WEBVTT 00:00:00.000 --> 00:00:02.517 I keep asking my friends, when does the 00:00:02.600 --> 00:00:04.877 asymptote arrive and when does the curve 00:00:04.960 --> 00:00:08.037 slow down? We've not seen it yes yet. 00:00:08.120 --> 00:00:12.357 There will be one, right? It is actually true that there is a 00:00:12.440 --> 00:00:14.397 limit to our craziness. >> [laughter] 00:00:14.480 --> 00:00:17.317 >> We have not found it yet. The next thing 00:00:17.400 --> 00:00:19.557 that's really interesting and terrifying 00:00:19.640 --> 00:00:21.597 also is recursive self-improvement, but 00:00:21.680 --> 00:00:24.517 we don't have it yet. What we do have 00:00:24.600 --> 00:00:28.797 Guys, Eric Schmidt just gave a banger interview during an AI conference last 00:00:28.880 --> 00:00:33.197 week. Eric talked about the upcoming AI and its dangers for humanity. And he 00:00:33.280 --> 00:00:36.837 said if we don't control AI now, it could become an existential threat to 00:00:36.920 --> 00:00:38.997 humanity in the coming years. Guys, I 00:00:39.080 --> 00:00:41.637 must say here that Eric is one guy who's 00:00:41.720 --> 00:00:43.917 been warning us about AGI for a very 00:00:44.000 --> 00:00:46.277 long time now. In a recent interview, he 00:00:46.360 --> 00:00:48.997 literally said that uncontrolled AI has 00:00:49.080 --> 00:00:51.077 the potential to become a bigger problem 00:00:51.160 --> 00:00:53.717 for us comparatively to nuclear bombs. 00:00:53.800 --> 00:00:58.397 So, I mean, he's been the CEO of Google and when he says AI is a bigger threat 00:00:58.480 --> 00:01:02.037 than nukes, it should be enough to open everyone's eyes. Let's watch the 00:01:02.120 --> 00:01:04.357 interview and I'll give my opinion as we go. 00:01:04.440 --> 00:01:08.877 So, let's start with a a question that uh 00:01:08.960 --> 00:01:13.077 I'd love to hear you expand on, which is we're living through a historic moment 00:01:13.160 --> 00:01:15.877 right now. Right? Um 00:01:15.960 --> 00:01:18.917 could you define the moment we're in and 00:01:19.000 --> 00:01:24.197 give us sort of a a state of the union of what's going on in AI? 00:01:24.280 --> 00:01:28.237 We're 10 or 15% into the impacts of 00:01:28.320 --> 00:01:31.677 this. And you can see it. 00:01:31.760 --> 00:01:36.477 You can feel it. And some of it will happen, some of it 00:01:36.560 --> 00:01:39.397 will take longer, right? So, for example, hardware takes longer than 00:01:39.480 --> 00:01:44.597 software. So, the robots take longer than digital systems uh on traditional 00:01:44.680 --> 00:01:46.397 hardware, things like that. 00:01:46.480 --> 00:01:49.797 Um we've not um 00:01:49.880 --> 00:01:53.877 the next thing that's really interesting and terrifying also is recursive 00:01:53.960 --> 00:01:57.677 self-improvement. >> Mhm. It's not happening yet. 00:01:57.760 --> 00:02:01.997 And so, it's easy to convince yourself that you're going to have human agents 00:02:02.080 --> 00:02:05.357 uh sorry, computer agents that are human-like completely within a year or 00:02:05.440 --> 00:02:07.877 two. We don't have the science for that 00:02:07.960 --> 00:02:10.077 yet. People are working on it. I can 00:02:10.160 --> 00:02:13.077 describe how I think it'll play out. But 00:02:13.160 --> 00:02:15.677 we don't have it yet. What we do have is 00:02:15.760 --> 00:02:18.517 reasoning systems that are perfect 00:02:18.600 --> 00:02:22.757 partners for human beings for good and bad. Right? And that has a lot of 00:02:22.840 --> 00:02:26.957 implications. So if we stopped today, which we're not, and it's not stoppable 00:02:27.040 --> 00:02:30.877 or controllable by any government or any single individual or corporation, 00:02:30.960 --> 00:02:35.317 we would still have advanced humanity because of these reasoning agents. How 00:02:35.400 --> 00:02:39.317 fast do you imagine this is going to accelerate? 