WEBVTT - Here's Who's Winning the Global Fight for AI Talent 0:00:02.440 --> 0:00:13.400 Bloomberg Audio Studios, Podcasts, Radio News. 0:00:17.960 --> 0:00:21.120 Hello and welcome to another episode of the Odd Lots Podcast. 0:00:21.200 --> 0:00:22.560 I'm Tracy Alloway. 0:00:22.280 --> 0:00:23.400 And I'm Joe Whysenthal. 0:00:23.680 --> 0:00:25.639 Joe, have you watched The Three Body Problem? 0:00:25.960 --> 0:00:28.200 No, but I really want to, and I didn't read 0:00:28.200 --> 0:00:29.720 the book. So in case you're going to ask that 0:00:29.800 --> 0:00:32.000 I didn't, I want to do that too, but I 0:00:32.080 --> 0:00:33.319 intend to at some point. 0:00:33.760 --> 0:00:37.200 There goes my carefully crafted intro where we talk about 0:00:37.200 --> 0:00:40.000 the Three Body Problem. Okay, well, this will work well. 0:00:40.400 --> 0:00:43.640 As everyone knows except for Joe, there's sort of two 0:00:43.760 --> 0:00:46.080 types of people in the world when it comes to 0:00:46.080 --> 0:00:48.440 the Three Body Problem. There are those who see it 0:00:48.520 --> 0:00:52.760 as an allegory for climate change, so humans coming together 0:00:52.920 --> 0:00:56.120 to unite against a common threat, which, in this case, 0:00:56.200 --> 0:00:58.760 since you haven't read the book, is an alien. 0:01:00.240 --> 0:01:00.440 Yeah. 0:01:00.440 --> 0:01:02.800 A friend of mine this weekend told me like two 0:01:02.880 --> 0:01:03.440 plot points. 0:01:03.440 --> 0:01:06.600 Okay, good, good, good, yes, okay. And then there are 0:01:06.640 --> 0:01:09.520 also those who see it as sort of an allegory 0:01:09.800 --> 0:01:13.600 for the trade or tech war between the US and China, 0:01:13.800 --> 0:01:16.319 So the idea that humans are going up against a 0:01:16.400 --> 0:01:20.760 much more technologically advanced opponent, and in this scenario, I 0:01:20.760 --> 0:01:24.840 guess Earth is China and the aliens are the US. Well, 0:01:25.000 --> 0:01:28.280 today we are firmly in that second camp. We're going 0:01:28.360 --> 0:01:33.080 to talk about US China rivalry in tech, and in 0:01:33.120 --> 0:01:35.559 particular one area of tech AI. 0:01:35.920 --> 0:01:39.920 Right, so obviously AI AI AI, everyone talks about it 0:01:40.040 --> 0:01:42.560 all the time. We don't really know where it's going 0:01:42.600 --> 0:01:44.560 to go, but we know a few things in the meantime, 0:01:44.600 --> 0:01:47.000 which is that people are spending money like crazy on chips, 0:01:47.160 --> 0:01:50.520 but they're also spending money like crazy on talent. And 0:01:50.600 --> 0:01:53.880 anyone who is capable of doing sort of cutting edge 0:01:54.000 --> 0:01:57.440 research in AI, from what I can tell based on articles, 0:01:57.840 --> 0:01:59.760 like they basically just get to pick where they want 0:01:59.800 --> 0:02:02.480 to work, can basically pick their salary. There's a great 0:02:02.640 --> 0:02:05.160 article in the Information a couple of weeks ago about 0:02:05.360 --> 0:02:08.680 Facebook hiring top researchers without even doing an interview. It's like, 0:02:08.680 --> 0:02:10.800 if you know this stuff, someone will hire you and 0:02:10.840 --> 0:02:11.639 pay you a lot of money. 0:02:11.800 --> 0:02:14.160 Yeah, And I have so many questions in this space. 0:02:14.200 --> 0:02:17.280 So first of all, like who is an AI talent 0:02:17.960 --> 0:02:20.519 or what is an AI talent? Where do they come from? 0:02:20.639 --> 0:02:22.840 Is it the same as being a software engineer, but 0:02:22.880 --> 0:02:26.679 you have a slightly different area of expertise. I really 0:02:26.720 --> 0:02:29.320 don't know. And then secondly, I'm kind of curious how 0:02:29.360 --> 0:02:32.240 fungible the jobs are. From what you just said and 0:02:32.280 --> 0:02:35.880 the fact that companies are hiring without interviews and things 0:02:35.919 --> 0:02:38.880 like that, and that demand is so strong, it seems 0:02:38.960 --> 0:02:42.400 like you can just do AI anywhere, whether it's China 0:02:42.520 --> 0:02:44.320 or the US or somewhere else in the world, or 0:02:44.320 --> 0:02:47.520 whether it's a specific company versus another one. But so 0:02:47.600 --> 0:02:51.280 many questions on this AI talent war. I guess you 0:02:51.280 --> 0:02:52.280 could say totally. 