysis, like the percentage of chips that are going to be manufactured in Taiwan, 18 months from now is still going to be like 85 % or 90%. So yeah, it used to, you know, it was 100 % for a window. I am glad Intel isn't a flaming pile of crap anymore. And it's like nice to see Samsung also starting to get their act together. But if if really that's the only thing you care about, no, you will still have a global economic catastrophe if the lights go off in Taiwan in 2028.

1:49:28Jordan Schneider:Now, why should you care about Taiwan besides the chips? I mean, it's a it's a democracy. People have, you know, deserve to have self-determination. We can set principles aside if we really have to do that. But look, I think from a sort of geostrategic perspective, there's also this kind of concept of the first island chain. I think it makes the broader Pacific architecture that America has built over the course of the past 75 years in Asia a lot more tenuous if all of a sudden, you know, the kind of anchor to both Southeast Asia and Asia. Asia as well as North Asia becomes, you know, appeal and base.

1:50:16So yeah, they're wrong.

1:50:20Alex Tabarrok:So many reasons that Taiwan will remain important into the distant future, regardless of what happens with Intel and TSMC Arizona, et cetera. Agree with you on that. What do you think the next catalyst might be in the rare earth story? Because we're starting to see rumblings about funding for American indigenous supply chains there. Eventually that becomes at least a stocking horse, if not a poker chip on the table, on the bargaining table. But where are we on the rare earth stalemate that became an important bargaining chip somewhat recently, as you put it in the intro?

1:51:00Jordan Schneider:I mean, it's a really interesting question. I've read reports saying that even five years from now, the leverage is going to still exist. And I think there's a sort of broader question of like, this is the second largest country in the world and largest economy in the world. And it's going to stay that way for a really long time. And the sort of ambition to fully decouple such that China does not have leverage over the US economy and can kind of squeeze it coercively seems to me to be a very tricky thing. Now, you can spend lots of money to sort of make that less acute or sort of turn it down over time or protect the particular things if you're worried about specialized inputs into fighter jets or something that would be really catastrophic, like insulin getting cut off or something.

1:52:02Jordan Schneider:But this is like, A, a generational challenge of shoring this up, and B, there's also an offensive side of escalation dominance in deterring an adversary from squeezing on stuff like this also requires you to be able to sort of credibly send signals that you can take pain and cause pain in a way that would require the other party to back down and not necessarily use these tools. So we're at this awkward equilibrium. I think there's a lot of hard thinking that still needs to be done to like, like really conceptualize, like what's the right way to spend money and the right way to think about this relationship.

1:52:51Jordan Schneider:We actually just ran an essay contest on China talk about economic security. We're going to be, I have at 4 p.m. We're doing a megapod with three of the winners. So maybe I'll have a better answer for you after that. But it's a hard one.

1:53:10Erik Bernhardsson:Does Xi day trade?

1:53:14Alex Tabarrok:who's the best day trader in the global power you know here's the thing i got a piece coming out

1:53:21Jordan Schneider:about comparing american and chinese corruption um and the difference is that in china they still feel like they have to hide it okay well we'll be an interesting piece uh shifting gears shifting

1:53:38Alex Tabarrok:gears both literally and figuratively cars have you driven any chinese vehicles have you been outside in the byd are you planning doing car reviews on china talk i would love to see one have you been in a zika i keep seeing these they look amazing zikers i watched a lot of youtuber

1:53:56Jordan Schneider:i watch a lot of uh youtube and billy billy reviews okay um i have been in a byd in norway of all places. They had, like in the main, like, like park, they had a giant setup. you know, it's really impressive. Yeah. I, you should get some, you should get some American car manufacturers on and yell at them. It's pretty embarrassing. You know, one of the more interesting developments, I don't know if you guys have been tracking this, the Waymo using a Zeker body. Yeah. Crazy.

1:54:27Alex Tabarrok:And I would have thought there were important restrictions because if, if Zeker is not here, BYD is not here you would think that like the supply chain would be restricted but they must

1:54:35Jordan Schneider:have found a way around so the trick is that it's there it's the chassis and the battery and literally nothing else yeah um so they are saving money on that but like the the rules right now are just on all of the sort of like electronics and connected anything so um waymo is filling up the entire car with stuff that is not Chinese.

1:55:01Alex Tabarrok:Yeah, well, that's probably more reassuring that Xi Jinping won't be teleoperating. You'll have a Waymo employee beaming in if you see a traffic cone out of place.

