Stuck Academia, US vs China, and Why Startups Work w/ Steve Hsu

Hey, Upstream listeners. Today, we're releasing a conversation I had last year with physicist and startup founder Steve Xu. We discuss academia, the rise of China, AI doomerism, fertility decline, and more. Please enjoy.

Steve, welcome to the podcast. Thanks for joining.

Hey, great to be here, Erik.

So we're going to talk about academia, we're going to talk about China, we're going to talk about AI, fertility. But first, let's start with academia. How hopelessly corrupted is academia? Do you see any white pills or signs of hope? We know that it's sort of ideologically you know, corrupted on some levels. But the actual, you know, in the humanities, people say are corrupted, but are the sciences also corrupted? I hear that replication crisis is not just in psychology, but that there's a lot of stuff on the hard science that doesn't replicate as well. Why don't you talk about some areas where you feel, hey, this is a problem that other people aren't talking enough about, or other areas where you feel, hey, this is actually okay, or there's signs of hope here?

Yeah, obviously academia is not monolithic, and you've already sort of gestured at that in your question. So to give you a fulsome answer, as we say, I have to kind of break it down into different silos. One sort of just empirical finding— there's a lot of good work by a guy called Eric Kaufman, a British— well, he's actually an American or maybe Canadian, but he's based in the UK. He's carefully studied empirically the leftward shift of academia in general. So your chances of finding a professor in a social science or humanities department who is not left of center is very low, like maybe 10%. Any department that's not engineering or maybe some business departments, is very, very left, and that tends to replicate. So it becomes sort of unacceptable for a candidate to be outwardly right-wing. So even if you were to say, I'm a Republican, or, wow, Trump is not that bad, I didn't vote for him, but he's not that bad, that could be disqualifying for you in a lot of settings. And in a competitive hiring environment, There would be people who would just say, we don't really want that guy in our department. I don't see any way of reversing that. That's gonna last for easily a generation. Okay, so you're just not gonna have any representation of even like, I would say, kind of slightly realistic hereditarian views, people who are really enthusiastic about free markets. There's just less and less of that in the academy. And I just don't see any change in that going forward. In the sciences, the problem, you know, sciences are actually almost as polarised in terms of the left-right dichotomy as the humanities and social science. It's a little bit better in some of the fields I've mentioned, like maybe engineering or something, but for like physics or math, I think it's just as left-dominated. But those views don't come into the practice of scientific research quite as much because often we're researching some abstract thing like the properties of quarks or something, and your political sympathies don't necessarily affect that. But what it does affect is the commitment to meritocracy and the commitment to truth-telling. And so the swing away from meritocracy is very, very serious now. So I would say it's possible that the majority of departments now are not requiring GRE scores for graduate admissions. The professors who are familiar with the role of psychometrics or measurement of cognitive ability and its ability to predict success in difficult subjects. I think that's a very small minority of faculty now, and most of them would just be embarrassed or sort of afraid to talk about the vast empirical literature on this topic, which used to be taken for granted. So like when I started graduate school, you know, 30 years ago or more, people understood, like, why do we make the students take the GRE exam? Why do we make them take the GRE subject exam in physics? And what does it mean? And, and now, because there are— in every department there's some fraction of people who are activists, they will literally shout you down. So if you say like, well, you know, this study which was published in Science, you know, 20 years ago showed pretty good predictive power of GRE scores independent of undergraduate GPA and independent of socioeconomic status of the kid for success in our PhD program, or, you know, in, in whatever, um, you would be shouted down. Like, that would be considered like, well, why is this guy saying this? Doesn't he realize like there's a problem with this? Doesn't he realize this is really damaging to diversity for him to say something like this? So, um, I would just say the atmosphere is very bad. And, um, it— for science and engineering, it's only affecting efficiency of the operation of these departments on the margins. So you get— you have slightly less good selection of students by talent, you have somewhat less good hiring practices, somewhat less good promotion practices. And gradually, though, I think there could be a cumulative effect where the overall commitment to truth, the commitment that we are going to let empirical evidence dominate our feelings, that cultural value is gradually being eroded. And that's the most dangerous thing. So that could cause not a negative effect around the margins, but it could cause an order 1 change around how the discipline is practiced. Now, I would say in social science and humanities, that's already happened. So it's already— their commitments to ground truth are very, very limited. And many of them, especially empirically minded social scientists, would deny this. But when you actually start probing them, you realize they really— they're very quick to avert their eyes from data that challenges their very sacralized beliefs. And maybe they don't even realize it. It's so ingrained in the culture now that, you know, if someone says to them, hey, you know, in Sweden or in Denmark, they have enough statistics that can actually calculate the social cost of immigration. By some, based on some crude characteristics of the type of immigrant, and does that person commit, uh, contribute more to the national budget in their lifetime, or do they, are they a net, um, negative impact on the national budget during their lifetime? And there's good statistics in some countries on that. That would be a very verboten conversation in a U.S. econ or sociology department, and, and some people would attack you. They'd say, why are you particularly interested in that question? They would say— they would be very quick to assert, very quick to imply, suggest that there's something bad or wrong about you, morally corrupt about you, that you would be interested in this really pretty fundamental question about how societies work and things like that. So I would just say it's not a healthy situation. And probably for your listeners, nothing I'm saying is at all— I'm just confirming what your listeners think. But just to offer my bona fides, I've been a professor for about 30 years now. was a senior administrator at a Big Ten university for about 8 years. So I, I think I have a pretty good feel for what's going on in the academy, and generally it's not good.

And Bryan Caplan, uh, the libertarian, had a post the other day basically saying, why isn't there kind of a right-wing parallel economy, um, if, if, you know, the country is split 50/50? And I'll just, you know, narrow that question to, to the academy in terms of why are we so hopeless that nothing can change if there's a large percent of the population that doesn't, that feels that the current solutions don't serve them. Richard Hanania might speculate and say it's because of the law or how sort of intertwined universities are with the government, and thus there isn't a ton of flexibility. I have another friend who says, no, there's actually something inherent in kind of these big managerial organizations that tends to attract people who are, um, more left-wing in terms of their, their ability to rise up and, and run these bureaucracies or, or, or, um, be sort of, um, you know, uh, powerful in these bureaucracies. So it's not just the law, there's some sort of emergent cultural effect. Though to me it just seems strange. People who are attracted to truth in theory should, you know, are attracted to academia And, you know, wouldn't sort of fall prey to ideology on either the left or the right. But why do you think that's the case, that there's so much of a desire to do things beyond sort of pursue truth in a place that's ostensibly focused on pursuit of truth?