0:02:52.320 --> 0:02:54.760 And there's two things. So I sort of consider myself 0:02:54.760 --> 0:02:56.600 a bit of an AI talent because I think I'm 0:02:56.600 --> 0:02:58.400 pretty good at coming up with chet GPT. 0:02:58.639 --> 0:03:01.720 You are, actually I listeners. I have learned a lot 0:03:01.760 --> 0:03:04.679 from watching Joe enter his prompts, and I still find 0:03:04.720 --> 0:03:07.120 it incredibly endearing that you say please and thank you. 0:03:07.720 --> 0:03:10.760 Well, it's important for when AI becomes sentient that they're 0:03:10.760 --> 0:03:13.560 going to remember who said please and thank you. But 0:03:13.760 --> 0:03:16.600 beyond that, you know, there's this other element, and you 0:03:16.639 --> 0:03:19.360 already sort of alluded to it. But it's clear that 0:03:19.840 --> 0:03:24.600 for whatever reason, countries feel like AI, almost as if 0:03:24.639 --> 0:03:28.040 it's a commodity there it must be some every country, 0:03:28.160 --> 0:03:30.840 or there's this narrative being pushed by the industry, and 0:03:30.880 --> 0:03:34.240 maybe it's just a narrative to sell chips or subscriptions 0:03:34.280 --> 0:03:36.720 to the open AI APIs. 0:03:36.200 --> 0:03:36.600 Et cetera. 0:03:36.880 --> 0:03:40.080 But there seems to be this narrative that every country 0:03:40.160 --> 0:03:45.360 must have some sort of homegrown AI strategy data center 0:03:45.480 --> 0:03:49.480 or something like. Something about this technology seems to engender 0:03:50.080 --> 0:03:52.480 political and nationalistic anxieties. 0:03:52.960 --> 0:03:55.480 Yes, I think that's absolutely true, and we're back to 0:03:55.600 --> 0:04:00.160 sort of the three yady geopolitical tension point. But I 0:04:00.200 --> 0:04:02.480 am very pleased to say that we in fact have 0:04:02.680 --> 0:04:05.000 the perfect guest to talk about all of this. We're 0:04:05.040 --> 0:04:07.160 going to be speaking with Damien ma He is the 0:04:07.280 --> 0:04:10.760 managing director at macro Polo, which is the think tank 0:04:10.840 --> 0:04:14.800 at the Paulson Institute, and they publish something called the 0:04:14.880 --> 0:04:19.960 Global AI Talent Tracker, so actually keeping track of where 0:04:20.000 --> 0:04:22.520 AI talent is coming from, how much there is, and 0:04:22.560 --> 0:04:25.359 where it's going. So Damian, thank you so much for 0:04:25.400 --> 0:04:26.000 coming on all. 0:04:25.920 --> 0:04:28.120 Thoughts, Thank you so much, it's great to be here. 0:04:28.440 --> 0:04:31.320 How long have you guys been doing this talent tracker? 0:04:31.400 --> 0:04:34.640 And what was the genesis because for me chat, GPT 0:04:35.000 --> 0:04:37.200 and all the chatbots seem to have come out of 0:04:37.200 --> 0:04:40.960 nowhere almost basically a year ago. So how did you 0:04:41.080 --> 0:04:43.440 get an early start on tracking AI? 0:04:44.040 --> 0:04:46.280 Well, the original conception is that we thought a little 0:04:46.279 --> 0:04:48.320 bit hard about, you know, what would you need to 0:04:48.400 --> 0:04:53.000 have a robust AI ecosystem or an AI industry, And 0:04:53.040 --> 0:04:56.080 we thought there are three key pieces. You need. Obviously, 0:04:56.440 --> 0:05:00.320 compute power, so things like chips and the infrastructure need. 0:05:00.320 --> 0:05:03.440 Obviously a lot of training data. Data is obviously everywhere now. 0:05:03.480 --> 0:05:05.480 And we thought the last piece that people haven't thought 0:05:05.520 --> 0:05:08.360 too much about is human capital because it is a 0:05:08.480 --> 0:05:12.320 very human capital intensive area and discipline because it's highly 0:05:12.360 --> 0:05:15.200 complex and complicated and you need to highly trained people 0:05:15.240 --> 0:05:16.919 to be able to do it. So we thought, nobody's 0:05:16.960 --> 0:05:19.560 really looked at the human capital side of things. Is 0:05:19.600 --> 0:05:21.320 there a way to do that? And then so we 0:05:21.440 --> 0:05:24.960 sort of found this one conference that's widely known in 0:05:25.000 --> 0:05:28.320 the AI community as one of the most prestigious, and 0:05:28.400 --> 0:05:31.080 so we looked at papers and researchers that went to 0:05:31.120 --> 0:05:34.120 that conference. This was back in twenty twenty during the pandemic. 