1:55:12Erik Bernhardsson:How much attention do you pay to their humanoid antics? I feel like they're just leaking out videos of just all this silly stuff to kind of just get us a little bit comfortable with it. But if you actually go there on the ground, it's like droid army you know they've got like sure a hundred thousand humanoids marching and in unison and then they just put out the video of it like dancing to michael jackson like collapsing and getting dragged off stage

1:55:37Jordan Schneider:You know, it's interesting because there isn't a market yet, right? Like these are not, these are toys and these are like things you sell to universities for research. But at some point there will be. And then there's going to be a really interesting challenge of like whether or not, I think there's like some level of consensus now that the Western makers have like smarter software. But if you can like, you know, it's one thing to fast follow the models themselves. But like if you can only if you can apply them like at 100 X scale, right, then, yeah, it could be really dramatic. I mean, you know, I don't I'm not doing like the episode, you know, Star Wars Episode One, like Droid Army.

1:56:26Jordan Schneider:Like, well, that might take it. That might take a few more cycles.

1:56:30Alex Tabarrok:Can you imagine a Chinese humanoid company trying to distill Tesla Optimus after they have a fantastic data breakthrough and they're just having all the Optimus work in cubes, like trying to farm the data out of it? I just feel like it's such a bad night. Because you have to physically distill the data. So you buy a million Tesla Optimus robots and lock them in a warehouse so that they can move boxes for you so that you can distill it into your own robots or something. This is a company. This is an RL environment. I love it. I should start this. In China? Oh, no. I mean, they're just like a reseller.

1:57:11Alex Tabarrok:In the U.S. Yeah, yeah, yeah. You need a hair dryer to remove the Optimus sticker and slap on some other sticker. That's where the money is being made these days. But you got to hide that corruption.

1:57:23Erik Bernhardsson:And it's really great that we have a software advantage and they have a hardware advantage on this. That's really bullish for us. Yeah, because software is like they would never be able to copy like software off of a hard drive and put it on their hardware.

1:57:37Jordan Schneider:Well, and that's the thing is like this is the difference with AI, right? is at least America has way more chips and way more compute as long as Taiwan still exists. But that will not be the case for the humanoid embodied AI takeoff. The thing you need to actually use the fancy model is almost certainly going to be manufactured in China and manufactured at a scale that could be multiple orders of magnitude bigger than in the West. Now, this may change like once this becomes economic.

1:58:13Alex Tabarrok:Yeah. Right. So America can. I mean, like you look at you look at space like like we have multiple companies that land rockets, very capital intensive. It's an industrial process. You would assume that China would win that. But they are. But we have a lead in in master orbit still. And so it's not impossible. But we do tend to let our leads languish like we have in the car industry.

1:58:37Jordan Schneider:uh seemingly and it'll be interesting because there'll be this lag where um china like because they already have all these humanoid companies that like have the ability to scale will scale faster um once the kind of like economic utility of these things turns on yeah definitely uh if you

1:58:54Erik Bernhardsson:if you believe in ai progress do you have to believe in a taiwan invasion because assuming assuming you know we keep making progress or we have accelerating progress doesn't that just

1:59:08Alex Tabarrok:increase the prize for yeah but it might increase the defensive capabilities sure yeah and i think

1:59:14Jordan Schneider:there's there's like plenty of other things you can do to be obnoxious um and there's so much there's look this is this is why chamath is wrong about everything is because it's not all about technology like there are other risks and incentives at and like institutional dynamics and historical dynamics at play and there are a whole lot of downsides to starting something that you don't know how like you don't know how it's going to end which is a lesson that she has now got to relearn from Putin over the past five years so

1:59:50Erik Bernhardsson:what last thing we touch on it briefly and then I know we're over time what happens with manis what's your prediction are they gonna are they gonna buy the company back from from meta and just head home is there any precedent for something like this

2:00:06Jordan Schneider:i don't really understand i mean like all of the employees like do they have to give the tech back they weren't really buying the tech they were buying the it was like an acqui hire everyone's in singapore except for the two founders so like maybe the founders just get screwed out of all this money and uh you know the employees continue on their merry way in singapore i think it's uh uh it's a messy one and uh i don't know like meta eats a loss yeah something i hope bill

2:00:37Alex Tabarrok:girl before you go we have an answer to the question of telephone pole camouflage tyler do you want to do you want to give us a little update on this uh can we can we do yeah let's

2:00:47Jordan Schneider:So, yeah, telephone pole camouflage basically began after the 1996 Telecommunications Act. So it basically, like, restricted the ability of local communities to regulate, like, the actual placement of the telephone pole.