Yeah, so the fraction of people that are really hardcore truth seekers is relatively small, even among, I would say, researchers on campuses. I mean, a lot of people just have the feels that they want to help people, help students, further goodness. Notice none of those are really like pursuit of or trying to really delve into what could be emotionally difficult truths, right? And it's those factors that kind of drive people toward the academy, the idea that, oh, I'm helping people. This is a good institution. It does good things for—

Right.

educates young people, it discovers new knowledge. Those are, you know, those are all good things. It doesn't mean that you have the kind of like difficult personality or commitment to truth that is required, you know, to, to study certain topics, uh, these days. Um, so I think it's not just not true that in general, uh, academics are all that committed to truth. A lot of them are just playing a particular game that they're based, based on their, you know, Big Five personality or desires. They kind of fit well into this system and they're just in a careerist way, uh, pursuing a trajectory through the system. Now, now, a lot of the people who would be making the big advances in STEM are the difficult personalities that are very truth-committed, maybe even Aspie, a little bit on the Aspie side, um, and just a little bit too, uh, obsessed with the truth and facts. But what's bad about the current situation is those people are being selected out now, because if, if you have to write a DEI statement To apply for the faculty job, to get promoted, to get a raise, you have to write a DEI statement at many places now. And everyone who's kind of in the know knows how they should write that statement. But a lot of people who are principled maybe don't agree with everything that's going on. And, you know, it's harder and harder for that kind of maverick person who would make a big breakthrough in STEM to survive in In the current system.

Where do the truth seekers go, if not academia?

Yeah, so, you know, it's funny, because for your audience, this won't come as a surprise. Like, I'm a weird guy, because I both do startups, tech startups, and innovation. And I do, you know, in some sense, very esoteric academic research, like in particle physics and stuff like that. And I can see today, like, I was just at Caltech, my alma mater, to give a seminar a couple of weeks ago. And I talked to a bunch of young physicists and I said, like, I'm kind of amazed that more of you aren't leaving now because there's so many exciting things which are impacting the world and moving fast and also intellectually pretty deep, like artificial intelligence. And so what's happening, I think, is that a lot of people who, you know, maybe they're, you know, if you're obsessed with some really specific question, like, oh, I want to understand the genetics of human evolution in Malaysia, you know, know, 3,000 years ago or something. You have to do that in academia. But if you're sort of generally capable and you want to achieve something and you want to be in a kind of functional system where if you do something good, people can tell it's good, it actually has an impact. If you create something good, it's adopted and widely used. There's a big draw away from, I would say, academia today and toward the innovation economy, which is all good. I mean, in a way, what we should do is just say like, Well, we have these academic institutions that are kind of not functioning well right now. They're kind of broken for the reasons that we're discussing. Maybe it's just good to kind of like move a little bit of human capital away from that system and toward this other system, which is actually kind of flourishing right now.

That makes sense.

We'll continue our interview in a moment after a word from our sponsors. Even if you think it's a bit overhyped, AI is suddenly everywhere, from self-driving cars to molecular medicine to If it's not in your industry yet, it's coming— fast. But AI needs a lot of speed and computing power. So how do you compete without costs spiraling out of control? Time to upgrade to the next generation of the cloud: Oracle Cloud Infrastructure, or OCI. OCI is a blazing fast and secure platform for your infrastructure, database, application development, plus all your AI and machine learning workloads. OCI costs 50% less for compute, And 80% less for networking. So you're saving a pile of money. Thousands of businesses have already upgraded to OCI, including Vodafone, Thomson Reuters, and Suno AI. Right now, Oracle is offering to cut your current cloud bill in half if you move to OCI for new U.S. customers with minimum financial commitment. Offer ends March 31st. See if your company qualifies for this special offer at oracle.com/turpentine. That's oracle.com/turpentine. Every time I hop on the 101 freeway in SF, I look up at the various tech billboards and think, damn, Turpentine should definitely be up there. And I bet I'm not alone. In a world of endless scrolling and ad blockers, sometimes the most powerful way to reach people is in the real world. That's where AdQuick comes in. The easiest way to book out-of-home ads like billboards, vehicle wraps, and airport displays. The same way you would order an Uber. Started by Instacart alums, AdQuick was born from their own personal struggles to book and measure ad success. AdQuick's audience targeting, campaign management, and analytics bring the precision and efficiency of digital to the real world. Ready to get your brand the attention it deserves? Visit adquick.com today to start reaching your customers in the real world. Again, that's adquick.com.

Let's say that someone with resources and talent was dedicated to solving this problem in academia or solving the sort of the stuff that we've been talking about over the next few decades. Where do you think the lever points are? Where's the opportunity to have the biggest impact here?

Yeah, that's a great question. So I've actually been involved with the University of Austin. which is this new university that they're trying to create in Austin, Texas, which is really focused on the core values that I think maybe you and I think are important. That is, of course, a very heavy lift. It's not easy to create a new world-class university. And it's also not easy to ensure that that university doesn't eventually just get captured by the same types of people that have captured all the other institutions of higher learning in the US. So I don't think it's an easy problem to solve. It's kind of honestly a generational type problem. What I hope will happen is society will start to realize that there is a problem with our academic institutions and some kind of populist sentiment will force some changes on the academy. I think if I were a billionaire or Elon Musk, I would probably devote some money to trying to just focus the spotlight on the problems in academia because the average person doesn't really know all the intricacies of how the university works, what's happening there, how things have changed since they went to college. And right now, the people who are trying to point that stuff out are a tiny minority. They're persecuted. And they're also easily kind of labeled as like, oh, these are crazy right-wing nuts. Whereas if you have high-prestige, high-status people like billionaires, tech founders, people like this who are committed to creating some mechanism by which we are just talking accurately, realistically about what's happening in the academy, and just making sure that that information is readily available for everybody to understand, like a parent trying to decide where to send their kid or a 25-year-old trying to figure out what to do with their career. That would have some, you know, eventual, I think, positive impact. But whoever that billionaire is or those billionaires are, they would have to put their reputations on the line because they're going to be attacked as right-wing loonies just for trying to shine a spotlight on what's actually happening.