0:05:34.279 --> 0:05:37.400 Was when we first launched the initial tracker. That gave 0:05:37.480 --> 0:05:39.880 us our idea that's a proxy for sort of the 0:05:39.920 --> 0:05:44.240 top twenty percent of global AI talent. So this is 0:05:44.279 --> 0:05:47.360 not all AI talent, This is not everybody in the world, 0:05:47.680 --> 0:05:49.560 but this is really sort of what we might call 0:05:49.560 --> 0:05:51.919 it the cream of the crop, top twenty percent, and 0:05:52.000 --> 0:05:55.120 within that there's also the top two percent. So we're 0:05:55.160 --> 0:05:57.200 looking at really kind of the elite people, which is 0:05:57.320 --> 0:06:00.440 probably the type of people that's being thought of, you know, 0:06:00.480 --> 0:06:03.799 most fiercely, because people want the top talent real quickly. 0:06:03.839 --> 0:06:04.560 What's the conference. 0:06:04.800 --> 0:06:07.960 It's called the Newer IPS. It's a conference that's held 0:06:08.160 --> 0:06:10.360 I think, I think every year, but we didn't track 0:06:10.400 --> 0:06:12.440 it every year. We tracked in twenty twenty and then 0:06:12.480 --> 0:06:14.239 we did it again and we looked at the twenty 0:06:14.320 --> 0:06:17.440 twenty two. We were trying to see, you know, had 0:06:17.480 --> 0:06:20.440 there been any changes after the three year pandemic, to 0:06:20.440 --> 0:06:23.600 see if there were different mobility patterns. This is a 0:06:23.600 --> 0:06:28.960 conference that's mainly focused on neuro networks, large language models, 0:06:28.960 --> 0:06:31.160 so a lot of things that are currently really pushing 0:06:31.240 --> 0:06:34.360 up frontiers of a generative AI. So we thought that 0:06:34.400 --> 0:06:37.120 those are the kinds of people that would probably want 0:06:37.160 --> 0:06:39.279 to work for the Googles that open AIS and you know, 0:06:39.360 --> 0:06:41.760 buy dues of the world, and so that seemed like 0:06:41.800 --> 0:06:44.960 a good sampling. Again, we don't pretend that this is comprehensive, 0:06:45.200 --> 0:06:47.839 but it is sort of the elite twenty percent sample. 0:06:48.080 --> 0:06:51.440 Just real quickly. Since you say you're able to distinguish 0:06:51.560 --> 0:06:54.360 between the top twenty percent and the top two percent, 0:06:54.839 --> 0:06:56.640 how do you do that part? I mean, it can't 0:06:56.640 --> 0:06:58.760 just be people who attend the conference, Like how do 0:06:58.800 --> 0:07:01.120 you sort of grade or figure out like who is 0:07:01.160 --> 0:07:04.559 this specific ultra elite AI engineering talent? 0:07:04.960 --> 0:07:07.600 So we looked at authors whose papers got accepted, and 0:07:07.600 --> 0:07:11.680 within that acceptance there's a oral presentation. You don't get 0:07:11.720 --> 0:07:14.520 accepted to oral presentation unless you're really really good. So 0:07:14.520 --> 0:07:16.280 they are only about two percent of people that got 0:07:16.320 --> 0:07:19.240 accepted at oral presentation, So that to us was sort 0:07:19.280 --> 0:07:20.800 of the proxy for the two percent. 0:07:21.240 --> 0:07:23.760 This kind of leads into what I was wondering, which 0:07:23.800 --> 0:07:28.040 is what makes a really good AI engineer? Like what 0:07:28.160 --> 0:07:30.640 is it that would lead them to be someone who 0:07:30.680 --> 0:07:32.320 presents at a conference like this. 0:07:32.920 --> 0:07:35.560 I mean, Joe just said, you know, he's a really 0:07:35.560 --> 0:07:37.080 good prompt engineer. 