2:01:00Alex Tabarrok:So a city can't say, you can't put a telephone pole there, you can't put a cell phone tower there, but they responded.

2:01:07Bill Clerico:But they basically made it so the local governments would say, you have to camouflage it, right? So now we see there's all different forms of these.

2:01:18Jordan Schneider:There's trees, there's cacti, there's like church steeples.

2:01:22Alex Tabarrok:Yeah, yeah.

2:01:23Jordan Schneider:Yeah, bigger of these. A lot of chipmunks run up and down those poles too, right? And camouflage can add over 100 ,000 to tower construction prices, which is still cheaper than losing the site. Tree towers can cost up to double.

2:01:39Alex Tabarrok:Hey, two can play this game. Two can play this game. You have a soundboard too. Give me a second. I got my buttons. We have a guest with buttons. Let's see. Can he get it to work? Uh-oh. Oh, no. Oh, no. Another one.

2:01:53Jordan Schneider:That's a narrative violation. What? Let's find out.

2:01:58Erik Bernhardsson:Your buttons aren't working.

2:02:00Alex Tabarrok:Anyway.

2:02:02Erik Bernhardsson:But bring them next time. Bring them next time. We'll do it again soon.

2:02:05Alex Tabarrok:It's been too long. We'll talk to you soon, Jordan. Have a great day. Great to see you. Hey, it's working. There we go. Hey. Very well done. Thank you so much for taking the time. Have a great rest of your day. We'll talk to you soon. We went over time, so we will bring in our next guest immediately. We have Christina Lee Storm from Secret Level. Welcome to the show. Sorry for the delay.

2:02:26Erik Bernhardsson:What's happening?

2:02:28Alex Tabarrok:Hey, how's it going? We were having a little too much fun with our soundboard. We were having fun with a regular. I love it. It's great. Playing sound cues, all sorts of stuff. But since it is your first time on the show, please introduce yourself a little bit.

2:02:40Christina Lee Storm:Hi, I'm Christina Lee Storm. I am head of studio over at Secret Level, which is an AI native studio. And I also am co-founder over at Playbook PLBK. I identify myself as a producer, like an independent film producer.

2:02:55Alex Tabarrok:Yeah. How are you thinking about the blurry line of how AI works its way into productions these days? You have Toy Story with CGI. You also have a little set extension and maybe even a little CGI creeps into a Nolan movie every once in a while. But AI is very different. People are thinking maybe it'll generate the whole movie. Maybe it'll just sort of, you know, make a green screen a little bit sharper. How are you thinking about the opportunities and the trailers?

2:03:24Christina Lee Storm:Yeah, I mean, it's interesting because we've had technology in the entertainment industry for a while. I was a consulting producer on a film called Jurassic Punk, which was about the birth of computer graphics. It was about a gentleman who was from Canada. His name was Baz Steele Steve Williams, and he actually created the first walk cycle, T-Rex for Jurassic Park, and he was a total rebel. And in the documentary, we talk about how he just felt like, you know, we could do this with computer graphics. At the time, Jurassic Park was like, hey, we're going to do it animatronics. We're going to do it old school.

2:04:02Christina Lee Storm:We'll do it on set. And there was an ability to actually now like, wait, hold on a second. We could create this. And so when Spielberg and Kathleen Kennedy saw the first walk cycle as he was playing it on his computer screen, you know, just sort of he knew when they were going to walk by. It changed everything. And that was the birth of computer graphics. So we know that technology has always been a part of that story. And so I think, you know, it is quite a lot of people like to use the headlines like disruptive, but I think it's just an ongoing process of how things should sort of evolve. And me being my background is a traditional production producer.

2:04:46Christina Lee Storm:I've been an independent producer for a while. I think it's just it's part of the process. And I think before, I think a lot of producers would just offline, you know, any kind of technology or any specialty to different departments or groups. And today, I think anyone who's going to really survive and thrive into the new world order of this, you know, everything's sort of shifting and changing. We have to actually lean in. And there's this interesting like technology, storytelling, you know, convergence, entertainment that is exciting. And I think, you know, that's. Yeah.

