Justin: So I think many people outside of the academic system don't really understand the rankings. It also has to be split between like undergrad and graduate. So, so without a doubt, Tsinghua undergraduate students are the best in the world.

I would say on, on average much better than, than the best universities in the US. But this doesn't actually mean the, the education quality itself is higher. It just means this, the filtering system, they can filter better students. But then for graduate it's still without a doubt that the front like most of the frontier research is still in the US although China is catching up very quickly. That's why we see that most of the PhD students in the US are actually Chinese because they're high quality students.

Steve Hsu: Welcome to Manifold. This is a special episode. We're here at Tsinghua University in Beijing, China. I have four guests in the room with me today. They are all Americans with some knowledge of this university.

First, we have Han Feizi, who has been a guest before on the podcast. We'll link to that episode in the show notes. He is a resident of Beijing. He is a former banker and also a novelist, and he is going to be part of the conversation, but he's not the main focus of the conversation. Welcome to the show, Han Feizi.

Han Feizi: Hello, everyone.

Steve Hsu: The main guests today are three individuals who are actually part of Tsinghua University. First, we have Gabriel, who's an American and who is finishing his undergraduate degree here. Gabriel, say hi.

Gabriel: Hello.

Steve Hsu: Next, we have Justin, who did his undergraduate degree in the University of California system but is doing a PhD here in AI-related research here at Tsinghua. Justin, say hi.

Justin: Hi, everyone.

Steve Hsu: Finally, we have Alex, who is a professor, also AI researcher here at Tsinghua University. Alex, say hi.

Alex: Hello.

Steve Hsu: Okay. We're actually recording in a research seminar room at the Yau Mathematical Sciences Center at Tsinghua.

And Yau as many of you physicists will know, it's the Yau of Calabi-Yau manifolds, and he won the Fields Medal and is one of the great geometers in the history of mathematics. But we're, we're borrowing a room in his research center right now. The acoustics, I think, might be less than ideal, but hopefully, my engineers will fix this in post-production.

Steve Hsu: So Gabriel, let's start with you. You are finishing your undergraduate degree here -in computer science.

Gabriel: Yes.

Steve Hsu: Tell us about your journey to Tsinghua. Why are you at Tsinghua and not at a US university? How has your experience been here?

Gabriel: To preface, I think it's important to state that I grew up in Hong Kong. I went to international school in Hong Kong and yeah, so most of my classmates, they went abroad to do university or just stayed in Hong Kong. So the main options were Hong Kong, Canada, the UK, and the States. I also applied to the States during my undergraduate application season.

In my opinion, it's it's been trending upwards that more people are considering Mainland universities. And my personal reasons were, one, yeah, I just one aspect was I have never got an education in mainland and also like language-wise, I was much weaker in Mandarin. So I felt like not only was Tsinghua, is Tsinghua really good for the field I'm interested in, so computer science, but also another aspect of it would be I would be able to learn about China as a system and just being in that new environment.

And I think, yeah, now four years down the line, I definitely see that I got a lot out of my college experience here.

Steve Hsu: Let's start with the basic thing, your Mandarin language capability.

Gabriel: Yes.

Steve Hsu: So you probably studied it at, in high school

Gabriel: I did

Steve Hsu: through the Hong Kong system. It sounds like you feel you weren't fully fluent when you went off to college.

Gabriel: Definitely not. I studied it actually as a first language. I did the IB system, so International Baccalaureate. I technically got a bilingual diploma. We only kind of practiced writing and reading, whereas spoken Mandarin, because the, like, playground language in my school was just English, so I never really got the opportunity to really converse that much in Mandarin. That was lacking.

Steve Hsu: And how was the Were the classes you were taking here in English or in Mandarin?
Gabriel: When you say here, Tsinghua?

Steve Hsu: Tsinghua.

Gabriel: Ninety-five percent of my classes were in Chinese Mandarin.

Steve Hsu: And was that a shock to you?

Gabriel: Yeah, first year was pretty hard. That was one of the biggest challenges I faced especially my first year. I, I realized later on that it wasn't nearly as big a problem as I thought because your language skills improve a lot when you start using it daily, and I just never had that in Hong Kong.

Steve Hsu: But you could already read?

Gabriel: Yes, but like slowly.

Steve Hsu: Okay.

Gabriel: Yeah.

Steve Hsu: Was it an active effort to get your, like, the number of characters that you could recognize up to what a typical. I mean, these are some of the

Gabriel: Right

Steve Hsu: top students in China, right?

Gabriel: Right.

Steve Hsu: This is like the average student here is kind of like one in a thousand

Gabriel: Yes.

Steve Hsu: talent level.

Gabriel: So to preface, I think my ability in Chinese was so I could converse with people, but it's like when I first came, it was so awkward to the point where they'd say like three sentences and I'd have to think about what I have to say. Like, I wouldn't be able to form like a multi-line dialogue, and I would actually have to think really hard about what I was saying.

But I realized like two months down the line, just because it was really hard, but after two months it just got much more natural for me.

Steve Hsu: And in terms of like lecture notes, like if you got some fat set of lecture notes all in Manda all in Chinese characters, what how did you handle that?

Gabriel: Yeah. So pre-GPT era, so that's when I first entered college because, yeah, I used a lot of translation software, but also you just really had to try really hard just reading.

Steve Hsu: Wow.

Gabriel: Yeah.

Steve Hsu: So we'll, we'll come back to this with some of the other guests here today. But if you compared notes, and the ideal case would be someone you went to IB high school with in Hong Kong that you know well, that went to a Western university to also study computer science. Is there a case like that where you can compare notes about what you learned here and what the experience was like compared to your friend?

Gabriel: When you say compare notes, like physical notes or just like talking about what we studied?
Steve Hsu: Just talking about what you learned, what the activities were like, research projects. Oh. Just comparing your two

Justin: Okay.

Steve Hsu: CS educations.

Gabriel: I reckon I have a friend at NYU who did undergrad for CS and als another one at Boston.
I think course-wise, just like the courses we took, like computer networks, computer architectureoperating systems. So these courses we all took separately, and we all kind of know the basics of it. But regarding like opportunities, they're vastly different.

As in, in Tsinghua, if you just want to dip your feet into research, you kinda just text the teacher. It's, it's that simple. Whereas in NYU specifically, I know it's actually a really competitive opportunity, and most people graduate without even getting into it at all. Yeah.

Steve Hsu: So you were able to get involved in research when you needed to?

Gabriel: Yes, really easily. Yeah. Tsinghua has a thing called, student research training, and that's advertised to undergraduates to just test out the research environment.

Steve Hsu: Yeah. Got it. Now, my understanding from talking to CS graduates in the US is that generally the hardest courses for them are maybe like a senior level machine learning class

Gabriel: Mm-hmm

Steve Hsu: and because there's a lot of statistical methods and stuff like, or statistical thinking that they're maybe not used to. And then also there's usually an algorithms class that maybe uses the book by Rivest.

Gabriel: Mm-hmm

Steve Hsu: Those are typically challenging. So there are a lot of people who are good at, at least this is a, you know, pre-Claude Code era, but there are a lot of kids who are actually good programmers, but they never really fully understand the material in, say, the senior level machine learning class, so they never really fully understand all the algorithmic thinking, which, which is kind of basically discrete math in that course.

Do you have any experience like that? Like, are the Tsinghua kids all really good and they can master that material easily?

Gabriel: I went through college with AI, right? I know first, firsthand 'cause all my friends are, like, Chinese. I know, I lived with Chinese people in my dorm. We just use AI for everything. I can't say that they know all the material, like, from a very basic level.

Steve Hsu: Yeah.

Gabriel: I think, yeah, the AI use and, like, the just understanding the overall concept, but maybe not, like, the specific details, I think that's universal.

Steve Hsu: Wow. Okay. So the, the thing I was trying to get at is you see, at the more elite US schools, all the CS majors can really understand the more mathematical stuff.

Whereas in a lot of like standard US CS schools, you might be a good programmer, but you actually don't understand the math. Oh, okay. And you, and you just like look back at your college CS degree and be like, "Oh, that was all useless stuff that I don't actually use now working at Oracle."

Gabriel: Right.

My question was like, oh, would Tsinghua be different because the students are so highly selected? But it sounds like because of AI that my question's like totally irrelevant now.

Steve Hsu: Right. I, I get what you mean. I think for a lot of the classes, yeah, we were forced to learn like, theoretical math, like the basics, like discrete math and stuff. yeah, ch- students at Tsinghua pick on, pick up on that like really easily.

Gabriel: Yeah. We also had like hard machine learning courses and like AI, and yeah, they, they got the fundamentals down really easily. But I think I was more implying as a sense of like, since AI progresses so quickly, and a lot of times like these new methods we don't learn in class, but regarding those methods, like for example, what GP has done for large language models

Steve Hsu: Yeah

Gabriel: those aren't covered even in like basic AI classes. Regarding those techniques and stuff, yeah maybe it's not that Chinese people can't understand it, but it's just because we didn't learn it in class.

Steve Hsu: Yeah.

Gabriel: Yeah. So we didn't really get into it.

Steve Hsu: I was just reflecting on my experience in industry dealing with a lot of coders who maybe often they don't have a CS degree, but even if they do have a CS degree, these sort of quote, "hardest classes" that CS majors take, a lot of them basically didn't master it and totally forgot it after being in industry for a while.