0:07:36.640 --> 0:07:36.720 So. 0:07:38.520 --> 0:07:39.600 They would let me present. 0:07:39.760 --> 0:07:41.240 Joe, I'm sure your invites. 0:07:40.920 --> 0:07:44.000 In the mail, you know, like really curate the questions. Well, 0:07:44.040 --> 0:07:45.240 but I think that's really good. 0:07:46.000 --> 0:07:48.640 It's not just curating the questions, right, it's like actually 0:07:48.680 --> 0:07:50.880 coming up with the natural language models and things like that. 0:07:50.960 --> 0:07:53.200 Okay, yeah, so so I think it's a really good question, 0:07:53.240 --> 0:07:55.720 and I'm not sure the distinction is huge. I think 0:07:55.760 --> 0:07:59.160 the foundation of AI is all computer science. Most AI 0:07:59.200 --> 0:08:02.280 people would call them those computer scientists first and foremost, 0:08:02.680 --> 0:08:05.240 or people that have a lot of mathematical training. And 0:08:05.280 --> 0:08:07.680 in fact, I think some of those people I think 0:08:07.720 --> 0:08:09.560 back into two thousands and two thousand tens, we're the 0:08:09.560 --> 0:08:12.040 same people that got attracted to big finance, right and 0:08:12.120 --> 0:08:15.240 went to build algorithms for you know, trading desks. Those 0:08:15.280 --> 0:08:17.400 are probably a similar type of people now they're just 0:08:17.400 --> 0:08:20.680 doing AI. And the AI specific apply part is being 0:08:20.680 --> 0:08:23.640 able to train large amounts of data and be able 0:08:23.680 --> 0:08:26.320 to write out algorithms. But those are the things that 0:08:26.360 --> 0:08:28.960 you would get from computer science training with a bit 0:08:28.960 --> 0:08:31.720 of sort of a you know, added AI specific component 0:08:31.760 --> 0:08:34.240 to it. And I think the neuro networks thing is 0:08:34.320 --> 0:08:36.920 probably you know, one distinguishing characteristic is trying to really 0:08:36.960 --> 0:08:39.160 figure out how do you make the computer mimic the 0:08:39.240 --> 0:08:44.040 human brain in a way. But fundamentally it's just mathematics, 0:08:44.120 --> 0:08:47.679 quantitative computer science. All those things you know eventually can 0:08:47.760 --> 0:08:49.000 become AI scientists. 0:08:49.600 --> 0:08:52.280 So there's a certain type of person who is seeking 0:08:52.320 --> 0:08:55.199 out the hardest or maybe most lucrative sort of real 0:08:55.240 --> 0:08:58.000 world math problem or computer science problem. At any time. 0:08:58.559 --> 0:09:01.320 Maybe in the two thousand they were going to Wall 0:09:01.360 --> 0:09:03.640 Street to figure out the best way to create new 0:09:03.679 --> 0:09:07.040 securitized products and derivatives. In the twenty tens, they went 0:09:07.080 --> 0:09:10.560 to Facebook and Google to figure out the ways to 0:09:10.640 --> 0:09:12.920 pack the most number of ads on a smartphone or 0:09:13.160 --> 0:09:15.840 get you to click on them. And now apparently they're 0:09:15.880 --> 0:09:19.120 going into AI research. So let's start with what the 0:09:19.240 --> 0:09:23.199 data shows. Big picture. When you started first started collecting 0:09:23.240 --> 0:09:26.120 the data in twenty twenty, where were they coming from 0:09:26.160 --> 0:09:26.839 and where were they going? 0:09:27.240 --> 0:09:29.040 A lot of them came out of China in the 0:09:29.120 --> 0:09:32.440 United States in twenty twenty, that was pretty clear. Most 0:09:32.440 --> 0:09:35.160 of them ended up in the United States by far, 0:09:35.559 --> 0:09:39.240 and we're still seeing that in our latest update in 0:09:39.440 --> 0:09:42.400 twenty twenty three. Although I would say the big surprise 0:09:42.600 --> 0:09:44.760 was that China has done a really good job really 0:09:44.880 --> 0:09:48.840 ramping up its domestic supply of top AS scientists, so 0:09:48.880 --> 0:09:52.600 they're producing nearly half of the world's top tier AI 0:09:52.640 --> 0:09:55.840 scientists now, and many of them are actually also staying 0:09:55.800 --> 0:09:58.800 in China. And the reason is, I think it's