2:05:24Alex Tabarrok:On the topic of Jurassic Park, grade this prediction. Jurassic Park, massive success. The franchise has grossed, I think, billions, maybe over a billion dollars. It's a fantastic financial success. But interestingly, Jurassic Park, of course, is owned by Universal. But dinosaurs are not particular intellectual property. It's not like Superman. The T-Rex is not something that a studio can own. Oddly, no other studio has created like the Superman to Spider-Man. There's not the Marvel Universe to the DC Universe. I would have expected a reaction series to Jurassic Park and that IP library. at some point, never really developed.

2:06:10Alex Tabarrok:But in the age of AI, where it's increasingly easy to generate imagery of dinosaurs, and you don't have any intellectual property concerns because you're not recreating a particular actor, I predict that there will be a massive surge in dinosaur-themed movies. What do you think?

2:06:27Erik Bernhardsson:You need to create the Dino Prize. The Coogan Dino Prize. The Dino Prize.

2:06:30Alex Tabarrok:What do you think? Is there anything here? Am I thinking about it correctly?

2:06:34Christina Lee Storm:I think that you're on the right track in terms of like dinosaurs are not, you know, there's not a particular IP, but here's the thing. And this is sort of where it does fall back to traditional methods. It's like the storytelling is everything. Like you could have, there's a lot, there's a lot of different ways, you know, that you can tell and share stories, but it's really, what is the specific, you know, what are the specific things with the character, what's happening, what's the world. And so, yeah, someone could come up with a dinosaur type movie, but what is it? Why, why is that special?

2:07:09Christina Lee Storm:Why is that undeniable for someone to say, yes, I want to watch that. And I want to, and I want to engage in that. So I think the concept of that is, is good. I think, you know,

2:07:19Alex Tabarrok:potentially half baked, but we'll get,

2:07:22Christina Lee Storm:well that's the difference between really great. No matter what the technology is, you have to still be a really good storyteller.

2:07:30Alex Tabarrok:Yeah.

2:07:31Erik Bernhardsson:Please. article in the wall street journal this morning about a new film called hell grind that's i guess premiering at can uh they're saying that it cost half a million dollars to make and around 400 000 of that was uh compute yeah um i'm sure there were you know there with all these things there ends up being costs that aren't kind of captured uh on on maybe a napkin sort of a deep seek moment Yes. But but is this the beginning of like, do you expect hundreds of these over the next year? I imagine you're hearing and seeing a lot that are in the works and working on some yourself. But yeah.

2:08:11Christina Lee Storm:Yeah. So, I mean, I think there are, you know, can I actually didn't go this year. I usually go every year. I knew that there was going to be a lot of announcements about different projects and things. at secret level, we actually will have an announcement next week. I was like, should I come on the show today? Because next week there's going to be... I'm happy to come back. But at the end of the day, I think, let's just be real. Let me just be real. There's a lot of things that are being pioneered. What are the costs? What's happening here? How do we align with even guild-related things that are issues?

2:08:51Christina Lee Storm:You know, how can we, it's not just like, let me just get this out. You know, I think there's something even bigger to that degree. You know, I think people, a lot of people could say, yeah, I can make a movie. But I think it's really, again, I'm going to fall back on story. I'm going to fall back on what is the process in which that's going to happen. Like at Secret Level, we do, we have our own proprietary workflow pipeline that really, really allows us to scale. And I think those costs, when those get flagged, it's like, yeah, what's the total cost in it? There are a lot of costs. And so I don't think, I think it's really, I think when people talk about like the budget and they're just focused on the budget, I think they've missed the mark on the bigger.

2:09:35Erik Bernhardsson:So what is your, maybe it's too early to say, but what do you think is going to be the winning formula? Obviously, it's still an experimental time, but you said story matters a lot. Obviously, technology matters a lot, what models you're using. But where are you thinking of taking traditional filmmaking approaches versus reinventing your approach?

2:09:57Christina Lee Storm:Yeah, I think there's a couple things here. Great question, Dredi. I think that there was a time where independent film was on the rise. It was awesome. We saw really great filmmakers come out of that, going to Sundance and whatnot. And I think that that's really important to feed and bring about like creativity, really great story. So I do believe that with the tools, we have more of an opportunity for independence to sort of bring that to the forefront. But again, you know, story, it's about the story. Like, is it going to be a really great story? I think the other piece is, is will we see this?