Justin: Like they just memorize.

Steve Hsu: Yeah. They got through the class somehow, but they didn't really when you start talking to them about it in the real world, they don't really actually know what you're talking about.

Gabriel: Mm-hmm

Steve Hsu: But like I would guess most MIT CS majors actually do really master that material as undergrads and remember it years later. So that's, that's the distinction I was trying to get at.

Gabriel: I've never talked to an MIT student, but like, I would assume that a lot of us, like even at the top universities yeah, just there's a big population of them who just don't bother understanding classes that deeply. Yeah. Because really a lot of people, the attitude in industry is like, "Yeah, we got the degree." That's all you can

Steve Hsu: Okay.

Gabriel: Yeah.

Steve Hsu: That's a, that's another thing physics physics people would find weird about CS because it's got such a career

Gabriel: Yes.

Steve Hsu: driven approach. Like a lot of the people are not as deeply interested in the core concepts and are maybe they're just looking for like a signaling kind of certificate that lets them get the job.

Gabriel: Right. Right. I think in computer science at least, yeah, a lot of us just The main goal is the degree.

Steve Hsu: Yeah.

Gabriel: Whereas like the underlying architecture, it's actually kind of rare to see someone that in tune with understanding like the very like lower level of everything. Yeah.

Steve Hsu: Yeah. Someone told me that like one of the best CS curricula in the US is the Berkeley one because it actually forces you to like actually understand how to build the machine from, you know, very low level considerations all the way up, and not all the some of the top CS programs don't necessarily force all the students through that that sort of conceptual roadmap.

Gabriel: I mean, I know the computer science department, one of their, for undergrad, their operating system class, they're forced to, like, code a virtual CPU from scratch.

Steve Hsu: Yeah.

Gabriel: And that's considered, like, one of the hardest projects they have to take. I, I wasn't forced to do that 'cause I do software engineering.

Steve Hsu: Yeah.

Gabriel: But yeah, I think if I were forced to do that, then yeah, I would have to understand things from the basic.

Steve Hsu: Yeah. Got it. Let me jump to Justin. Justin is a graduate student here at Tsinghua, and as I mentioned, he did his undergrad degree in the University of California system. He and I actually communicated a little bit because he was trying to decide what to do for his graduate career. And I actually just made the comment to him that I thought Tsinghua was exceptional in the sense that a lot of the there was a lot of research going on here, the students were really good, and also the professors seemed to have quite a lot of connection to industry.

You know, comparable to, at least in my perception, like at Stanford, like so many professors, they and their students are involved in startups. And my perception was that that was true here too. Justin, I don't know how much influence my remarks had, but Justin ended up coming here and you've been here one year now? How long have you been here?

Justin: I got here last year in August.

Steve Hsu: So about almost one year.

Justin: Yeah.

Steve Hsu: Okay. So tell us how your experience has been.

Justin: My experience here has been really good. I think the research environment and the resources I got through my advisor are probably one of the top ones. So in particular I always have access to GPUs when I need to run experiments. I won't, I won't say which ones, but

Steve Hsu: Yeah. Made by a company that starts with N?

Justin: Yeah. Yes. Or,

Gabriel: or with H, the best eSense.

Justin: Yeah, the best, the best eSense. Yes. basically, my professor is one of the top professors at Tsinghua, which I got in because I'm American. And it's also great because I get to meet and collaborate with the best students in all of China as well. So it's just a, it's a really great opportunity overall.

Steve Hsu: So let me unpack that a little bit. So when you say, "I got in because I'm an American," do you mean you were able to work with one of the top professors in your department because you're an American? Did that play a role in it?

Justin: Yeah. So there's two separate systems for international students and domestic students, and even for the Hong Kong, Taiwan, and Macau residents. So, so the domestic students, they have limited quotas, and they have to compete with, all the other dom-domestic students, so the competition is just very intense.

And then because there aren't many international students applying in the first place, then I have much less competition.

Steve Hsu: Do you mean admission into the PhD program or the seat in the lab of your advisor?

Justin: Both, actually.

Steve Hsu: Both. Okay. So there's a quota. Is there even at the level of, like, advising students, there's kind of a quota for international students?

Justin: Yeah. Alex can speak on this more later because he's a professor here. But I think for most professors, they have two domestic quota every year, and then one to two international quota every year as well.

Steve Hsu: Okay.

Justin: And then the Taiwan, HK, Macau, they count for international. They are forced to do a master's first and cannot directly apply to PhD.

Steve Hsu: And whereas you, are you already in the PhD program?

Justin: Yeah. I for international since I applied as an American, I can directly apply to PhD.

Steve Hsu: Okay.

Justin: I don't have to do master's first.

Steve Hsu: Now, I have the same question for you that I had for Gabriel, which is, if you have a friend who's in the US also at the same roughly stage of doing their PhD in CS, assuming you do have such a friend. How do you guys compare your two experiences?

Justin: So, so I can't say for CS, but I do have friends who did do undergrad at Berkeley and then are now doing some are doing stats, some are doing physics PhD in the US system.

And I think in the US system, at least for stats and physics, the, the advisor relationship is much more individual because in particular in China, it's very common for these professors in the traditional dep-departments to have dozens of students. For example, my professor, he has fifty students.

Steve Hsu: 50 PhD students or is it mix of master's and PhD?

Justin: Mostly PhD and then some postdocs. Wow. Maybe like 30 to 40 PhD and then okay a couple postdocs. That's

Steve Hsu: huge. Yeah.

Justin: Okay. Yes. Extremely huge. Yeah. And on top of that, my professor is also running a startup. So basically, students aren't even, aren't even able to reach him. Like, we have Fun, funny story. So, so I was walking to the lab, and then I saw my senior, who's a fifth-year PhD about to graduate, and he's like, "Oh, I can't find our advisor."

I'm like, "Oh, he's teaching this class. Let me send you the classroom so you can go find him there." Yeah. So even the, the most senior students have a very hard time, like, contacting our advisor.

Steve Hsu: And That, that sounds like a bad thing to me. It doesn't sound like you're directly learning from the professor, so are you learning from postdocs or more senior grad students in the group?

Justin: Yeah. So I'm, I'm not learning directly from the professor. Luckily, the senior students and the other PhD students, they're all they're already independent researchers since they've gained those skills in undergrad. I do working with other senior, PhD students and postdocs. So I'm

Alex: Can I just I directly

Steve Hsu: Yeah, Alex, jump in. Yeah.

Alex: Yeah. So, the way the traditional Chinese professorship works is a little bit different. So basically, when you come in as an assistant professor, you work under the lab of an associate professor. So it's kind of like this joint lab. So because your advisor, Justin, is like a full or very high-level professor, he has some assistant professors working under him, so you're getting more of the advising from those assistant professors.

Justin: Yes.

Alex: And, and the other thing is so Andrew Yao is one person here who set up a couple of departments, the IIIS and the AI College, which more closely emulate the US system where the assistant professors are fully independent. But most Chinese universities have this Yeah system.

Justin: Yeah. So I'm in the computer science department, and Alex just talked about the separate independent colleges.

So in the traditional departments such as computer science they have a, they have a big professor, and then they have several professors under him in the same lab so for example, my professor, he's the head of the lab, and then under him there are five other professors

Steve Hsu: Got it.

Justin: Who each do different directions.

Steve Hsu: Got it.

Justin: But even then, because there's still so many students, then like one because our lab is now doing embodied AI, then this one professor has to manage around 30 students, so it's, it's also not much time to advise everyone.

Steve Hsu: Right. This system, the traditional one sounds a little bit more like the German or Japanese system, where there is often a senior professor and then some junior professors, and some monster group attached to those professors.

But the American system, usually assistant professors are pretty much independent from day one and generally would have much smaller groups.

Justin: Yeah. I think that's true. They might have taken the system directly from, from the Germans.
Steve Hsu: Yeah. So how do you, how do you feel about your fellow students? Like this is, many people would say the top university China, undergraduates are highly selected. In the past, I think it was true that the top undergrads from Tsinghua would all come to America or maybe to Oxford or Cambridge or something for their PhDs, but my understanding is a lot of them now stay here.

How do you feel about your fellow grad students, and how many of them turn down an opportunity to say go or, or prefer to be here rather than going to some U.S. university for their PhD?

Justin: I would say now there's a pretty decent amount because now the conditions of China, in China are quite good for them.

There are several top undergrads in both Yau Class and in computer science department who say they don't want to go to the US even though it's still the best.Because number one, it's too far and they prefer to stay close at home, or like it's a new environment. It's too much of a change for them.

Steve Hsu: Yeah.

Justin: So for example one of my underclassmen, he's a fourth year and he, he wants to stay, he wants to stay in China for PhD, and he, he already got an offer. And there's another one who didn't get the offer from the lab, so I asked Alex to help him refer him to another professor, and he got that offer.

But at first, the professor advised him to apply to US universities, and he really didn't want to do it. He said it wasyeah, it was too much of an ask for him

Steve Hsu: Okay. Some background for the listener. So we're recording in the S.T. Yau Mathematical Sciences Institute, so that's Y-A-U. That Yau is a mathematician who was a professor most of his career at Harvard, and then came to China and now has set up several research institutes here.

And he created an undergraduate college where the students who are in that specially selected sub-college are typically pursue they're gonna pursue a PhD in math or theoretical physics. There's a guy called Andrew Yao, who also was a professor in the US, I think for a long time at Berkeley and Princeton whose main field is theoretical computer science, and he won the Turing Prize.