00:02:39.400 --> 00:02:42.997 Um the there's a thing which I call the San 00:02:43.080 --> 00:02:46.917 Francisco consensus. And the reason I call it is cuz everyone in San Francisco 00:02:47.000 --> 00:02:51.477 believes this. Uh everyone I know anyway, which is that 00:02:51.560 --> 00:02:53.637 it it it's easy to understand. 00:02:53.720 --> 00:02:57.877 Uh this is the year of agents, which we can discuss why agents will take over 00:02:57.960 --> 00:03:00.197 everything this year. Mhm. During this 00:03:00.280 --> 00:03:05.077 year, the scaling of the use of agents and reasoning will sort of grow at this 00:03:05.160 --> 00:03:08.357 enormous rate. Everybody's out of hardware, everyone's out of electricity. 00:03:08.440 --> 00:03:11.717 It's a real boom, right? It's like the biggest boom I've seen, and I've been 00:03:11.800 --> 00:03:15.677 through three or four of these in my career. 00:03:15.760 --> 00:03:20.517 Um the in this thinking, once you have recursive 00:03:20.600 --> 00:03:24.517 self-improvement, where the system can begin to improve itself, you have 00:03:24.600 --> 00:03:28.757 intelligence learning on its own. And it will in this argument, it will learn 00:03:28.840 --> 00:03:32.517 faster than we can cuz we're biologically limited. 00:03:32.600 --> 00:03:36.397 And the way this is expressed in San Francisco, and I'll give a simple 00:03:36.480 --> 00:03:38.037 example, you have a tech company with a 00:03:38.120 --> 00:03:43.157 thousand fantastic AI researchers. Um 00:03:43.240 --> 00:03:46.437 so one day they turn on AI research, 00:03:46.520 --> 00:03:50.637 that is an AI research agent. Well, how many AI research agents do you 00:03:50.720 --> 00:03:55.917 have? >> Mhm. Well, as many as you're limited by electricity, right? 00:03:56.000 --> 00:03:58.317 You don't have to feed them, they don't need housing, there's no more housing in 00:03:58.400 --> 00:04:00.837 San Francisco, you know, all that kind of stuff, right? You don't have those 00:04:00.920 --> 00:04:04.477 problems. You don't have an HR department for them, if you will. 00:04:04.560 --> 00:04:07.557 Uh and you don't have to pay them, you just have to feed them electricity. So, 00:04:07.640 --> 00:04:08.997 how many could you have? Well, maybe a 00:04:09.080 --> 00:04:11.637 maybe a million of these agents. 00:04:11.720 --> 00:04:14.837 Now, in AI, the way you determine you've 00:04:14.920 --> 00:04:18.397 made progress is you have clear metrics 00:04:18.480 --> 00:04:20.037 that the reasoning or testing or 00:04:20.120 --> 00:04:22.197 whatever the evaluation framework is 00:04:22.280 --> 00:04:26.317 better, right? So, that's what happens. 00:04:26.400 --> 00:04:29.717 So, in that scenario, the slope goes like 00:04:29.800 --> 00:04:34.077 this because you're already at this slope, then you add more people, then 00:04:34.160 --> 00:04:38.437 you get the agents and you go like this, and this is essentially um a 00:04:38.520 --> 00:04:40.797 superintelligence moment. 00:04:40.880 --> 00:04:42.437 The belief in San Francisco is this 00:04:42.520 --> 00:04:45.837 occurs within 2 to 3 years. 00:04:45.920 --> 00:04:49.037 The evidence in favor 00:04:49.120 --> 00:04:51.797 goes something like this. Um Claude code 00:04:51.880 --> 00:04:56.077 came out a couple months ago, the the latest one, Opus, whatever it is. 4.5, 00:04:56.160 --> 00:04:58.957 yes, thank you. And 00:04:59.040 --> 00:05:01.437 everyone I know in the Bay Area that's 00:05:01.520 --> 00:05:04.637 doing software says it was 80/20, now 00:05:04.720 --> 00:05:06.517 it's 20/80. 