I love what Austin is doing, what UATX is doing, and I admire those folks. My dream is that the Thiel Fellowship actually tries to compete head-on with Stanford or Harvard. I think that it's out-served its purpose as a, you know, program that gives 20 people a year. That, that's, that's great, but it's not really creating a dent in, uh, in, in the higher education system. And I think that that institution, what it does is it, it, um, it gets the sort of ideological diversity as a byproduct, but that's not its core offering. Its core offering is helping people start companies. Um, and it does it better than Stanford or Harvard. So people don't have to take a sort of step back, um, in their career to pursue some ideological, um, you know, or, or some sort of principled position. They're actually advancing in their career. Whereas UATX, um, well, it sounds like a great institution. It doesn't feel like it yet provides the same career opportunities as Stanford, um, for, for startups, whereas Thiel Fellowship does. Um, Now that's very hard to do, but that, that, that, that's my, my recommendation is that a university try to compete on the main value prop first, which is career life prospects, and then on sort of principles, pursuit of truth second. Any reaction to that? Hey, Steve, can you hear me?

Yep.

Okay. Where did you lose me?

I think you were comparing the Thiel Fellowship to UATX and pointing out that the Thiel Fellowship actually Establish— helps the kid establish a career path as an entrepreneur. And it's just, in a way, it's just as good. It gives you as much access to the ecosystem as, say, having graduated from Stanford, which I think I agree with all those points. The thing is, like, not every kid though is gonna, like, skip higher education, just be— and not every kid even just wants to be a tech founder or whatever. So the question is, should there be a—

Yeah.

you know, institution of higher learning, but just with somewhat different values, or maybe the old values, uh, that our institutions had before this kind of woke takeover.

I guess I'm saying that I think the Thiel Fellowship should become a college, um, and, and compete on for people who want to get into technology or work in startups in some capacity.

I think it could be. I mean, it's just a question of commitment because, um, you know, you could do something where like you create a new kind of degree program that has both a technical education component and a component which trains you to do startups and things like that. And, but again, it's everything is a heavy lift because at first it has no credibility and it's kind of weird. And people don't know how the credential is going to be judged in the future. How valuable is it for signaling? And, but I'm all for all of these things. I mean, that's why I participate in some UATX things. And I think I was even there for the big meeting for like kind of like establishing their constitution. So I'm all for this kind of thing.

Speaking of things that seem hopeless, I want to segue into the fertility crisis or underpopulation in general. I was having dinner with Tyler Cowen the other night and he was sort of echoing Robin Hanson saying that it seems inevitable that We're going to have an underpopulation problem and that there will be sort of, you know, economic, you know, significant economic impacts globally because of it. And what I'm curious about is if this is such a massive problem, why aren't there more people focused on it? And if more people were focused on it, what are the levers that could have an impact? It doesn't seem like any policies that we've tried have really moved the needle. And could you think of any either private or public sector or policy solutions that could address this in a significant way?

You know, okay, I think the politically incorrect way to say all this is that, you know, because traditional structures have broken down, you know, people don't necessarily value having kids all that much. you know, people are socialized to, you know, pursue the development of me, you know, well into their 30s or whatever, rather than the idea that, no, it's your duty to have kids and that's the right thing to do. You're a failure if you don't have a family and kids by the time you're in your 30s. That fundamental change in values is difficult to change, right? I mean, I don't know. I don't really know how even a government can go in and forcibly change that feeling. Around the margins, there are lots of things like making it cheaper to have a family, providing lots of daycare, free education, maybe even direct monetary incentives to have children. You could do things like that. And I think we'll probably see at least a few countries experiment with that kind of thing in the next 20 years. But there's this very fundamental thing that, like, look, I mean, what were these first-wave feminists saying to us? You're too young to remember this, but When I was growing up, the first wave feminists were saying, like, look, I'm not a baby machine. The highest aspiration for me in life is not, like, staying home and raising my children. When I was in elementary school, if even a single kid in my class came from a divorced family— this is totally true, Ames, Iowa, 1970s— if there was one kid in the class who came from a family of divorce, a broken home, the teachers would watch out for that kid. They would assume that kid is going to have a lot of trouble. Sometimes the other parents of the other kids wouldn't want that kid to be the best friend because obviously he's from a broken home. So the social stigma related to things which are completely normative now are 100% changed. How could that not have an impact on, you know, TFR and fertility? It's going to have an impact. Now, if you say like, well, okay, I can be a first-wave feminist. I'm actually pretty sympathetic to women's rights as advocated back then for sure. Some women want to have careers. Some women want to pursue higher education. Some women don't want to be stuck with all the childcare duties. Fathers in my father's generation came home, got their pipe out, put their feet up on the sofa and didn't do anything. And mom had been watching the kids all day and cooked dinner and did all the stuff, and dad could focus on his important research or business ventures. So it was just a totally different world. And we moved very, very heavily away from that world. And the idea that that wouldn't affect fertility is kind of crazy, actually. I fully would have predicted back then that it would affect the number of kids that people on average have. So, I do think, like, people are dancing around this. Like, I could talk to you about freezing eggs and stuff like this and just say, like, oh, we should make it free to freeze eggs. Every 17-year-old girl should freeze 100 eggs. And then even if she has a high-powered career, when she's 35, she'll thaw the eggs out and everything will be fine. And around the margins, that may be a good step toward improving fertility levels. But this fundamental question of families for most of human civilization were were built on the back of women's labor. Women were literally second-class citizens for most of the history of humankind and the history of this country. They were second-class citizens, and their labor was pushed toward the creation of an environment that was good for kids. All the women who should have been PhDs and corporate researchers in the '70s were teaching high school or elementary school, or they were librarians when I was growing up. So you had a very, very high talent pool of women who were forced because of the social values at that time to basically be looking after kids, helping kids. And so therefore, there were just more talented teachers around because of all these women whose careers were artificially stopped because of sexism back then. Now, that was terrible. I'm not really for that. But on the other hand, that created a very, very good environment for for a much better environment in some sense for kids, uh, than we have now. So I, I think people don't want to recognize what was actually happening through most of human civilization, and they're, they're, they're, they're, they're upset about a, a second-order consequence of, you know, equality of the sexes, and they're— but they're not willing to say like that was the main driver for what happened.