He came back to China and has now established a special Yao class, Y-A-O class, of some of the most talented math and CS students in China. So we have several very highly selected subpopulations on campus. The, the overall population of students here is highly selected, but then you have even more highly selected groups that, like, the kids in those classes would have been, like, gold medalists at the national level math Olympiad or physics Olympiad or CS Olympiad.

So it's a very select population of students here. And the question I wanna ask Justin is, if you take one of those groups of kids and you say, "Oh, look at the top twenty kids graduating in a particular year from one of those highly selected groups," what fraction are gonna end up in the US and what fraction are gonna stay here?

Justin: Yeah. So for the S.T. Yau class,

Steve Hsu: Well, they're, they're signed up to do a PhD here from the beginning.

Justin: Yeah.

Steve Hsu: So

Justin: they have to stay in

Steve Hsu: China. Yeah. Let's do Y-A-O class.

Justin: Yeah. Yeah. So the and the Andrew Yao class, the CS class that class is it, it started off with thirty students I forget what year, maybe back in two thousand and seven.

But back then, one hundred percent of them, like every single one of them went to the US for PhD or, or master's or just went there for work. And then eventually it expanded to almost around a hundred students now because they combined three different classes. They combined they had, they had three directions.
They have theoretical computer science, they have quantum computing, and they have an AI direction. Each of them were like around thirty students per class. They just combined it all. So they expanded the Yao class to around a hundred students now.

And before COVID, it was still either ninety percent or a hundred percent went to the US for PhD or work.
And now it's around fifty per- fifty to sixty percent go to the US, and then the rest stay in China.
Steve Hsu: Okay. So that's, that's an interesting trend, right? And very pronounced. Of the kids in those with that background, what fraction are actually going to get a PhD versus they just go and start a company or do something in industry right away?

Justin: Yeah. So right now, or I guess based on historical trends, maybe we could say at least seventy percent go to do PhD, maybe even ninety percent. Yeah.

Gabriel: I'd say ninety.

Steve Hsu: Yeah. Wow. See, I don't know if that's still true in the US. Like, I wonder what fraction of the top, say, MIT CS, CS grads, like maybe a lot of them are just going out and trying to start a company or join a company right away Or join one of the big AI labs

Justin: the Yale class kids who don't go to do PhD, before they would just go into quant, like what all the MIT kids do. Yes. I think more recently there are some that go to the top labs like DeepSeek or ByteDance , but it's still the most common route for them to do PhD in the US or continue PhD in China.

But it seems to be they're, they're starting to change perception as well. Now, many of the the fourth-year undergrads in Yale class are thinking that, "Oh, PhD is probably-- There's no point to do it anymore," 'cause they've already published so much as undergrads. They have a PhD. They don't even need another one.
Steve Hsu: Right. Just to elaborate on that for the audience. So through you, I've met a bunch of these kids, and a lot of these kids, they're senior in college, and they've published a handful of papers already. Yeah or they've been one of the first or first few authors on a number of papers

Justin: Yeah

Steve Hsu: already. So they might even be I, I don't know what the typical expectation is for US PhD students, how many papers they should have published, but, but these kids seem very accomplished to me.

Justin: if we were to put a number on how many papers like a PhD program would expect, it's around three papers at top conferences.

Steve Hsu: Yeah.

Justin: And most of them, they, like, have way more than three papers.

Steve Hsu: You're saying a lot of them hit that when they're undergrads.

Justin: Most of them, most first, first authors as well.

Steve Hsu: All right.

Steve Hsu: Let me jump to Alex. So Alex, you are the professor here. You work on AI and robotics. Tell us a little bit about your academic background and how you ended up here.

Alex: Yeah. I'm Alex. I did my PhD in Montreal. I guess as I was graduating, I was interested in both industry and faculty positions.

Steve Hsu: And can you just say you have a famous PhD advisor.

Alex: Yeah, it's Yoshua Bengio.

Steve Hsu: Yeah.

Alex: I was interested in faculty positions, and I had a coworker from Microsoft Research Asia, so that's in Beijing who took a faculty position at Peking. His name is Yi He. And I told him I was, you know, serious about becoming a professor in China. And he said, "If you wanna do it first, you know, use the connection to establish that you're serious."

And then he said, "If it's a department headed by Andrew Yao this is the place to start. And if that doesn't work out, you know, we could look for something else." And it did work out, and I ended up coming here to College of AI at Tsinghua. I've been here for about a year, and I really like it so far.

So I have, I guess, like, five incoming PhD students in my lab. I think they're really great. As Justin, I guess, alluded to, when they come in, they typically have many papers, or they have some other kind of substantial accomplishment. And I also have a, have a pool of interns here. So I would say the, the energy here is really, really good.

Like it's you're constantly getting people who, like, want to work on things or want to start projects. A lot of students will start really ambitious research projects even their freshman or sophomore year. Yeah.
Steve Hsu: My impression from being here a little bit is it's a super high energy place. It reminds me of places like Caltech and MIT, where people really wanna do stuff, and there's just tons of talent just kind of floating around.

Justin: One thing I that did surprise me a little bit, which I think is kind of interesting, is I, I feel like you have a lot of students here who, who could do just very purely theoretical CS or physics research, but they do, like, empirical AI or kind of like deep learning type research.

Whereas I feel like maybe that's less common in the US unless it's changed recently. 'cause yeah I feel like in the US, at least when I went to school, maybe there was a little bit of a perception that, like, if you could just do purely theoretical research, you should do that to, like, use that comparative advantage. I, I don't know what you think about that.

Steve Hsu: The, the parallel in physics would be some kids are gonna be theoreticians, and some kids are gonna be experimentalists working in the lab. Uh-huh. And working in the lab is a little it's, it's obviously, it's more empirical, and it's a little more like in the nitty- Close. Yeah. Yeah, in the nitty-gritty details. And, yes, there is some kind of like classification, like some kids end up in one bucket than the other.

Gabriel: Mm-hmm.

Steve Hsu: And, but sometimes, like the very best, this is just physics experience, the very best experimentalists are kids who have the chops to do theory

Alex: mm-hmm

Steve Hsu: but they also have the hands to go in the lab and really do stuff. And those people are the ones who really can push the field forward

Alex: Mm-hmm

Steve Hsu: by themselves. So there may be an analog to that in CS.

Justin: Mm-hmm. So, yeah, I have a comment about theory in China versus the US. So when I was in the US, I was more involved on the theory side on AI as well, in the math and stats departments. And through that, I I got to know a ton of Chinese students because actually most math departments and stats departments, even CS departments, most of, most of the PhD students are Chinese And there seemed to be a common consensus that the, the theoretical research in the US was much more mature in China.
So even if there are students in China who wanted to do theory they would have to go to the US to do theory, to do frontier theory research, or if they wanted to stay in China, then they would work on more empirical research or engineering research.

Steve Hsu: So theory in the US is ahead of theory in China. Is that what you said?

Justin: Much like by quite a lot. Okay. Which is a common consensus among all the Chinese researchers. And S.T. Yau even said, like in a, in a speech, that China is behind by 40 years in math mathematical research, yeah. Okay. Theory research,

Steve Hsu: yeah. Good. Alex, I wanted to drill down on what why were you interested in becoming a professor in China?

Justin: Well, I guess I, I just felt like, okay, objectively here I can get really strong students. I also feel like the overli or sorry, the underlying economic strength in China, like the industry is very good. And I feel like eventually, you know, it, it will support like world-class academics and, you know, I think it's already kinda starting to happen.

But I think there's a lot of room for growth in the academia in China. I mean, I guess also even just as a little kid living in China, working in China was something I was interested in doing.

Steve Hsu: Yeah.

Alex: So the fact that I got a good opportunity to do it is pretty exciting.

Steve Hsu: I mean, I definitely feel like here there's a can-do spirit

Alex: Mm-hmm

Steve Hsu: and a kind of an upward trajectory to everything, whereas in the US it's like, can we, like, maintain our position? Are we just declining a little bit? It's just very different

Alex: Mm-hmm

Steve Hsu: vibes in the two places. How do you feel about like some of these crazy college rankings like US News and stuff, they're already putting Tsinghua like ahead of all the other, like, engineering like I think like in the engineering ranking for universities, like some of these crazy rankings have Tsinghua number one in the world. Is that crazy or is it, is it, is it reasonable?

Alex: Well, here's one way I might think about it. So, you know, if you bracket people by like their age or seniority, I think if you take the people who are like, let's say, twenty to thirty, I think at Tsinghua they're as strong or maybe even as, about as strong or maybe even a little stronger than the best US universities.
But then if you kinda go towards the more senior cohort, like the people who have like thirty or forty years of experience, I think if you compared that cohort in the US to China, I think we don't have as many, like, extremely senior, extremely

Steve Hsu: Yeah

Alex: experienced people. And I think that also goes to what Justin was saying about our theoretical research being a little bit less mature. I, I think that's an area to work on.

Steve Hsu: Yeah.

Alex: In terms of what my students say in terms of their preference rankings, they usually say like the top five schools in the US are like pretty competitive with Tsinghua. So like, I think it's like Harvard, MIT, Princeton, Berkeley, Stanford, roughly that pool.

Steve Hsu: Yeah.