pretty simple, 0:09:58.920 --> 0:10:01.720 is that China is obviously been focusing on its own 0:10:02.040 --> 0:10:05.640 AI industry, and as we already said, you know, people 0:10:05.679 --> 0:10:08.400 go where the jobs are, and if you look at 0:10:08.760 --> 0:10:11.720 the major economies where they're focused on building out AI 0:10:11.760 --> 0:10:15.960 industry opportunities, it's probably the United States and China. And 0:10:16.000 --> 0:10:18.560 if you look at Europe, actually I think punches way 0:10:18.559 --> 0:10:20.840 below its way in terms of having an AI industry, 0:10:21.440 --> 0:10:24.280 and so they you know, they don't tend to attract 0:10:24.720 --> 0:10:27.600 as many top tier AI talent as China or the US. 0:10:28.160 --> 0:10:33.920 And if you look within top US institutions where top 0:10:33.960 --> 0:10:38.240 A italent work, it really is almost a Chinese American doopoly. 0:10:38.800 --> 0:10:42.960 Chinese origin and American AI scientists are seventy five percent 0:10:43.400 --> 0:10:46.000 of the top aalent within US institutions. 0:10:46.840 --> 0:10:50.640 What are the factors that would go into say a 0:10:50.679 --> 0:10:54.560 computer scientist who has been educated in China and they're 0:10:54.600 --> 0:10:58.800 surveying the different opportunities available to them, what are the 0:10:58.840 --> 0:11:02.040 factors that would go into them making a decision, like 0:11:02.120 --> 0:11:07.600 are there immigration considerations? I imagine pay and renewneration would 0:11:07.679 --> 0:11:10.000 have to factor into that how easy is it for 0:11:10.040 --> 0:11:13.720 them to switch from China to the US. 0:11:14.120 --> 0:11:16.640 I think the skills and you know, and the training 0:11:16.760 --> 0:11:18.560 is fairly similar if if you come out of a 0:11:18.600 --> 0:11:22.400 top program, whether it's Chinhua in China or or you know, 0:11:22.480 --> 0:11:26.520 Stanford in California. I think the key from what we're seeing, 0:11:26.559 --> 0:11:28.800 you know, one key indicator of where people end up 0:11:29.120 --> 0:11:31.320 for work, you know, is really where they go to 0:11:31.360 --> 0:11:34.160 graduate school. That's probably not a surprise. If you're going 0:11:34.200 --> 0:11:37.360 to do your master's or PhD somewhere, you generally start 0:11:37.360 --> 0:11:40.000 to search for job opportunities you know, near you, around you, 0:11:40.600 --> 0:11:43.360 unless you happen to be in a country in an 0:11:43.400 --> 0:11:46.599 area where there's not a lot of opportunities post graduation. 0:11:47.080 --> 0:11:50.240 And of course, when you're considered an elite AI talent, 0:11:50.280 --> 0:11:53.480 you generally have a terminal degree, usually a PhD, but 0:11:53.559 --> 0:11:56.240 at least a master's, so, you know, I think where 0:11:56.280 --> 0:11:58.560 you choose to go to graduate school is really important, 0:11:58.559 --> 0:12:00.760 and we see that in the data. You know, those 0:12:00.760 --> 0:12:03.200 who come to the United States or graduate school, by 0:12:03.240 --> 0:12:05.199 and large tend to stay in the US to work 0:12:05.559 --> 0:12:08.560 unless there's some very lucrative opportunity that attracts them back 0:12:08.600 --> 0:12:11.320 home or somewhere else, But generally there's a bit of 0:12:11.320 --> 0:12:14.080 a path dependence between you know, graduate school and staying 0:12:14.120 --> 0:12:15.160 in that country to work. 0:12:15.600 --> 0:12:18.280 There has been a lot of anxiety for years in 0:12:18.320 --> 0:12:21.600 the tech industry where you see CEOs and leaders complaining 0:12:22.040 --> 0:12:24.880 that the US immigration policy has made it too hard 0:12:25.160 --> 0:12:28.120 to keep talent who has graduated in the United States, 0:12:28.160 --> 0:12:30.000 and there's this idea of like, hey, if they're going 0:12:30.040 --> 0:12:32.080 to come here for education, why are we not reaping 0:12:32.120 --> 0:12:36.520 the benefits of the US educated talent. It does seem 0:12:36.559 --> 0:12:39.200 like from your data that still many are staying in 0:12:39.200 --> 0:12:42.920 the United States, But the numbers have changed since twenty twenty. 