2:10:41Christina Lee Storm:I kind of share with this because I've been a former surviving studio executive and also a producer. And I think that what's neat is, you know, is the chasm going to widen? Like, how do we sort of bring that? Do we want to bring the chasm closer together so it's not just traditional, you know, studio fair? Or do we want to like really see these independent voices come out and we can see those play out? And I think as time tells and how people will refine their storytelling abilities, I think that's really important. I think there's also, to be honest, there's a way in which filmmakers who use AI tools are using it at their best abilities.

2:11:26Christina Lee Storm:Is it almost a slight variation than maybe a traditional approach from, let's just say, like a studio, you know, pipeline type film? I think, and that reminds me of like, you know, when George Lucas came out, like, you know, he was breaking new ground in technology. It was like, this doesn't exist. So how do we create something that doesn't exist? And I think we have to allow for that creativity to birth.

2:12:01Alex Tabarrok:Totally. I want to take a question from the chat. What is your favorite movie? Oh, gosh. I have a lot or maybe something just with a great story I love Inception Inception, that's a great movie have you seen that? I have there we go

2:12:17Christina Lee Storm:I like to say, and I'm going to embarrass him our founder, Jason Zada from Secret Level I call him like he's like the blossoming Chris Nolan and he's just really creative, the way he approaches a story. Does he use a phone? Christopher Nolan famously

2:12:38Alex Tabarrok:no phone.

2:12:39Erik Bernhardsson:He did an interview with

2:12:42Alex Tabarrok:no email either. Printed out emails if somebody needs to email him. No phone. Also doesn't like maps because they reorient you. They don't always face north and he doesn't consider that a map. Very interesting lifestyle.

2:12:55Christina Lee Storm:I would say Inception, I would say let's go really old school. It's a wonderful life. and then if I get my like just real old school I love like great character based when Harry met Sally I'm all over the place I like being that way so people can't say oh like sci-fi or she you know godfather

2:13:20Alex Tabarrok:I saw when Harry met Sally at an outdoor movie theater at one of the screenings you buy tickets to seeing an old movie outside with a bunch of people tons of fun

2:13:29Erik Bernhardsson:last question from the chat What is your single favorite piece of fully AI-generated content? It could be like 30 seconds long, a minute long. I can't really expect anything else. I think for us it's like the Harry Potter Balenciaga, which funny enough was like a year and a half ago maybe at this point.

2:13:53Christina Lee Storm:So if you haven't seen it, yes, you know, there's a secret level. the heist.

2:14:01Alex Tabarrok:Okay. Watch the heist.

2:14:02Christina Lee Storm:When Jason shared his first 30 seconds, I was like, okay. Okay.

2:14:07Alex Tabarrok:We got something.

2:14:10Christina Lee Storm:We're approaching some really interesting things. So you can see a lot of things at secret level.

2:14:15Alex Tabarrok:Yeah. I love it. Amazing. Well, thank you so much for taking the time.

2:14:20Erik Bernhardsson:Really great to meet you. It's great to meet you. Make sure to share your announcement next week with us. Yeah. I will. Can't wait. Have a great rest of your day. Cheers. Bye.

2:14:28Alex Tabarrok:Bye. Up next, our last guest, Eric from Modal Labs with some big news. It's been a minute. It's been a minute. I'm excited to catch up. How you been?

2:14:38Jordan Schneider:Good.

2:14:39Alex Tabarrok:You been busy?

2:14:40Jordan Schneider:What'd you do? Tell us what happened. We just announced a C round today, which is very exciting. How much? How much? How much? 4.65 post value.

2:14:55Jordan Schneider:I'm racing$355 million. So general catalyst led it together with Redpoint. Fantastic.

2:15:02Erik Bernhardsson:What was the specific catalyst that led to this round?

2:15:07Jordan Schneider:There's been a lot of stuff that's been going on. I think one particular exciting moment in the last 12 months, or really the last six months, has been we now have a product called Sandboxes that's just growing insanely fast, almost 2x every month for the last six months. So Sandboxes, if you don't know, basically the idea is you can take typically LM-generated code and execute it in a safe environment. So it powers a lot of reinforcement learning, powers a lot of Vibe coding apps, a lot of background agents. But yeah, I mean, in general, we've seen tremendous growth also with inference. Modul has always been kind of, we started a company five years ago with the idea of building pretty general purpose infrastructure.