Justin: But I think below that it's pretty widely agreed that generally Tsinghua is somewhat preferred.

Steve Hsu: Got it.

Justin: Actually swap out Harvard for CMU.

Gabriel: Yes, they call it the Big Four.

Justin: Big Four. Big Four: CMU, MIT, Stanford, Berkeley, and then now you add in Princeton.

Gabriel: Oh, yeah, for NLP and then UIUC for like five

Justin: I think it'll depend a little bit on the area, 'cause in my opinion, if you wanna do certain things like embodied AI, I think you can still make the case for Tsinghua even over those universities, just 'cause, I mean, you have so much of a robotics industry here compared to the US.

Steve Hsu: Yeah.

Justin: It's just my opinion.

Steve Hsu: By the way, my, my experience in having been a physics researcher for a long time now is that, yes, the older generation, you would very seldom find a truly world-class guy here, because most of those guys would take if they could get jobs in the US or elsewhere, they or even in Europe, they would go there.

Justin: Mm-hmm.

Steve Hsu: But it's the younger group where you're starting to see really world-class talent that stays here.

Justin: Mm-hmm

Steve Hsu: I think one of the tipping future tipping points will be when they feel confident that they can really fully compete without relying on people who went out to the US and got their PhD or postdoctoral training and came back.

They can fully rely on the people that are just trained one hundred percent within China. I think famously, like the of the DeepSeek authors, like in one of their early papers that caused like the DeepSeek moment, I think all of those people had been like hundred percent trained in China or something like that.

Alex: Mm-hmm.

Steve Hsu: So that was like maybe one of the first indications of that tipping point.

Alex: Mm-hmm. I, I would say for the professors who we hire in our department I think the majority have done their PhDs overseas, often in the United States at good top universities. But some have also done their PhD here in China.

Steve Hsu: Yeah. So I think that's the, that's the tipping point that we're crossing right now.

Justin: I think it's just a joke, but they call the the domestic PhD student the Tuboshe, right?

Steve Hsu: Tubaozi.

Justin: No, Tubosi. Oh.

Alex: The Tu in Yang. Tu is you know, a, a rough country bumpkin. Can you say it again? And what the meaning is, Tuboshe? Tu means

Steve Hsu: dirt. Tu

Justin: means dirt. Yeah.

Steve Hsu: Okay. Like a farmer.

Justin: Yeah. Yeah, a farmer bosi and a yang bosi. A yang bosi would be an international type of bosi and Sophisticated the tu bosi which is You know, the, the, the redneck version of China.

Steve Hsu: Yeah.

Justin: I think it's just a joke. Yes.

Steve Hsu: And it'll change over time.

Justin: Yeah.

Steve Hsu: So this was before my time in theoretical physics, but there was a moment. So Oppenheimer was the first American who really learned quantum mechanics.

Justin: Mm-hmm

Steve Hsu: So at the time when Oppenheimer, if you remember the biopic, which was really, a really good film, he had to go to Europe actually to get that education. And when he came back from Europe, he was the only guy, he was the only American professor who could really teach quantum mechanics at the frontier level.

Alex: Mm-hmm

Steve Hsu: And he started the first school of real, really mature school of theoretical physics, and he literally split his time. He would spend half the year at Berkeley and half the year at Caltech because he was so in such demand. They wanted him in both places.

Alex: Mm-hmm

Steve Hsu: And that was So that was, like, just prior to World War II, and that was when America was just, like, nowhere in terms of cutting-edge science. But then we rapidly, just one more generation, we caught up and went to the lead. Yeah.

Alex: Yeah. I, I had heard the same thing that for a while there was a perception that like, okay, the US has good industry, they can do good applied stuff, but if you wanna do like theory or basic research, you gotta be in Europe.

Steve Hsu: Yeah. It's a hundred percent analogous to the current situation. So the experimentalists in America were good. They could get stuff working. We had the Industrial Revolution. We had become by then the number one industrial power. But in terms of the high level theory, we did not have it.

Alex: Mm-hmm

Steve Hsu: And it almost, it seems like there's a very parallel thing going on right now with China vis-à-vis the rest of the world.

Justin: Yeah. I have a comment about the rankings. So I think many people outside of the academic system don't really understand the rankings. It also has to be split between like undergrad and graduate. So, so without a doubt, Tsinghua undergraduate students are the best in the world.

I would say on, on average much better than, than the best universities in the US. But this doesn't actually mean the, the education quality itself is higher. It just means this, the filtering system, they can filter better students. But then for graduate it's still without a doubt that the front like most of the frontier research is still in the US although China is catching up very quickly.

That's why we see that most of the PhD students in the US are actually Chinese because they're high quality students.

Steve Hsu: Do you get the feeling that, we're turning back to Gabriel now, who is just finishing up his undergraduate degree. Do you feel like the competition you had to deal with in the last four years here is pretty much the toughest CS competition at any university in the world, average level?

Gabriel: Yes. Every single one of my classmates are they're brilliant, right? 'Cause they had to go through the Gaokao filter, and they're like the top in their province. but you made a comment during lunch about what you think about once you get past this Gaokao filter, does this mean that they're just gonna be the top of their class at Tsinghua?

Like, relatively speaking, I did the IB. I was good at the IB. I wasn't top one in all of Hong Kong, right? But that if, if you go according to your original theory, that would imply I'd be the last place in my cohort. Honestly, that's what I was expecting, but that's not what ended up happening. As in my grade right now, like even after having to acclimate myself to the Chinese environment, I'm like smack in the middle, like fifty percent. My take on this is, yes, they're all brilliant students, but a lot of them, they're like much more acclimated to the testing environment.

And the main issue for them is not that they can't get their heads around the classwork, but rather it's like social skills or just like really just self-discipline. The really, really impressive people Chinese people at Tsinghua are the ones who aren't just smart 'cause they all are, but they're like socially adept as well.
Like, once you get both of those, yeah, those are what you get as like the top PhD students that stay in Tsinghua or they go abroad.

Steve Hsu: Yeah. I mean, the some of these superstar kids that Justin, you introduced me to last time I was here, those kids clearly like they're clearly smart. They've done a bunch of research.

They've published a bunch of research as undergrads. And, but when you interact with them, they're pretty polished. Like, you could tell like and some of those people had already, like, raised money for their startups. Yeah. Even they're, they were like seniors in college, and they had already raised money.
Now, now part of that ability to raise money here is I think that investors here still have quite a lot of respect for academia. And so like they'll just like write a check, I think. It seems like they'll write a check because you're some big professor at Tsinghua or some big professor at Tsinghua is saying, "This is my best student," and the, the venture investors will just write a check.

It's a little bit different. In the US maybe that would happen like around Stanford or something, but, but even then, like the VCs are thinking more about like, "Can this guy actually run a business? Are they a hungry entrepreneur? I don't really care what this old professor says." It, it seems like there's a slight difference in the cultures.

I don't know that you would necessarily know about that, but that, that seems that's my impression.
Justin: Yeah. I have some interesting comments to say about that. So in particular about the respect for academia or academics in China, that's actually a major reason why most of the students, even the top students, they actually pursue PhD instead of like going into industry or doing a startup, is because of the prestige of doing a PhD. Purely because of the social prestige.

Steve Hsu: The issue is whether it's because of that respect for academia that one of the best routes here to get venture funding is to distinguish yourself in academia and have some big professor say, "Hey, this is my most promising student. Give him money for a robotics startup."

Justin: Oh, yeah. That

Steve Hsu: Yeah, that's I think that happens I think that sort of happens a little bit, like around Stanford or something, because you have a confluence of one of the top CS departments and like lots of bags of money just around. But I don't think it's actually true at like Caltech or MIT

Justin: Right

Steve Hsu: or some of these other places. Like, you could be like a really awesome student, but there isn't necessarily a bag of money that's just like tossed your way for being really good at academic stuff.
Justin: Okay. I, I remember what I wanted to say about that as well. So, so in in Chinese society among ordinary people, there's also like an explicit worship of Tsinghua and Peking University undergraduates because of how hard it is to get in.

So like I had a friend I made in the US who's doing a math PhD. I met him at a conference in the US and he's doing a math PhD at the University of Bonn in Germany. And he came back to Beijing to visit, and I invited him here. And when he came here, he was like how to say? He was praising the undergraduate students of Tsinghua. He was like so amazed. He's like, "Oh, like we treat these people as gods in China."

Steve Hsu: Yeah.

Justin: The social prestige of the Tsinghua and Peking University undergraduates is at the very top.

Steve Hsu: Yeah. I, I mean, I would almost guess that if, if your if your prestige that you derive just from being an undergrad here is enough, then you wouldn't feel like you have to get the PhD to pad it, right?
Back in my day, this is a totally different era, but There was a time when Harvard, MIT, and Caltech were by far the most prestigious undergrad degrees to have. And it used to be said among people at those schools, like, these are the, the only schools where you just need that undergraduate degree. If you just say, "Hey, BS Caltech," or "BS, SB MIT," or, you know, whatever, Harvard, you know, Summa at Harvard, that's it.

You don't need to go and get your PhD 'cause people just assume you're smarter than most PhDs, actually. I think that's all gone now. That's like a bygone world of 50 years ago or something. But you could end up like that in China too, I think. So, so you could have more and more kids who are coming out of Tsinghua, and they're just like, "I'm just gonna start my company. I'm not gonna go to, like, live in Pittsburgh at CMU for four years and, you know, just to and write some more papers just to get a PhD behind my name."

