Applied Intuition：为十亿台机器赋予智能

Kazer Peter，欢迎来到 Z 播客。

Kazer Peter, welcome to the Z podcast.

嗯，谢谢邀请我们。

Well, thanks for having us.

你的名字只是众多之一。

Your name is just one of many.

我觉得我们彼此认识太久了，久到我不想承认。

I feel like we've all known each other for too long, more than I'd like to admit.

是啊，很久了。我们很幸运都是首轮的第一批投资者。当然，支票大小不同，但……

Yeah, long time. We're lucky to both be the first investors or among the first investors in the first round. Of course, different check sizes, but...

而且在那之前我就投资了你。

And I was an investor for you even before then.

没错。

Exactly.

那我们以此作为过渡。今天有很多要聊的。我们要聊公司历史上最大的一次发布，但首先，不如先更新一下状态？Applied Intuition 是做什么的，给那些……

So let's do that as a segue. We have a lot to talk about today. We've got the biggest launch in company history to talk about today, but first, why don't we just give an update or a status? What does Applied Intuition do for those who...

是的，对于不了解的人来说，Applied Intuition 是一家物理 AI 公司。我们把智能赋予机器。这是最简单的描述方式。各种机器，汽车、卡车、坦克、无人机，你能想到的。它是一个物理移动的东西，我们让它变得智能。公司历史是，我们最初从制造构建智能的工具开始，然后进入了智能本身。从某种意义上说，就像一家非常无聊的 AI 公司，因为公司 83% 是工程人员。我们靠做出真正优秀的产品来获胜，不是靠销售之类的。我觉得我们不够格做一家销售驱动的公司。但确实，有超过一千名工程师，总部在硅谷，但我们在全球有 18 个办公室。我们的使命是把智能赋予十亿台机器，我们认为这能对社会产生深远影响，无论是大家常说的安全这类烦人的问题。如果你真的和经历过车祸、矿难或农业事故的人聊过，那些都是非常棘手的状况。除了修复这些问题，如果你能释放生产力，我想我们在数字世界已经看到了这种释放，每个人都超级兴奋，出现了万亿美元市值的公司。我坚信，当我们回看 25 年，就像现在回看互联网，人们实际上，如果你看最初的互联网公司，它们做服务或分析，那些很有趣，但真正当你 25 年后回看，那些大型巨头公司是亚马逊，给你送东西的，苹果，这些才是真正成熟的公司。我认为当我们回看这 25 年的智能革命，那些影响物理世界的公司可能实际上会比影响数字世界的公司更大。

Yeah, for the people who don't know, Applied Intuition is a physical AI company. We put intelligence on machines. That's the simple way of describing it. And all types of machines. So cars, trucks, tanks, drones, you name it. It's a physical moving thing. We make it intelligent. And the history of the company is we originally started by making the tools that would make the intelligence, then we got into the actual intelligence itself. In some ways, like a very boring AI company, in the sense that 83% of the company is engineering. We win by making really great products. It's not like a good sales or something like that. I don't think we're good enough for a sales-enabled company. But yeah, over a thousand engineers, based in Silicon Valley, but we have offices globally, 18 offices. And our mission is to put intelligence on a billion machines, and we think that can have a profound impact on society, both in the kind of pesky things everyone talks about, safety. You know, if you really talk to somebody who's been in a car accident or in a mining accident or in a farming accident, those are real gnarly situations. Beyond just fixing that, if you can unlock productivity, I think we've seen the unlock in the digital world and everyone's super excited about it, and you have trillion-dollar companies emerging. I'm a pretty strong believer that when we look back 25 years, if you look back at the internet now, people actually, if you look at the original internet companies that are doing serving or doing some analytics, those are interesting, but really when you look back 25 years from now, the big monolithic companies are Amazon that delivers you stuff, Apple, these are the true kind of companies that come of age. And I think when we look back 25 years in this intelligence revolution, the companies that impact the physical world might actually be bigger than the companies that impact the digital world.

我想请你谈谈以下这点：当你最初创办公司时，我觉得对公司的批评是，哦，这就像让汽车自动驾驶，对吧？自动驾驶汽车。但有点像，好吧，有特斯拉和 Waymo 在造自己的自动驾驶汽车，然后还有六到八家重要的汽车公司，然后公司永远不可能变得那么大，因为客户就那么多。

I would love for you to talk about the following, which is when you first started the company, the knock on the company, I think, was, oh well, it's like making cars autonomous, right? Self-driving cars. But it's kind of like, okay, there's Tesla and Waymo building their own self-driving cars, and then there's like six or eight other car companies that matter, and then the company just could never get that big because there just aren't that many customers.

是的。

Yeah.

那么，人们应该如何看待有多少会移动的东西，物理 AI、物理智能的概念会在哪些地方变得重要？

So, how should people think about how many things are there that are things that move where the idea of physical AI, physical intelligence are going to matter?

是的，我的意思是，即使今天，即使你对我们持有这种看法，汽车业务大约占我们业务的 30%。所以 70% 已经是非汽车业务。我认为如果你再快进 10 到 20 年，即使是制造商本身作为客户群也会变得很小。我认为我们的使命，就是不断思考十亿台机器变得智能，你想想存在的所有类型的机器。汽车只是一个容易的例子。我觉得它深入人心是因为我们都开车，而且这是一个大市场。但我认为它会成为业务的少数部分。我的意思是，它已经是少数业务了。我认为它会越来越成为少数业务，但这不一定意味着它会变小，对吧？汽车仍然巨大，是全球 GDP 的一部分。汽车大约占所有 GDP 的 3%。我认为我们总是这样想，当你试图实现使命时，最初制造商是向消费者分发那种智能的渠道，但然后你开始涉足国防，开始涉足建筑、采矿和农业，突然制造商很重要，但也许采矿运营商实际上非常重要，或者战争部门非常重要，突然他们成为客户，而所有这些也是我们的客户。

Yeah, I mean, even today, even if you put that view on us, automotive is like 30% of our business. So 70% already is non-automotive. And I think if you fast forward another 10, 20 years, even the manufacturers themselves as a customer base will be a small amount. I think our mission, just keep thinking about a billion machines becoming intelligent, and you think about all the types of machines that exist. Automotive is just an easy one. I think it sticks in people's heads because we all drive cars and it's a big market. But I think it'll be a minority of the business. I mean, it is a minority business. I think it'll be increasingly a minority of the business, but that doesn't necessarily mean it'll be small, right? Automotive is still huge just as a part of the globe's GDP. Automotive is something like 3% of all GDP. I think the way we always think about it, like as you try to get to your mission, initially the manufacturers were the distribution to that intelligence to consumers, but then you start working in defense and you start working in construction and mining and agriculture, and suddenly the manufacturers are important, but maybe the mining operator is actually really important, or the department of war is really important, and suddenly they become customers, and all of those are customers of ours as well.

是的。我认为如果你把 AI 分为数字 AI 和物理 AI，对吧？数字 AI 当然是构建软件、优化广告和创建视频之类的。这些都很有趣也很好，但真正谈到全球经济时，那是物理 AI。然后我们谈论的是制造业、采矿、物流和运输，所有这些……

Yeah. I think if you split AI into digital AI and physical AI, right? Digital AI, of course, is building software and optimizing ads and creating videos, that sort of thing. That's all interesting and good, but really where you talk about the global economy, that's physical AI. And then we're talking about manufacturing and mining and logistics and transportation, all of these things that...

供应链。

Supply chains.

是的。供应链。没错。

Yeah. Supply chain. Exactly.

嗯，我想接着这个再深入一点。今天会动的东西，或者说历史上会动的东西，都是某种形式上有驾驶员或操作员在控制的，对吧？飞机必须围绕驾驶舱里的人来设计。船也必须围绕操纵它的人来设计。在一个自主的世界里，我们是已经知道哪些东西会动，还是会发现当不需要人坐在驾驶座上时，会有很多新东西被造出来？

Well, I mean, let's build on that for a second. Things that move today, or historically, are things that have human beings at the wheel or at the controls in some form, right? Airplanes have had to be designed around a human in the cockpit. Boats have had to be designed around a human steering things. In a world of autonomy, do we already know what the things are that move, or are we going to discover that there are a lot of new things that are going to get built when you don't need a human in the driver's seat?

我觉得两者都有。你要记住，港口里的运输系统，或者矿场里的铲运机、小松的运土机，这些设备的设计寿命是 20 到 25 年。所以那些产品的买家可能还没收回全部投资回报。无论新产品有多好，他们都不会马上买新的。所以我们战略的一部分是，你必须让那些现有设备变得智能，因为它们不会消失。第二部分就是你说的：如果驾驶室里没有人，机器可以更小，形状也可以完全不同。比如地下采矿，真正的约束其实是人，因为人需要呼吸，而且那里非常危险。所以你可以造出完全不同的机器。我们这两方面都在做。

I think both. You have to remember, you take a haulage system on a port, or a catap, or a Komasu dirt mover in a mine, those are made for 20-25 years. So the buyers of those products might not have gotten their full ROI on them. They're not immediately going to buy something new no matter how much better it is. So one part of our strategy is you've got to make those things intelligent because they're not going anywhere. The second is what you're talking about: if you don't have a human in the cab, the machine can be smaller, it can be shaped in very different ways. You talk about mining underground, the constraint actually is the human because the human needs to breathe and it's very dangerous. So you can build a very, very different machine. We're doing both of those things.

对。

Right.

然后我们还没谈到的一点是，我们都在谈论系统内部的智能，但系统级的智能才是真正的解锁点。我们已经在做这样的工作，比如，嘿，让我们看整个港口、整个矿山、整个采石场。这些异构的机器组合，它们可以互相通信，可以优化并提高效率。当一台机器出故障或有问题时，矿山的其他部分不必停下来。如果是人驾驶的，我们甚至不知道机器会出故障，因为没有分析；人并没有接入机器的核心系统。所以像知道刹车系统什么时候会坏这样简单的事情，其实非常重大，因为你可以提前做准备。哦，这里的磨损比其他矿山更严重。这只是举个例子。但另一个宏观观点是，如果你看农业，美国农民的平均年龄是 58 岁。35 岁以下的农民比例不到 10%。那会发生什么？粮食需求持续增长。稀土材料的需求也在持续增长。这些需求只会增加，但作为瓶颈的人力却在减少。卡车运输也是一样。所以你真的可以解锁更多的效率。

And then the thing we're not talking about is we're all talking about intelligence almost within a system, but the system-level intelligence is where the unlock is. We're already doing work like that where you say, hey, let's take an entire port, let's take an entire mine, let's take an entire quarry. This heterogeneous mix of machines, they can all talk to each other and they can optimize and be efficient. When one machine goes down or has an issue, the rest of the mine doesn't have to stop. When it's human-driven, we don't even know the machine is going to go down because there's no analysis; the human is not plugged into the core systems of the machine. So a simple thing like knowing when a brake system is going to break is actually huge because you can start preparing for it in advance. Oh, this wear and tear is higher than in other mines. Just using an example. But the other macro point is if you look at agriculture, the average American farmer is 58 years old. The number of farmers under 35 is less than 10%. So what's going to happen? The need for food growth is continuing to grow. The need for rare earth materials is continuing to grow. These demands are only growing, but the humans who are the bottleneck are decreasing. Trucking is the same way. So you can really unlock a lot more efficiency.

所以一种思考方式是，想象一下如果食品成本因为效率大幅提升而下降，下游影响会是什么？想象一下货物运输，从每英里几美元降到每英里 20 美分。我觉得这个解锁空间非常大。而且这不一定需要所有机器都从头重新设计。

So one way to think about this is imagine if the cost for food decreases because it's way more efficient. What's the downstream impact? Imagine goods being transported, instead of a few dollars a mile, it's 20 cents a mile. I think the unlock is very, very big. And that doesn't necessarily need for all the machines to be redesigned from the ground up.

对，对，明白了。有道理。

Right, right, got it. Makes sense.

然后也许再问一个问题：给我们介绍一下公司目前的规模和范围吧。

And then maybe just one more question: give us a sense of the scope and scale of the company today.