2:15:45Jordan Schneider:We felt that a lot of infrastructure wasn't really built for AI. And so when you look at all these companies out there building sort of AI applications, they all need different things. Some of them need sandboxes, inference, training, batch jobs, like all kinds of stuff. And our goal is to provide all of that and really kind of build a new sort of, almost like a new cloud in a way that supports all these types of very cool applications.

2:16:07Erik Bernhardsson:Very cool. I feel like it's been almost a year since you were on, which is like, you know, 15 years. It's back in October. October. Well, there we go. Feels like a year ago, even though it's maybe less. Yes. What have been the key inflection points since then? Sandboxes sounds like one, but I imagine a lot of this is the space is moving so quickly. If you get a signal from two or three customers, hey, we have this problem. Can you build a solution? And then you can just see tremendous growth in a very short period of time. And so in some ways, you have an existing portfolio products that are all, you know, growing, but are you trying to constantly be making, you know, new bets based on customer needs?

2:16:58Erik Bernhardsson:Or is it is that maybe not even the approach?

2:17:01Jordan Schneider:I think one of the benefits of building infrastructure is like, you can sort of, you know, look at customers, of course, and see what they're doing. But I think with infrastructure, you can also sort of argue a little bit more from like first principles, like what are the building blocks that people are going to need in terms of compute? So So sandboxes was actually something we launched three years ago. And it really started taking off like nuts in the summer last year. And similarly, I think with, you know, in France, it's sort of similar. We launched it almost four years ago. We're working on some other really cool stuff around trading and some other products.

2:17:34Jordan Schneider:And I think one of the benefits of infrastructure is you can have a little bit more sort of a first principle, like, you know, let's build the right building blocks and then let our customers sort of figure out what to use them for.

2:17:43Erik Bernhardsson:how how do you think about uh planning like planning planning demand add a zero to last yeah yeah add two add two zeros uh yeah but yeah it just feels like the you know the best this is the hardest like demand planning challenge that really any set of businesses have ever faced in history uh yeah it's rough it like i mean for us it's like we basically look at

2:18:11Jordan Schneider:Like the last couple of months, figure out, you know, we're growing 30, 40 % every month. You just, you know, take, you know, take the power of three. That's how much you're going to need in three months. That's how many GPUs to go out and get those GPUs. You can typically get GPUs with like about three years of, sorry, three months in advance.

2:18:29Alex Tabarrok:Sure.

2:18:29Jordan Schneider:But yeah, that's a big part of how we think about the business today. It's like, we need to get the GPUs now that we're going to need in three to six months, which is a lot.

2:18:35Erik Bernhardsson:Yeah. Yeah. In some ways, it's not too dissimilar to the challenges that consumer packaged goods founders facing. You're growing super quickly. But if you want to stay on that growth curve, it means you need to be committing to things now and committing capital.

2:18:54Alex Tabarrok:What's exciting to you across the semiconductor ecosystem? There's a whole bunch of ASIC startups that have been on the show. cerebrus ipo'd uh every hyperscaler is working on uh different chips at this point um what's most interesting what's under discussed what's on your roadmapper or or uh what have you already

2:19:19Jordan Schneider:sort of uh sunk your teeth into yeah i mean there's some like really cool alternative accelerators i'm quite bullish on tpus amd trinium like all these things yeah we see zero demand And from our customers for any of those things, to be clear.

2:19:34Alex Tabarrok:Because only the really big labs, Anthropa, can actually go and figure out how to run it. Is that the issue?

2:19:40Jordan Schneider:Yeah, I think the cost today of rewriting your software to run on those stacks is just like very high. Sure. So while I remain like very bullish on this sort of, you know, two, three of Verizon, you know, I do also want to like temper the expectation a little bit.

2:19:53Alex Tabarrok:It sort of has a fixed cost to rewriting for that chip stack. And if you're not doing billions a month in revenue, it's hard to amortize that cost.

2:20:01Jordan Schneider:That's exactly right. It's just not worth it unless you're operating at a very large scale. But I think that cost is going to go down over time. We're going to have software that basically lets you take existing CUDA-compatible stuff and run it on other alternative accelerators. So I think it would be good for everyone to have a little bit more competition in this space. But we also love NVIDIA, and that's what our customers want today.

2:20:19Alex Tabarrok:That's great. Do you have a view or any opinions or predictions about the compute futures market that's been talked about? I've seen these price charts of B200s going up and down. Everyone is trying to read the tea leaves, understand where we are in the various AI cycles based on it. But is that something that's relevant or important at all? Do you look at that data? Do you have your own internal data set that's more relevant to you?