00:05:06.600 --> 00:05:09.997 The best analysis I can come up with is 00:05:10.080 --> 00:05:12.677 it's not the Claude code part, it's that 00:05:12.760 --> 00:05:16.957 the that the underlying LLM can produce 00:05:17.040 --> 00:05:19.437 more reasoning over time, better quality 00:05:19.520 --> 00:05:22.557 tokens over time. It's a deeper thinker, 00:05:22.640 --> 00:05:26.557 right? And all the labs now are competing for that. This is not just the 00:05:26.640 --> 00:05:29.397 size of the context window, it's actually the reasoning skill and the 00:05:29.480 --> 00:05:33.797 length of which it can think. It can just think longer and produce more 00:05:33.880 --> 00:05:36.437 stuff, right? I'm going to pause here for a 00:05:36.520 --> 00:05:39.037 bit, guys. I've been hearing tech people 00:05:39.120 --> 00:05:41.277 on podcasts saying that the recursive 00:05:41.360 --> 00:05:46.277 self-improvement is exponential. That's wrong. Recursive self-improvement is not 00:05:46.360 --> 00:05:49.917 exponential. Just because a system can improve itself doesn't mean that it does 00:05:50.000 --> 00:05:54.357 so by a constant factor each time. In fact, technology generally tends to show 00:05:54.440 --> 00:05:56.397 diminishing improvements as it matures. 00:05:56.480 --> 00:06:00.117 Yes, a better model can make improvements faster, but this ignores 00:06:00.200 --> 00:06:03.917 the current state of things that is being improved upon. And as it improves, 00:06:04.000 --> 00:06:08.517 it gets harder and harder to improve, which can cancel out the fact that the 00:06:08.600 --> 00:06:10.957 models are better and better at improving. 00:06:11.040 --> 00:06:14.157 >> I watched this stuff when I was I moved about 00:06:14.240 --> 00:06:18.597 the Bay Area when I was 21. And I was a programmer in high school 00:06:18.680 --> 00:06:21.917 way back when. And I was pretty good programmer. 00:06:22.000 --> 00:06:24.037 And I watch what it does and I go, my 00:06:24.120 --> 00:06:26.877 god, I'm over, you know. There's there's 00:06:26.960 --> 00:06:32.117 not a thing that I could do that it cannot do. So, when they wrote a C 00:06:32.200 --> 00:06:35.157 compiler in Rust, 00:06:35.240 --> 00:06:39.957 I got it's over, you know, like declare success. So, I think part of this part 00:06:40.040 --> 00:06:43.477 of this is because the people who are building it are also seeing the 00:06:43.560 --> 00:06:46.237 diminution of their own skill. 00:06:46.320 --> 00:06:49.197 They're being forced to go from programmers, which is what I'm very 00:06:49.280 --> 00:06:54.277 proud to have been, to being the the director of a programming system. 00:06:54.360 --> 00:06:59.237 Right? And the most likely scenario, by the way, there's a lot of implications for 00:06:59.320 --> 00:07:02.477 this. One is that it's always been true, 00:07:02.560 --> 00:07:06.517 speaking as your local arrogant programmer, that the very top 00:07:06.600 --> 00:07:10.557 programmers were worth 10 times more than the ones right below. There's 00:07:10.640 --> 00:07:14.637 something special about the mathematical reasoning skills of programmers. 00:07:14.720 --> 00:07:17.397 Those people will become more valuable, 00:07:17.480 --> 00:07:21.677 not less valuable, because these systems need to be controlled by humans at the 00:07:21.760 --> 00:07:25.517 moment. Those people will be capable of grasping the parallelization and 00:07:25.600 --> 00:07:29.317 activities of this. But I do have a proposal for universities, those of you 00:07:29.400 --> 00:07:31.717 who are associated with universities, you should stop everything else you're 00:07:31.800 --> 00:07:36.197 doing in the university right now and design a course for freshmen, men and 00:07:36.280 --> 00:07:37.997 women, uh 00:07:38.080 --> 00:07:40.957 uh who starting in September, 00:07:41.040 --> 00:07:43.997 which is a prompt engineering class. 