是的，超过一千名工程师。这些工程师显然包括经典的软件和 AI 工程团队，但我们也有真正懂安全系统的工程师。我们还有真正懂硬件的工程师。因为我们一直在回避的关键点是，所有这些事情都很难，因为它们最终必须面对真实世界，而真实世界有更多的复杂性和更多的问题。我们有能够将我们的模型部署到 50 多个平台上的工程团队。这听起来甚至有点微不足道，因为当你想到模型时，你会想到通过浏览器或手机部署，一切都被抽象化了，因为你有 iOS、Android、Windows、Linux，所有这些系统都已经处理好了。但在真实世界里，你没有这些。所以我们也有能胜任这项工作的工程团队。我们的一个亮点是，在公司历史上我们筹集了超过十亿美元，而且这些钱都存在银行里。我总是说这有个星号——这并不意味着我们下个月不会花掉它。

Yeah, north of a thousand engineers. Those engineers are obviously the classic software and AI engineering teams, but we also have engineers who really know safety systems. We also have engineers who really know hardware. Because the important thing we're kind of stepping around is all this stuff is hard because it ultimately has to meet the real world, and the real world has way more complexity and a lot more issues. We have engineering teams that can deploy our models onto 50-plus platforms. Even that sounds trivial because when you think about models, you think about deploying them through a browser or on a phone, and everything's abstracted away because you have iOS, Android, Windows, Linux, all these systems that have already taken care of it. In the real world, you don't have that. So we have engineering teams that can do that as well. Our claim to fame is we've raised over about a billion dollars in the company's history, and all that is sitting in the bank. I always say that with an asterisk—it doesn't mean we're not going to spend it next month.

好消息，坏消息。

Good news, bad news.

是的，好消息，坏消息。但我们正处于这样一个阶段：巨大的市场就在我们周围，我们可以决定要多么积极地追求这些市场，因为坦率地说，我们有十年的执行和部署到生产的经验。我认为我们工程团队的标志就是让产品投入生产。这确实是一件大事。

Yeah, good news, bad news. But we're at that phase where these giant markets are around us, and we can make the decision how aggressive we want to pursue those because of a decade of frankly execution and deployment into production. I think the hallmark of our engineering team is putting products into production. That really is a big deal.

你怎么看待规模？

How do you think about scale?

是的，我认为这大致就是让智能惠及十亿台机器的使命。我们就是这样想的。然后思考哪些类型的机器我们会产生最大的影响，并首先关注这些领域。但我们会做到的。

Yeah, I think that's roughly the mission of bringing intelligence to a billion machines. That is how we think about it. And then thinking about what are the types of machines that we'll have the most impact on, and focusing on those areas first. But we'll get there.

对，很好。让我们更深入地探讨数字 AI 和物理 AI 之间的差异，并进一步了解我们今天所处的位置。取得了哪些进展？物理 AI 的一些主要瓶颈是什么？请详细说说。

Right, good. Let's go deeper into the differences between digital and physical AI, and more so into where we are today. What progress has been made? What are some of the major bottlenecks in physical AI? Unpack some of that.

是的，我认为很多时候人们认为物理 AI 的进展基本上只限于两个用例，因为它们显而易见且有趣：自动驾驶出租车和人形机器人。它们非常直观，让你兴奋，而且有点科幻。我认为这些非常有趣，我们和其他人确实在这些领域做了实际工作。我认为所有其他领域也会同样重要。我的意思是，想想港口里发生的事情——那里有巨大的解锁空间。而这正是我们真正关注的领域：所有其他角落和缝隙。

Yeah, I think a lot of times people think about the progress in physical AI as limited to basically two use cases, just because they're obvious and interesting: robotaxis and humanoids. They're very visceral, they excite you, and they're kind of sci-fi. I think those are very interesting, and there is real work being done by us and other people in those domains. I think all the other domains are going to be just as important. I mean, just think about what happens on a port—there's a huge unlock there. And that's the area we're really focused on: all the other nooks and crannies.

如果你看看，我们之前聊过思科的崛起，网络如何从最初的单台机器，到公司联网，再到整个国家联网。AI 也在发生类似的事情。AI 正在达到某种程度，主权 AI 现在成了讨论话题。主权 AI 实际上关乎物理 AI，因为那才是你在谈论国防中的 AI、在移动的物理机器中的 AI。如果只看 Waymo 和 Pony 的例子，它们试图在其他国家部署，不是美国、不是欧洲、不是中国，在每一个这样的地方，它们都更加犹豫，不会说“好，没问题，你们的机器人出租车可以在我们国家不受限制地运行”。所以，如果你回顾互联网的弧线，当第一批互联网公司出现时，没人真正考虑主权问题。就像浏览器无处不在，互联网无处不在，这几乎是它的力量所在。然后当社交媒体出现时，有了一点“嘿，其实不是所有社交媒体”的警惕，然后中国不允许 Facebook 进入。接着进入线上线下结合的下一个层面，阻力更大。Uber、DoorDash 这些，突然本地玩家被非常积极地扶持。当我们进入物理 AI 时，我认为会有巨大的阻力。而且还有一个更大的地缘政治主题，是分裂多于全球化。你会看到这种需求，即 AI 应该以某种方式本地化，我认为这必须纳入我们的战略。我们是技术提供商，所以我们可以在全球提供这种技术。我认为这在对话中被低估了。

If you look at like we've talked before about the rise of Cisco and how networking went from individual machines to companies getting networked and then entire countries getting networked. There's a similar thing happening with AI. AI is getting to that level where sovereign AI is now a discussion. Sovereign AI really is about physical AI because that's where you're talking about AI in defense, AI in the physical machines that are moving around. If you look just at the example of Waymo from America and Pony from China trying to deploy in other countries, not America, not Europe, not China, every one of those spaces, they're way more hesitant of saying, 'Yeah, thumbs up. Your robo taxis can run unfettered in our country.' And so, if you look back at the arc of the internet, when the first internet companies come, nobody really thinks about sovereignty at all. It's like the browser goes everywhere, the internet goes everywhere. That's almost the power of it. Then when social media emerges, there's a bit more of 'hey, actually not every social media' and then you have China not allowing Facebook to come in. Then you get into the next level of online-offline stuff, there's more resistance. The Ubers, the DoorDashes, suddenly there are local players who are being favored very aggressively. When we get to physical AI, I think there's going to be huge resistance. Also, there's a larger geopolitical theme of more fracturing than globalization. You're going to have this demand that AI should somehow be localized, and I think that has to play into our strategy as well. We're a technology provider, so we can provide that technology across the globe. And I think that's something that's understated in this conversation.

是的。关于数字 AI 和物理 AI，还有几点。在数字 AI 中，最先进的技术是你可以有效地在互联网的全部数据上训练模型，然后也许用一些雇佣专家收集和精炼的额外数据来增强。对吧？这现在是个热门领域。但通常你谈论的是基于互联网数据的 foundation model。在物理 AI 中，互联网数据也有用。然而，要真正构建物理 AI 的 foundation model，还需要大量私有数据收集。当我们谈论矿山、物流或任何其他领域时，用于训练模型的数据不一定可用。所以我们必须自己做很多工作，实际去收集这些数据。然后另一个关键因素是安全，对吧？如果你谈论构建智能手机应用，你不一定关心安全关键应用。但当你谈论移动一台重达数吨的机器，或者想象一个人形机器人可能倒在你的孩子身上，你会非常关心安全以及安全的评估。这确实是物理 AI 的前沿，真正证明一些最先进模型的安全案例。

Yeah. A few other things on digital versus physical AI. In digital AI, the state-of-the-art is you can train models effectively on the entirety of the internet and then maybe augment that with additional data that's been collected and refined with some hired experts. Right? This is sort of a hot field right now. But generally you're talking about a foundation model that's built on internet data. In physical AI, the internet data is useful too. However, to actually build a foundation model in physical AI, there's also a lot of private data collection. When we're talking about mines or logistics or any of these other fields, the data that's useful for training models there is not necessarily available. So we have to do a lot of work ourselves actually going out and collecting that data. And then the other key factor is safety, right? If you're talking about building a smartphone app, you don't necessarily care about safety-critical applications. But when you're talking about moving a machine that weighs many tons, or think of a humanoid which could fall over on your children, you care a lot about safety and the evaluation of that safety. And that is really sort of getting to the state of the art of physical AI and really proving out the safety case around some of these state-of-the-art models.

是的。我认为，你谈到人形机器人的数据收集本身就是一个小的兴趣领域，但当你谈到在韩国这样的地方收集数据时，那里有朝鲜，他们不允许地图公司，更不用说允许美国公司进来收集数据了。多年来我们已经弄清楚了，无论是在中东、拉丁美洲，如何进入这些国家，与政府合作，获得收集专有数据的许可。所以，在某种程度上，它类似于其他数字 AI 系统，你的专有数据集、缩放定律，所有这些都是一样的。只是应用方式非常不同。而且几乎可以这样想，这些模型的传播非常不同，因为不是每个人都能通过手机访问它们。所以，讽刺的是，这实际上对我们有利，因为一旦我们拥有庞大的专有数据集，我们一直在构建，我们已经拥有数百 PB 的数据，然后我们有自己的工具，比如合成数据工具、神经模拟，我们可以用我们自己的工具和我们自己的专有数据，这使我们能够构建业内最好的系统。这里有一个先有鸡还是先有蛋的问题，为了构建自主的物理事物，你需要大量数据，而为了收集这些数据，你需要大量物理自主事物在周围运行收集数据。所以，一旦你拥有一个巨大的物理事物网络在运行，你就有了让它们全部工作的数据。这里面有飞轮效应吗？启动这个飞轮的难度有多大？

Yeah. And I think like you talk about humanoid data collection has been its own little area of interest, but when you talk about collecting data in places like Korea, where they have North Korea, they don't allow mapping companies, let alone allowing an American company to come in and data collect. We've figured out over the years whether it's the Middle East, whether it's Latin America, how to get into these countries, work with the governments and get the thumbs up to collect proprietary data. And so in the way that it's similar to other digital AI systems, your proprietary data sets, scaling laws, all that stuff is the same. It's just applied in a very different way. And it's almost like the way to think about it is the diffusion of these models is very different because not everyone can just access them through a phone. And so that ironically actually plays in our favor because once we have a massive proprietary data set where we've been building, we already have hundreds of petabytes of data, and then we have our own tools which are like synthetic data tools, neural sim, we can use our own tools with our own proprietary data, and that allows us to build some of the best systems in the business. Is there a chicken and egg thing which is like in order to build an autonomous physical thing you need a lot of data, and to gather that data you need a lot of physical autonomous things running around collecting the data. So once you have a giant network of physical things running around, you have the data that makes them all work. Is there a flywheel aspect of that? And what's the level of difficulty involved in booting up that flywheel?

这很难，但也不难。我的意思是，坦率地说，我认为我们拥有地球上最大的数据收集车队之一。所以这就是你启动它的方式。那只是钱、资源和技术知识，但可能只有不到五家公司拥有这种技术知识。所以并不是极其晦涩。我认为更困难的是，你如何真正拥有那个能在许多不同硬件上工作并经过适当测试的模型，因为你在 Cruise 看到了，对吧？Cruise 是一家做了出色自动驾驶工作的公司，然后一次事故，通用汽车拥有他们，他们非常害怕，然后退缩了。所以，把这些东西投入生产实际上比看起来更难。我认为我们相信合成数据会很重要。所以我们在大约 5 年前就开始了我们的合成数据团队，加上，是的，现在更久了。而且我们坚信合成数据可以加速自主性的发展。我们刚刚看到了这一点，然后还有很多其他二级和三级的技术创新发生。

It's difficult, but it's also not difficult. I mean, I think we have one of the largest data collection fleets on the planet, frankly speaking. So that's how you bootstrap your way into it. That's just money and resources and technical knowledge, but it's not like there are probably more than five companies that have that technical knowledge. So it's not extremely obscure. I think what is more difficult is then how do you actually have that model which is going to work on lots of different hardware and is tested appropriately because you saw it in Cruise, right? Cruise was this company that did amazing self-driving work and then one accident, General Motors owns them and they get super scared and they pull back. So it's like just getting these things into production is actually more difficult than it seems. I think like we believed synthetic data was going to be important. So we started our synthetic data team like 5 years ago now, plus, yeah, more than that at this point. And like we're a strong believer that synthetic data can accelerate autonomy development. We've just seen that, and then there are lots of other secondary and tertiary technical innovations that happen.

