AI is all the rage right now. There are billions of dollars now flowing into the space, with large and small companies all competing to create the next big thing. But in addition to lots of money, building new AI models requires top-tier researchers. So, who's attracting the best? And what does it take to be considered top talent in AI anyway? On this episode we speak with Damien Ma, managing director at MacroPolo, the in-house think tank of the Paulson Institute. Damien helps put together MacroPolo's
Global AI Talent Tracker
, which monitors the flow of top-tier AI researchers around the world. We discuss who's winning the AI talent war so far, the purported talent drain in China, competition from India, and much more. This transcript has been lightly edited for clarity.
Key Insights from the pod:
Tracking AI Talent — 4:28
The top 20% vs the top 2% of AI Talent — 6:48
What makes a really good AI engineer — 7:21
Where are AI Talent coming from and where are they going? — 8:49
US.AI Talent and Immigration — 12:15
Chinese domestic AI talent — 14:20
How tight is AI talent labor market? — 16:42
Broadening AI graduate programs in China — 18:36
Chinese Internet Companies and AI — 20:08
Will AI ramp up US-China competition? — 22:27
Domestic Chinese rhetoric around the talent war? — 24:18
Have US universities increased AI research capacity? — 27:04
Will AI technical skills be automated in the near future? — 28:29
Is computing power a recruiting tactic for AI companies? — 30:47
---
Tracy Alloway (
00:17):
Hello and welcome to another episode of the Odd Lots podcast. I'm Tracy Alloway.
Joe Weisenthal (
00:22):
And I'm Joe Weisenthal.
Tracy (
00:23):
Joe, have you watched The Three Body Problem?
Joe (
00:25):
No, but I really want to, and I didn't read the book, so in case you're going to ask that I didn't. I want to do that too. I intend to at some point.
Tracy (
00:33):
There goes my carefully-crafted intro where we talk about the Three Body Problem. Okay, well this will work.
Well, as everyone knows, except for Joe, there's sort of two types of people in the world when it comes to the Three Body Problem. There are those who see it as an allegory for climate change. So, humans coming together to unite against a common threat, which in this case, since you haven't read the book, is an alien invasion.
Joe (
00:59):
An alien civilization. A friend of mine this weekend told me like two plot points.
Tracy (
01:04):
Okay, good. Yes, okay. And then there are also those who see it as sort of an allegory for the trade or tech war between the US and China. So, the idea that humans are going up against a much more technologically advanced opponent, and in this scenario, I guess Earth is China and the aliens are the US. Well, today we are firmly in that second camp. We're going to talk about US-China rivalry in tech, and in particular, one area of tech: AI.
Joe (
01:36):
So obviously, AI, AI, AI, everyone talks about it all the time. We don't really know where it's going to go, but we know a few things in the meantime, which is that people are spending money like crazy on chips, but they're also spending money like crazy on talent.
And anyone who is capable of doing sort of cutting edge research in AI, from what I can tell based on articles I read, they basically just get to pick where they want to work and basically pick their salary. There's a great article on The Information a couple of weeks ago about Facebook hiring top researchers without even doing an interview. It's like, if you know this stuff, someone will hire you and pay you a lot of money.
Tracy (
02:11):
Yeah. And I have so many questions in this space. So first of all, who is an AI talent? Or what is an AI talent? Where do they come from? Is it the same as being a software engineer but you have a slightly different area of expertise? I really don't know.
And then secondly, I'm kind of curious how fungible the jobs are. From what you just said, and the fact that companies are hiring without interviews and things like that, and that demand is so strong, it seems like you can just do AI anywhere, whether it's China or the US or somewhere else in the world, or whether it's a specific company versus another one. But so many questions on this AI talent war, I guess you could say.
Joe (
02:52):
Totally. And there's two things. So I sort of consider myself a bit of an AI talent because I think I'm pretty good at coming up with ChatGPT prompts.
Tracy (
02:58):
You are actually. Listeners, I have learned a lot from watching Joe enter his prompts. And I still find it incredibly endearing that you say, “please” and “thank you.”
Joe (
03:07):
Well, it's important for when AI becomes sentient, they're going to remember who said “please” and “thank you.” But beyond that, there's this other element, and you already sort of alluded to it, but it's clear that, for whatever reason, countries [treat] AI almost as if it's a commodity.