Justin: Yeah. So I, I have some more interesting comments about that. So on, on the startup side, instead of doing an undergraduate what they do now, more practically speaking, is they, they just continue doing PhD, and then while they're doing their Ph their PhD, they can pull funding by using the name of a PhD student and the prestige of their advisor to attract funding and do the startup.

And secondly, it's like, oh, why not just quit? Why still do a PhD? think that's, I think that's something quite common in Chinese culture. They, they like to get the next best thing.

Steve Hsu: Yeah.

Justin: You know, they continue to strive or compete in the rat race.

Steve Hsu: Yeah.

Gabriel: Yes. It's, it's nature. It's evolution.

Justin: Yeah.

Gabriel: Nature.

Gabriel: Like, it's because yes, even though they have the Tsinghua undergrad, but, like, since year one as an undergrad, they're just thinking, "Holy crap, we gotta get this. We gotta go get the master's." And to get the master's, you have to be, like, top 60% of your undergrad class, and that's all they thought about Wow like, for three years.

Steve Hsu: Wow.

Gabriel: Yeah.

Steve Hsu: So it's like always there's always one more hoop to jump through or one

Gabriel: more yeah. And the reason

Steve Hsu: Yeah

Justin: is not even because they're like, "Oh, once I get the master's, then I can get a better job." It's because, "Oh look, that's what everyone's, everyone else is doing. Might as well do as well," you know?

Steve Hsu: Wow. Alex, you were

Alex: I, I also see a lot of students who I feel like can kind of hustle around the neijun. Is that how you said it?

Gabriel: Yeah, yeah. Neijun.

Alex: Yeah. So like, even if they did badly on the Gaokao, they went to a lower ranked undergrad, they still work hard, get some good papers, go to a top PhD or go to a good company.So I don't know. I don't feel like it's the end of the world for everyone if Yeah

Steve Hsu: And I feel that's healthy. Like yeah you know, back in the day, the way we used to say it in the US is you could be a total fuck up in high school, but you could still go to a community college for a couple of years and learn, you know, calculus and physics and stuff, and then transfer to a pretty good state university, transfer to the University of Illinois, which is actually a really good engineering school.

And then like, so there's no, there's no point at which you're totally out of it, you know? Or even if you didn't have a good undergraduate degree, you're like, "Oh, but I'll let me go work at Lockheed Martin, and if I'm really good at my job, I can still make my way up." So I think it's healthy to give people many ways to succeed, even if at one particular stage, at age 18, they were not up to snuff.

Alex: Mm-hmm.

Steve Hsu: They were playing too many video games. Yeah. But you can, you can still make it up at some other stage, and I think that's just healthy. I would be alarmed if, like, there was no way to the top except by jumping through all the perfect hoops, you know

Alex: Mm-hmm

Steve Hsu: at every stage of your life. That would, that would probably that's what's related to my question at lunch today that I was asking is like, is like it are there kids who were not quite as distinguished when they finished high school but managed to get in here, but they still turn out to be the top kid when they're actually allowed to do research or something like that? So that it's healthier if that's possible.

Justin: Yeah. So that, that's definitely possible. My, my lab mate that we met today is literally one of them.

Steve Hsu: Yeah.

Justin: He's like the top, he's the top researcher in our lab as a first year, and also published so many papers as an undergrad, and he didn't go to Tsinghua or Peking.

Steve Hsu: Yeah.

Justin: Yeah, but, but actually the, the perfect hoop right now in, in Chinese society is not like Tsinghua undergrad to Tsinghua PhD. It's, it's, you do Tsinghua undergrad, and then you go to MIT, Stanford, or Berkeley for PhD.

Steve Hsu: Yeah.

Justin: And then, and then you stay there preferably. Yeah.

Steve Hsu: Hmm. And then eventually you come back

Justin: Or

Steve Hsu: not.

Justin: Yeah. I, I think most of them don't wanna come back. So, so it's like in China, the perfect group isn't to go from like like you do you perform really bad in high school, and then you go to a bad university, and then you go to Tsinghua. Yeah. It's like they actually prefer you go, like, to the US in the end. That's, that's still like a okay perception in Chinese society interesting as the, as the optimal path.

Steve Hsu: So I my purpose of having this discussion with you, with you guys here is that, like for most people in the US, they're even people who are in technical subjects or academia, they have this sense that like US is competing with China, and it's, it's a serious competition, and the Chinese have their strengths, and the Americans have their strength, and maybe they are starting to beat us in some important ways.

But very few Americans understand, like, what is a Chinese university like? What is the talent selection system in China? What is the preference stack of a twenty-two-year-old very bright Chinese kid? Like, I think most Americans don't understand any of that, and that's what I was trying to elaborate. Is there some aspect of our discussion that I didn't cover that you think the audience, the Manifold audience would, you know, be informed by if we discussed it?

Any-anything that you think we didn't cover in that bundle of stuff that we should talk about?

Gabriel: I think we should emphasize more just the hoops, like firstly, how to get into undergraduate. But actually before that, how to get into high school. That's a good point that I never even thought about in Hong Kong. So first of all, to get into high school in China, you have to take the Zhongkao, so the middle school test.

Steve Hsu: Yes.

Gabriel: And if you fuck that up, fifty percent of people fuck that up. Not fuck that up, but, like, just don't do that great. They just

Steve Hsu: don't get into

Gabriel: a top high school. Not even a top high school. They don't get into high school. They go to a vocational training school. And when you get into vocational training school, you don't even take the gaokao, and then you just go directly to what they call a daxue, where you learn, like, practical skills, like factory work and stuff. So once you're on that track, it's kind of impossible for you to jump back into academia.

Steve Hsu: Yeah.

Gabriel: Let's say you do get into high school, then you take the gaokao. And with the gaokao yeah, you still have to do good enough to even get into a university, not even a so they separate them based off So there's vocational schools, there's Arbin, there's Eben. And then if you're in the top five percent, that's a two one one.

And then if you're in the top one to two percent, you're a ninety-five. And then ninety-five is thirty something of them. And then the top two are Tsinghua, Beida.

Steve Hsu: Okay, just to elaborate. So nine they have These are classifications

Gabriel: Yes.

Steve Hsu: Of the universities.

Gabriel: Yes, of four-year universities.

Steve Hsu: Right. And of four-year universities. And if you score top five percent on the Gaokao around, yeah

Gabriel: depending on province.

Steve Hsu: Roughly. Yeah. You can get into one layer.

Gabriel: You can get to the two one ones.

Steve Hsu: Okay.

Gabriel: Lesser than a ninety-five, but it's still considered a, like, damn good school.

Okay.

Yeah.

Steve Hsu: But the next layer, which is how many schools are in the next layer?

Gabriel: Ninety-five is like thirty something.

Steve Hsu: Okay. So top thirty-ish universities in China.

Gabriel: Yes.

Steve Hsu: Roughly speaking, all the kids getting into those are top one to two percent one, yeah, exactly on Gaokao.

Gabriel: Yeah.

Steve Hsu: Which is actually already crazy because on the US exams, like the ceilings are usually only like, well, you the, the, the apps, if you actually get a perfect score on the SAT, it's like a few per thousand kids.

Gabriel: Yeah.

Steve Hsu: Or one per thousand kids. But you're, you're getting toward the ceiling of what the American system can resolve just to get into one of these top thirty schools

Gabriel: Yes.

Steve Hsu: In China, or at least to get into Beijing or Tsinghua University, right? So this, the system here is very, very elitist, right? Yeah. Elitist and meritocratic based on one test.

Gabriel: Yes, exactly. So most people get into university based on the Gaokao. Yeah.

Steve Hsu: Yeah.

Gabriel: And then I think for grad school applications, it's So if you're at a big university already there's actually two systems. One is based off your undergraduate GPA. If you're good enough, you get to bao yan, so stay in the school or or actually you can bao to other schools.

So let's say you're in like some other ninety-five. If you're really good, like top one of your school, you get to go to Tsinghua, Beida. Or if you're already in Tsinghua, Beida, you just gotta be good enough. And the other system is called kaoyan. So that's more the social mobility part of it, and that's basically we don't even look at your undergrad, like GPA.

You can think of it like another Gaokao before or after your undergrad deg-degree. But that's like ex-- that's extremely hard as well, but because less people take it, they consider that's why they consider master degrees like less prestigious than undergrad degrees.

But yeah, if you're not from like a top university where you can bao yan into the best universities, you can technically take that test. But even that, you gotta be like top five-ish percent to even be considered.
Steve Hsu: Okay. So there's a GPA-based and also a further exam-based way

Gabriel: Yes.

Steve Hsu: to get into the graduate programs here.

Maybe that's enough about like this Not everybody's actually that interested in like, well, what the hell, the minutiae of like the Chinese academic system.

Steve Hsu: Let's talk a little bit about the, about competitiveness between China and the US, right? Which is a topic that's often we often discuss on this podcast.

Like, from what you guys can see within what you could argue is a top university in China, or at least the top technological university in China, and the companies around it here in Beijing Any observations about where you think US-China competition is going, like in technology and AI, anything? Yeah.

Alex: One thing is I think the US is really being held back by its lack of high-quality infrastructure because I think you have a lot of universities in the United States, and they're held back by the fact that they're not located in a city with, like, top-tier industry.