显然，Transformer 革命对自动驾驶产生了巨大影响，基本上 2021/2022 年之前自动驾驶所做的一切都与之相关，但你几乎可以说那只是起点，而且和今天作为起点又有所不同。比如那四五年里其实已经做了很多工作，最明显的是特斯拉，但还有其他公司也在做。

Obviously the transformer revolution hit self-driving massively, basically everything done in self-driving pre-2021/2022 is relevant, but you're almost like that's kind of the starting point, but it's also different than today being the starting point. Like there those four or five years are actually there has been a lot of work done. You can see it most clearly with Tesla, but there's other folks in that process.

从历史上看，实际的技术，我这里只是简单说一下，模仿学习是当时的方法，就是收集大量数据，然后模型基本上模仿人类驾驶员的行为。

The actual techniques historically, and I'm just simplifying here, imitation learning was the way, which was collect a bunch of data and then the models would basically imitate what human drivers do.

现在真正最前沿的是在你们的工具中进行端到端的强化学习，形成闭环。

The real state-of-the-art right now is end-to-end reinforcement learning in a closed loop in your tools.

所以说系统自我学习有点简化了。它识别自动驾驶系统中的问题所在，然后你基本上找到类似的数据，或者合成创建类似的数据，然后你闭环，看看在同样的场景下是否表现越来越好。

And so it's a little simplified to say the system learns itself. It identifies where the issues in the self-driving system are, and essentially you then find data like that or you synthetically create data like that, and then you close that loop and you see are you performing in those same scenarios better and better.

我认为如果快进几年，那将是一个完全闭环，没有人类干预。现在仍然有，比如我们看到的雾气误差？我们仍然看到现实世界中影响自动驾驶的错误。

I think if you fast forward some years, that will be a completely closed loop, like with no humans intervening. Right now you still have, like, what's the fog error that we saw? We still see errors in the real world that impact self-driving.

哦，是的。

Oh yeah.

所以，瓶颈是什么？瓶颈有很多，但每当你处理物理系统时，不可避免地会遇到很多棘手的硬件问题。可能从过热到传感器轻微校准偏差，或者我们昨天看到的一个有趣问题，基本上是传感器起雾，比如雾气影响传感器。但这些都是你必须解决的事情，才能让这些东西在现实世界中非常可靠地工作。

So it's like, well, what are the bottlenecks right? And the bottlenecks, there's plenty of them, but whenever you're dealing with physical systems, you inevitably hit a lot of gnarly hardware problems. And it could be anything from overheating to sensor being slightly miscalibrated, or a funny issue we saw yesterday was basically a fogging sensor, like fog impacting a sensor. But these are the things that you actually have to solve for this stuff to work very reliably in the real world.

是的。所以我想问你一个问题，你们可以决定是否要回答。这可能是个机会，也可能讨厌这个问题，那就是你惊讶吗？Cruise 是一家非常成功的硅谷自动驾驶初创公司，早期与特斯拉并驾齐驱，团队非常顶尖，然后他们被通用汽车收购，这很出名。

Yeah. So I want to ask you a thread a question, and you can decide whether you guys want to engage on it or not. It might be an opportunity or might hate the question, which is were you surprised? So Cruise was a super high-flying Silicon Valley autonomy startup that was kind of running neck and neck with Tesla early on, and so forth, and you know, very top-end team, and then they famously got bought by General Motors.

我个人最早的投资之一。

One of my first distributions personally.

这就对了。Y Combinator 的公司。

There we go. Y Combinator company.

Y Combinator 的公司。

Y Combinator company.

而且你知道，团队顶尖，据大家所说，他们进展出色。他们被通用汽车收购，成为通用汽车的自动驾驶项目。通用汽车至少科技圈里得到了很多赞誉，被认为是投资最大的传统汽车制造商。

And you know, a top-end team, and they were, by all accounts, making excellent progress. They got bought by General Motors. They became the GM autonomy program. GM got a lot of praise, at least in tech circles, for being like the legacy automaker with the biggest investment.

在消息公布之前，我给 Peter 打了电话。我说，嘿，Cruise 刚被收购了，你知道，他也是通用汽车家族的。我们都是通用汽车家族的。Peter Guest 说，英伟达？我说不是。他说，苹果？我说不是。我说，通用汽车。

I called Peter when before it was announced on that. And I said, hey, Cruise just got bought, you know, he's also GM family. We're both GM families. And Peter Guest, he said Nvidia? I said no. He said I said go fish is Apple. I said no. I said General Motors.

对。所以这对来自通用汽车的人来说很惊讶，他们愿意收购，而且据大家所说，他们进展出色，然后发生了事故。是受伤还是死亡？

Right. So that's surprising to people who are from GM that they were willing to buy that they did it, and then by all accounts they were making excellent progress, and then they had this accident. There was an injury or fatality?

不是死亡，是重伤。有人被拖了 20 英尺。

It wasn't a fatality, it was a serious injury. Somebody was dragged for 20 ft.

是的，重伤，负面新闻，然后通用汽车 CEO 在董事会上终止了 Cruise 项目。我知道至少一些 Cruise 高管对后果非常不满。他们那样反应，你惊讶吗？

Yeah, serious injury, bad press, and then they put a bullet, the GM CEO on board put a bullet in the Cruise project. And I know that at least some of the senior Cruise people were extremely upset by the aftermath. Was it surprising that they reacted the way that they did?

所以，充分披露，通用汽车是我们的客户，我上过通用汽车学院，所以我们非常喜欢这家公司。但顺便说一句，讽刺的是，我正在读一本非常著名的书，我以前从未读过，叫《晴天可见通用汽车》。

So, full disclosure, General Motors is a customer, and I went to the General Motors Institute, so we have a lot of love for the company. But incidentally and ironically, I'm reading this very famous book which I had actually never read before, called On a Clear Day You Can See General Motors.

德洛雷安的书。你读过吗？

And Delorean's book. Have you read it?

就像人们常做的那样。

As one does.

就像人们常做的那样。

As one does.

你读过那本书吗？

Have you read that book?

我读过，很多年前。这是史上最伟大的书名之一。我们应该暂停一下，说约翰·德洛雷安是汽车行业的天才。

So I have, years ago. It's one of the great all-time book titles. And we should just pause to say John Delorean was like, he was the super genius of the car industry.

他本来要成为通用汽车的下一位总裁。

He was going to be the next president of General Motors.

通用汽车的总裁，后来他创办了自己的汽车公司，就像《回到未来》那样，然后整个事情因为各种原因崩溃了。

Of General Motors, and then later on he started his own car company, which like Back to the Future, and then that whole thing collapsed for a variety of reasons.

但是，是的，他像个传奇。他是当前鲍勃·卢茨这类创新的主要推动者之一。你要记住这一点。

But yeah, he was like a legend. He was one of the main principal drivers of innovation of the current Bob Lutz this category. And you got to remember this is right.

但抱歉，重复一下书名。重复一下书名。

But sorry, repeat the title. Repeat the title of the book.

《晴天可见通用汽车》。

On a Clear Day You Can See General Motors.

为什么书名叫这个？因为有很多……

And why was that the title of the book? Because there's a lot of...

这是一个非常庞大的综合体。

It's a very large complex.

是的。综合体。是的。

Yeah. Complex. Yeah.

就像一个民族国家。

It's like a nation state.

是的。

Yeah.

它确实意味着，我的意思是，我认为我们有时几乎轻率地说，但这些公司就像国家的延伸。现代汽车是国家的延伸。丰田是国家的延伸。大众汽车简直是，大众董事会成员是政府成员。所以这些是国家的延伸，几乎每个……而且有句老话，对通用汽车好就是对美国好。你不能低估通用汽车在美国公司历史中的重要性。斯隆的《我在通用汽车的岁月》和《白领历险记》。如果你管理一个大型工程组织，你应该读那本书。我们所说的现代公司不是凭空出现的。斯隆和凯特林，凯特林是工程主管，创造了这个，你知道，有层级和副总裁，以及如何做职能和矩阵组织？就在那里。真的，就像源代码一样。约翰，你知道，德洛雷安来了，他说，他写道，他要当总裁，他对公司非常不满。但有争议的是，通用汽车当时做得很好。通用汽车就像，我们说通用汽车在《财富》100 强中排名第一，它就像第一、第二和第三。它是一切，被视为美国最好的公司。所以有人公开批评公司，所以他写了一整本书，他辞职时对通用汽车被领导的方式非常不满。他写了这本书，然后他，比如，他后悔了，他说，我不想出版那本书。所以他多年来一直争取让合著者不要出版这本书。

It means really, I mean it is like, I think we say that sometimes almost flippantly, but these companies are like extensions of the state. Hyundai is an extension of the state. Toyota is an extension of the state. Volkswagen is literally, Volkswagen board members are members of the government. So these are extensions of the state, and almost every... And there used to be an old saying, what's good for General Motors is good for America. And you cannot underrate how important General Motors is in the history of the American corporation. Sloan's My Years at General Motors and Adventures of a White-Collar Man. If you run a large engineering organization, you should read that. This thing that we talk as a modern corporation didn't just emerge. Sloan and Kettering, Kettering is the head of engineering, created this, you know, with levels and vice presidents, and how do you do functional and matrix organizations? It's there. Really, like the source code comes along. John, you know, comes on Delorean, and he says, he writes, he's going to be president, and he's so fed up with the company. But what was controversial was GM was doing really well at the time. GM was like, when we say GM was number one in the Fortune 100, it was like number one, two, and three. It was everything, and it was seen as the best company in America. So somebody to openly criticize the company, and so he has a whole, he writes this book as he quits out of how annoyed he was, GM is being read led. He writes this book, and then after he, like, so up, he's like, I don't want that book published. And so he fights for years for his co-author not to publish the book.

合著者还在出版这本书。所以这是对一家大公司的真实洞察。我碰巧在读这本书，尽管我 20 多年前在通用汽车工作过，对这家公司了解很多。令人震惊的是，这不仅仅是通用汽车的问题，大多数主要制造商内部实际上仍然这样运作。大家要记住的重点不是经营这些公司的人愚蠢——他们并不愚蠢。这有点像当你向国防部销售时，人们会说‘你为什么要这么做？’就像，嗯，分销定义了业务。

The co-author still publishes it. So it's a real true insight into a large corporation. I'm incidentally just reading it out even though I've worked at GM 20 some years ago and had a lot of knowledge about the company. What's shocking is it's not only about GM; most of the major manufacturers actually still operate that way on the inside. The point for everyone to take away isn't that these people who run these companies are stupid—they're not stupid. It's kind of like when you're selling to the Department of War and people say, 'Why are you doing that?' It's like, well, the distribution defines the business.

就像，所以分销就是这是一个消费品，这个数据可能过时了，但 20 年前我在安全系统部门工作时，我记得通用汽车总是向你灌输美国历史上五大消费者诉讼中，三个是汽车行业的。我们占了大多数，所以你必须极其小心。但我们公司内部有这些奇怪的事情，你不能——不是红、黄、绿。而是像紫色，或者你总是得解码，因为你知道为什么吗？因为当他们打官司时，他们会说‘你让一个标记为红色的安全系统投入生产了。’我说‘不，它标记的是洋红色。’所以你能想象这有多令人恼火吗？每次你都像……

Like, so like the distribution is this is a consumer product of this stat might be outdated, but when I worked in safety systems 20 years ago, I remember GM used to pound into your head of the top five consumer lawsuits in American history, three are automotive. We got the majority right, so it's like you have to be extremely careful. But we had these weird things like inside the company you couldn't—it wasn't red, yellow, green. It was like purple or you'd always have to decode it because, you know why? Because when they go to lawsuits, they're like, 'You let a safety system that was marked red go to production.' I was like, 'No, it was marked magenta.' So like, can you imagine how infuriating that is? Every time you're like...

橙色是什么意思？这是否意味着我必须……

What does orange mean? Does this mean I have to...

所以快进到你遇到那个系统。嗯，福特很长时间的口号是‘质量是第一要务’，对吧？安全是工作……

So fast forward to you're meeting that system. Well, for then the Ford slogan for a very long time was that it was 'Quality is Job One,' right? Safety is job...