There's this narrative being pushed by the industry, and maybe it's just a narrative to sell chips or subscriptions to the OpenAI, APIs, etc. But there seems to be this narrative that every country must have some sort of homegrown AI strategy or data center or something. Like, something about this technology seems to engender political and nationalistic anxieties.
Tracy (
03:52):
Yes, I think that's absolutely true. And we're back to the Three Body geopolitical tension point. But, I am very pleased to say that we in fact, have the perfect guest to talk about all of this.
We're going to be speaking with Damien Ma, he is the managing director at MacroPolo, which is the think tank at the Paulson Institute, and they publish something called the Global AI Talent Tracker. So actually keeping track of where AI talent is coming from, how much there is, and where it's going.
So Damien, thank you so much for coming on Odd Lots.
Damien Ma (
04:26):
Thank you so much. It's great to be here.
Tracy (
04:28):
How long have you guys been doing this Talent Tracker and what was the genesis? Because for me, ChatGPT and all the chatbots seem to have come out of nowhere almost basically a year ago. So how did you get an early start on tracking AI?
Damien (
04:43):
Well, the original conception is that we thought a little bit hard about what would you need to have a robust AI ecosystem or an AI industry? And we thought there are three key pieces.
You need, obviously, compute power, so things like chips and the infrastructure. And you need a lot of training data. Data's obviously everywhere now, and we thought the last piece that people haven't thought too much about is human capital, because it is a very human capital intensive area and discipline because it's highly complex and complicated and you need highly trained people to be able to do it.
So we thought, nobody's really looked at the human capital side of things, is there a way to do that? And then we sort of found this one conference that's widely known in the AI community as one of the most prestigious. And so we looked at papers and researchers that went to that conference. This was back in 2020.
During the pandemic was when we first launched the initial tracker that gave us our idea that's a proxy for sort of the top 20% of global AI talent. So this is not all AI talent, this is not everybody in the world, but this is really sort of what we might call the cream of the crop, top 20%. And within that, there's also the top 2%. So we're looking at really kind of the elite people, which is probably the type of people that's being fought over most fiercely because people want the top talent.
Joe (
06:03):
Real quickly, what's the conference?
Damien (
06:04):
It's called the NeurIPS. It's a conference that's held I think every year, but we didn't track it every year. We tracked in 2020 and then we did it again, and we looked at 2022. We were trying to see, had there been any changes after the three year pandemic to see if there were different mobility patterns?
This is a conference that's mainly focused on neural networks, large language models, also a lot of things that are currently really pushing the frontiers of generative AI. So we thought that those are the kinds of people that would probably want to work for the Googles and OpenAIs and Baidus of the world.
And so that seemed like a good sampling. Again, we don't pretend that this is comprehensive, but it is sort of the elite 20% sample.
Joe (
06:48):
Just real quickly, since you say you're able to distinguish between the top 20% and the top 2%, how do you do that part? I mean, it can't just be people who attend the conference. Like, how do you sort of grade or figure out who is this specific, ultra elite AI engineering talent?
Damien (
07:04):
So we looked at authors whose papers got accepted, and within that acceptance, there's an oral presentation. You don't get accepted to oral presentation unless you're really, really good. So there are only about 2% of people that got accepted to the oral presentation. So that to us was sort of the proxy for the 2%.
Tracy (
07:21):
This kind of leads into what I was wondering, which is what makes a really good AI engineer? Like, what is it that would lead them to be someone who presents at a conference like this?
Damien (
07:32):
I mean, Joe just said, he's a really good prompt engineer, so…
Joe (
07:38):
So they would let me present?
Tracy (
07:39):
Joe, I'm sure your invite’s in the mail.
Damien (
07:41):
Like, really curate the questions well. But I think that's really good…
Tracy (
07:45):
But it’s not just curating the questions, right? It's like actually coming up with the natural language models and things like that?
Damien (
07:51):
No, no, no. Yeah, so I think it's a really good question, and I'm not sure the distinction is huge. I think the foundation of AI is all computer science. Most AI people would call themselves computer scientists first and foremost, or people that have a lot of mathematical training.