So like you have Carnegie Mellon, top computer science school, but it's in Pittsburgh. UIUC is in Illinois. I guess Georgia Tech is in Atlanta. And I feel like, you know, in China, because you have the high-speed rail, it feels like you can get to any city within a day and just saying like, "Hey, go to Shanghai or go to another city," that's like nothing for me.

But if I have to, you know, take a flight, that's kind of like a big trip. So I feel like you have a lot more, like, intercity collaborations in here, and I feel like more of that talent is unlocked. So I do think in the US, there's a risk that it's really gonna become overly dependent on, like, just the Bay Area and maybe, maybe just New York.

Steve Hsu: Or

Alex: Boston.

Steve Hsu: Yeah.

Alex: But there's a chance that if that Bay Area companies lose the remaining competitive edge, the US might be in serious trouble.

Steve Hsu: Right. So let's break that down a little bit. So here they have a high-speed rail system

Alex: Mm-hmm

Steve Hsu: which is awesome. And I don't think people really appreciate in the States what that means. So like, you know, like every hour or every thirty minutes, like there's probably a train, high-speed train between here and Shanghai.

Alex: Mm-hmm.

Steve Hsu: And you get on the train, it's not stressful. It's not like the airport where you gotta go through. I mean, there's a little bit of security, like they are you do put your bags on a conveyor belt at one point.

Justin: Yeah.

Steve Hsu: But it's, you, you get to the train station, like not, you don't have to get there hours ahead, right? Mm-hmm. You can kind of cut it a little tighter. You get on the train, super comfortable. There's even a business class car where you can lie flat if you want

Alex: Mm-hmm

Steve Hsu: take a nap. You get there, you're arriving at the city center

Alex: Mm-hmm

Steve Hsu: of the other city. So just door to door you have access to many other cities with five million plus people within like a few hours of here, and you could do it as a day trip. You could go there and come back, and that just makes it Easier for you to collaborate across In specific cities. And do you think though that means there's less of a concentration?

'Cause I still feel like even in China, like most of the tech is, okay, around here, Haidian, where we are.
Alex: Yeah.

Steve Hsu: A lot of the big com- a lot of the best, most promising companies are here. There's a bunch in Shenzhen.

Alex: Mm-hmm

Steve Hsu: There's some in Hangzhou because Alibaba is there and DeepSeek is there, and then Shanghai. I guess Shanghai has SMIC. Mm-hmm. But it, it's still pretty localized. Is there a sense of like these other big cities that most Americans wouldn't be that familiar with, there's, there's plenty of high-tech activity there going because of this infrastructure?

Alex: That's an interesting question. I mean, I would've said Hangzhou is actually decently far from Shanghai, right? It is,

Steve Hsu: yeah.

Justin: So

Steve Hsu: Well, about an hour on the high-speed rail. Yeah.

Justin: Yeah.

Steve Hsu: Yeah.

Gabriel: So I have some comments on that. So in particular to pro to promote more even technol technological development across China's geography the government has a national plan called East Data, West Compute

Steve Hsu: Yes

Justin: where they build data centers out in the western provinces of China.

Steve Hsu: Yeah, where there's lots of solar as well.

Justin: A lot of solar, a lot of cheap land, cheap electricity.

Steve Hsu: Yeah, yeah.

Justin: There's a bunch of subsidies there to provide more jobs and development.

Steve Hsu: Yeah.

Justin: and, and also for national security reasons because chip manufacturing is such a capital-intensive industry.

Before it was mostly concentrated in Shanghai, the government actually forcefully moved some to the central regions in China, such as,Chengdu to develop more chip chip industry there. Yeah. So Chengdu is also considered a, a developing tech hub there, al- although not in AI, but in chip manufacturing

Steve Hsu: Yeah

Justin: and military technology.

Steve Hsu: So I mean, aside from infrastructure, I think the, the government because the government, things are more state-led here than they are in the States, in the US. If it's state-led, then they might say like, "Yeah, let's encourage this industry to be in Chengdu," or, "Let's encourage this industry to be in Chongqing," or something like that.

So it does get spread out more than just letting like everything like the free the market maybe just concentrates everything in the Bay Area, and the government doesn't try to do anything to counter that. Here, they would maybe try to do something to spread it out more, and maybe that's the thing that's manifested, here.

Alex: There's another thing which really confused me before I came to China, but which I've just started to understand, which is if you look at, like, the total market cap of, let's say, the Chinese companies, it's so much smaller than the market cap

Steve Hsu: Yes

Alex: of the American companies.

Steve Hsu: Tiny, tiny, yeah.

Alex: But then if you look at, like, measures of productivity, like what they actually produce, the Chinese companies seem to produce more in the aggregate, depending on how you measure it.

Steve Hsu: Yeah.

Alex: But they definitely do. So I think what's going on, you guys can let me know if you agree, but the market cap is kind of like an integral of the future, profits or, like, the future dividends they return. So if you have a situation of low competition, like you have a few monopolies, they're very profitable.
Maybe they don't produce as much, but they, they return huge dividends, and they have a big market cap. Whereas I feel like in China, they try to induce a high level of competition, and they help new companies to keep entering. So you have a huge number of companies. You have relatively low profit margins

Steve Hsu: Yeah

Alex: but you have a lot of production.

Steve Hsu: So let's, let's hear from Han Feizi on

Han Feizi: In, in, in economics terms, what China is doing is flattening the supply curve. It's getting more and more supply in there so that the supply curve is almost horizontal. And when it looks like that, then, you know, if you've taken you know, beginning microeconomics what it is, is the supplier surplus, that little triangle to the supplier surplus, nearly zero.

Alex: Mm-hmm

Han Feizi: It just goes away. Everything is that huge triangle that is, you know, above the price line, which would be the consumer surplus.

Alex: Mm-hmm

Han Feizi: So that's what you have in China. In the US system what, what, what, you know, you don't have a very flat supply curve. You only have a few participants in the market. So the consumer surplus is less, and the supplier surplus is a lot more. It's like a monopoly market.

Alex: Mm-hmm

Han Feizi: and, and, you know, there are benefits and, you know, costs and benefits of both systems. It's, to me, it's just whatever surplus you, you, you just maximize surpluses however you do it. And it appears to be, you know, when you have Tesla is worth ten times more than BYD.

But you know, the electric vehicle penetration in the US is, like, what? Five percent of the market when it's, like, sixty percent of the market in China. That, that, that's a failure. You know, that, that's pretty much a failure of the EV industry in the US. It's not a, you know, a success that Tesla is worth ten times what BYD is.

In terms of how an economist would look at it. How a businessman would look at it, he would say, "Well, Tesla far outperformed BYD."

Steve Hsu: Yeah. I, I agree with you, Alex. It's a question of what the society is optimizing for

Alex: mm-hmm

Steve Hsu: and if the society says, "We want the consumers to benefit, so we wanna maximize competition between companies

Alex: Mm-hmm

Steve Hsu: we're not gonna allow monopolies to extract monopoly rents from the system," well, then there aren't as many great stock investments. 'Cause what's my best stock investment? Get into some company that's gonna become a monopoly

Alex: Mm-hmm

Steve Hsu: that's gonna generate wild earnings numbers year after year after year in a predictable way. Then yeah, that thing suddenly becomes a trillion-dollar company. but that's not necessarily great for the consumers, right? So it's, it'sBasically that conflict. The question, though, is, like, if you're racing to get to AGI, maybe you wanna be the system that will allow the AGI monopoly to whoever wins the race, and then that guarantees your companies win the race.

Because so much capital ends up pursuing those opportunities in the US, and so little capital is pursuing the AI opportunities in China by comparison. I mean, certainly compared to the rest of the world, it's a lot of resources. But compared to America, the amount of resources flowing toward AI is, is like one-tenth here as in the US. And yet they're still able to kinda, kinda keep up.

Gabriel: I believe that's something what Nick Land argues.

Steve Hsu: So yeah. So Nick Land would He has a term techno-capitalism, which is that, you know, capital produces awesome technology that technology makes a lot of money, which produces more capital, and then you just have this feedback loop that's running out of control.

And part of it, his observation is it's out of the control. There's no like It may seem like Elon is the genius or, you know, this guy is the Sam Altman is the But actually what's happening is this machine is just like, it's just like pushing capital toward more tech development, and then more tech development creates more capital, and the thing just works on its own, and the people are kind of irrelevant, the individual people. It's just that the dynamics will eventually then lead to, like, super intelligence or something.

Alex: I guess one thing is I feel like the Chinese firms, because they're a lot smaller, like the Alibaba, they're a little bit more afraid to pursue like a novel like product market fit. Like something where they don't know if there's a market yet.

So like one example is like Anthropic kind of took the lead in making like a, a complete software program to help you with coding. So even though like the Qwen model, for example, you know, the intelligence level is like roughly the same, the coding ability is roughly the same because they didn't have like the fully fleshed product, people don't want to use it as much.

But I kind of feel like once they observe that like this is something people will pay for

Steve Hsu: Yeah

Alex: they'll build the fully fleshed out product, and I think they'll catch up.

Steve Hsu: I agree. I think we're seeing that right now that, you know, fast following by Moonshot with Kimi and with Qwen and a coding rig for Q-Qwen.

That fast following could lead to a competition where the, a lot of the profit margins are competed away for Anthropic.