是的。对。没错。这就是汽车行业的一记组合拳。是质量和安全。质量和安全，而质量之所以变得重要，是因为日本人真正重新设定了这个舞台，而那是另一段完整的汽车历史。我们可以聊一个小时的汽车历史。但关键点是硅谷公司遇到了这个不可动摇的对象。

Yes. Yeah. Exactly. And that's the one-two punch of automotive. It's quality and safety. Quality and safety, and quality really becomes because the Japanese really reset that stage, and that's a whole separate automotive history. We could talk about automotive history for an hour. But the punch line is you have the Silicon Valley company meeting this immovable object.

现在有一个平行宇宙，Cruise 就在那里，即使作为通用汽车的一部分。所以，我认为你总是必须把它放在公司的背景下来看，比如那一年工会谈判在哪里进行。如果你是工会，你会说‘你不能为我们赚十亿美元，但你却在资助这个害死人且草率的东西。’所以，我不是说确切就是这样，但这是一个多变量问题。

There is a parallel universe that cruises out there right now, even as a part of General Motors. So, I think you always have to take it into the context of where the company is, where union negotiations are happening literally that year. And if you're the union, you're like, 'You can't make a billion dollars for us, but you're funding this thing that's killing people and it's sloppy.' And so, I'm not saying precisely that's what happened to be very clear, but it's a multivariate problem.

我的另一个大胆观点是，你知道，我在两家公司都工作过，对吧？谷歌和通用汽车。这些公司的相似之处远多于不同之处。

My other hot take is, you know, I worked at both companies, right? Google and General Motors. Those companies are way more similar than they're different.

相似得多。真的，人们不需要知道这个。谷歌的职级体系和通用汽车的一样。我以前在谷歌会议上说，‘嘿，实际上我在通用汽车认识的一些工程师比这里的工程师更好。’人们会看着我，就像我在教堂里说没有上帝一样。他们就像‘你怎么敢，你这个底特律来的金属弯曲猴子。’我说，不，实际上制造现代内燃机极其复杂。这不是简单的玩意儿。所以宏观观点是，这是很多因素的综合。我认为安全始终是他们清单上的首要事项。我确实认为，你知道，我们雇了很多 Cruise 的人。我认为他们处理那个特定政府问题的方式，那种情况下你必须以特定方式跳舞，而他们只是没有跳对，这就给了政府官僚……

Way more similar. Literally, people don't need to know this. The Google leveling system is the same as the General Motors leveling system. And I used to say, you know, this inside of Google meetings is like, 'Hey, actually some of the engineers I knew at General Motors are better than the engineers here.' And people would look at me like I'm saying there's no God in church. It's like they're like, 'How dare you, you metal-bending monkey from Detroit.' It's like, no, actually making a modern combustion engine is extremely complex. It's not just like, you know, it's not simple stuff. And so the macro point I think is it's a bunch of things. I think safety is always at the top of their list. I do think, you know, we've hired lots of Cruise people. I think the way they dealt with that specific issue with a government, you got to dance a particular way when that happens, and they just didn't dance exactly right, and that just gives government bureaucrats...

更多弹药去追捕。而且你是像通用汽车这样的大目标，你必须——你知道，这让我想起，你们看过《好家伙》那部电影吗？你知道，最后一场戏之一，‘旭日之屋’，你知道，所有老老板都走进法庭后面，他们就像——那就是发生的事。他们就像，董事会说‘我们该怎么处理 Cruise？’‘我们能做什么？’‘凯尔是个好人。’但然后就像‘响起旭日之屋。’人们跑过旧金山办公室。开玩笑的。别把这个做成 AI 视频。我会收到凯尔刻薄的短信。

More ammo to go after. And you're a big target like General Motors, you got to—you know, it reminds me, you guys ever see that movie like Goodfellas? You know, there's one of the last scenes, 'House of the Rising Sun,' you know, all the old bosses go in the back of the courtroom, and they're like—and that's what happened. And they're like, the board was like, 'What are we going to do about Cruise?' Like, 'What can we do?' It's like, 'Kyle's a good guy.' But then it's like, 'Cue the rising sun.' People running through a San Francisco office. Just kidding. Don't make that an AI video. Gonna get a mean text from Kyle.

所以，我认为存在一个本可以生存下来的宇宙，但这很难。所以 Applied Intuition 做的很多事情就是，正如你所说，那种舞蹈——就像如何成为这些公司的优秀合作伙伴。

So, I think there is a universe that would have survived, but it's tough. So then a lot of what Applied Intuition does is kind of, as you said, that dance—it's like how to be a great partner to these companies.

没错。要记住他们自己非常现实的问题和限制。

Exactly. Bearing in mind their own very real issues and constraints.

我认为通用汽车还有商业模式的问题，对吧。所以 Cruise 在追求自动驾驶出租车概念，但通用汽车从个人汽车所有权中获利，而这些事情可能有点奇怪。所以我认为这也是其中的一部分。

I think General also had the topic of business model, right. So you have Cruise was going after the robo-taxi concept, but GM makes its profits from personal car ownership, and those things can be a bit odd. So I think that was also a bit of the equation.

好的。

Okay.

是的。而且我认为当时并不清楚。我的意思是，顺便说一句，你知道，你 2013 年在 YC 讲过话。我当时在观众席。我当时是合伙人，你说了一些我认为在这里非常递归的话。我们在互相喂设备。在新科技业务中，关键的事情是：实际上每个人都能搞清楚技术，尽管那仍然很难。有时构建真正复杂的东西仍然很难。重要的是你何时以及如何将它们部署到市场。‘何时’变得非常重要。你早两年就注定失败。你晚两年，竞争对手太多。你必须恰到好处地击中时机。而且我认为……

Yeah. And I think it wasn't clear. I mean, by the way, you know, you actually spoke at YC in 2013. I was in the audience. I was a partner at the time, and you said something which I think is very recursive here. We're feeding each other devices. It's the key thing in new technology business: actually everyone kind of figures out the technology, though that's still hard. It's still hard sometimes to build really complex things. It's when and how you deploy them into the market. The 'when' becomes really important. You're two years early and you're doomed. You're two years late, there's too many competitors. You have to hit it at the right spot. And I think...

就像，我的意思是，一个有争议的说法是，我实际上认为 Cruise，你知道，他们肯定比 Waymo 快得多。他们起步晚得多，而你说的是在最终拔掉插头时并驾齐驱。所以谁知道长期会发生什么。我们在同一个等式中的假设实际上是分销。你让制造商来做。比如我们现在在日本运营自动驾驶卡车。它们运载商业货物。它们，你知道，那里有安全驾驶员，但它们是自动驾驶的。但你不会知道，因为品牌是五十铃。

It's like, I mean, a controversial thing to say is like, I actually think Cruise, you know, they were certainly moving at a much faster pace than Waymo. They started way behind, and you're talking about neck and neck when ultimately the plug was pulled. So who knows what happens in the long term. Our hypothesis in that same equation is actually the distribution. You let the manufacturers do that. Like we run self-driving trucks right now in Japan. They carry commercial loads. They're, you know, they're safety drivers there, but they're autonomously running. But you won't know that because the brand is Isuzu.

嗯。

Mhm.

那是客户。在这种情况下，我们与五十铃合作之所以如此好，是因为那家公司已经存在了近百年，对吧？如果我没记错的话，是二战前的公司。

That's the customer. And why it's so good for us to partner with Isuzu in that case is that company's been around for almost a hundred years, right? If I'm not mistaken, pre-World War II company.

而且他们，你知道，他们了解政府，他们有测试跑道，他们懂安全，他们非常了解自己的卡车。所以当我们去为他们提供智能以及与他们实体机械的整合时，那是一个绝妙的组合拳。我认为今天世界已经准备好消费现实世界中的 AI，这在很大程度上是因为 ChatGPT 和 Anthropic 以及所有这些，你知道，已经发生的一切。所以人们不再问什么是自动驾驶汽车，这要归功于 Waymo 和 Tesla。

And they are, you know, they know the government, they have test tracks, they know safety, they know their own trucks very well. So when we go and provide them with the intelligence and the integration into their physical machinery, that's a fantastic one-two punch. I think today the world is ready to consume AI in the real world, and that's a lot because of ChatGPT and Anthropic and all these, you know, everything that's happened. So people are no longer like, what's a self-driving car? And it's because of Waymo and Tesla.

所以市场已经准备好消费，我认为你只需要以最好的方式满足市场。

So the market is ready to consume, and I think you just have to meet the market in the best way possible.

而我们的观点一直是，你要通过那些现在运营经济的人。无论是矿业运营商、战争部门还是制造商，我们在每个垂直领域都与合适的合作伙伴合作。但这与 Tesla 或 Waymo 根本不同，它们将是垂直的，而我们真正在玩横向。我认为我们思考自己公司的方式，我们有点像芯片制造商，你知道。我们实际上看起来、说话和行事都很像一家硅谷公司，除了我们显然不制造芯片，但你知道，我们有设计胜利，然后我们有非常庞大且长期的合作关系，一旦我们进入，我们就进入了，很难把我们赶出去。所以你需要深度信任，我们的合作伙伴确实有很多深度信任，而且我们非常非常了解他们的市场。Jensen 知道的是他了解他的客户。这就是 Nvidia 做得好的原因，除了显然他们制造了非常复杂的技术。

And our view of that has always been you go through some of the people who run the economy right now. Whether it's a mining operator, whether it's a department of war, whether it's the manufacturers, and we work with, you know, within each vertical with the right partner. But that's a fundamentally different view than a Tesla or a Waymo, which are going to be vertical, where we're really playing the horizontal. And I think the way we can always think about our companies, we're kind of like a chipmaker, you know. We actually look and talk and walk a lot like a silicon company, except we obviously don't make chips, but you know, we have design wins and then we have really large long-term relationships, and then once we're in, we're in, it's really hard to take us out. So you need deep trust, our partners have really a lot of deep trust, and we know their markets really, really well. The things that Jensen knows is he knows his customers. That's why Nvidia does well beyond the fact obviously they make a very complex technology.

那么，这些传统汽车公司如何为未来做准备？他们是在进行更多收购吗？他们是在建设、与你合作吗？他们将如何与科技原生公司竞争？

So, how are these legacy car companies preparing for the future? Are they making more acquisitions? Are they building, partnering with you? How are they going to compete with tech-native companies?

这就像说政府如何处理 AI？这是一个非常广泛的话题。而且每个制造商，比如你拿本田、日产、丰田，三家有着悠久历史的日本制造商，他们的处理方式都截然不同。他们大致处于从“我们要自建”到“我们要购买”的光谱上。而且比以往任何时候，“我们要购买”都是常见的答案，因为他们一直在尝试，而我们一直都在那里。对于想要自建的人来说，我们提供工具，我们会稍微谈谈我们在这里发布的新产品。而对于那些只想购买的人来说，我们向他们出售实际安装在机器上的智能。所以我们满足客户在旅程中的任何位置。更微妙的版本是，你知道，现实是每个产品都是不同的产品，所以你可以投入的硅片数量和美元数量，客户愿意支付什么，所有这些都取决于长期实际能得到什么。所有这些都将完全自主，但中间的步骤非常像我们在 PC 上看到的，你有一个缓慢的升级过程，直到有一天，就像现在，没有人真正看笔记本电脑的规格，甚至可能坦率地说你的手机规格，但从 85 年到 2002 年、2005 年并非如此，那时人们终于完全停止关注规格，然后他们真正转向笔记本电脑。但那里也有一个类似的 20 年时间跨度。

It's like saying how are governments dealing with AI? It's such a broad topic. And each manufacturer, like even you take Honda, Nissan, Toyota, three Japanese manufacturers with long legacies, they all approach it very differently. They're roughly in a spectrum of we're going to build to we're going to buy. And more than ever, we're going to buy is the common answer because they've been trying and we've been there the whole time. For the folks that are going to build, we provide them tools, and we talk a little bit about our new product that we're announcing here. And then on the ones that just want to buy, we sell them the actual intelligence that goes on the machines. And so we meet the customer wherever they are in their journey. The more nuanced version of that is, you know, the reality is every product is a different product, and so the amount of silicon and amount of dollars you can put towards sensors, what the customer is willing to pay, all that depends on what actually gets in the long horizon. All these things will be fully autonomous, but the intermittent steps are very much what we saw in the PC, where you have this slow step up to one day that'll be like now, nobody really looks at laptop specs, and even maybe frankly your phone specs, but that's not the case from basically 85 to 2002, 2005, where finally people stop actually specking at all and then they're really moving to laptops. But there's a similar kind of 20-year horizon there.