00:07:44.080 --> 00:07:47.437 Why Why Why university? Why not high school? 00:07:47.520 --> 00:07:49.557 God, you're so aggressive, Peter. 00:07:49.640 --> 00:07:51.517 >> [laughter] >> All right, let's start with 00:07:51.600 --> 00:07:55.837 universities. You can improve my idea. I I thought 18-year-olds would be would be 00:07:55.920 --> 00:07:58.317 young enough. Maybe you think it's younger. 00:07:58.400 --> 00:08:02.997 Here's the most important thing. Spend Spend Spend a quarter or a semester, the 00:08:03.080 --> 00:08:07.237 first thing they learn in university is how to use these tools. 00:08:07.320 --> 00:08:10.237 Universities are completely opposed to my idea, as usual. 00:08:10.320 --> 00:08:14.277 >> Yeah. Because it violates every one of their tenets. But if you think about the 00:08:14.360 --> 00:08:16.477 student, and I mean every student, 00:08:16.560 --> 00:08:18.997 liberal arts, you know, math, whatever, 00:08:19.080 --> 00:08:21.757 they're going This platform will be the 00:08:21.840 --> 00:08:24.517 expression platform for their art, their 00:08:24.600 --> 00:08:27.077 music, their writing, and so forth. Why 00:08:27.160 --> 00:08:28.997 wouldn't you teach them immediately? 00:08:29.080 --> 00:08:32.037 Peter, improve my proposal. No, I I just 00:08:32.120 --> 00:08:35.117 I just feel like we're that AI is going 00:08:35.200 --> 00:08:38.957 to impact every student in high school 00:08:39.040 --> 00:08:41.877 today. Yeah. And that they're they're 00:08:41.960 --> 00:08:45.117 living an unnatural life by not engaging 00:08:45.200 --> 00:08:50.477 with it. And when they hit universities, for those that still exist, um 00:08:50.560 --> 00:08:53.437 Well, plus your kids are that age, and so they're literally right now 00:08:53.520 --> 00:08:56.517 >> Yeah. doing exactly what you're describing, so. People here who have 00:08:56.600 --> 00:08:59.237 teenagers, you know what I'm talking about, cuz they're all in it already. So 00:08:59.320 --> 00:09:02.957 So I think you're That's an improvement to my argument. There's a problem of age 00:09:03.040 --> 00:09:07.237 restriction. Uh you really have to think about vulnerable teenagers with this 00:09:07.320 --> 00:09:10.637 technology. Um I did [clears throat] some some analysis 00:09:10.720 --> 00:09:13.277 of where the real problems are with this stuff. 00:09:13.360 --> 00:09:17.197 Um A simple summary is that at some point 00:09:17.280 --> 00:09:20.837 there will be jobs impact from this stuff. 00:09:20.920 --> 00:09:23.677 We're seeing it in software, and we're seeing it in certain customer service 00:09:23.760 --> 00:09:27.157 industries, not across the board. At some point that will happen. That's an 00:09:27.240 --> 00:09:28.237 issue. 00:09:28.320 --> 00:09:33.197 Another one is how do we as a country 00:09:33.280 --> 00:09:35.237 maintain our moral values while we're 00:09:35.320 --> 00:09:37.757 also racing against China? Another one 00:09:37.840 --> 00:09:41.037 is the impact on um um 00:09:41.120 --> 00:09:45.597 uh young people. It is not okay for 13-year-olds to be committing suicide 00:09:45.680 --> 00:09:48.557 because of an LLM. It's just not okay. It should 00:09:48.640 --> 00:09:51.997 It's It's It needs to be addressed. It needs to be addressed right now. For 00:09:52.080 --> 00:09:55.557 sure. Uh and there's all sorts of other issues. The other one I came up with was 00:09:55.640 --> 00:09:59.157 in agent orchestration. Agents can be combined. 