Alex: Mm-hmm

Steve Hsu: I feel like most investors, global investors, there's still a huge US side bias where they just don't wanna put money behind the Chinese companies for whatever reason.

I mean, it could be like, oh, the Chinese government will never let these guys make as much money as it's could be something that equilibrates out like in the next ten or twenty years, or it could just be like, just stuck like that for a long time.

Justin: I, I think one major component would be China, China trying to develop its own semiconductor supply chain. I that's probably the most important piece in China's AI development.

Steve Hsu: Yeah.

Justin: Because at any point in time, if they continue to rely on Nvidia, it's not a reliable source of compute for them.

Steve Hsu: Yeah. Now we've discussed there was a group of American AI researchers and journalists who came through here, came through Beijing. I think you, you met with them, right, Justin?

Justin: Oh, yeah.

Steve Hsu: Yeah. I met with

Justin: them.

Steve Hsu: And I think I, I, I looked at all the reports that they read, or that they wrote, based on their trip here. And, you know, a lot of them, I think, thought that the GPU sanctions or controls were good because they kept reporting that the Chinese companies the number one thing they heard from the Chinese side AI researchers is, "We, we wish we could have more Nvidia GPUs." Do, do you wanna comment on that? Is that, is that a fair assessment of the situation?

Gabriel: Yeah, I would say that's pretty fair. Basically, all Chinese would want more NV-uh, Nvidia GPUs. Yeah.

Steve Hsu: Is, is the issue the money? Like so there's two things sort of going on here. One is the Chinese companies don't have as much money, which we just discussed.

So even if they had the GPUs, they might not be able to do the monster training runs that OpenAI and Anthropic can do. The other issue is just if you have the money, you still can't buy the GPUs because the US government is not allowing it. Like which, which of those two factors is actually more decisive?
Justin: it, it depends on the company. So for the startups, it's definitely the money. And for the large companies, then, then it might be the supply, although it seems they they are still able to get a ton of GPUs.

Steve Hsu: Yeah. See, one of my one of the guys that I've had on the podcast, a guy called TP Huang, who's a software developer himself, but also a very close observer of AI and military competition between US and China.

He, he said his reaction to Nathan Lambert and all these guys, Jasmine Sun. jasmine Sun, I think I introduced her to you. She, she was one of the reporters who was here, one of the writer. So his response is like, "Look, they only interviewed basically startup companies. They didn't actually. Did they meet with, like, Alibaba people?

Did they meet with ByteDance people?" Because TP Huang would say those guys have very deep, much deeper pockets, and it's unclear whether they really are GPU poor. So that, that, that was TP Huang's reaction. I don't know the answer myself, actually.

Gabriel: Yeah. So, so for my lab mate who's currently working at ByteDance, he's able to get, like, any GPU he wants.

Steve Hsu: Yeah.

Gabriel: So, so it doesn't seem like they're GPU poor.

Steve Hsu: Right. I mean, one of the secondary things I would invite people to study is the amount of NVIDIA sales to Taiwan, Malaysia, Singapore, all of these countries which don't produce any models. Also, as far as I can tell, don't, like, produce a lot of inference tokens, right?

So what are, what are these GPUs doing in these countries? Even Taiwan. So Taiwan is, is one of the biggest in terms of numerical purchases of GPUs from NVIDIA. But where do those where do those GPUs go? I think they get put in a suitcase, and the guy gets on a plane for Shanghai, and that's where I think the GPU actually ends up.

Because I can't go to Taiwan and find anybody training any monster model. I can't go to Taiwan and find anybody running a monster model providing tokens. So what the hell are these GPU, you know, billions of dollars of GPU sales to Taiwan, to Malaysia, to Singapore? Where are these things? Singapore does have some data centers and but, like, the data centers in Malaysia, like, I wonder if they're just doing computations for ByteDance people or your friend or, or whatever.

So, so I, I actually, I actually don't know, for the people that have the money in China enough to pay for the compute, do they have trouble getting it? Do they have trouble getting the latest NVIDIA GPUs? And whether in, in country or, like, virtually by, by running the jobs in Malaysia or running the jobs in Taiwan or something like that, I don't know the answer to that question. I don't think that group that came here really got to the bottom of that.

Gabriel: I'll give, I'll give an anecdote from the founder of ZAI

Steve Hsu: Yeah

Gabriel: who's also a professor at Tsinghua. His name is Tang Jie. And so I did ask him, "Oh, why don't you use B200s?" And he simply said, "Oh, you couldn't get any." But they did have H100 stock problems over years back, so they could use that to train their models.

Steve Hsu: So yeah, so I just don't know the answer. So maybe that, in his case, he can't get what he would like to get, the B100s yeah but he can't.

Alex: Yeah.

Steve Hsu: Yeah.

Alex: One thing I can add is, you know, in model training, it's kind of a high-risk thing. So I think even if like the Ascend GPU is of equally good quality, people are afraid of switching to something that's less established and

Steve Hsu: Yeah

Alex: less tested. I think another thing we might see in this computing catch-up process is like maybe the first thing to really catch up will be like the gaming GPU. Then you'll see the inference GPUs catch up, so more people will use the Chinese Ascend for gaming, or sorry, for inference. And then maybe post-training fine-tuning will catch up, and then pre-training might be the very last thing to catch up.

Steve Hsu: I think one of the important things is that now that DeepSeek is fully optimized for the Huawei architectures, at least for inference like that is a big development. And my understanding is the pre-training which requires the really excellent networking that NVIDIA provides, is increasingly a smaller and smaller portion of the total compute that's involved in AI.

Because there's a bunch of inference, which is generating tokens. And there's a bunch of stuff like in post-training where it isn't really you're not really updating all the weights, you're just

Alex: Yeah

Steve Hsu: updating small subsets of the weights, and so the bandwidth aspects of the hardware are not as important.

Justin: Mm-hmm

Steve Hsu: So I think that the place where NVIDIA has its biggest advantage is like shrinking as a fraction of the total amount of compute involved in AI.

Justin: Mm-hmm

Steve Hsu: That, that's my impression. So any other observations you wanna make about competitiveness between US and China? It doesn't have to be about AI or tech, it could be about just like the day-to-day. When Han Feizuo, when you were on the podcast earlier, we talked about the convenience of how, how nice it is to live in China, like the food delivery, the good food, everything's so inexpensive.
Anybody wanna comment on that aspect of it? What, what your daily life is like here compared to what it would be like in Montreal or somewhere else?

Justin: Oh, I guess before we go on talk about daily life, I think, I think one province and one university to definitely watch out for in China is this university in Anhui called USTC, University of Science and Technology of China.

This university is producing some of the world-class, like chip engineers and one of the top memory companies in China called ChangXin Memory Technologies, CXMT, is just based out of there as well.

Steve Hsu: Yeah. I, I think for people in physics, we've been aware of USTC for a long time because although it's in, out in the middle of nowhere, it's been one of the most excellent Chinese universities for a long time.

In fact, in fact, I would actually say, you know, Beida, Tsinghua, and then USTC, at least from the physics perspective, are probably the, the top universities. USTC has been good for a long time, and they also had a genius program for a long time, so they were admitting kids to USTC at age 15 or 16 for, for a long time.
And so, like a lot of the best people that we would see in the US were not Tsinghua or Beida kids. They were actually USTC kids who had gone to college when they were 15. So that, that's that's been around for 30 years or more.

Justin: Yeah, that definitely makes sense on the physics and I think even the electrical engineering side.

Steve Hsu: Yeah, yeah.

Justin: And then on the AI side, it would be Tsinghua, Peking, and then Shanghai Jiao Tong

Steve Hsu: Yeah

Justin: which has their ACM class. Yeah.

Steve Hsu: And there's also Zeda is supposed to be very Zhejiang.

Justin: Oh, Zhejiang, yes. Yeah.

Steve Hsu: Zeda is supposed to be quite good.

Justin: Yeah. So

Alex: So the urban layout and the way the roads work is super weird if you're from the US, for example.
So a lot of people like to ride these little two-wheel scooters around, like they're always electric. They're super cheap but it's just funny to see a whole group of people riding around in these things. But to make it work, you know, they have big bike lanes, and then they have really wide sidewalks, and then people park their two-wheelers in the sidewalk.

Yeah. On the sidewalk. And, and then also the urban layout of Beijing is super weird because they basically stuffed all the universities into the top left corner of the city. So you got like Tsinghua, Chinese Academy of Sciences, Peking, Renmin. So this, I don't know, it's very unusual to just have all the universities stacked together. It's, it's

Steve Hsu: a little bit like Boston, Cambridge in the US where it's like there's so many colleges in that town, and this Haidian, this, this northwest, is it northwest part of Beijing is like that. Yeah. It's like that's where all the universities are, and they're pretty close to each other.

Alex: Yeah.

Gabriel: But not just only the univers-versities, but the political centers here as well, so

Steve Hsu: And the tech companies.

Speaker 4: Yeah.

Steve Hsu: Yeah. You know, I was gonna say that the, this electric bike culture is very unique here. I, I was talking to Kaiser Guo, Kaiser Kuo. And he says he gets the best way to actually get around Beijing is just to ride around in his forgot the name, but it's like Electric Turtle or there's some name for what he has. It's just some classic, like, electric bike, and he's just rides it around Beijing, and you avoid the traffic that way.