广义上讲，当我们谈论机器和机器变得智能时，对吧，从根本上说，机器是这些不同组件的集合，它们被整合在一起，对吧？而做最终整合的往往是那个把品牌标志放在上面的公司，品牌名称，但很多很多公司都在构建进入这些机器的技术。所以我们现在有一堆可以进入这些机器的技术组件和平台，但我们也出售核心技术，这些技术也可以用来帮助开发它们。

Broadly when we talk about machines and machines becoming intelligent, right, fundamentally a machine is a collection of these different components that are integrated, right? And whoever does that final integration is often times the company that puts their badge on it, the brand name, but many, many companies are building technology that goes into those machines. So we now have a bunch of technology components and platforms that can go into these machines, but we also sell the core technology that can be used to develop help them as well.

而且如果你看，顺便说一句，在一台推土机、联合收割机或柴油卡车的引擎盖下，它们都有康明斯发动机。但没有人说，好吧，因为这些家伙都买康明斯，这意味着卡特彼勒不是一家好公司。这就像，不，那只是他们购买的组件。他们有不同角色。所以，当你观察这些垂直领域中的任何一个，它只是一个复杂的人员网络。这就是为什么我总是说芯片类比实际上非常有效，因为那些公司都不制造芯片，但他们都购买芯片。所以我认为这是一个很好的思考方式。

And if you look, by the way, under the hood of a dirt mover or a combine or diesel truck, they'll have Cummins engines in them. But nobody says, well, because all these guys buy Cummins, this means that Caterpillar is not a good company. It's like, no, that's just a component that they buy. They have a different role. So, when you look in any of these verticals, it's just a complex web of folks. That's why I always say the chip analogy actually works quite effectively because none of those companies make chips, but they all buy chips. And so I think that's a good way to think about it.

所以自动驾驶汽车。你知道我们一直在谈论自动驾驶汽车，我想整个事情大概始于 2005 年左右，最初是 DARPA 大挑战赛，然后 Google 在那之后不久参与了该项目。

So self-driving cars. So you know we've all been talking about self-driving cars for like, I think the whole thing started like around what, 2005 or something with the DARPA Grand Challenge originally, and then Google engaged in the program shortly after that.

是的。00 年代末。

Yeah. Late double O's.

00 年代末。所以基本上差不多快 20 年了。而且在过去 20 年里有很多预测说自动驾驶汽车随时都会到来。所以我想坏消息是我们今天坐在这里，大多数汽车还不是自动驾驶的。好消息是现在有自动驾驶汽车了。

Late double O's. So almost basically around a little less than 20 years maybe. And there have been lots of predictions over the last 20 years of like self-driving cars are imminent at any moment. So I guess the bad news is we're sitting here today and most cars are not self-driving. The good news is there are now self-driving cars.

是的。

Yeah.

所以在它们部署的地方，大多数汽车的驾驶方式，你知道，已经变得，你知道，就像旧金山的人们，我认为，现在把进入它们视为常规。

And so the way most cars are driving all over, you know, in the places they're deployed, it's become, you know, like people in San Francisco, I think, treat it now as routine that they get into.

而且我认为你可以说，我认为 Tesla，这有点像 AGI 的事情。就像，你知道，如果我们谈论 20 年前，我们现在看到的一切都像是令人惊叹的 AGI。目标不断移动。Tesla 的东西令人惊叹。你可以看看很多制造商。Blue Cruise、Super Cruise、BMW、Volvo 的 Pilot，它们都是相当令人印象深刻的系统。它们不是完全自动驾驶，对吧？

And I think you can call, I think Tesla, it's kind of like the AGI thing. It's like, you know, if we're talking 20 years ago, everything we're seeing right now is like mind-blowingly AGI. The post keeps moving. The Tesla stuff's amazing. You can look at a bunch of manufacturers. Blue Cruise, Super Cruise, BMW, Volvo's Pilot, they're all quite impressive systems. They're not full self-driving, right?

但是，是的，嗯，这是完全自动驾驶 X，无论远程监控正在发生什么。我们的 Tesla，我们在洛杉矶有一个家，你们可能还记得洛杉矶发生了一场大火。然后电力，而且在那之前加州电网就已经不堪重负了。所以，实际上事实证明 Cybertruck 擅长的事情之一是它们是非常好的电池。

But yeah, well, it's full self-driving X, whatever remote monitoring is happening. The Tesla we have, we have a home in Los Angeles, and you guys may recall there was a large fire in Los Angeles. Then the power, and then the California power grid was buckling even before that. And so, it actually turns out among the things Cybertrucks are good at is they're very good batteries.

是的。是的。

Yeah. Yeah.

所以实际上我们有 Cybertruck 作为房子的备用电池。而且截至去年，无论 FSD 发布了什么，我忘了具体是哪一个，但有一个至少很多人认为它真的迎来了转折点。

And so literally we have Cybertrucks as our backup battery for the house. And as of last year, whatever the FSD released, I forget the exact one, but there was one where it at least a lot of people thought it really turned the corner.

14。

14.

而且那东西真的能自己开。我昨天跟一个开 Model Y 的人聊过，他让车全程自己开，一路沿着 1 号公路穿过大苏尔。

And like that thing drives you. I talked to somebody yesterday who has a Model Y who let the thing do the full route all the way up Highway One through Big Sur.

对。我觉得平均接管次数真的很低，每次接管之间的行驶里程非常高。我觉得里程数能达到几千英里。

Yeah. I think mean disengagements are really low, and miles per disengagement are really high. I think miles is like in the thousands.

是啊，这非常了不起。

Yeah, which is very impressive.

对。对于没开过 1 号公路大苏尔那段的人来说，那是一次压力很大的驾驶。他说全程都很棒。

Yeah. The big for people who haven't driven the big Highway One, Big Sur, that's a stress-filled drive. He said it was great the whole way.

总之，要不是他把方向盘拆了，我也不会跟他聊这个，那车可能就直接冲下悬崖了。所以，直接冲下悬崖。

Anyway, so and I wouldn't have been talking to him had it not been would have gone right off the cliff because he unbolted the steering wheel. So right off the cliff.

嗯，然后你知道特斯拉正在推出他们的机器人出租车，已经开始在现实中出现。所以一方面，这些车存在。另一方面，你知道，99.9999999% 的汽车仍然不是自动驾驶。

Um, so um, and then you know Tesla's rolling out their robo taxi, you know, is starting to show up in the wild. So on the one hand those exist. On the other hand, you know, 99.9999999% of cars are still not self-driving.

然后我想再说一个，就是自动驾驶卡车。媒体上一直反复出现这种恐慌，比如卡车变成自动驾驶，然后就业，你知道，所有这些卡车司机都会失业。而今天坐在这里，我不认为，我不知道。现在路上有没有自动驾驶卡车，车上连安全员都没有？我觉得答案可能仍然是没有。

And then I would say maybe just one other would be the self-driving trucks. There's been this recurring kind of panic in the press of like the trucks become self-driving and the employment, you know, all these truck drivers be out of a job. And sitting here today, I don't think I don't know. Is there are there any trucks on the road that are self-driving that don't have at least a safety driver in the truck? And I think the answer is probably still no.

对，银，如果我们把这里提到的几个点分开来看。一个是关于个人拥有的车辆，为什么它们没有更普及。部分原因是制造商不擅长部署技术。部分原因是他们想注重安全，但大部分原因是成本。你在中国的所见，中国是一个不同的电动汽车生态系统，主要是因为它们不在乎利润。当你谈论一个不在乎利润的企业时，它改变了整个行业的整个计算方式。但你所看到的是 L2++ 系统。所以我们可以把整个自动驾驶的讨论简化为方向盘后面有没有司机？

Yeah, silver, if you so let's split the multiple points that we brought up here. One is on the personally owned vehicles and why are they not more ubiquitous. Part of that is the manufacturers are not good at deploying technology. Part of that is they want to be safety conscious, but most of it is cost. What you're seeing in China, which is a different EV ecosystem mainly because they don't care about profits. When you're talking about a business that doesn't care about profits, it changes the entire calculus of the entire industry. But what you're seeing is L2++ systems. So we can simplify the entire self-driving conversation to is there a driver behind the steering wheel?

对吧？

Right?

所以方向盘后面仍然有司机，但通常像特斯拉那样到处都能开。它们价格低于 1000 美元，对吧？

So there's a driver behind the steering wheel still there, but generally like Tesla drives everywhere. They're like sub $1,000, right?

那是一个激进的方案，包括芯片、传感器、整套系统、软件，所有东西。

There's an aggressive that's chip, sensors, the package, the software, everything.

我们预计一旦价格降到大约 500 美元，汽车制造商实际上会免费补贴，他们就直接给你。这在导航系统中发生过。如果你们还记得，导航系统曾经是一件大事，你要花 4000 美元、3500 美元才能得到一个导航系统，然后突然它变成免费，并且成为标配。我认为那会发生。有一件奇怪的事情，实际上进入你的一部分汽车要花 x 美元，而进入所有汽车要花 x 加上一点点增量，因为它只是固定成本，而且考虑到车辆数量、装配线的方式、以及你进行认证、所有这些测试制度等等。所以我认为你会等待，然后很多，每一个汽车制造商，无一例外，甚至是最低价的汽车制造商，都在研发 FSD 的竞争对手。

We anticipate that once you get to like $500, the automotive OEMs will actually subsidize it for free, they'll just give it to you. This happened in nav systems. If you guys remember, nav systems used to be a big thing where you pay $4,000, $3,500 to get a nav system, and then suddenly it became free and it just became default. I think that'll happen. There's a weird thing which is actually getting into a subset of your cars costs x dollars, and to get into all the cars costs x plus just a small incremental amount because it's just a fixed cost, and the way that how many vehicles and the way the assembly line comes and the way you have homologation, all these testing regimes, all this stuff. So I think you'll have wait and then a lot of every single OEM without exception, even the lowest dollar OEMs, are working on an FSD competitor.

所以它会到来，但就像，你知道，一个很好的类比来思考个人拥有生态系统中的自动驾驶，就是手机。

So it'll come, but it is just like, you know, with a good analogy to think about self-driving in the personally owned ecosystem is mobile phones.

我们有卫星电话，然后有高通砖头手机。然后有摩托罗拉 Razr，从 90 年代末到 2000 年代末，每个人都在想移动什么时候来？有一个巨大的，然后它来了，到 2007 年，从 iPhone 发布，大约四年后，你有了 Uber、Instagram、WhatsApp、Snapchat，对吧？

We had the satellite phones, then we had the Qualcomm brick phones. Then we had the Motorola Razors, and from the late '90s to the late 2000s, everybody was like when's mobile going to come? There was a huge like, and then it comes, and by 2007, from the iPhone launch, it's like four years when you get Uber, Instagram, WhatsApp, Snapchat, right?

那些是杀手级应用。

Those are the killer applications.

所以我认为会有非常类似的等待，然后它基本上在每辆车中都无处不在。嗯，如果你问我那个数字是什么，28 年 SOP，29 年，开始生产，29、30 年，然后到 30 年代初，它会开始变得非常便宜甚至免费。

So I think there's a very similar kind of wait and then it's just basically ubiquitous in every vehicle. Um, if you had to ask me for what that number is, 28 SOP, 29, start of production, 29, 30, and then by the early 30s, it'll start becoming very cheap to free.

到 30 年代初，通常你买一辆车，你就默认它有自动驾驶。

Routinely by the early 30s, you would just you buy a car and you just assume self-driving.

完全正确。或者它有驾驶员在座位的 L2++ 系统，非常具体。

Exactly. Or it has the driver in seat L2++ system being very specific.