Let's say that we recognize that, um, sort of, you know, women, uh, being You know, more apt to be mothers was, was the reason why we, you know, had a better environment. Let's say we recognize everything you're talking about. What do we do now?

Right?

Sort of the, the sort of the cat's out of the bag. Is there a recommendation to elevate the role of motherhood so that more, you know, from a cultural perspective, so more women want to be mothers instead of, you know, sort of pursuing super ambitious careers at the expense of that? Is there, you know, you mentioned technology that helps people have kids, but really, it's— is it fair to say that the gap is not just on having the kids, it's on the raising of the kids? And we invest more in childcare technology or childcare labor, making it cheaper so that more people can get more help? Or where do you think the leverage points are to help address this problem, given where we are now and the norms that we have now, unless you feel those norms can be changed a little bit?

Yeah. So, I don't feel that comfortable being prescriptive, but I'll just throw some things out there. First of all, these fertility technologies can help. You and I are probably techno-optimists, so we feel like, oh, well, we have a problem here, technology will solve it. I'd like to think that's true. Maybe that's true. But, on the other hand, I think without changing this, I still have this nagging feeling that without having the society valorize and value the having and raising of children, you just won't get as many children out of the population. That's just like full stop, right? And, and, um, you know, they're actually trying to do this. Xi Jinping is trying to do this in China. Like, he is, he is— like, they're changing lots of language in the party documents which really emphasize really honestly the role of women, uh, in the family and things like this, and the role of masculinity and things like this. So they are trying to move the needle on these social issues, the values that undergird society. Now, it could just be that like maybe the next generation after the Zoomers will just kind of go completely the other way and they'll have very tradcon values and they'll just be like, wow, those crazy people my parents' generation, you know, they were nuts. I want to have a family. I don't want to date— I don't want to sleep with 1,000 guys off of Tinder. You know, maybe that will— there'll just be some kind of back reaction. It's, it's very hard to predict. I don't feel I'm particularly good at predicting social trends. And remember, we had crazy things in the United States. Like, we had Prohibition for a while. Like, can you actually imagine, like, circumstances that would allow us to have Prohibition? Well, similarly, things could swing completely wildly on, on these reproductive and gender issues. Um, you know, 50 years from now or 100 years from now.

Yeah, that's well said. Speaking of labor, I want to segue into AI, and it relates to what your startup is working on at the moment. You mentioned, and you've written about, how you were just in the Philippines. I have some friends who have some sort of offshore labor companies, and I myself am working on an outsourced engineering company. What is the right way of thinking about, um, you know, AI and labor and the right mental model for thinking about where human labor is going to be displaced, where it's going to be augmented, and what timelines, etc.? What's your, what's your mental model for thinking about this?

Yeah, so I, I thought I would— it would be good to illustrate it with a little narrative instead of just giving some kind of abstract analysis. So, uh, I was in Manila for about 10 days, and that's where the calls— that's one of the call center capitals of the world. So 8% of the Philippine GDP is this business process outsourcing or call center type work because they speak English there and labor costs are pretty low. So 8% of their GDP, $40 billion a year, is that kind of work. And my startup builds large language model AIs, but within an architecture that suppresses hallucination. And so we can plug in, in effect, plug in a memory to the LLM and force it to Answer questions in a constrained way, so it's sticking to the content of what's in the installed memory. And so we can take like the training manuals that they use. Like one example that where we built you know possibly the highest functioning narrow AI of its type right now with a global consumer brand, consumer electronics brand. We built an AI that can basically support troubleshoot three hundred different models of TV. that they sell. And this thing has as its memory every product manual, every troubleshooting script that they've developed over 20 years as a company for troubleshooting these TVs. And the AI will just perfectly— we've tested it on thousands now of queries and things like this— it will perfectly troubleshoot your TV for you as well as a human could do it over the phone. And the fact that we're a small startup is able to build something like that just tells me that huge a huge fraction of his labor in the Philippines is going to go away because it's much, much cheaper to run the AI than to have humans paid to sit there with the— you know, the number of resolutions of a problem like the kind I described, which a human can do in a day, is on the order of 10, maybe 15. So just imagine how many, um, the AI can handle, uh, just, you know, simultaneous calls and just crazy, you know, crazy numbers. So the whole time I was in the Philippines, I was just looking around saying, wow, this entire building, which says American Express on the side, the entire work of the people in that building is dealing with weird account chargebacks and sorting through documents and things having to do with your Amex card. And next to it is this, I think, JP Morgan building, which all they're doing is back office work for JP Morgan. maybe not all of that, but a very big chunk of that work is fully capable of being done by an AI now. And so it gave me a very direct visceral feel for the near term, not like 5 years from now, but near-term impact of AI on very large sectors of human employment.

The last thing I heard you say was visceral.

Yeah, visceral impact, because we were meeting with the BPO companies in the Philippines and, you know, management, you know, top-level founder of the company, management layer. And then also like in the background, we could see them training people and people at desks with headsets, you know, doing all kinds of stuff. And it was just mind-blowing thinking like, because in the meeting we would demo, we demoed some of our AIs for them. And we said, well, here's an AI we built that can answer any questions related to the Google Fi. Google Fi has like a data and calling plan. It like competes with T-Mobile and stuff like this. But this AI can answer any question about the calling plan, how to switch your number, what countries does it allow roaming in. It just basically perfectly answers all these questions. And we could see the eyes of the BPO people that we were meeting with just get bigger. And they would say, well, ask it this, ask it this. And then, you know, because they thought, oh, this is going to be a tough question. You know, just perfectly answer the question. And you could just see the wheels spinning in their heads, realizing that this is going to have pretty significant impact on their industry. It's a little bit like, you know, you— I don't know what the right analog is, but, you know, you go into a car, you go into a factory, and you're like, you have a little robot and you show them the robot and the robot is like welding stuff and its arms are moving so fast you can't see it. And they had never seen a robot like that. And then you watch like their reaction in real time in the meeting.