2:20:48Jordan Schneider:We look at it a lot. I mean, also, like, I feel like we're very plugged into the market. We talk to Neoclots all the time. We get capacity all over the place. So we have, like, a very good, like, pulse of the market. I think the market is going to remain quite tight. I think fundamentally, like, also, like, that's a big part of what we do is, like, we offer, you know, that as a product. Like, don't think about capacity. Come to us instead. And so that becomes our problem is, like, managing that capacity for thousands of companies at the same time. Like, we built a multi-tenant product that sort of aggregates all that demand.

2:21:16Jordan Schneider:And so if you need a thousand GPUs, you can come to us and we'll give you to that, give you those GPUs, like often within minutes, because we have like a very big pool we can sort of tap from. Yeah. But yeah, like, you know, I think GPUs are going to be tight. We look at the prices, they keep going up. At some point, obviously, I think it's going to normalize like most markets do, but it might remain tight for the next year or two.

2:21:35Alex Tabarrok:When I think about the big applications of a big pool of compute, I think training and inference of LLMs, coding models, agentic workflows, I think image, video, audio generation. What's next? What do you think the next big driver of demand is? Is it world models? Do those fit in? Are those structurally different? Is there something else that you're tracking where you're seeing customers come to you with sort of like a different product, but it fits in the same shape? so you can work with them as a business partner?

2:22:07Jordan Schneider:Yeah, we have a very wide range of customers. So we have everything from Suno, which generates using AI. Cognition is training coding models using reinforcement learning. Ramp uses us for background agents. We have a lot of vibe coding platforms like Lovable. We also have drug discovery companies like Chai using us to simulate molecular dynamics. And we have weather forecasting companies, robotics. So we have very general, we look at across a lot of different verticals. Obviously, the big application in the last couple of years has been LM inference, and that keeps growing a lot. But we've also seen some really cool applications with diffusion models.

2:22:48Jordan Schneider:In terms of what I look forward to, I think that probably personally the coolest thing I kind of expect to happen in the next couple of years is speech to speech. like imagine just like talking to a computer in order to figure that out we need to get the latency down you know for for all the three components of speech to speech you know doing the lm and the you know text to speech and stuff like that so that's something very cool but i'm also really cool about i think it's incredibly cool if we can i don't know cure cancer or something

2:23:14Alex Tabarrok:like that yeah yeah it's interesting because uh i'm with you on that on that timeline speech to speech being uh you know somewhat impressive but you're often not actually hitting a real reasoning model. There's so much speed up. We probably need to be 100x faster, 1 ,000x faster on the response time to really have a breakout moment. And then you watch Google I.O. this week and you see with Omni a glimpse of, okay, it's more like a FaceTime call with the AI and it's generating a video in real time. And you know how long it takes to crank on video models to get an eight second clip it's minutes and minutes uh and and and they're still have errors and stuff and we're still not even at the the super the superhuman element so lots to do lots of servers

2:24:02Erik Bernhardsson:to rack lots of gpus to set on fire i'm sure yeah what and cpus yeah any any plans to actually you know go full stack get your own you know land power powered shells etc is that like not the highest and best use of your guys' talents?

2:24:24Jordan Schneider:We think of our values like adding a big software layer on top of the underlying computer, on the underlying compute layer. So we think of ourselves as almost like a cloud, like one layer up from the existing clouds. The existing clouds are very good at running computers in the cloud and offering that through an API. We would love to keep tapping that as much as we can. If we can't get the capacity we need, we might have to build it ourselves. Ultimately, it's a kind of an economics question or a practical question. I wouldn't rule out any sort of move. The idea of racking computers and plugging in cables and dealing with fires and data centers and whatnot, to me, that's not super appealing.

2:25:03Jordan Schneider:But at the end of the day, we'll do it if we have to. Absolutely.

2:25:06Erik Bernhardsson:Yeah. Makes sense.

2:25:07Alex Tabarrok:Well, congratulations.

2:25:09Erik Bernhardsson:Yeah, congrats to the whole team on an awesome milestone and incredible progress. It felt like a year. It's only been since October.

2:25:16Alex Tabarrok:Only a couple months?

2:25:17Erik Bernhardsson:Yeah, it's like 10 AI years. I love it. Congratulations.