比如 Cybertruck 或特斯拉，特斯拉车主今天拥有的将成为普遍。

Like Cybertruck or Tesla, what Tesla owners have today will become common.

对。将成为默认。

Yeah. Will be default.

那么接下来的问题，另一面是为什么我们到处都没有一堆 Waymo？

So then the question then, the other side of this is why don't we have a bunch of Waymos everywhere?

具体来说，Waymo 有不同的技术。这里不深入细节，特斯拉和许多中国及应用派非常倾向于这种端到端模型架构。这是一种新的自动驾驶方式。Waymo，找不到更好的词，不是那样的。这并不意味着他们没有学习。只是没有一个端到端系统。不是一个单一的模型。他们方法的一个倾向是它确实依赖高清地图。因此，有一个地理围栏的概念。我认为 Waymo 正在努力消除这个瓶颈，以便在地理上更快扩张。但今天的现实并非如此。另一件事是，当你有研究人员，这确实来自 Alphabet 的研究组织，他们没有施加商业限制。所以传感器是定制的且昂贵。车和里面的算力，它们在经济上不可行。他们尝试了很多来降低成本。但这有点像，从非常便宜的东西开始并使其功能更丰富，比从过度设计的东西开始然后试图削减并使其非常非常便宜要容易得多。这就是大辩论。谁会先到达那里？特斯拉的全自动驾驶，还是 Waymo 的成本和地理普及。但你知道我们不辩论什么吗？它会不会发生？

Specifically Waymo has a different technology. Without getting into the nuances here, Tesla and many of the Chinese and applied were very much in this end-to-end model architecture. This is a new way of doing self-driving. Waymo, for the lack of a better word, is not that. It doesn't mean they're not learned. It's just there's not one end-to-end system. It's not one monolithic model. One of the proclivities of their approach is it does depend on HD maps. Therefore, there is a geo-fencing concept. I think Waymo is trying hard to remove that bottleneck so they can expand geographically faster. But the reality of today isn't there. The other thing is when you have researchers, which really was coming out of an Alphabet research organization, they didn't put commercial constraints. So the sensors are bespoke and expensive. The cars and the compute that are in there, they're just not economically feasible. And they've tried a lot to get that down. But it's kind of like it's a lot easier to go from something that's really cheap and make it more featureful than something that's overbuilt and then trying to trim and make it really, really cheap. And that's the big debate. Who's going to get there first? Tesla with full self-driving or Waymo with cost and geographic ubiquity. But you know what we're not debating about? Is it gonna happen?

对吧？

Right?

你知道我们不辩论什么吗？比如是否需要发生重大的技术突破？这些都不是。所以现在我们显然处于自动驾驶的工程方面，这只是一个磨砺到每英里美元效率的过程。而一旦它便宜了，你猜怎么着？所有汽车制造商都很聪明。他们就会采用它。并不是汽车制造商抵制，因为他们认为消费者不想要它，或者他们不理解技术。

You know what we're not debating about? Like is there a big technical breakthrough that needs to happen? None of those things. So now we're clearly in the engineering side of self-driving, which is just this grind down to like dollar per mile efficiency. And the moment that it's cheap, guess what? All the OEMs are smart. They'll just adopt it. It's not that the OEMs are resistant because they don't think consumers want it or they don't understand the technology.

这是因为他们想要一个价格区间，以便在规模上保持微薄的利润率，而这个规模在 V1 阶段已部署到 100 多个国家。所以如果你只是做一个小规模部署，情况就非常不同。最后我想说的是，斯巴鲁或铃木的买家与特斯拉的买家有着非常不同的品牌期望，包括消费者的年龄以及他们认为会发生和不会发生的事情。所以这也是原因之一。所以如果你是铃木，你会想，我的买家不想要这些东西，所以我不会把它硬塞进车里。这不是因为他们技术能力不行，只是领域不同。

It's because they want a price envelope which allows them to keep their razor thin margins at scale, which is deployed across 100 plus countries in V1. And so if you're just doing a small deployment, it's very different. And the last thing I would say is the buyer of a Subaru or a buyer of a Suzuki have very different brand expectations than a buyer of a Tesla, including the age of the consumer and what they think will happen and won't happen. So that's also the reason. So if you're a Suzuki, you're like, well, my buyer doesn't want this stuff, so I'm not going to jam it into the car. It's not because they're not technically competent. It's just a different area.

你觉得什么时候会变成常态？比如说美国最大的 200 个城市，走在外面，理所当然地认为机器人出租车能来接你，这会是常态吗？

When do you think it'll be routine? Let's say the 200 biggest American cities. Would it be routine to walk outside and just take it for granted that a robot taxi can come pick you up?

现在是 26 年。我的意思是肯定到 30 年。好吧。是的，肯定到 30 年。我想说，那里的大变量实际上是，因为 Waymo 会说，每个城市的成本收益已经可行了，而且像一家基本上拥有无限资本的公司，为什么他们不已经在 200 个城市运营？但然后你看到他们的推出计划相当激进，你会觉得，那可以做到。所以如果我激进一点，我会说 28 年。

It's 26 now. I mean certainly by 30. All right. Okay. Yeah. Certainly by 30. And I would say the big variable there really is because what Waymo will say is that the dollars and cents per city already work, and it's like, well, a company that has basically unlimited capital, why are they not already in 200 cities? But then you see their launch schedule is pretty aggressive, and you're like, that can get there. So maybe if I was being aggressive, I would say 28.

嗯，好的。

Yeah. Okay.

我想说 30 年可用，但常态化可能要到 32 年。是的，33 年。

Like I would say available in 30, but routine maybe like 32. Yeah, 33.

当然。因为有一个规模扩大，有一个数量。

Sure. Because there's a scale up. There's a volume.

而且，你知道，你住在洛杉矶，所以五年前我去洛杉矶，人们会说，“什么是 Applied Intuition？我不知道自动驾驶汽车是什么。”在过去的几年里，现在他们都了解自动驾驶了。有些人甚至知道 Applied Intuition，因为他们从其他制造商那里听说过。我想再快进两到四年，每个人都会知道。

And also, you know, you live in LA, and so like five years ago I'd go to LA, people would be like, "What's Applied Intuition? I don't know what self-driving cars are." In the last couple years now, they all know self-driving. And some of them even know Applied Intuition because they know from the other manufacturers. I think you fast forward another two to four years, everybody knows it.

那是不是意味着每个人都只坐 Waymo 了？

Does that mean everyone's taking Waymos exclusively?

对吧？实际上答案是否定的。现在有一个巨大的，巨大的，如果你看数字，如果你是 Uber，你会害怕。我的意思是他们正在蚕食拼车市场。

Right? That answer is no actually. Now there is a huge, huge, if you look at the numbers, if you're Uber, you got to be scared. I mean they're just eating into ride sharing.

嗯，是的，但要达到 100% 的普及，我的意思是那是另一回事，它必须非常便宜。

Um, yeah, but to get 100% ubiquity, I mean that's another, it has to be extremely cheap.

那长途卡车运输呢？

And what about long haul trucking?

那是乘客方面。长途卡车运输是完全不同的经济模式，完全不同的商业模式。现在有很多公司，我想说大概超过五家，在运营有司机的长途卡车，在美国和中国之间运输货物。如果算上中国，可能达到两位数。所以它存在，但你不知道它、它不在你脑海中的原因是它不是消费品。不像 Waymo 和特斯拉那边，投资者愿意基本上给你一些市值调整以换取潜力，他们说卡车运输业务就像，你买车是凭感情，你买卡车是用计算器。

So that's the passenger side. The long haul trucking is completely different economics, completely different business model. There are many companies right now, I would say probably north of five, that are running long haul trucks with drivers carrying loads between America and China. If you had China, it's probably getting into double digits. So it's there, but the reason you don't know it and the reason it's not top of mind is it's not a consumer product. And unlike on the Waymo and Tesla side where investors are willing to essentially give you some market cap adjustment for the potential, they say the trucking business is like, you buy a car with your heartstrings. You buy a truck with a calculator.

是的，这是一个计算器业务。

Yeah. It's a calculator business.

所以这纯粹是成本收益，所以我认为，作为自动驾驶卡车提供商，如果你做整个事情，就像一些公司那样，我们不是，你必须展示每一英里，我将为你节省这么多美元，而且这绝对是肯定的，因为买家不成熟，他们只是说，好吧，我已经有员工能开车了，他们就是不倾向于。现在我们在日本玩，我们做卡车运输需求不是随机的。今天有大量的劳动力短缺和崩溃的人口状况，所以几乎每个部门都有需求，这就是我们选择那个市场来真正增长的原因。但我认为你可以拿更冷门的，比如什么时候所有的采石场，你知道，就像你移动水泥，你移动泥土，不是采石场，采石场，岩石石头。

And so it's like pure dollars and cents, and so I think, as the provider of self-driving trucks, if you're doing the whole thing like some of the companies are, which we're not, you have to show every mile, I'm going to save you this many dollars, and it's like for sure, for sure, for sure, because the buyer's unsophisticated and they're just like, well, I already got a staff that can drive, and they're just not inclined. Now where we're playing in Japan, it's not random that we're doing trucking demand. There's a massive labor shortage today and an imploding demographic situation, and so there's a demand from almost every sector, and that's why we've picked that market to really grow. But I think you can take even more obscure, like when will all quarries, you know, literally like where you're moving cement, you're moving dirt, not quarries, quarries, rock stone.

岩石石头水泥。那些什么时候？

Rock stone cement. When are those?

我可以告诉你，拥有这些东西和经营这些东西的人今天就想要它。

I can tell you the people who own those things and run those things want it today.

对吧？

Right?

所以实际上，我们生产这些东西的速度不够快。

So it's literally, we can't make the stuff fast enough.

对吧？

Right?

但人们不谈论的宏观观点，我认为所有事情都会发生，但那很快发生，而且相当快。在立法和政治经济中的宏观观点是 AI。你真的看到在数字 AI 中有很大的抵制，因为会计师们说，我不知道这会对我的工作产生影响，而风投们，我相信你所有的同事都非常害怕，但在我们的领域里，他们在争论是否需要我们。

The macro point though that people don't talk about, I think all the stuff's going to happen, but that happened pretty soon and happened fairly soon. The macro point in legislation and in the political economy of this conversation is AI. You really see this big pushback in digital AI because accountants are like, I don't know this is going to happen to my job, and VCs, I'm sure all of your associates are very scared, but in our universe they're debating whether they need us.

是的。是的。是的。在我们的领域里正好相反。就像，我会遇到这些运营商，他们说，我们会给你一切。如果你能做到这个，我们会给你一切。所以接下来就靠我们积极地去实现。

Yeah. Yeah. Yeah. In our universe it's the other way around. And it's like, I'll meet these operators and they're like, we'll give you everything. If you can do this, we'll give you everything. So then it's just up to us to get there aggressively.

嗯，你知道长期以来对卡车运输的恐惧，出于某种原因触发了媒体对末日级别失业的想象。比如会不会有……

Well, you know the fear for a long time has been for trucking, for some reason triggers the press's imagination on apocalyptic levels of job loss. Like will there...

但那太错了。继续说。

But that's so wrong. Go ahead.

卡车司机不够，你猜怎么着？没人想当该死的卡车司机。

There's not enough truck drivers, and guess what? Nobody wants to freaking be a truck driver.

为什么？解释一下。

Why is that? Explain that.

因为这是一份糟糕的工作。就像你……

Because it's a terrible job. It's like you're...

顺便说一句，我在一个以卡车停靠站为主要特色的小镇长大。所以我知道答案，但为什么卡车驾驶现在……

By the way, I grew up in a town where the main feature was a truck stop. So I know the answer, but why is truck driving now...

就像你在问我，你知道吗，这就像跟我孩子说话，他说，为什么我不能把手放在炉子上？就像因为会烫伤你的手。他说，但为什么？就像问了第三个为什么之后，就像，来吧伙计，我们换个难的方式。

It's like you're asking me, you know what, this is like talking to my kid who's like, why can't I put my hand on the stove? It's like because it's going to burn your hand. It's like, but why? It's like after the third why, it's like, come on buddy, let's do this the hard way.

是的，来吧。那什么是难的？

Yeah, let's. So what's hard?