And in fact, I think some of those people back in the 2000s and 2010s were the same people that got attracted to big finance, right? And went to build algorithms for trading desks. Those are probably a similar type of people. Now they're just doing AI and the AI specific apply-part is being able to train large amounts of data and being able to write out algorithms.
But those are the things that you would get from computer science training with a bit of sort of added AI specific component to it. And I think the neural networks thing is probably, one distinguishing characteristic is trying to really figure out how do you make the computer mimic the human brain in a way. But fundamentally it's just mathematics, quantitative computer science, all those things, you know, eventually can become AI scientists.
Joe (
08:49):
So there's a certain type of person who is seeking out the hardest or maybe most lucrative sort of real world math problem or computer science problem at any time. Maybe in the 2000s, they were going to Wall Street to figure out the best way to create new securitized products and derivatives.
In the 2010s, they went to Facebook and Google to figure out the ways to pack the most number of ads on a smartphone or get you to click on them. And now apparently they're going into AI research. So let's start with what the data shows big picture. When you first started collecting the data in 2020, where were they coming from and where were they going?
Damien (
09:27):
A lot of them came out of China and the United States in 2020. That was pretty clear. Most of them ended up in the United States by far. And we're still seeing that in our latest update in 2023. Although, I would say the big surprise was that China has done a really good job really ramping up its domestic supply of top AI scientists. So they're producing nearly half of the world's top tier AI scientists now.
And many of them are actually also staying in China. And the reason is, I think it's pretty simple, is that China is obviously been focusing on its own AI industry. And, as we already said, people go where the jobs are.
And if you look at the major economies where they're focused on building out AI industry opportunities, it's probably the United States and China. And if you look at Europe, Europe actually, I think, punches way below its weight in terms of having an AI industry. And so they don't tend to attract as many top tier AI talent as China or the US.
And if you look within top US institutions where top AI talent work, it really is almost a Chinese-American duopoly. Chinese-origin and American AI scientists are 75% of the top AI talent within US institutions.
Tracy (
10:46):
What are the factors that would go into, say, a computer scientist who has been educated in China and they're surveying the different opportunities available to them, what are the factors that would go into them making a decision? Like, are there immigration considerations? I imagine pay and remuneration would have to factor into that. How easy is it for them to switch from China to the US?
Damien (
11:14):
I think the skills and the training is fairly similar if you come out of a top program, whether it's Tsinghua in China or Stanford in California, I think the key from what we're seeing, you know, one key indicator of where people end up for work is really where they go to graduate school. That's probably not a surprise.
If you're going to do your Master's or PhD somewhere, you generally start to search for job opportunities near you, around you. Unless you happen to be in a country [or] in an area where there's not a lot of opportunities post-graduation. And of course what, when you're considered an elite AI talent, you generally have a terminal degree, usually a PhD, but at least a Master's.
So I think where you choose to go to graduate school is really important. And we see that in the data too. Those who come to the United States for graduate school by and large, tend to stay in the US to work unless there's some very lucrative opportunity that attracts them back home or somewhere else. But generally there's a bit of a path dependence between graduate school and staying in that country to work.
Joe (
12:15):
There has been a lot of anxiety for years in the tech industry where you see CEOs and leaders complaining that the US immigration policy has made it too hard to keep talent who has graduated in the United States. And there's this idea of like ‘Hey, if they're going to come here for education, why are we not reaping the benefits of this US educated talent?’ It does seem like from your data that still many are staying in the United States, but the numbers have changed since 2020, yes?
Damien (
12:43):
Yes. They have gone down a little bit. We didn't go into really exploring exactly what happened over the last three years, in part because I think many people realized the pandemic years have been a little strange, whether it's for economic data or just general mobility for people where people work, how people work.
So there's going to be a lot of distortions in those last three years. But there has been a relative decline, especially among the Asian talent. It's not just China. India has also done a better job retaining its own top tier AI talent. South Korea, interestingly, that's not on our data set yet, but we're about to publish regional South Korea. They've retained 90% of their talent. They've not let anybody leave. And they've been really good at doing that.
nd places like France have actually done a very good job on retaining their talent. So, I can't say definitively what the reason is, whether countries have stepped up their game to retain domestic talent or there's been other things that happened in the pandemic that's triggered it, or there could be immigration challenges and so on. I think maybe in the future when we do the next iteration, we will have more clarity to see the pattern. So I'd be a little hesitant to give definitive conclusions at this point.