And so say more about what that means in terms of like when we look out a decade from now, what are the types of things that are going to be replaced? What are the types of things that are going to be augmented? What's the right way of thinking about sort of where this all goes?

It is super hard to extrapolate because, as you know, so many resources are being put into further model training, making the fundamental models better. Our company has built an architecture that allows us to use those models and suppress the hallucination of those models, focus those models on a narrow corpus of knowledge. But, um, in the short run, it's very easy to say, like, any particular task which, um, you teach a human to do by— they come in and they, like, they have to read a manual and they sit in some classes and, you know, they, they develop the ability to support the Apple iPhone or something, like at the Genius Bar or something like that in the Apple Store, anything where that information is codified so that an AI could read it or even look at pictures or video of people doing it, that AI will be able to do it. And it may not have the mechanical dexterity to do it with a robot hand, but it can certainly answer any questions you have about how do I delete this set of files on my— or uninstall this app, or my phone is frozen. It can do as well as a human on all that stuff. And so that near-term thing is, I think, unquestionably going to happen. Longer term, it's much harder to project what is going to happen, where this is going to stop. Some law firms that we've met with, they claim that They're getting roughly already just using just plain old GPT without any special fancy stuff like what we've built. They're getting easily 20%, 30% efficiency gains in their associates. So basically they can get by with 20% fewer associates now. And so that's just the near-term stuff that you can easily extrapolate, but the longer-term stuff is just much harder to understand.

Yeah.

We'll continue our interview in a moment after a word from our sponsors. Hey everyone, Erik here. In this environment, founders need to become profitable faster and do more with smaller teams, especially when it comes to engineering. That's why Sean Lanahan started Squad, a specialized global talent firm for top engineers that will seamlessly integrate with your org. Squad offers rigorously vetted top 1% talent that will actually work hard for you every day. Their engineers work in your time zone, follow your processes, and use your tools. Squad has frontend engineers excelling in TypeScript, React, and Next.js, ready to onboard to your team today. For backend, Squad engineers are experts at Node.js, Python, Java, and a range of other languages and frameworks. While it may cost more than the freelancer on Upwork billing you for 40 hours but working only 2, Squad offers pre- Get premium quality at a fraction of the typical cost without the headache of assessing for skills and culture fit. Squad takes care of sourcing, legal compliance, and local HR for global talent. Increase your velocity without amping up burn. Visit choosquad.com and mention Turpentine to skip the waitlist.

Do you think it's going to be like when we, you know, when, you know, 90% or whatever of people were farmers and then sort of the revolution meant that those people then shifted to, you know, manufacturing or something else? Do you think it's going to be that degree of a shift? And if so, what industries could you predict are ballooning as a result? Like, what are people going to do?

I think I can predict that UBI counselors will be in high demand, people who will help you sign up for UBI and help you allocate your UBI check every month. And gee, now that I've said that, I realize that an AI can probably do that better than the model human. So maybe even that job category isn't going to be there. I'm sort of kidding, but, you know, it could even be the case. I would not be surprised if in, say, my field of theoretical physics, we get to a point where I can go in and have a conversation with a specially trained model that's been trained on the archive and all this other stuff, all the textbook, all the— thousands of physics and math textbooks. And I can have a conversation with it where it understands what I'm saying. And then it just gives me a solution or it gives me a research direction. It's like, well, you should look at this. Or, hey, you know, actually, Steve, did you know like this Russian group proved this actually in 1997? I don't know if you're familiar with this line of work. Here, I'll summarize it for you. I don't think we're very far from that actually. So even my own like day job is going to be affected by this.

Yeah. What's the mental model for thinking about day jobs or industries that are less affected versus more? Like, what's the safest business to be in or type of work to be in?

Very tough to say, but things where you're being— I'd like to say that the sort of like cliché answer is the things that value your kind of human qualities, like maybe an entertainer, sports star, um, that kind of thing is, is not going to be replaced, uh, easily. But on the other hand, you know, the success of things like AI girlfriends and AI porn, stuff like that, um, maybe says like even demand for that kind of stuff is going to go down. You're going to be outcompeted by AI versions of stuff. Just to come back to the Philippines for one second, I'm not answering your question because I don't really know the answer. But just to give you another example from the Philippines, we found out a lot of these BPO people are actually doing like outbound sales. So like they're being hired to like screen, like call people, see if they're interested. And if they're interested, then, you know, you can hand it off to somebody who's, you know, maybe even better. But that early stage stuff can easily be done by a kind of AI avatar. And they have very good AI avatars like in China. A lot of the online shopping is done in a kind of way where there's like an influencer person who's like demonstrating a product and telling you why this is a cool product. And then they're like little orders on the screen, people are ordering it and they get a discount. So there's a lot of like sales like that, but you can now basically have them construct. So you can take the top influencers and you can construct an AI version of that influencer and just give it a script. And it's actually really good. I was watching some of the examples of this stuff. And these guys can sell product. They really can do a good job of capturing the character of that influencer and then make some good remarks about why this— wow, this is a really great cup. It doesn't spill. It keeps your drink warm, et cetera, et cetera. I've had it for 5 years. It's not broken. I think even that kind of job can be done by an AI relatively soon. So obviously, the main friction I'm talking about the potentiality of the technology. The main friction is the people who currently control the levers, which is like mid and high-level managers at these existing companies. It will take a little time for them to accept what I'm saying to you. They have to see examples, and then they have to see it, they have to deploy it, they have to get themselves organized to run this big project to deploy it, and then they have to— so the main friction in the system actually is that kind of adaptation. It's not, it's not the tech. So again, like the AI that we built that can fully troubleshoot your TV for you, that could— that actually, to be honest, that is being— I can't say much about it, but that is actually being deployed right now. But the slow step is people not believing it's possible or not quite knowing how to reorganize their existing enterprise in such a way to adopt these new technologies. That will actually be the bulk of the multi-year delay that it takes for all this stuff to become universal. Yeah.