0:12:43.040 --> 0:12:45.719 Yes, yes they have, you know, gone down a little bit. 0:12:46.000 --> 0:12:49.839 We didn't go into really exploring exactly what happened over 0:12:49.840 --> 0:12:52.680 the last three years, in part because I think many 0:12:52.679 --> 0:12:55.480 people realize the pandemic years have been a little strange, 0:12:56.120 --> 0:12:59.000 whether it's for economic data or just general ability for people, 0:12:59.000 --> 0:13:01.640 where people work, how people work. So there's going to 0:13:01.679 --> 0:13:04.000 be a lot of distortions in those last three years. 0:13:04.160 --> 0:13:07.240 But there has been a relative decline, especially among the 0:13:07.280 --> 0:13:10.280 Asian talent. It's not just China. India has also done 0:13:10.280 --> 0:13:12.800 a better job retaining its own top tier AI talent 0:13:13.200 --> 0:13:16.640 South Korea. Interestingly, that's not on our data set yet, 0:13:16.640 --> 0:13:20.920 but we're about to publish Regional South Korea. They've retained 0:13:21.000 --> 0:13:24.040 ninety percent of their talent, they've not let anybody leave, 0:13:24.240 --> 0:13:26.080 and they've been really good at doing that. And places 0:13:26.120 --> 0:13:28.440 like France have actually done a very good job on 0:13:28.520 --> 0:13:31.680 retaining their talent. So I can't say definitively what the 0:13:31.720 --> 0:13:34.680 reason is. Whether countries have stepped up their gain to 0:13:34.800 --> 0:13:37.600 retain domestic talent, or there's been other things that happen 0:13:37.640 --> 0:13:39.960 in the pandemic that's triggered it, or there could be 0:13:40.240 --> 0:13:44.320 immigration challenges and so on. I think maybe in the future, 0:13:44.320 --> 0:13:46.760 when we do the next iteration, we will have more 0:13:46.840 --> 0:13:49.160 clarity to see the pattern. So I'd be a little 0:13:49.200 --> 0:13:51.880 hesitant to give definitive conclusions at this point. 0:14:07.080 --> 0:14:10.400 Tracy, If France does a really good job keeping their talent, 0:14:10.640 --> 0:14:13.400 who will fill the niche of blowing up trading desks 0:14:13.400 --> 0:14:17.360 with exotic derivatives? If all those it called polytechnique in 0:14:17.480 --> 0:14:19.840 sciences PO graduates going to AI. 0:14:19.680 --> 0:14:23.800 Onstay, Yeah, Yes, it is always a French person working 0:14:23.800 --> 0:14:27.520 in equity derivatives with a mathematics degree. You're absolutely correct, 0:14:27.720 --> 0:14:31.600 but on the degree topic. So I hadn't realized that 0:14:31.680 --> 0:14:35.600 in China and Damian, I think this factoid was in 0:14:35.680 --> 0:14:38.280 one of the reading materials that you sent. But Chinese 0:14:38.360 --> 0:14:42.480 universities have launched more than two thy three hundred undergraduate 0:14:42.560 --> 0:14:47.120 programs since twenty eighteen, when the Ministry of Education designated 0:14:47.200 --> 0:14:52.520 AI as a separate major that's distinct from computer science. So, 0:14:52.920 --> 0:14:55.520 first of all, how common is that that you would 0:14:55.560 --> 0:14:58.640 get a separation between computer science versus AI? Is that 0:14:58.760 --> 0:15:00.840 the standard in other parts arts of the world or 0:15:00.920 --> 0:15:04.680 is it still relatively new? And then secondly, presumably this 0:15:04.720 --> 0:15:08.400 is part of China trying to build up its domestic 0:15:08.520 --> 0:15:13.080 AI talent pool and eventually its capabilities in this area. 0:15:13.200 --> 0:15:15.080 What else is it doing on that front? 