Joe (
14:07):
Tracy, if France does a really good job keeping their talent, who will fill the niche of blowing up trading desks with exotic derivatives, if all of those École Polytechnique and Sciences Po graduates go into AI instead?
Tracy (
14:20):
Yeah, it is always a French person working in equity derivatives with a mathematics degree. You're absolutely correct. But on the degree topic, so I hadn't realized that in China, and Damien, I think this factoid was in one of the reading materials that you sent, but Chinese universities have launched more than 2,300 undergraduate programs since 2018, when the Ministry of Education designated AI as a separate major that's distinct from computer science.
So first of all, how common is that? That you would get the separation between computer science versus AI. Is that the standard in other parts of the world or is it still relatively new? And then secondly, presumably, this is part of China trying to build up its domestic AI talent pool and eventually its capabilities in this area. What else is it doing on that front?
Damien (
15:15):
Yeah, so that's why one of the reasons we think that China has really seen this boom on top AI talent is you have just kind of a graduating class in 2022. If you start in 2018, some of them are graduate students, some of them are undergraduates. So they've really pushed really hard to grow the AI talent, but not all of them are the top 20%.
But I think China looks at it as a way that they're going to need a lot of AI specific technicians. China's not really thinking about AI in the generative AI sense. I think there are definitely some startups and folks pursuing things like ChatGPT chatbots. But my understanding is that China's probably going to focus much more on industrial applications of AI manufacturing: robotics, probably healthcare, biotech. I'm going to bet that's going to be a huge application for China.
And I think for obvious reasons, generative AI is probably not as copacetic with the governance system in China ultimately. And I think that's a pretty clear thing that I think everyone knows, but I think they're really looking at how to apply artificial intelligence to energy, to industry, to advance manufacturing, or things like climate.
That's where China's really focused on. And I think they feel like they need a lot more people, not just the cream of the crop, but middle level technicians, people that are just familiar with being able to run data or to run Python or to just check all the data. So I think they're viewing AI as a very wide, expansive way of creating certain jobs.
Tracy (
16:42):
Yeah, I can't imagine China's ambition here is to have like 5,000 different chatbots. Like there is clearly a tendency towards industrial, real world applications of this technology. On which note, do you think there's currently enough places for AI graduates or specialists to actually go within China? Because in some respects, it feels like this might be a very hot degree, people are being encouraged to do it, but, at the moment, companies aren't necessarily at the same sort of level. It feels like there's sort of a mismatch in the evolution of this at the moment.
Damien (
17:20):
I think you're absolutely right. So we've seen the kinds of bubbles before, that the new hottest sector in China, everyone goes there because they think that's where the opportunities are. And then, China already had what we would call a college bubble for the last 10 years. And that's why you have really high youth joblessness in China
The way I think about how China works in that respect specifically is that they're basically two different cycles in China. There is a policy-induced cycle and then there's an actual market cycle that comes after that. So right now we're in this policy driven, like, you know, ‘You guys gotta come in and we really like AI, we're going to create all these programs and you should just get AI.’
And then, you know, parents are like ‘Whoa, whoa, that seems like the good new thing. And, that's what the government's promoting. So all my kids that are going to do computer science, they're going to add the AI component to it.’ So that's sort of the policy induced cycle.
And then after that, once the bubble happens, it will kind of eventually get into a market cycle where it'll correct a little bit and then people will be like ‘Oh, well actually we probably now have an oversupply of a lot of these middle AI technicians that will have no jobs. What are we going to do with them? We don't know.’ So I think this is a pattern that happens in China a lot, and I wouldn't be surprised if it happens with the AI talent pool as well.
Joe (
18:36):
So there's a lot of interesting threads to pull on already in this conversation. And I want to return to the non-chatbot applications of AI, like, how can we make better robots and factories and drug discovery, etc.
But I want to ask another question. So, okay, all these new institutions or graduate programs have been launched in China and more and more universities offering degrees in AI or computer science or related fields. In my mind's eye, if I imagine what a top AI researcher [is], I imagine maybe they have a PhD from MIT or Stanford or something like that.