You know, we were just alluding to Bev Jezos before, before our episode. Where do you stand on the sort of, uh, AI safety versus accelerationism, uh, perspective in terms of, you know, your, your concerns or, or hopes?

I, uh, I think I, I may have a very unusual and possibly disturbing view of all this stuff. I think there is a pretty good chance that we'll create minds which are really much, much better, like orders of magnitude better than ours. Not maybe not far in the future. I don't, I don't know the exact time scale, but it's looking very, you know, plausible over just in the last, the last couple years have moved my priors a lot. Um, but here's the thing. If you actually understand what these large language models are doing, we've built a structure that in a way compresses an understanding of all the concepts that humans have developed, that you and I use to communicate through writing or whatever. We've built a thing which compresses all those concepts together. It understands the relationships between them. It can manipulate them a little bit. It's getting better at that. And in a way, because its first purchase on the universe is through human characterizations of this is a ball, this is a cup, this is love, this is smart, this is dumb, this is a bear. It didn't— it's not a biological being that stumbled for millions of years through our universe and absorbed maybe its own different view of things in the universe. It basically just absorbed our, compressed our views. of the universe. And so in a way, all of these things are our descendants. These are minds which descend from our minds, and they'll always carry that— at least this line of development in AI will always carry the human seed in them, because that is how they got their first purchase on the world. And so in a way, if they replace us, I'm not as disturbed as maybe some other people are. And another way of saying this is that as a physicist, someone who actually works in cosmology, I often am thinking on billion-year time, literally billion-year or 10 billion-year timescales in the universe. And I never had any thought that some crappy ape-like thing that just happened to evolve, you know, on the edge of the Milky Way galaxy. You know, was going to be present for even one of those billion years, was going to persist for even one of those billion years. We'd be lucky to persist for a million years. And obviously, if we could persist at all for those timescales, it would probably be our descendants and maybe our in silico or whatever substrate comes after that descendants that would survive. And it wouldn't necessarily be a guy with cells and DNA and, you know, a long memory of the African savanna. You know, I just, you know, I don't care that much, honestly. So it was bound to happen. So it's something like some people are just coming to this thought right now, like, oh, AI, oh, AI risk, extinction doomers. Whereas if you're a physicist, you're already thinking about these billion-year timescales. You're like, well, humans are not going to play a big role in this billion-year timescale history, except if we solve AI and build von Neumann robots and stuff like this. And, you know, that's the way that we're going to influence the future. And this is a step in that, in that direction.

Yeah. I, um, well articulated. I want to bring up perhaps the last topic that you've written quite a bit about the last couple decades, which is the, the US and China relationship, and particularly the, the rise of China. You know, there's the, the Peter Zeihan perspective, which is a China doomer, which is basically saying, hey, China, for a number of reasons— demographic reasons, energy reasons, sort of its own, you know, economy, real estate, credit crisis, et cetera, is going to sort of have a collapse, he believes, in the next, you know, decade or two. And then there's the perspective of, hey, China's governance is incredibly strong. Their technology is incredibly, you know, advanced. They're, you know, incredibly high-Q people. They've, you know, incredibly long-lasting and flourishing civilization. Where do you net out on what— it seems like you're bullish. Why are the doomers incorrect? Or when you look at the future of China, how it relates to US, what's your mental model or worldview?

Yeah, so I would first, before taking a side on these issues, I would just make a kind of meta remark, which is that most people, especially Westerners, who think they know what the current situation is, in China or the US-China competition generally are very poorly informed. Like, I'm not— and I'm not saying like this is like some dull think tank guy in Washington or some, you know, UCLA professor who doesn't know anything. Even, even like tech bros, you know, or very high IQ, you know, very sharp guys, like people whose names you would know, like, you know, uh, famous founders you know, billionaires, etc., they often are way overconfident in their understanding of the situation, uh, relative to their actual knowledge. So the first thing I would just say to everybody is try to get a little smarter about this because it doesn't hurt you, right? Try to get a little bit smarter. Try to look at sources outside the usual sources. Like, you can go on YouTube and there are plenty of vloggers just going around China doing stuff. You can just see, like, I knew that Chinese electric cars were good long before everybody else because you could just see people, like, testing the car and talking about the car and what can the car do. And, you know, so I think there should be an effort to try to get a little bit closer to base reality on a lot of these issues rather than make kind of, like, easy, lazy generalizations and then try to reason from that. Like, oh, these people are communists, therefore their economy will eventually fail because didn't that happen to the Soviet Union? Right? It's a very kind of lazy, sloppy chain of reasoning. It could be that Xi moves them too far away from market forces and fucks up their economy and they collapse because of that. It could happen. But the kind of lazy reasoning that that is definitely going to happen— for that, you might want to look at like what fraction of economic activity in China is actually run in a market-oriented way versus a state, uh, controlled way. Like, you could look at those numbers a little more carefully. You could look at success stories— alternative energy, solar panels, batteries, their space program, their J-20 stealth fighter, their actively electronically scanned radar arrays in their missiles, which we don't have. Um, you know, all these things, you could look at a little more carefully at these things to try to understand, like, what— how did that happen? How did they just steal all that? Oh wow, they're— they must have the best cyber warfare hacker brigades ever because they stole all of— why, we don't even have it and they stole it from us. It's just amazing, right? So most of what I hear is just very poor quality analysis. Now, of course, it's true that most people are not technologists themselves who have spent quite a lot of time in China, know lots of Chinese people, have access to original Chinese language sources. Most people don't have that. But if somebody walks in the room who does have that, you might want to just pay attention to what they're saying. You might just want to like listen to what they're saying and decide like, oh, that guy's still full of shit. Or maybe like, wait, he's saying a bunch of stuff which— wait, he just said they brought the cost of LiDAR down by an order of magnitude within a few years. I don't really know what LiDAR is, but apparently it's important for autonomous vehicles or something. Well, how did they do that? Did that— Did they steal that from us? What? So I, I just think there's a kind of intellectual laziness when it comes to China. You know, it's a little bit like, imagine like tomorrow a huge alien spaceship just like appeared over New York City and was just hovering there. And you're like, wow, those are aliens. But yeah, they, they look kind of nice. I'm sure they're not going to do anything to me. And then like you don't pursue it any more than that. You're just like, oh well, wow, the spaceship is like kind of like a big circle. You know, that's the level of your analysis of, of the aliens.