我开玩笑的。只是想确保每个人都知道我第一次没有那样做。

I'm kidding. Just wanted to make sure everybody knows I did not do that the first time.

是的。是的。

Yes. Yes.

那么难在哪里？为什么当卡车司机是一份困难的工作，或者为什么孩子们长大后不想做这个？

So what's hard? Why is being a truck driver a difficult job, or why would kids not want to do it when they grow up?

让我用一个非常清晰的类比。人们说“没人想工作了”，然后指着麦当劳说那里有那么多职位空缺。不，实际上，那些以前在麦当劳工作的人现在去 DoorDash 和 Uber 了，因为对他们来说更好。他们可以自己决定上下班时间，没有老板，不用站着，还能在订单间隙刷手机。这就是原因。这不是随机的。市场是高效的。所以在卡车司机的例子里，为什么有人不想连续 4 到 8 天离开家人做长途运输？更尖锐的例子是澳大利亚：为什么人们不想真的买张机票去矿场工作，或者去海上石油钻井平台？那些工作存在。如果你想要一份六位数薪水的工作，它们就在那里。即使薪酬如此丰厚，还是不够，因为人们会想：“你知道吗，我喜欢和家人在一起，我愿意接受收入上的一些减少。”

So let me use a parallel analogy which is very clear. People say, 'Nobody wants to work anymore,' and they point to McDonald's having all these job openings. No, actually, those people who used to work at McDonald's now DoorDash and Uber because it's better for them. They can start and end their hours when they want, there's no boss, they don't have to stand on their feet, and they can surf their phone between orders. That's the reason. It's not random. The market is efficient. So in the truck driving example, why does somebody not want to be away from their family for 4 to 8 days in a row doing long-haul trucking? A sharper example is in Australia: why don't people want to literally buy a plane ticket to go to a mine and work, or go to offshore oil rigs? Those jobs exist. If you want a job that pays six figures, they exist. Even with such lucrative pay packages, it's not enough because people are like, 'You know what, I like being around my family, and I'm willing to take an incremental decrease in cost and how much money I make.'

而且，我觉得今天比以往任何时候都更，像背痛、暴露在阳光下、癌症这些，人们现在很在意。

And also, I think today more than ever, things like back pain and being exposed to the sun and cancer—people care about that now.

事情是这样的。你告诉我我理解得对不对，但我相信商业长途司机的预期寿命比同龄人少 10 年。我认为这是几个因素共同作用的结果。一个是营养和睡眠的结合。很难吃得好、锻炼身体。

This is the thing. Tell me if I have this right, but I believe commercial long-haul drivers have a life expectancy 10 years less than their peers. And I think it's a consequence of several things. One is some combination of nutrition and sleep. It's very difficult to eat well and exercise.

如果你是长途卡车司机，你的睡眠评分是多少？我猜是八？

What's your sleep score if you're a long-haul trucker? Let me guess, eight?

没错。所以肥胖、心脏病、高血压等等都非常高。第一和第二，我觉得振动对身体影响很大。第三是你提到的癌症，但我认为卡车司机左臂的黑色素瘤发病率要高得多。

Exactly. And so obesity, heart disease, hypertension, and so forth are all very high. One and two, I think the vibration is very difficult on the body. And then the third is you mentioned cancer, but I think truck drivers have a much higher rate of melanoma on their left arm.

没错。有照片显示一个开了 30 年车的卡车司机，半边脸和另半边脸不一样，因为暴露在阳光下。

Exactly. There are photos of a truck driver who's been driving for 30 years—one half of their face is different from the other half because they're exposed to the sun.

一个更有趣甚至更惊人的数据：采矿业占全球劳动力的 1%，却占工伤死亡的 8%。你觉得人们听到这样的数据还会争先恐后地去矿场工作吗？大多数大型矿场定期都会有人死亡——一年一次、两次、三次。如果你去过矿场，你会看到一切都围绕安全展开，因为一旦你经历过同事死亡，你就会想：“我在这里干什么？”

A more interesting or even more stark stat: mining is 1% of the labor pool globally, but 8% of work-related fatalities. Do you think people are rushing to work in mines when they hear stats like this? Most major mines have a fatality regularly—once, twice, three times a year. And if you ever visit a mine, you'll see that everything is based around safety because once you experience one of your co-workers dying, you're like, 'What am I doing here?'

是啊。

Yeah.

就像还有其他工作我可以做。所以我理解你是在向观众列举：“但这些不是好工作。”

Like there are other jobs I can take. So I understand you're trying to enumerate for the audience, 'But these are not good jobs.'

是啊。最好的证据是这不是一个采矿播客。这不是一个关于长途卡车运输有多棒的播客吗？它们就是没有吸引力的工作。

Yeah. And the best evidence is this is not a mining podcast. Isn't this a podcast about how long-haul trucking is so great? They're just not attractive jobs.

是啊。甚至卡车司机自己也不想让孩子成为卡车司机。这就是为什么他们希望孩子至少从事更安全的工作。

Yeah. And even truckers don't want their kids to become truckers. That's the reason why they want their kids to be in a safer line of work, at the very least.

但尽管如此，在自动驾驶的长途卡车里，还会有多久需要安全驾驶员，或者甚至只是有人在驾驶室里处理到达时的情况？

But notwithstanding all that, how long will there be safety drivers in long-haul trucks that are self-driving, or even just somebody in the cab to deal with what happens when they arrive?

我们现在知道有几家公司有移除驾驶员的目标。他们正在努力让驾驶员离开。不深入我们自己的细节，说实话，不会太久。我们说的是几年。

We know multiple companies that have driver goals right now. They are working to get drivers out right now. Without going into our own details, to be honest, it's not long. We're talking a few years.

我觉得在长期来看。

I think on the long end.

是的，在长期来看。问题是，软件技术只是问题的一部分。但另一部分是你需要的硬件冗余以及这些冗余所需的验证。在很多情况下，这实际上可能是一个漫长的过程。就像，他们正在将完全冗余的转向系统、完全冗余的制动系统投入生产——这些还没有进入高价值生产。一旦你进入高价值生产，质量就上去了，然后经过验证，现在你才能真正实现这些价格下降。

Yeah, on the long end. And the thing is, there's a software technology thing which is one part of the problem. But the other part is the redundancies you need in hardware and the validation necessary for those redundancies. In many cases, that can actually be a long pull. It's like, they're productionizing a fully redundant steering system, fully redundant braking system—that's not in high-value production yet. And once you get that in high-value production, now you've got the quality up, and then that's validated, and now you can actually do these near the price downs.

没错。

Exactly.

你看到一个有十亿个这种小送货机器人到处跑的世界吗？

Do you see a world where there are a billion of these little delivery robots running around?

是的，我想是这样。我们这次宣布的产品叫 Dana。你可以把 Applied Intuition 做的一切简化为两个部分。我们主要讨论了装在机器上的模型——我们称之为机载软件或机载 AI。然后是离车 AI，也就是设计和开发这些系统的工具。真正装在机器上的模型——我们的愿景，送货机器人就是一个很好的例子——就像高中生或初中生能做 iPhone 应用一样，他们也应该能做自主系统。为什么九年级学生不能在家里做一个送货机器人？他们没有实际环境来首先开发场景。他们会定义需求：“嘿，我想让这个机器人在我高中校园里绕着这四栋楼走。”然后，一旦你定义了需求，场景就会生成。所有场景都可以用，比如说，高中的卫星图像来生成。然后你需要训练机器人，所以你需要一些数据。你从哪里得到这些数据？也许有足够的公开数据可以训练一个相当基础的机器人。现在你从网上得到了数据，也许 YouTube 视频和其他几个地方。突然机器人表现不好，所以你需要把它部署到实际机器上。你把它部署到机器上，机器人撞到了墙。发生了什么？循环闭合了。那个用于设计和开发的平台就是我们正在推出的。它叫 Dana。

Yeah, I think so. The product we're announcing around this time is called Dana. You can simplify everything that Applied Intuition does into two buckets. We've been talking mostly about the models that go on the machines—we call that onboard software or onboard AI. Then there's offboard AI, which is the tools to design and develop these same systems. The models that actually go on the machines—our vision for that, and the delivery robot is a great example—is like a high school kid or a middle schooler can make iPhone apps, so they should be able to make autonomous systems. Why can't a ninth grader make a delivery robot in their home? Well, they don't have the actual environment where they would first develop the scenarios. They would define the requirements: 'Hey, I want this robot to go on my high school campus around these four buildings.' Then, once you define the requirements, you have the scenarios get made. Where are all the scenarios that can be made by using, let's say, a satellite image of the high school? Then you have to train the robot, so you need some data. Where do you get that data? There's maybe enough publicly available data to train a fairly rudimentary robot. Now you've got that data from online, maybe YouTube videos and a couple other places. Suddenly the robot's not doing well, so you need to deploy it onto the actual machine. You deploy it onto the machine, and the robot runs into the wall. What happened there? The loop closes. That platform for designing and developing is what we're launching. It's called Dana.

Applied Intuition 总部所在的那条街是哪条？

Which is the street that Applied Intuition is headquartered on?

嗯，这源于我们的工具背景。如果你看看数字 AI 世界中工具的变化，看看 Claude 对我们所有人的影响，从 Mixpanel 到 GitLab、GitHub，所有这些现在都变成了非常不同的、几乎可以说是 IDE 的东西。我们认为物理世界也会发生同样的事情。所以这就是我们正在构建的，也是我们正在推出的，而且我们已经在内部使用它来开发我们的自动驾驶系统，你知道，我们正在所有这些不同垂直领域处理地球上最复杂、规模最大的系统。所以我们相当确信它确实非常有用。我们看到了巨大的生产力提升。但我们也认为其他公司会用它来构建自己的系统，因为这关系到构建智能机器这一使命。

Uh, and this comes from our tooling background. And if you look at how tooling has changed in the digital AI world, if you look at what Claude did to all of us, remember from Mixpanel to GitLab, GitHub, all these now everything has moved into a very different, almost IDE, frankly speaking. We think the same thing is going to happen in the physical world. So that's what we're building, that's what we're launching, and we already use it in-house to develop our autonomy system, which is, you know, we're working on the most scaled, complex systems on the planet in all these different verticals. So we're pretty confident that it's actually quite useful. We've seen massive productivity gains. But also, we think other companies will use this to build their own systems because it gets to that mission of building intelligent machines.

从根本上说，Dana 是我们面向物理 AI 的智能体平台，也是我们过去近十年构建和开发的一切。Dana 中可用的每一个工具、每一项技术，而且通过智能体界面使用起来非常容易。所以以前可能需要几天或几周才能完成的工作流程，现在很多情况下几分钟就能完成，对吧？

Fundamentally, Dana is our agentic platform for physical AI, and everything that we've built and developed over the past nearly a decade. Every tool, every technique that's available in Dana, and it's actually very easy to use with the agentic interface. So workflows that used to maybe take days or weeks to run, you can now run those in minutes in many cases, right?

而这只是降低了构建这些系统的门槛，也降低了开发自主系统的门槛。

And this just lowers the barrier to entry to building these systems, and just lowering the bar of what it means to develop an autonomous system.

自主性在软件领域实际上仍然相当特殊。这并非因为我们讨论过的那些原因，而我们非常激进地降低了这一门槛。这有点像那句老话：如何在软件中做出伟大的产品？要么提高安全性、便利性，要么降低成本，而我们想用 Dana 尝试同时做到这三点。

Autonomy is still actually quite in the scope of software, it's quite exotic. It's not because of the things that we've talked about, and we've just brought that down very, very aggressively. And it's kind of like the old adage of how do you make a great product in software? It's like you either increase safety, convenience, or cost, and we want to try to do all three of those things with Dana.

而我们的希望正如你所说，孩子们可以开发自己的机器人，这延伸到人形机器人。所以我们不仅仅在谈论地面系统或那些，所以你有的人形机器人，你可以做无人机。

And our hope is just like you said, kids can develop robots for their own use, and that extends to humanoids. So we're not just talking about land-based systems or ones that are, so you have humanoids, you can do drones.

目前编写无人机软件并部署它相当冷门，几乎像业余爱好，我们想让它绝对，你知道，也许不是小儿科，而是青少年级别的游戏。

The fact that right now writing drone software and deploying it at the time is quite obscure and almost hobbyist, we want to just make that absolutely, you know, maybe not child's play but like teenager play.