When you look at the institutions in China, has there been any sort of broadening out of the number of schools that are capable of producing either those top 20% or top 2% talent beyond just the sort of like handful of schools that we for a long time understood as the elite schools?
Damien (
19:30):
There have been a little bit. And when it comes to Asia specifically and China, I think they have 11 of the 14 top AI institutions in Asia. But in terms of just top in general, China has climbed quite a bit. A place like Zhejiang University, Shanghai Jiao Tong, which are not your traditional names that you would hear.
Joe (
19:51):
Yeah, I've never heard of either.
Damien Ma (
19:52):
It's not PKU, it's not Tsinghua. And interestingly, you enter into 2022, Huawei is actually one of the top 25 institutions for AI research globally. So they've invested a lot in hiring top AI talent for obvious reasons.
Tracy (
20:08):
Oh, this is actually exactly what I wanted to ask you next, which is, you mentioned Baidu as well earlier in the conversation, but in terms of domestic destinations for AI specialists, is the idea here that a lot of the existing internet companies in China, that they're going to devote more development and more resources to this particular technology as we've seen here in the US, but also that maybe some of those big consumer internet companies, the ones that had a very rough few years during Xi Jinping's big crackdown on disorderly capital expansion, that they're going to pivot as well?
Damien (
20:49):
So I think that's basically correct. Baidu, as far as I'm concerned, has basically become an AI company. And I think they made that strategic change many, many years ago. And one of their big focuses is, I think, like Tesla, autonomous driving. And no one has really been able to crack that.
I think that's sort of the AI frontier that everyone's really focused on is, is how to solve vision, right? Because everyone's now focused on how to solve language, which is what generative AI and a lot of the products we see today is kind of language based, but vision is a really tough nut to crack.
And Baidu is the one in China that's really been trying to solve it. And I'm not sure their progress is any better than Google or anybody else. But in terms of some of the software companies like Alibaba [or] Tencent, Tencent has been doing a lot of AI investments and obviously ByteDance. So there's been a lot of that.
But what we're also seeing, we did a recent piece where we looked at where Chinese VC money has been going, venture capital, whether venture capital's going to a lot of these places, but in fact, venture capital actually has invested less in software in the last few years, but actually invested more in sort of hard tech, you know, hardware. So similar things to like the advanced manufacturing side.
So I really think in the next few years, we're going to see a lot of money, private and public, going into these advanced manufacturing hard tech side of things that will have AI applications. And I think there'll be some startups in China that probably we haven't heard of today that's going to put a lot of money into AI. But, the big guys are doing it. But Baidu is probably the one that's the most prominent in trying to solve the autonomous vision problem. And they will be a big employer in China, for sure, for AI talent.
Joe (
22:27):
So going back to the other industrial applications of AI, already there's this just tremendous anxiety in the US and Europe about whether there's any way to catch up with China's advanced manufacturing prowess. Whether we're talking about cars, whether we're talking about batteries, certainly whether we're talking about certain types of chips.
Should the US be concerned perhaps that here, chatbots are the shiny new thing and everyone wants to work on a better chatbot, and in the meantime, China gets even better at automated factories, particularly I imagine, with better vision technology, that factory floor robots could be safer or could be more agile, etc. Like, do you see a sort of further widening of the nature of the US-China competition as a function of where the AI talent is going?
Damien (
23:17):
I'm not sure I can give you a very satisfying answer. I guess, the way I would think about that, something that would be emblematic of both advanced manufacturing and AI applications, sort of the software and hardware. I think the key for both countries, and I think all countries, is probably going to be in robotics, that's sort of the new frontier of, whether it's the optimist, humanoid robot, China's got, I'm guessing, like half a dozen robotics startups already.
So if one country [or] one company succeeds in that arena and is able to really blend that hardware and software and make it work and commercially viable, I think that could send a lot of strong signals about the relative capabilities of each country.
Tracy (
23:59):
Are you going to start a Robotics Talent Tracker?
Damien (
24:03):
Robots is, that's going to involve a lot of supply chains. So it's a little tougher than just looking at the people. You’ve got to bring in the chips, you’ve got to bring in the engineers, the mechanics. So it's more than just the AI scientists when it comes to robots. But interesting for sure.