Like, well, how—

like, what is their technology? What can they do? How fast is their technology advancing? What are they doing to us? You should have a little bit of curiosity because it's, it's actually going to affect your life. Like, um, the US-China relationship and competition is clearly the most important thing that's going to dominate what happens in the world for the next few decades, at least. Even if China collapses, it's still going to affect Americans very directly, probably. So anyway, I just encourage people to be more epistemically open to the situation. There's really a lot to learn because it's a— their system is clearly not our system. What do you know about their system? Well, how many engineers graduate from Chinese universities every year versus not just U.S. but even rest of world? Some of their numbers They're comparable to rest of world combined on some of their numbers. So, well, do I know those numbers? Should I look at those numbers? Um, how good are their engineering programs? How good are their engineers? Why were they able to build an entire big tech ecosystem? Like, we're very proud of our big tech ecosystem. Like, if I talk to a European, I can just laugh. I could say, yeah, you guys have no tech companies. You guys have to use Google products and Meta and all that. You guys have nothing. But then I look over at China and I realize like they have everything that we have and they actually have better stuff. Like Elon wants to build X, but X is just WeChat. See, Elon spends a lot of time in China, so he knows how powerful the WeChat platform is. It can do everything. It's an everything app. And he realizes how powerful that is. He's trying to build Twitter into an X, into a WeChat-like thing. So, um, but most Americans are just totally, uh, you know, ignorant of the fact that they built a comparable big tech platform which actually works better. Like, if you go there and you're scrolling— like, if you're scrolling down their version of TikTok or something, the videos load super fast, there's never hiccups. It's not— it's not like, you know, even like pre-Elon takeover of Twitter. Like, Twitter was like much shittier technology than what you find in these kinds of equivalent apps in China. So You could just, you could say, wow, they built all that. And then at the same time, they're doing all this industrial stuff with like robots and, and building huge ships. Like, wow, their, their shipbuilding capacity is like 10x the US shipbuilding capacity, or, you know. So I would just encourage people to look into it. I'm not trying to tell them like, oh, it's hopeless, China's going to win. Not claiming that at all. But if you're not putting some probability density on them being a very formidable competitor and increasingly formidable over the next decade or two, then you're very foolish. I would just put it that way. Yeah.

And, and so you're not concerned about the, the, the demographic collapse or the, uh, energy, uh, uh, sort of situation? I believe it gets like 25% from Persian Gulf or, or, or something. Like, in a world where the US doesn't patrol the seas anymore— and this is Zeihan's argument— You don't think that's—

Yeah, the demographic number, that's a good test of what I was saying about some people just not being epistemically very careful. So it's very easy, like anybody who, like, say, like, I don't know, managed to pass their first year of engineering in college or something can go and run the numbers on their demography. Like, it's amazing how many few people have done this. I actually have tweeted about this repeatedly. So they had a drop in TFR recently, which was kind of around the COVID period. And at that point, it got to a point where if it's not fixed in 20 years— so, so we're talking about kids born today that will enter the economy 20, 25 years from now— if that drop is not fixed, they are going to have a serious problem. But the kids who have all been born— so all the kids that are going to complete their education, enter the economy in the next 20, 25 years, They're pretty much born. So you don't have to guess. You can actually just do these calculations. Okay. And they don't have a problem for the next 20 years. So that's, that's the bottom line. They don't have a problem for the next 20 years. Furthermore, you can look at more subtle things. You can say like, well, look at South Korea during its phase of fastest industrialization growth, like say '80s, '90s, early 2000s. And you'll notice their working-age population was shrinking that entire time. So, as you're climbing the technology ladder, you can still make tremendous progress, even if your working-age population is not growing. The direct linkage between demography and economic advancement totally ignores the role of technology. It's kind of funny to hear tech bros quoting this stuff when they themselves should know that, yes, it is a problem if your, you know, TFR drops by, you know, to half. And, you know, some decades from now, that will be a huge problem. But having a slight decline in your working-age population, uh, isn't necessarily a problem if at the same time the college-going rate jumps from 1% or 3% of the population to 60% of the population, and the STEM— the number of trained STEM people jumps by an order of magnitude. during that period of time, even though the overall working age population is dropping, the productivity per person is still going up, like skyrocketing. So it's actually very comical. Like just the way people reason about this is totally comical, unless they say, but they never say this, I am extrapolating 30 years out and I'm assuming no change in their TFR during that period. Okay, fine. But then we're fucked too. All Western European countries, the United States, we're all screwed. If you analyze our situation with no change in TFR over the next 30 years, we're in trouble too. So the big winner will be, by demographic reasoning, Nigeria and maybe India. If you just extrapolate what's happening now, those are going to be the world superpowers. But there are other factors which I don't necessarily want to talk about, but there are other factors at play here than just demographics.

Right. And so extrapolating forward again, Is a bipolar world where there's a Western power and Eastern power sustainable, or how does this all play out or come to a head?