0:15:15.200 --> 0:15:17.520 Yeah, So that's why one of the reasons we think 0:15:17.560 --> 0:15:20.480 that China has really seen this boom on top AI 0:15:20.560 --> 0:15:22.800 talent is you have just kind of a graduating class 0:15:22.800 --> 0:15:25.320 in twenty twenty two. If you start in twenty eighteen, 0:15:25.560 --> 0:15:27.320 some of them are graduate students, some of them are 0:15:27.840 --> 0:15:31.040 undergrad so they've really pushed really hard to grow at 0:15:31.040 --> 0:15:33.160 the AI talent when now not all of them are 0:15:33.200 --> 0:15:35.920 the top twenty percent, But I think China looks at 0:15:35.960 --> 0:15:37.240 it as a way that they're going to need a 0:15:37.280 --> 0:15:41.400 lot of AI specific technicians. China's not really thinking about 0:15:41.400 --> 0:15:44.200 AI in the generative AI sense. I think there are 0:15:44.200 --> 0:15:48.680 definitely some startups and folks pursuing things like chatbt Chatboss, 0:15:48.680 --> 0:15:51.280 but my understanding is that China's probably going to focus 0:15:51.360 --> 0:15:58.479 much more on industrial applications of AI, manufacturing, robotics, probably healthcare, biotech. 0:15:58.600 --> 0:16:00.000 I'm going to bet that's going to be a huge 0:16:00.120 --> 0:16:03.400 application for China. And I think for obvious reasons, generative 0:16:03.400 --> 0:16:07.360 AI is probably not as copesthetic with the governance system 0:16:07.760 --> 0:16:10.560 in China ultimately, and I think that's a pretty clear 0:16:10.600 --> 0:16:13.160 thing that I think everyone knows. But I think they're 0:16:13.200 --> 0:16:17.120 really looking at how to apply artificial intelligence to energy, 0:16:17.600 --> 0:16:21.480 to industry, to advanced manufacturing, or things like climate. That's 0:16:21.520 --> 0:16:24.040 where China's really focused on, and I think they feel 0:16:24.040 --> 0:16:26.480 like they need a lot more people, not just the 0:16:26.520 --> 0:16:28.560 cream of the crop, but sort of you know, middle 0:16:28.640 --> 0:16:32.160 level technicians, people that are just familiar with being able 0:16:32.240 --> 0:16:35.320 to like run data or to run Python, or to 0:16:35.480 --> 0:16:38.280 just check all the data. So I think they're viewing 0:16:38.320 --> 0:16:42.119 AI as a very wide, expansive way of creating certain jobs. 0:16:42.560 --> 0:16:45.480 Yeah, I can't imagine China's ambition here is to have 0:16:45.560 --> 0:16:49.520 like five thousand different chatbots. Like there is clearly a 0:16:49.680 --> 0:16:54.880 tendency towards industrial sort of real world applications of this technology. 0:16:55.320 --> 0:16:58.880 On which note, do you think there's currently enough places 0:16:59.160 --> 0:17:04.400 for AI graduates or specialists to actually go within China, 0:17:04.480 --> 0:17:06.760 Because in some respects it feels like this might be 0:17:06.760 --> 0:17:09.640 a very hot degree. People are being encouraged to do it, 0:17:09.880 --> 0:17:14.240 but at the moment companies aren't necessarily at the same 0:17:14.280 --> 0:17:17.000 sort of level. It feels like there's sort of a 0:17:17.040 --> 0:17:18.960 mismatch in the evolution of this. 0:17:19.160 --> 0:17:21.560 At the moment, I think you're absolutely right. So we've 0:17:21.560 --> 0:17:23.879 seen these kinds of bubbles before that you know, the 0:17:23.960 --> 0:17:26.840 new hottest sector in China, everyone goes there because they 0:17:26.840 --> 0:17:29.800 think that's where the opportunities are. And then you know, 0:17:29.880 --> 0:17:32.359 China already had what we would call a college bubble 0:17:32.720 --> 0:17:34.520 for the last ten years, and that's why you have, 0:17:34.600 --> 0:17:37.919 you know, really high youth joblessness in China. Though. The 0:17:37.960 --> 0:17:40.359 way I think about how China works in that respect 0:17:40.400 --> 0:17:44.000 specifically is that they're basically two different cycles. In China, 0:17:44.000 --> 0:17:46.919 there is a policy induced cycle, and then there's an 0:17:46.920 --> 0:17:49.760 actual Marcus cycle that comes after that. So right now 0:17:49.760 --> 0:17:51.840 we're in sort of this policy driven like you know, 0:17:51.920 --> 0:17:54.399 you guys got to come in and we really like AI. 0:17:54.520 --> 0:17:57.280 We're going to create all these