2:25:20Alex Tabarrok:We'll talk to you soon. Thank you so much. Cheers. Have a great rest of your day. Let's click through the timeline, see if there's anything that we missed before we jump off. DJ Kaus has an idea here. Found a$10 bill. It took five seconds to pick it up. That's$63 million in annualized ARR. This is the thinking you need to be deploying if you're going to be raising money potentially. No, don't do that.

2:25:51Erik Bernhardsson:I didn't know that Messi, the football player, had a prime copycat called Moss. Moss. And it's shutting down after 23 months in business.

2:26:02Alex Tabarrok:I wonder if that has to do with the logistics of shipping internationally because he is an international icon, but setting up distribution and retail presence across many countries very quickly. It's not quite as easy as dropping it on Amazon and flying off the shelves. A lot of those are sort of one country by country. And I wonder if that has a piece of what happens. So this is shutting down fully. Moss Plus. Interesting. Well, Mitchell Baldridge recommends setting up a Vanguard account. Not Fidelity, not Schwab, Vanguard. Smart advice, you might think. They do have the lowest fees. If you want to get up to speed on Vanguard, listen to the latest episode of the Acquired FM podcast.

2:26:46Alex Tabarrok:But he says wrong. It's not because they have the lowest fees. It's because their interface is so awful. You will never trade It's his mate. It has made his clients millions

2:26:54Erik Bernhardsson:Let's close it out with this video What you got in Catan with a billionaire?

2:27:00Alex Tabarrok:Yes

2:27:02Erik Bernhardsson:I think I think you'll like what is this face?

2:27:05Alex Tabarrok:Is this a face filter or a background replacement or both something funny is going on here face filter? Okay face

2:27:16Jordan Schneider:That's against the rules.

2:27:17Alex Tabarrok:How else am I supposed to AI generate my Christmas card? Not popular. I'm not joking. The way you walk, the way you understand me. Do you like it? No, it's blocking all the land. Wait a minute. Are you one of those paid protesters? The way you move, the way you just whisper me. Next turn, I'm gonna use 7 water on my data center.

2:27:40Jordan Schneider:That's not how that works. That's not a resource in the game. How else am I supposed to power it?

2:27:45Alex Tabarrok:The water stuff is really, it's so interesting that no one has moved to energy. Like natural gas, there are natural gas turbines that would be opposed, and yet people are focused on water. Stop giving them ideas. I know, I'm doing their job for them, I suppose. But it's like, I don't know why the water, I think it's just because like water's delicious and electricity is vague and abstract and you don't think about it. Like you can visualize a glass of water. It's hard to visualize a battery in the same way. But yeah, what is it? It's like dozens of LLM queries every day for a full year is equivalent to eating a single almond, something like that.

2:28:29Alex Tabarrok:But Matthew Ball is back at Xbox. Congratulations on the move. He announced it yesterday. We're gonna try and get him on the show because he's one of my favorite people, favorite authors. If you haven't read his book or his blog, He has a fantastic mind for future of technology. And I could not be more optimistic about the future of Xbox with him on the team. Tay Kim says, this hire is a literal game changer. Matthew Ball knows gaming and what needs to be done. This news makes me the most bullish I've been on Xbox in seven years. And I completely agree.

2:29:02Erik Bernhardsson:And to close it out, the White House is awarding$2 billion in grants.

2:29:06Alex Tabarrok:Oh, yeah.

2:29:06Erik Bernhardsson:quantum computing to IBM to nine quantum computing companies in taking an equity stake so their grants their grants overtaking an equity stake it's an investment it's an investment it's

2:29:19Alex Tabarrok:an investment and you American taxpayer will now own a basket of quantum computing companies

2:29:26Erik Bernhardsson:uh spaghetti computing is up that's not spaghetti computing it's righetti computing spaghetti computing is up 30 %

2:29:33Alex Tabarrok:who came up with that that seems like a Trump something he would say

2:29:37Erik Bernhardsson:I created that

2:29:39Alex Tabarrok:just now

2:29:40Erik Bernhardsson:I'm sure I'm not the first person to think of it but anyways folks thanks for watching

2:29:47Alex Tabarrok:leave us 5 stars on Apple Podcasts and Spotify sign up for our newsletter at tbpn.com and we will see you tomorrow Friday goodbye we love you

2:29:57Erik Bernhardsson:have a wonderful evening You