Tracy (
24:18):
So one thing I wanted to ask, because you are looking at this world very carefully and sort of watching what people are doing and saying, but what is the language that I guess policy makers in China are using around AI talent? Like, what sort of statements do you tend to hear? And I'm thinking back again to that famous disorderly capital expansion phrase that Xi Jinping deployed when he was cracking down on things like the education sector and consumer internet companies and stuff like that. But how is this whole dynamic, this talent war, couched among policy makers?
Damien (
24:58):
I think it's natural and it's [a] given that no country generally likes brain drain. Everybody wants to have brain gains. And I think that rhetoric aside, the actualization of that and how you set up your own country, how do you set up the environment and incentives, compensation, all sorts of things.
The thing about top tier talent in any arena, but particularly in computer science and these sort of frontier technologies, most of that talent, I would imagine, would want to be in the most competitive and dynamic industries. That's where they probably feel the most comfortable. That's where they want to make a difference. That's where they want to make an impact. And obviously the compensation, all that stuff follows that.
But I think they want to have the freedom to do the best cutting-edge work possible. So I think having dynamic industry is really important. And so I'll bring up the Europe example again. Europe doesn't seem to have that, which is why they've consistently been sort of underweighted when it comes to attracting top tier talent.
And if you look at the UK, which has been the main place in Europe where most top tier AI talent work, but in the U., most of them work for Google DeepMind, which is a US company, right? Having that industry is, I think, really, really important.
And so, in our current debate about regulating AI and industry, I think it's going to get controversial, it's going to get testy. We all know that. We all can see that. But I think we have to think about, if countries want to attract the top tier talent, they want to work in the most cutting-edge dynamic thing where they can do the coolest, the most transformative stuff possible. And if that's in America, great, but if China does that, maybe it's China. But you know, right now China still mainly relies on its own domestic talent. They're not really importing much foreign talent either. So to me, I think having that industry is really, really vital.
Joe (
27:04):
What are US universities doing? I imagine a high schooler graduating in 2024, probably way more than four years ago or even one year ago, are saying like ‘Oh, yeah, well this is what I want to do. I want to work in AI or something in this realm.’ Have we seen an expansion of what US universities are offering or capable of offering? Has there been that sort of supply side capacity increase here to take advantage of what is almost certain an increased interest in this industry?
Damien (
27:33):
Well, did you see the WSJ piece yesterday where all the Gen Zs are becoming plumbers and electricians?
Tracy (
27:38):
Oh, I did, yeah. A return to trades.
Damien (
27:41):
Yeah, I mean, frankly, if I were 18, I might consider that route. But my understanding is that a lot of the top tier technical schools or things that have a technical school reputation, whether it's Stanford, you know, Caltech, MIT, Carnegie Mellon, I mean, they definitely have AI programs.
I don't know if it's to the extreme volume that China has offered in a span of two or three years, but they've definitely added those. But again, the foundation really is computer science. So I think if you go in and study computer science or some sort of mathematics foundation, that's going to get you into AI one way or another much easier than if you just go straight into AI because you can't really think about AI without having any foundational knowledge from CS or mathematics.
Tracy (
28:29):
This might be a weird question, but it's related to the idea of people choosing to become plumbers or plasterers or whatever it might be. Do you sense a sort of note of caution among potential graduates in the sense that a lot of people in recent decades were encouraged to go into coding and become fluent in Python or Rust or whatever it might be.
And now, we've seen the rise of AI, we've seen models that can actually write your code for you pretty much. And a lot of software engineers are currently a little bit worried about their job security and the outlook for their skills. Does that impact the potential AI talent pool at all? Like, is there a sense that, okay, I can get into this, but then maybe in 10 or 20 years the AI is just going to be developing itself, right? Self-learning models are already a thing, so why get into it at all?
Damien (
29:27):
Oh yeah, that's a tough question. Can AI be so good that it doesn't need any human input anymore?
Tracy (
29:32):
Again, I've been watching the Three Body Problem, so a little bit of a sci-fi bent.
Damien (
29:36):
I don't know, I can't see that far into the future. But what I will say, I guess kind of the more realistic near term future, I think we said earlier that, if AI is able to really solve human language, which is obviously a big indicator of human intelligence, and that seems to be a lot of where the efforts are.