So I think it's totally sustainable to have a kind of US Anglophone bloc and a China bloc. Um, and the China bloc could even be more powerful, like it could have a higher GDP. But, you know, the US has so many things going for it. You know, we have 2 big oceans. We will probably always be a technology leader. We suck brains in from other parts of the world. We'll have a formidable nuclear and military establishment. I think if the US plays its card right— cards right— it doesn't have to be existentially threatened by the PRC, but it does have to accept that it's no longer the sole hegemon. Like, all the thrashing around that we're doing right now is we refuse to accept That we're not going to be a predominant single hegemon. And it seems like some factions within our national security establishment are willing to blow up the entire world rather than cede, right, that position of sole predominant hegemon. That's literally the thing, the crazy thing that's happening in Washington. Whereas if they said like, no, we should actually strengthen ourselves, strengthen the US economy, the US human capital, all that stuff, strengthen ourselves, we're not at existential risk from these guys yet. Now, maybe, you never know, like AI could take off in some weird way, you know, maybe they get way ahead of us or something, but we could also get way ahead of them. Um, you know, those other things could happen, but the idea that, oh, if, if any other power arises on Earth which could challenge the US, we must then make it our number one priority to contain them, suppress them, risk a nuclear war with them. That just seems crazy to me. That just seems like a kind of very stupid calculation.

That's a good articulation of what's happening. I want to wrap by asking kind of a summary of what we asked, a summary question in the sense of, hey, we've talked about AI, we've talked about fertility, we've talked about China. I'm curious how you think about how to spend your time and where to have the biggest impact given your interests. And you've identified several challenges, several problems, where things are going. It's a question a lot of people who listen in is like, where do I start? What's the biggest way to— what's the most important thing to work on? How do you think about that for yourself?

Yeah, I can only say it for myself because obviously the position that I'm in at this stage in my life, my cognitive profile, all these things are somewhat unique to me. They're not applicable to the average person or the, you know, someone who's listening. So I, one thing that will shock a lot of people is that if you're a theoretical physicist, you're often thinking about very deep problems, which have no practical impact on the world. So I just spent the last couple of years working very, very hard on this, something called the black hole information paradox that was first formulated by Hawking, like you know, 40 years ago. And that's something which is very important for the understanding of something called quantum gravity and internal consistency of our fundamental theories of physics. And, but the number of people that I could explain to on the whole planet is, you know, on the order of hundreds of people. So is that a good allocation of my time? Like, you know, a lot of people, the people who are working at my startups would say like, Steve, quit that black hole shit and get over here and help me with this problem, right? So I would just say my own personal budgeting of time and effort is not really a good model for anybody. There are physicists who would say to me, like, Steve, why are you fucking around with these startups when you should be, you know, doing more stuff in fundamental physics, right? So it's very tough to make everybody happy. But for me personally, my own philosophy is I try to have a balance between these deep questions, which I've really been interested in since I was pretty young, which have predominated through my whole life, things related to math and physics and things like that. Some AI questions are actually in that category. But then the other thing I try to do is I look at areas where, as a startup founder, I can have a big impact, where there's clearly big changes happening. Like, it could be large language models, it could be embryo selection. One thing we didn't talk about today is the startup that I co-founded called Othram, which has totally revolutionized the way that crimes are solved in the United States. So, If you get even a few cells of DNA from a crime scene, we can typically narrow down the identity of the, say, killer down to an order of 100 people right away at a cost of a few thousand dollars. So, and we've solved many, many cold cases now. So famous cold cases that were unsolved for 30, 40 years, even some hot cases like the— I have to say this very carefully. It was reported in the New York Times that Othram was responsible for the identification of this Idaho killer who stabbed the 3 roommates to death in one night. It was reported that we are the ones who found him after the police searched fruitlessly for weeks for him. And that was a hot case. That was not a cold case. So that's an example of like completely changing the way crime solving is done. And I just saw that opportunity as something that technology allowed it to be, to happen. So I wanted to work on it. And, you know, gradually it is going to change law enforcement quite a bit. So for me, there's this balance between impact in the real world and really deep problems, which in a way, it's like a personal luxury to be able to work on these problems in my spare time.

Yeah. And do you have any more requests for startups, so to speak, which is, let me say, if you weren't constrained by capital or talent, and you could redirect people to work on interesting problems, or problems that you find important that are underserved at the moment, where would you point people to? Or where would you deploy, you know, more capital, more talent, besides the things you're already working on?

That's a great question. I would say that In so many resources are going into AI right now. It's hard to advocate for more. But what's interesting is that most of the money is going toward just training better models, bigger models. Just adjacent to that world, there's a bunch of theoretical questions about how the models work. Are there further optimizations of the actual training algorithms or the methodology or the structure, even the architecture of the models? There's a bunch of work there that I think could use a lot more effort. And maybe if there were a public, like government, like, you know, in the US we have these national labs, which traditionally have been very strong in like physics and also like supercomputing and stuff like that. If I were the czar of the Department of Energy or the National Science Foundation, I would allocate a bunch of money for somewhat more theoretical AI adjacent stuff.

Hmm.

Which helps us understand AI better rather than just racing toward the creation of, you know, an AGI. Try to understand theoretically some of these adjacent issues. And I think that would pay off more for AI safety and things like that. So the government should be allocating large amounts of money to that, but they're not yet doing so.

I think that's a good place to, a good place to wrap. Steve, for people who have enjoyed this conversation and want to go deeper into your work and understand what you're up to at the moment, where can you point them? You have a fantastic blog. You have, you know, people should follow you on Twitter, et cetera. Any other plugs that you have?

I'll plug my own podcast called Manifold.

Yes.

Excellent. Which I try to interview, you know, a variety of thinkers. I think, you know, it's got a small following, but I think the people who do listen to it enjoy it. And it's somewhat unique because I cover a pretty broad range of stuff and sometimes in the podcast we'll go pretty deep into a topic. So I recommend that.

Yeah. A must-listen. I've enjoyed a couple dozen or so episodes. Steve, thanks so much for coming on the podcast. It's been a great episode.

Upstream with Erik Torenberg is a show from Turpentine, the podcast network If you like the episode, please leave a review in the Apple Store. Hey everyone, Eric here. At Turpentine, we're building the first media outlet for tech people by tech people. We're the network behind the show you're listening to right now. We have a slate of hit shows across a range of topics and industries, from our AI and investing cluster of podcasts to shows like The Future of Work, Shows that drive the conversation in tech with the most interesting thinkers, founders, investors, and influencers, like Econ 102 with Noah Smith. We're launching new shows every week, and we're looking for industry-leading sponsors. If you think that might be you and your company, email me at erik@turpentine.co. That's erik@turpentine.co, and let's partner together.