Alex: Yeah. Yeah. 'Cause you can kind of go anywhere with the electric scooter 'cause you can do sidewalks- bike lane, or the main road. Whereas cars and pedestrians are kind of limited to just one of the three.

Steve Hsu: So China's very safe. It's very safe. There's low crime rate and everything, but the, the most dangerous thing, I always tell my friends, "If you heard that I was killed in an accident in China, I was killed, I was killed by an elec like a delivery guy on an electric scooter who just hit me while I was on the sidewalk." And that's like the one dangerous aspect I find.

Justin: It's, it's also funny, most people don't, don't wear helmets.

Steve Hsu: Yeah. Well, the, the delivery guys do wear helmets. They do. They do. Yeah. Yeah.

Gabriel: I think they just made the they just changed a law to, like, enforce it, like, this week, I think.

Steve Hsu: Yeah. No, I just someone just told me you have to wear a helmet now.

Gabriel: I saw that. One of my friends got a ticket.

Steve Hsu: Yeah.

Gabriel: Oh, interesting.

Steve Hsu: Okay. So I think that electric bike culture is coming to the US, and it's an example of something that actually, although most Americans don't realize it, it originated in China, and now it's coming to the US. Yeah. So now you can buy inex- relatively inexpensive electric bikes in the US.

They're all made in China, and it's becoming a thing where, like, people who have a slightly longer commute Would get an electric bike

Justin: Yeah

Steve Hsu: and ride it. You know, and so that, that's, like, coming to the US now.

Gabriel: Yeah. Yeah.

Steve Hsu: Yeah.

Gabriel: I, I would imagine, like, the car culture is so ingrained in America that they just won't change it.
Steve Hsu: You know, I think it, it's only gonna come to certain, like, cities probably or, or campuses where people. But I see more and more electric bikes now. I mean, in Berkeley I saw a lot of electric bikes.

Gabriel: Ah, yeah. So the universities, they're starting to have these electric bikes because of the Chinese students there.

Steve Hsu: Oh, that could be it. That could be the vector. That could be like like yeah.

Justin: Yeah. Wait, weather's gotta be a factor, too, 'cause if you have ice on the road, electric scooter would be so scary.

Steve Hsu: Yeah.

Justin: They somehow

Gabriel: man they somehow manage it in Beijing, so

Justin: No, no, Beijing has really dry winter, so there's usually not ice on the road.

Steve Hsu: Yeah. It may not work in Montreal. Okay, anything else? Hot

Han Feizi: hot I've got one. Yeah. Just, just a practical thing for Gabriel. If you are a high school student in the US what, what kind of student would you recommend apply to Chinese universities? And what preparation do you think do you recommend they do, say, a couple years out?

Gabriel: Oh, that's a, that's an interesting question. I think they should definitely at least be interested in China. That's prerequisite. regarding Chinese language, I think it's definitely better to have prep. I wouldn't Say it's like completely bad not to have any prep at all, as in I know Tsinghua allows foreign students to do like a one-year language course, like before you go on to start your actual university degree.

So Tsinghua's actually actively recruiting a lot of just like foreign-born people. So that's, that's one of their goals 'cause they seem to have a lot of like Chinese people with foreign passports trying to just come back. But yeah, so they're actively trying to recruit these actually foreign-born people.

I, I guess the main thing is oh my, my friend did recommend me this. hopefully you're interested in STEM and you're actually interested in research. If you're here for just like a, a humanities major, you won't get nearly as much out of it, I think

Steve Hsu: My, my wife's a professor of literature and film. And she, she's gonna be on sabbatical at Tsinghua this fall. Oh, cool. So, so hopefully there is some humanities here. She's been actually a visitor at Beijing University and now at Tsinghua, so we'll see how they compare in terms of humanities.

Gabriel: Does she do like I don't know, oriental literature, or is it

Steve Hsu: her, her focus is mainly, yeah, modern Chinese literature.

Gabriel: Oh, then of course you get a lot.

Steve Hsu: Yeah. Yeah

Justin: yeah. I guess I have some extra advice for that, for those high school students. So make sure you really know how to speak Chinese, and you can make friends with Chinese people. Yes

Gabriel: you gotta be social. I think, I think you're a good example.

Justin: So, so there's a very clear split in Tsinghua.

Among the internationals, even among many of the so-called overseas Chinese who can speak a little bit of Chinese, they still don't interact much or at all with domestic Chinese students. And I feel like I can even see this case among Malaysian students who are they, they grow up in a Chinese schooling system.

They speak Chinese all their lives, but even then they have some trouble like interacting with the Chinese domestic students. So, so you can imagine how hard it is for people who don't speak Chinese well at all. But I think, I think Chinese students are very friendly, and, and they're very happy to be friends with the outside people.

So I, I would encourage people to really just reach out to Chinese people and try to, try to be friends with them. And, and, and a very interesting person I met here is, he's a, like a, just a pure, pure-blooded American, white American from Tennessee. And he was working in the battery industry in the US for a few years.

And one day his Chinese coworker was like, "You teach me English, I'll teach you Chinese." And he's like, "Okay, yeah, let's do that." And he studied Chinese for like a year, and then like very intensely for a year. And then he came to Tsinghua for his master's in material science, and he only exclusively hangs out with Chinese people.

His Chinese is amazing, like I would I would say in some, some ways better than my Chinese, which is so impressive. Yeah. And, and very embarrassing for me as well.

Steve Hsu: It's wild. I, I think though the hard part, at least for me, would be becoming fully fluent in the written language because, like, for me, it's really hard to memorize the characters and stuff like that. I just, I, I'm completely amazed by people who could come here in a year and suddenly learn how to read.

Gabriel: Yeah, I can't do that as well as to this, I'm struggling a lot with it. So, so in texting, You're gonna

Steve Hsu: have to wear your AI smart goggles wherever you go.

Gabriel: Yeah. Exactly. I need that. Luckily, there's AI now, so I just click a button and translate everything.
Steve Hsu: All right. We're almost out of time. In fact, we only booked this room until 3:00, so we're gonna have to get out of here.

Steve Hsu: But let me, let me ask you two AI guys Do you have any predictions that you might think would be surprising to the US listeners or the non-Chinese listeners? Any, anything about like robots, robots in the home, or who's gonna be leading the model race, LLM model race a year from now?

Anything that you think is sort of a little bit non-obvious for the listeners?

Gabriel: Maybe the most non-obvious one, although it's also the quite. It might be the most uncertain one is probably the, the phase transition in when they start mass producing EUV machines. Like, that is a very politically sensitive issue in China as well, so you can almost get no, like, high quality information about that. But, but it seems like they're gonna start mass producing that soon.

Steve Hsu: Okay. I wanna, okay, I wanna drill down on this 'cause I'm intensely interested in this question. Yeah. And I even like when I was in Shanghai, actually right after I taped with you . Han Feizi, last time I went to Shanghai, and I another Manifold listener who's Taiwanese Chinese brought me to dinner, and I met the top leadership of SMIC who are all Taiwanese. It's wild. They and their wives were like all Taiwanese people. Even they like we wouldn't really talk about EUV 'cause it was so sensitive.

Gabriel: Yeah.

Steve Hsu: you are asserting that we might see actual EU Chinese-made EUV machines in production soon. Is that what you're saying?

Gabriel: But this is based on information that people can rumors that people, other people can also find on the internet. So this is like, this is like. Okay. Low quality information.

Steve Hsu: But there's no, there's no like actual Tsinghua rumors like, "Oh, my, my lab mate went off to this huge facility that Huawei runs in near Shanghai where they're actually. Maybe he's working on EUV." Like that. You're, you're not hearing rumors like that.

Gabriel: I can okay. I may be able to ask around. I do know many of the the physics engineering students. Yeah. And many of them, they go to the, the SME manufacturers like CMXD, YTMC Yeah. Side carrier.

Steve Hsu: Yeah.

Gabriel: Huawei as well. Yeah.

Steve Hsu: When I met with Taylor Ogun, this hedge fund guy in Shenzhen, he has he was tracking this greenfield, the site that Huawei.

And, like, he has a lot of inside access to rumors and stuff. So he claims there's this huge site where they're developing EUV stuff. Huawei is doing it. And so, but the question of like how far they are from actually shipping a machine that's used to make chips, I, I have no idea. But there is definitely intense effort.

Gabriel: Yeah. This

Steve Hsu: And there must be people going to staff that facility, right? Yeah. So, so

Gabriel: So, so in particular, Huawei does have a campus of like, like maybe one to two million R&D people.

Steve Hsu: Yeah.

Gabriel: Yeah.

Steve Hsu: Yeah.

Gabriel: On somewhere in the out-outskirts of Shanghai.

Steve Hsu: Yeah.

Gabriel: yeah. But, but, but we wouldn't be able to get any like real information on this because there, there, there were like research groups who tried to dig into it, and they were raided by the government.

Steve Hsu: Oh, I see. So the government's specifically trying to keep a lid on

Gabriel: it. Yeah. So any yeah. Anyone who actually wants to do like rigorous analysis and research on this,

Steve Hsu: Yeah

Gabriel: they, they don't want to do it. They wanna stay away from it yeah because it's a red line.

Steve Hsu: Yeah. Interesting. Okay. Well, maybe now is a good time to call it. Awesome.

Gabriel: yeah.

Steve Hsu: thanks to you guys. I hope for our listeners this has been informative, and now you know more about what life is like at China's top technological university. Thanks for listening. Bye