Large language models and trying to figure out, how to mimic human language, human thought through language. I would say one of the areas that's probably going to be in trouble lot is translators. That whole area seems like it's going to be probably, for lack of better term, disrupted quite a bit.
Or if you think about somebody that needs to do research in different languages, maybe in two or three years I can read Japanese as easily as anyone else, just get it quickly translated on some AI software. And I can be pretty fluent in reading Japanese. That doesn't mean you shouldn't be studying foreign languages. There are a lot of intellectual benefits to that, but I think as a research tool and as the ability to kind of use it as a way to understand the world, once AI really gets to that point, there are going to be a lot of disciplines like translation, interpretation, those kinds of things. It doesn't seem like there's going to maybe be a huge need for that sort of stuff.
Joe (
30:47):
So in the earlier part of the conversation, we talked about three necessary components to have a domestic AI industry. One is talent, one is infrastructure, and then the other one is just the pure compute. And we see companies like Facebook, like they tout as an advantage, ‘We just acquired so and so many H100s from Nvidia and we're spending $10 billion.’
And I kind of get the impression that having a lot of computing power is a recruiting tactic, and that if you're a top AI researcher, you want to be at the place that has the most advantageous sort of raw computing capacity. We know that there's a lot of restrictions on some of the cutting edge semiconductors going into China and Jensen Wong of Nvidia has talked about this and the constraints there.
For a potential talented AI researcher maybe from China or studied in China, does that factor into it? The fact that, at least for now, it looks like, still without question, that the US institutions, whether we're talking about Meta, whether we're talking about Amazon, Microsoft with OpenAI, have the most computing power to play with, for lack of a better term?
Damien (
32:01):
That could certainly be one attractive factor. But I can't remember where I read it, but I was shown an interesting survey on one of the Chinese social media sites where apparently our AI Talent Tracker got some traction in Chinese. And so a bunch of AI people in China weighed in.
And if I remember correctly, don't quote me on it, but I think one of the main things that stood out was that, one of the things that really attract that kind of talent is the research environment where they're able to have the freedom and the ability to have free thought and be able to, you know, pursue things that they think are really interesting, that are really worthwhile.
So that stood out to me as a really important factor beyond the compute prowess and beyond compensation, obviously. But I think it seems like, you know, at least the United States still seems to really have that culture by default. And I think that's a really important ingredient that people shouldn't forget about. Again, I just think top tier talent tend to want to be unencumbered, unrestricted because they want to pursue things that they think are really, really, really interesting and groundbreaking. And that's just the way they work. And so you’ve got to give them that environment to work in.
Tracy (
33:10):
Alright, Damien, that was such an interesting conversation. Thank you so much for coming on Odd Lots. And it is the Global AI Talent Tracker and you can look it up online. It's got some really good charts and sort of interactive elements that you can play around with. So thanks Damien for coming on and walking us through the latest work that you've been doing!
Damien (
33:29):
Thank you so much. Great talking to you.
Tracy (
33:44):
Joe, that conversation answered a lot of questions for me. It was just interesting to talk about the patterns that we're seeing play out. I think it's kind of funny that in many ways, this is a new technology that everyone is excited about, but it's kind of playing out the way a lot of stuff has played out historically where the US has a lead at the moment, and then China is like rapidly on its heels and trying to build out its own capacity. And then Europe is in the background publishing thought pieces and new pieces of regulation about it. It's kind of funny.
Joe (
34:18):
It's exactly right. I'm really interested in this idea that, you know, I do think that in the US, if you say “AI” at this point, either people think about the text generators or the image generators, which are amazing. But this idea, and we've been, and I think we're doing some more episodes coming up on it, but like there's also a lot of excitement that there's more to AI than just human language.
And we talked about it a little bit on the food automation episode. The idea that like, if robots could sort of have the same framework, where they're fed tons of data and then make better decisions, so the arms aren't swinging or a slight deviation on the assembly line doesn't disrupt them, then that could be incredibly powerful if they had enough training data about all of these different scenarios that they face.
And so it's interesting to see that China, which seems to be, leading the world in many ways in terms of electrical engineering capacity, that's also in alignment with where a lot of the AI researchers are going.