programs and you should 0:17:57.280 --> 0:17:59.960 just get AI. And then you know, parents are like, well, 0:18:00.080 --> 0:18:02.480 well that seems like a good new thing, and that's 0:18:02.520 --> 0:18:05.239 what the government's promoting. So all my kids that are 0:18:05.240 --> 0:18:06.840 going to you know, do computer science, they're going to 0:18:06.880 --> 0:18:08.960 add the AI component to it. So that's sort of 0:18:09.040 --> 0:18:12.159 the policy induced cycle. And then after that, once the 0:18:12.200 --> 0:18:14.560 bubble happens, it will kind of eventually get into a 0:18:14.600 --> 0:18:17.360 market cycle where it'll correct a little bit. And then 0:18:17.400 --> 0:18:19.480 and then people will be like, oh, well, actually we 0:18:19.840 --> 0:18:22.480 probably not have an oversupply of a lot of these 0:18:22.520 --> 0:18:25.080 you know, you know middle AI technicians that will have 0:18:25.119 --> 0:18:27.360 no jobs. What are we going to do with them? 0:18:28.160 --> 0:18:28.679 We don't know. 0:18:29.000 --> 0:18:31.560 So I think this is a pattern that happens in 0:18:31.640 --> 0:18:33.680 China a lot, and I wouldn't be surprised if that 0:18:33.760 --> 0:18:35.640 happens with the AI talent pool as well. 0:18:36.119 --> 0:18:39.480 So there's a lot of interesting threads to pull on 0:18:39.640 --> 0:18:42.600 already in this conversation, and I want to return to 0:18:43.280 --> 0:18:46.199 the non chatbot applications of AI, like how can we 0:18:46.240 --> 0:18:49.680 make better robots and factories and drug discovery, et cetera. 0:18:49.840 --> 0:18:51.800 But I want to ask another question. So, okay, all 0:18:51.920 --> 0:18:56.159 these new institutions or graduate programs have been launched in 0:18:56.359 --> 0:19:00.320 China and more and more universities offering degrees and or 0:19:00.320 --> 0:19:03.880 computer science or related fields. In my mind's eye, if 0:19:03.920 --> 0:19:07.120 I imagine what a top AI researcher, I imagine maybe 0:19:07.119 --> 0:19:11.360 they have a PhD from MIT or Stanford or something 0:19:11.440 --> 0:19:14.080 like that. When you look at the institutions in China, 0:19:14.200 --> 0:19:17.199 has there been any sort of broadening out of the 0:19:17.320 --> 0:19:21.720 number of schools that are capable of producing either those 0:19:21.760 --> 0:19:25.679 top twenty percent or top two percent talent beyond just 0:19:25.720 --> 0:19:27.800 the sort of like handful of schools that we've for 0:19:27.840 --> 0:19:30.520 a long time understood as the elite schools. 0:19:31.119 --> 0:19:33.240 There have been a little bit. And when it comes 0:19:33.240 --> 0:19:35.600 to Asia specifically and China, I think they have the 0:19:35.600 --> 0:19:39.480 eleven of the fourteen top AI institutions in Asia. But 0:19:39.640 --> 0:19:42.440 in terms sort of you know, just top in general, 0:19:42.960 --> 0:19:46.800 China has climbed quite a bit. Places like Due John University, 0:19:47.359 --> 0:19:50.080 shanghaijel Tone, which are not your traditional names that you 0:19:50.080 --> 0:19:50.480 would hear. 0:19:50.920 --> 0:19:52.000 Yeah, I've never heard of either. 0:19:52.240 --> 0:19:56.119 It's not Pku, it's not chin Hua. And interestingly, this 0:19:56.280 --> 0:19:58.880 is an interesting you enter into into twenty twenty two. 0:19:58.960 --> 0:20:02.600 Huawei is actually one of the top twenty five institutions 0:20:02.600 --> 0:20:05.960 for AI researcher globally, so they've invested a lot in 0:20:06.040 --> 0:20:08.520 hiring top AI talent for obvious reasons. 0:20:09.280 --> 0:20:12.080 This is actually exactly what I wanted to ask you next, 0:20:12.119 --> 0:20:15.240 which is you mentioned I do as well earlier in 0:20:15.280 --> 0:20:20.800 the conversation. But in terms of domestic destinations for AI specialists, 0:20:21.280 --> 0:20:24.280 is the idea here that a lot of the existing 0:20:24.680 --> 0:20:28.320 internet companies in China that they're going to devote more 0:20:28.359 --> 0:20:32.680