00:09:59.240 --> 00:10:02.997 I I've always been worried that when you put the agents together, especially if 00:10:03.080 --> 00:10:05.517 they're from non uh non-compatible 00:10:05.600 --> 00:10:07.997 vendors, you get unpredictable effects. 00:10:08.080 --> 00:10:12.677 >> Mhm. Yeah. So So these are problems to be solved. So we herald the future and 00:10:12.760 --> 00:10:14.157 we solve the problems that it brought. 00:10:14.240 --> 00:10:18.357 So we're we're going to talk about China and government and jobs. 00:10:18.440 --> 00:10:22.397 But before we do that, um I want to say I'm I'm in this savor-the-moment's kind 00:10:22.480 --> 00:10:25.397 of mode right now because I feel like the world to a year from today will be 00:10:25.480 --> 00:10:27.917 nothing like the world today. And all 00:10:28.000 --> 00:10:29.757 everything we're doing right now, 00:10:29.840 --> 00:10:32.637 I've enjoyed so much for so long. And I 00:10:32.720 --> 00:10:37.077 just want to savor the moment. But reminisce for 1 minute about the fact 00:10:37.160 --> 00:10:40.997 that while you were running Google, the Transformer got invented there, 00:10:41.080 --> 00:10:45.957 the TPU got invented there, Demis Hassabis solved protein folding, 00:10:46.040 --> 00:10:49.157 which is now universally used. You know, it does the 00:10:49.240 --> 00:10:50.917 work in I think it's does the work in an 00:10:51.000 --> 00:10:54.317 hour that used to take a PhD student 4 00:10:54.400 --> 00:10:57.197 years 4 years. Yeah. It's like 300 00:10:57.280 --> 00:11:01.557 million times more efficient. Um all of that and all the diaspora from that, all 00:11:01.640 --> 00:11:04.237 the people working in the field in San Francisco as you mentioned, they all 00:11:04.320 --> 00:11:08.757 were your people. Yeah. And so you were there at the creation of everything 00:11:08.840 --> 00:11:11.597 we're experiencing right now. Is there anything anything like that 00:11:11.680 --> 00:11:14.717 profoundly strikes you about that moment? I mean, did you even realize at 00:11:14.800 --> 00:11:18.717 the time that I think when you're making history, you typically don't know it. 00:11:18.800 --> 00:11:22.357 Um I give a lot of credit to Larry and Sergey. 00:11:22.440 --> 00:11:24.077 Because they were ahead of me. I'm I'm 00:11:24.160 --> 00:11:27.597 an operating CEO. And they pushed pushed 00:11:27.680 --> 00:11:29.877 and pushed for excellence. 00:11:29.960 --> 00:11:31.837 And I'll give you an example. In the 00:11:31.920 --> 00:11:34.797 early years of Google, Larry would uh my 00:11:34.880 --> 00:11:37.037 favorite interaction was one day I said 00:11:37.120 --> 00:11:40.237 we need to hire some people doing Java. 00:11:40.320 --> 00:11:44.837 And Larry and Sergey said, "This is the stupidest idea we have ever heard." And 00:11:44.920 --> 00:11:47.557 and I could never tell with them were they were being serious or not or they 00:11:47.640 --> 00:11:51.317 were just joking with me. But their argument was that real programmers were 00:11:51.400 --> 00:11:52.957 programming one level lower. Today 00:11:53.040 --> 00:11:55.117 Google has many thousands of everything. 00:11:55.200 --> 00:11:59.957 But they but they were so precise and so driven to excellence 00:12:00.040 --> 00:12:03.597 >> in technology. Yeah. >> Right? That I could not fool them. I 00:12:03.680 --> 00:12:07.877 couldn't market around them. I needed to have the technical expert. And they say, 00:12:07.960 --> 00:12:09.357 "Oh, that's boring. Don't do that. 00:12:09.440 --> 00:12:10.907 That's another one of your ideas." 