Tracy (
35:18):
Yes, absolutely. And I know I brought it up a number of times now, but that's why the consumer internet crack down was so interesting to me, because China explicitly said like ‘We don't want all this money pouring into another new online retailer. We have enough of those. Why don't you take that money and invest it in chips, or something tangible like that.’
And so, I do think we are seeing that tendency right now, that focus on like real world applications, industrial applications, manufacturing, that you don't necessarily see in the US and other places in the West because, as you know very well, Joe, it's fun to play around with the chatbots and they've become the public face of this entire new technology.
So that's probably one area where China does have an advantage. But the other thing I think, so first of all, Damien talked about the brain drain aspect of it and the idea that, well, a lot of China AI talent does end up in the US because they go to university in the US and then they stay there and there's demand for their services, etc., etc., although maybe that will change soon.
Tracy (
36:26):
But then the other thing I was thinking is, you brought up that question of compute power and whether or not that's sort of a carrot for AI developers, I also wonder about data and data restrictions in China and what data sets they're playing around with. Specifically for the large language models, but maybe for other things as well. That could maybe be a competitive advantage, if you're really interested in this area, maybe you want to go to a place that has bigger and more wide ranging data sets like Damien was kind of alluding to.
Joe (
36:59):
Totally. The other thing I think is really important to watch, I remember like 20-25 years ago, you know, when if you just looked at the raw number of people graduating with an engineering degree, it was like exploding in China and there was a lot of sneering in sort of western publications. It's like ‘Oh, these are trash degrees.’ Like, yeah, people graduate with a degree in engineering, but it's like pretty mediocre talent and you know, not really that good.
And we sort of have to take some of these numbers with a grain of salt. I get the impression that's changed dramatically a lot of these schools. And so the fact that, you know, that you can sort of come up with this objective measure of talent, which is who gets to speak at these big conferences.
And if there is a broadening out of the number of degree-granting institutions that are represented in that top 2% or top 20%, that strikes me as like a very important trend to watch. And so these universities in China that, you know, I'm not familiar with any of them, but if there's like, you know, beyond just the sort of the equivalents of the MIT or Stanford are also contributing to that elite, that strikes me as a very key indicator to watch.
Tracy (
38:09):
Absolutely. And Neural Information Processing Systems conference organizers, if you're listening, Joe's interested in going, so send him an invite.
Joe (
38:18):
Please, yeah. I'll demonstrate some of the great poems and songs. No, I've done some, and like I had AI come up with a new verb tense for me. It was very impressive. So I come up with creative stuff.
Tracy (
38:29):
Oh, that's interesting. You didn't tell me about that one.
Joe (
38:32):
I didn't want to bore you with all my...
Tracy (
38:34):
It's not boring!
Joe (
38:35):
Alright. Alright. I'll show you, I'll show you that one.
Tracy (
38:36):
Wait, have you started using Claude?
Joe (
38:38):
Yeah. I love Claude.
Tracy (
38:39):
It's better, right?
Joe (
38:40):
There's something about it. I don't know objectively, but this is also another interesting question. So while we're talking about this, this is like another interesting thing I'm wondering about, which is what if it turns out that some of the sort of moats that we associate with software do not end up applying as well to AI?
Tracy (
38:59):
No, absolutely, yeah.
Joe (
38:59):
Yeah. So it's like, for whatever reason, because I like the interface, I like the way the nature of the language it speaks, I started using Claude a lot more in a way that I couldn't ever just imagine, say like going back and forth between like… Once I used Google in 2000, I never went back to Yahoo after that, you know? Or something like that. I've been using Google ever since. It does make me wonder whether it'll turn out that a lot of institutions with sufficient talent, with sufficient compute, can kind of do the same thing and switching costs aren't that high.
Tracy (
39:29):
Yeah. I was wondering about this as well because the premise of this entire conversation was [that] there’s like a war going on. And people are trying to develop their AI capabilities really fast because, first one wins kind of. But it does seem like some of these programs, like the moats might not actually be that high. And once you crack like one level, it might be kind of fungible in other ways. I don't know. I guess it'll be interesting to see.
Joe (
39:56):
Definitely.
Tracy (
39:57):
All right. Shall we leave it there?
Joe (
39:58):
Let's leave it there.
You can follow Damien Ma at
@damienics
.