00:12:10.990 --> 00:12:14.437 >> [laughter] >> Right? We want a new idea. Wow. And I 00:12:14.520 --> 00:12:15.597 give them a lot of credit for it. 00:12:15.680 --> 00:12:18.517 >> But what what about the TPU in particular? Like tell that to me cuz I I 00:12:18.600 --> 00:12:21.557 didn't even hear about it until much later. 00:12:21.640 --> 00:12:25.317 And it takes you years to to build design and build your own internal 00:12:25.400 --> 00:12:29.077 chips. And now it's it's about to explode. I don't know how much is public 00:12:29.160 --> 00:12:33.277 or not, but it's it's just >> the TPU version one was essentially a 00:12:33.360 --> 00:12:35.357 matrix multiplier of a particular kind. 00:12:35.440 --> 00:12:38.397 When they went to version two, they changed the algorithm in a complicated 00:12:38.480 --> 00:12:39.877 way. And it's particularly good for 00:12:39.960 --> 00:12:42.037 inference. 00:12:42.120 --> 00:12:45.877 Whether it's brilliance or just luck, 00:12:45.960 --> 00:12:48.037 those decisions made 10 years ago set up 00:12:48.120 --> 00:12:50.477 the TPU as the perfect inference engine. 00:12:50.560 --> 00:12:53.277 And for everybody's benefit, inference 00:12:53.360 --> 00:12:57.517 is what the reasoning steps that I'm describing on. So Google is particularly 00:12:57.600 --> 00:13:01.997 well positioned. As you know, Nvidia purchased Croc um 00:13:02.080 --> 00:13:04.517 for the reason of getting that in Yeah, trying to catch up to what you thought 00:13:04.600 --> 00:13:06.437 of 10 years ago. >> interesting about Nvidia, if you look at 00:13:06.520 --> 00:13:09.797 them, I was looking at the Ruben architecture. 00:13:09.880 --> 00:13:14.117 They managed to do Nvidia managed to do what Intel could never do. Intel could 00:13:14.200 --> 00:13:18.477 never get control of the complete server architecture. And they tried. 00:13:18.560 --> 00:13:23.037 Nvidia has managed to build real supercomputers that you can really buy 00:13:23.120 --> 00:13:24.957 with enough time and money and so forth. 00:13:25.040 --> 00:13:28.037 And it will really be delivered to your to your And and it just do the whole 00:13:28.120 --> 00:13:30.477 thing. These are major industrial 00:13:30.560 --> 00:13:35.077 achievements and that's why both companies will do incredibly well. Okay, 00:13:35.160 --> 00:13:38.877 in the following part Eric talks about robotics and how he fears that China 00:13:38.960 --> 00:13:42.997 might defeat the United States in the robotics race. Guys, one thing is pretty 00:13:43.080 --> 00:13:47.117 clear at this point. China is advancing much more rapidly on the hardware level 00:13:47.200 --> 00:13:51.277 compared to the United States. I mean, if you look at the electric vehicle 00:13:51.360 --> 00:13:55.277 production and advancement, they've outpaced the US with a huge margin. 00:13:55.360 --> 00:13:59.597 Guys, China is fundamentally a socialist country. Their long-term goal is to 00:13:59.680 --> 00:14:03.317 reduce labor for the benefit of the people. Now, if that actually 00:14:03.400 --> 00:14:07.877 materializes, which is of course a real question in itself, but at their core 00:14:07.960 --> 00:14:11.997 they believe that the productivity of the working class is a shared national 00:14:12.080 --> 00:14:15.797 good. So, I mean, it's really important for the United States to understand the 00:14:15.880 --> 00:14:17.877 dynamics of our competition with China. 00:14:17.960 --> 00:14:21.077 I I read an a Time op-ed piece you wrote 00:14:21.160 --> 00:14:25.237 last night. Um China can dominate the physical AI 00:14:25.320 --> 00:14:29.797 future or uh what did you Can you summarize that for us? It was an important 00:14:29.880 --> 00:14:34.357 conversation. >> in the geopolitical context, um they 00:14:34.440 --> 00:14:36.917 and I've said this many times, I'll say it again. 00:14:37.000 --> 00:14:41.797 The American competitor, not enemy, but competitor is China. 00:14:41.880 --> 00:14:44.077 I I think it's a By the way, I think it's an important distinction for you to 00:14:44.160 --> 00:14:47.797 make, so thank you. >> Not Not enemy, competitor. 00:14:47.880 --> 00:14:51.837 Um and how do understand them as a competitor? 00:14:51.920 --> 00:14:53.877 They have lots of money. They're very, 00:14:53.960 --> 00:14:56.717 very smart. Their work ethic is equal or 00:14:56.800 --> 00:14:59.757 stronger than ours and they dominate key 00:14:59.840 --> 00:15:01.277 industries. 00:15:01.360 --> 00:15:04.437 Right? I- I- With respect to robotics, 00:15:04.520 --> 00:15:07.317 we somehow decided it was okay for them 00:15:07.400 --> 00:15:11.997 to dominate the electric vehicle industry. This was an error. T- To be 00:15:12.080 --> 00:15:15.717 very clear, it's an error. Why do you not understand it's an error? Because we 00:15:15.800 --> 00:15:19.037 don't allow their cars in. And spend some time outside of this 00:15:19.120 --> 00:15:21.637 country in Chinese cars, trust me. 00:15:21.720 --> 00:15:23.277 They are real competitors, they've done 00:15:23.360 --> 00:15:26.477 a great job. As I understand it, 00:15:26.560 --> 00:15:31.277 China is capable of vertical integration and build these these gigafactories at a 00:15:31.360 --> 00:15:35.197 scale that we can't for all sorts of reasons. That's got to get addressed. 00:15:35.280 --> 00:15:39.197 Mhm. So, if you want to compete, and I want to compete and win with China, and 00:15:39.280 --> 00:15:43.037 the and the and I compete not enemy, I want us to have the same kind of 00:15:43.120 --> 00:15:47.397 system. So, in robotics, it turns out robotics, 00:15:47.480 --> 00:15:50.317 um you can understand robots as essentially actuators, these little 00:15:50.400 --> 00:15:54.517 stepper motors, book click click click click click, and a brain. Ignoring the 00:15:54.600 --> 00:15:58.277 appearance and the Google, you know, eyes and all that kind of stuff. 00:15:58.360 --> 00:16:00.397 It turns out that the industry of the 00:16:00.480 --> 00:16:02.997 electric vehicle produces the same kind 00:16:03.080 --> 00:16:06.397 of motors and the same kind of systems. 00:16:06.480 --> 00:16:09.157 They have an expertise that we don't. 00:16:09.240 --> 00:16:13.637 My own view is that at least for very low cost, 00:16:13.720 --> 00:16:17.357 China is going to win that. And that was what I was trying to say in that piece. 00:16:17.440 --> 00:16:21.837 Mhm. And I worry about that. Now, today these are not particularly useful, you 00:16:21.920 --> 00:16:24.757 know, they're they're fun toys, they're a replacement for the dog if you get mad 00:16:24.840 --> 00:16:26.757 at your dog. Sorry, I love I love dogs, 00:16:26.840 --> 00:16:30.797 but you get the idea. Um so so we we we 00:16:30.880 --> 00:16:34.757 need to address this. But at the moment, it sure looks to me 00:16:34.840 --> 00:16:39.717 like the robotic hardware of China is the winner at the low end. 00:16:39.800 --> 00:16:42.277 I'm not talking about the high end, I'm not talking about the expensive stuff, 00:16:42.360 --> 00:16:44.237 I'm not talking about industrial robots. 00:16:44.320 --> 00:16:46.317 And if you're confused, watch the 00:16:46.400 --> 00:16:51.317 Unitree um robot dance with the humans. 00:16:51.400 --> 00:16:52.877 Yeah. It came out about a month ago. 00:16:52.960 --> 00:16:56.077 >> Unitree here in the Tech Hub, and the co-founder will be on stage with us 00:16:56.160 --> 00:16:58.157 later today. Pay pay attention to them. 00:16:58.240 --> 00:17:01.877 They're they're very very impressive. Uh and I spent some time with them last 00:17:01.960 --> 00:17:03.957 time I was in China. Uh and they're one 00:17:04.040 --> 00:17:07.077 of many. And the way China works is that 00:17:07.160 --> 00:17:12.077 they have brutal competition. Um brutal brutal it's like unbelievable. 00:17:12.160 --> 00:17:15.557 I was talking to my friend, we teach at Stanford, and And said, you know, in 00:17:15.640 --> 00:17:17.757 China, we don't have the board dinner. 00:17:17.840 --> 00:17:20.877 We have a 2-hour meeting, we get back to work. And there's no preamble. We're 00:17:20.960 --> 00:17:23.677 not, you know, we don't say hello, how are you, how's the family, that kind of 00:17:23.760 --> 00:17:27.437 stuff. We're boom, boom. Right? Uh, it's 00:17:27.520 --> 00:17:31.117 just cultural. Yeah. >> And the workaholic the the the work 00:17:31.200 --> 00:17:35.277 ethic, the precision, and the scale that is possible in China is a real 00:17:35.360 --> 00:17:36.717 competitor. 00:17:36.800 --> 00:17:39.757 Um, I don't want to lose the robotic 00:17:39.840 --> 00:17:44.557 revolution, in my view, the way we lost the electric vehicle revolution, at 00:17:44.640 --> 00:17:48.957 least on >> the low end. Interesting. Cuz I had the Chinese model very much has a a 00:17:49.040 --> 00:17:51.837 well-built out supply chain with many vendors in the loop. But we you know, 00:17:51.920 --> 00:17:56.437 we've been on a worldwide tour of all of the humanoid robotics companies. And I 00:17:56.520 --> 00:17:59.917 just like coincidence, I guess. But when you look at the Gigafactory and look at 00:18:00.000 --> 00:18:03.317 Elon's vertical integration, and also Brett Adcock at Figure, same thing, it's 00:18:03.400 --> 00:18:06.277 all vertically integrated. Why? Well, cuz there's no vendor. 00:18:06.360 --> 00:18:11.037 >> Yeah. I have no choice. I have >> it appears that to get to again, we're 00:18:11.120 --> 00:18:15.037 in this is the abundance group. This is the abundance club. The way you get to 00:18:15.120 --> 00:18:19.197 abundance is you drive prices down and you get vertically integrated. And Elon 00:18:19.280 --> 00:18:21.597 in our country pioneered that, to his credit. 00:18:21.680 --> 00:18:25.277 >> Yeah. Right? You know, the the old joke about Google was we would build anything 00:18:25.360 --> 00:18:29.477 including the buildings. Well, Elon is actually doing that. Yeah. 00:18:29.560 --> 00:18:33.917 >> Right? You know, he's And why? Not because he's insane, but because he 00:18:34.000 --> 00:18:36.437 actually That's how you drive cost down. 00:18:36.520 --> 00:18:41.437 >> Yeah. Guys, the Chinese are very good at scaling at incredible rates. For 00:18:41.520 --> 00:18:46.477 example, they built the Tesla Shanghai Gigafactory in only 168 days. People 00:18:46.560 --> 00:18:50.157 point to Boston Dynamics as evidence that the US is winning, but they're 00:18:50.240 --> 00:18:54.037 missing the speed and scale. Boston Dynamics is basically one of two 00:18:54.120 --> 00:18:55.997 advanced humanoid robot companies based 00:18:56.080 --> 00:18:58.277 in the US. And they've been working on 00:18:58.360 --> 00:19:01.157 Atlas for almost 15 years. China has 00:19:01.240 --> 00:19:05.837 dozens and dozens of manufacturers with advanced humanoid robot solutions that 00:19:05.920 --> 00:19:09.277 have sprung up in only two to three years. Today, they're producing 00:19:09.360 --> 00:19:12.757 thousands of robots annually and deploying them to their megafactories. 00:19:12.840 --> 00:19:16.917 At this rate, the US is going to lose the race just like we did with drones 00:19:17.000 --> 00:19:20.477 and electric vehicles. All right, guys, that's it for today's video. I'll see 00:19:20.560 --> 00:19:24.000 you guys next week with another one.