所以这指向一个充满更多实验和创业精神的世界，比如农业机器人，基本上每个领域，建筑、国防。你突然会有更多的人发挥创造力，提出想法，制造会动的东西。

So this points to a world of just like a lot more experimentation and entrepreneurship, like agriculture bots and basically every domain, construction, defense. You just all of a sudden have a much larger number of people who are applying creativity and coming up with ideas and making things that move.

是的。

Yeah.

你有没有看到，就像 Claude 一样，让工程师更高效或让更多人进入工程领域是一回事，但当这些智能体真正运行时，你进入的领域，就像 iPhone 的例子，你无法想象 Instagram 在 iPhone 之前。就像想象 2005 年在笔记本电脑上，你说 10 年后会有这个应用，你可以放照片。他们说，嗯，手机没有摄像头。就像，是的，但它会是社交性的，搞什么？就像 Facebook，你看，这很难。所以我们认为通过降低这个门槛，你会得到更多更有创意的自主产品，对吧？

And have you seen, like with Claude, it's like it's one thing just to make the engineer more efficient or bring more people into engineering, but then when these agents really run, you're getting into, it's just like the iPhone example of you couldn't imagine Instagram before the iPhone. It's like imagine 2005 on laptops, you're like in 10 years there's going to be this app where you can put photos. They're like well the phones don't have cameras. Like, yeah, but it's going to be like social, like what the hell? Like, so like Facebook, it's like, see, it's just hard. And so we think by lowering that barrier, you're going to get way, way more creative autonomy products, right?

是的。

Yeah.

我肯定会决定是否包含这个。所以，我的孩子在 Factorio 里构建自主机器人。哦，不错。这是他的项目之一。所以，但他一直在滚动，因为工具包还没可用。所以他实际上在训练模型。他在游戏中收集数据。而且实际上有一整支他开发的机器人军队。

I will definitely decide whether to include this or not. So, my kid is building autonomous bots in Factorio. Oh, nice. Is one of his projects. And so, but he's had rolling because the toolkit's not available yet. So he's actually training models. He's gathering data in the game. And actually has like a whole army of bots that he's developed that go.

是的。所以，然后他妈妈问为什么你玩那个游戏玩那么多？他解释说当然这是一个纯粹的教育过程和体验，但这是那种事情。就像是的，没有理由让自主性成为这种晦涩、困难、炼金术般的技术。而且我认为这不仅对社会有巨大影响，还让人们理解这些系统不是魔法。就像如果我能在周末用 Dana 在家里为自己开发一个 Roomba，对吧？那它就不会突然那么可怕了。我认为这很重要。

Yeah. So like, and then his mother is like why are you playing that game so much? And he explains of course it's a purely educational process and experience, but it's the kind of thing. It's like yeah, there's no reason autonomy should be this obscure, difficult, alchemistic technology. And I think not only does that have a huge impact on society, it also allows people to understand that these systems are not magic. It's like if I can develop a Roomba for myself in my house on a weekend using Dana, right? Then it's not suddenly so scary. And I think that's important.

而且我们可以以各种我们甚至还没想象到的方式支持人们，因为是的。绝对。

And we can support people in all kinds of ways that we haven't even imagined yet, because yeah. Absolutely.

是的。完全正确。我的意思是，想想残障人士，你知道我们总是认为人形机器人有像叠衣服这样非常重要的任务，这似乎是。所以我们专注于重要的任务，但当你允许这些工具存在时。我的意思是，我们创办了一家工具公司。我非常强烈地认为工具就像区分先进文明和不太先进文明的东西。而我们公司的第一个标志是猴子的头，然后我们找了一个设计师，他说这太蠢了。

Yeah. Exactly. I mean you think about folks with disabilities, you know we always think about humanoids as like this very important task of folding laundry, which seems to be. So we focus on the important task, but when you allow these tools to exist. I mean, we started a tooling company. I feel so importantly that tools are like what separates advanced civilizations from less advanced civilizations. And our first mark for the company was a monkey's head, and then we got a designer who said what this is stupid.

我说我觉得挺好的。

I was like I thought it was pretty good.

所以你之前谈到，当移动端技术变得非常好时，出现了一波这样的公司，你知道 Uber、WhatsApp、Snap、Airbnb 等，它们接连涌现。那么现在物理 AI 的基础设施技术正在成熟，有哪些用例或公司，显然很难预测未来，但你对哪些最兴奋？比如我们可能会谈论这里的快速接连涌现的等价物是什么？

So you were talking earlier about how when the technology got so good in mobile, there was a wave of these companies, you know Uber, WhatsApp, Snap, Airbnb, etc., that emerged in quick succession. And so now that the technology is getting there for the infrastructure for physical AI, what are some use cases or companies that you can, obviously it's hard to predict the future, but where are you most excited for? Like what could we be talking about the equivalent here of in quick succession?

我的意思是，我认为，你知道，中期我们希望 Dana，如果不是短期，真正让人形机器人变得更加真实。

I mean I think, you know, midterm we want Dana, if not the short term, to really make humanoids way more real.

呃，我是说，有多少？家里有上千种核心任务，从人形机器人到这些公司，简直——我是说，如果你和这些公司的人聊，每件事都很难。每一步都难：收集数据难，清洗数据难，训练模型或部署模型也难。而我们的目标是让一个高中生就能造出人形机器人——这就是我们的路径，我们认为那里潜力巨大。但那是显而易见的部分。我认为真正非显而易见的东西，我们回头看时会发现有趣得多。

Uh, there's, I mean, how many? It's like a thousand core tasks in a home, from humanoids, and these companies, it's like such—I mean, if you talk to people who work in these companies, everything is difficult. Every step of the way is difficult: collecting data is difficult, cleaning that data is difficult, training those models or deploying the model is difficult. And the bar being—I want a high school kid to make a humanoid—so that's our path, and we think there could be a lot there. But that's like the obvious stuff. I think the true non-obvious stuff is going to be—we'll look back—will be way, way more interesting.

而且我们在数据方面整合了一些核心要素，对吧？我们让模仿学习真正落地变得容易得多，让强化学习与之结合也容易得多。我们有预训练模型，可以作为很多任务的基线。世界模型、先进仿真技术——所有这些结合在一起，然后你就受限于你的创造力，比如，我想做什么？如果你考虑任何物理 AI 任务，那就是理解世界并操控某物，我们现在可以用这个工具更容易地构建它。我想有时人们会问，作为一家工具公司——比如你拿自动驾驶卡车来说，我们部署自动驾驶卡车，很多自动驾驶卡车公司都用我们的工具。我想有时人们会问，哦，你看，和 Dana 合作，你会不会助长所有这些竞争对手？那很好。

And there's some core ingredients that we're bringing together in data, right? We're making it way easier to actually get imitation learning to work, way easier to make reinforcement learning work in combination with that. We have pre-trained models that can be used as a baseline for a lot of things. World models, advanced simulation tech—all of these things come together, and then you're sort of limited by your creativity, like, well, what do I want to do? And if you think about any kind of physical AI task, it's—you are understanding the world and you're manipulating something, and we can build that—that can be built now much more easily in this tool. And I think sometimes people ask, like, us being a tooling company—like, you take self-driving trucks, we deploy self-driving trucks, and many of the self-driving trucking companies use our tools. I think sometimes people ask, oh, look, you know, with Dana, are you going to enable all these competitors? That's great.

对。

Right.

这完全没问题。看看 Google 对 Web 应用做了什么，当时互联网已经很大了。Google 依然通过搜索、YouTube 和其他 Web 应用成功了，其他人学习并使用开源产品，然后是闭源产品，最终是风投支持的产品。我们认为这里也可能发生同样的事情。

That's absolutely completely fine. If you look at Google and what Google did to web applications, there was a massive internet. Google still succeeded through, you know, search and YouTube and other web apps, and other folks learned and used open source products, and then ultimately closed source products, and ultimately venture-backed products. And we think the same thing could happen here.

我之前在一家机器人初创公司待过，你们很熟悉那家。他们当时在训练——经历一个训练过程，训练他们的机械臂做一个我觉得特别有吸引力的杀手级应用，那就是捡狗屎。

I was at a robotics startup a while back that you guys know well. And they were training—they were going through a training process, training one of their arms to do a particularly killer app that I thought was very appealing, which was picking up dog poop.

真的，你知道，一遍又一遍地训练。区别在于——

Literally, you know, training over and over again. The difference was for the—

所以，你知道，我不知道为什么不，对吧？为什么不让小机器人在你遛狗时跟着你，捡起狗屎？是啊。

And so, you know, I don't know why not, right? Why not have the little robot follow you around when you walk the dog, pick up the poop? Yeah.

而且我想，像你——我认识一个人，他造了——我忘了是谁。有人造了一个小草坪机器人，会到处捡单片叶子。

And I think, like you—I know somebody who built—I forget who it was. Somebody who built a little lawn robot that would go around and pick up individual leaves.

嗯。

Yeah.

因为你有那个问题，对吧？你把院子耙干净了，完全干净了，然后两小时后又有 14 片叶子，你会想——

Because you got that problem, right? You rake your yard, it's completely clean, and then like two hours later there's like 14 leaves, and you're like—

嗯，嗯。

Yeah. Yeah.

派小机器人去捡叶子。

Send up the little bot to pick up the leaves.

就像如果开发成本为零，人们就会这么做。我是说，你们记得早期的 iPhone 应用——热门的是啤酒应用或放屁应用。如果你想象那是在——

It's like if development costs are zero, then people will do that. I mean, you guys remember like the early iPhone apps—the hits were like the beer one or the fart app. If you imagine that in like—

嗯。

Yeah.

如果你想象那是在 98 年，你知道，用 Symbian 手机，不管那个操作系统是谁的，我想是爱立信还是谁，那是不可能的。你需要一个大约 50 人的团队才能为黑莓开发出啤酒应用。所以我认为类似的事情正在发生。我们，你知道，我们真的想成为其中的一部分，我们将促成这一点。如果它让——比如，我认为要让自动驾驶出租车变得超级容易还需要一段时间。

If you imagine that in '98 with, you know, the Symbian mobile, whatever the OS from, I think it was Ericsson or somebody, that'd be impossible. You need a team of like 50 people to develop like the beer thing for the BlackBerry. So I think there's a similar type of thing that's happening. We're, you know, we really want to be a part of that, and we're going to enable that. And if it makes—like, I think it'll still be a while before making a robo-taxi is super, super easy.

嗯。

Yeah.

但那会发生。但还有——我是说，可以部署在医疗保健领域的机器人数量——机器人种类几乎是——仅医疗保健就有居家护理——

But that'll happen. But there's—I mean, the number of bots that could be—the number of kinds of bots that could be deployed in healthcare is almost—healthcare alone is home care—

嗯，然后在建筑领域，你知道，所有体力劳动。

Um, and then in construction, you know, all the physical trades.

就像我们在 2007 年坐着说，让我们搞一个应用商店——会有什么类型的应用？我们会列出一个八个的清单，然后会有一个消息应用，然后会有一个相机应用。而现在你看应用商店，就像，你知道，有一个应用是为你去的酒店准备的，就像，你知道，用来点菜单上的食物。

It's like us sitting in 2007 and saying, let's have an app store—what types of apps? And we would come up with like a list of eight, and then there'll be a messaging one, and then there'll be a camera one. And it's like now you look at the app store, and it's like, you know, there's an app for like the hotel you go to, and it's like, you know, to order food off the menu.

对吧？

Right?

嗯。

Yeah.

嗯，有道理。

Yeah, makes sense.

嗯。我们之前聊过数字 AI 和物理 AI 的区别。我们暗示了大语言模型，但世界模型现在很流行。你谈谈它们与物理 AI 相关的现状，以及我们应该如何看待它们？

Yeah. We were talking, you know, earlier about the differences between digital AI and physical AI. We were sort of hinting at LLMs, but world models are, you know, in vogue right now. Why don't you talk about sort of the state of them as it relates to physical AI and how we should think about them?

首先，世界模型意味着大约 100 种不同的东西，我们最近有一个团队在 CVPR，我和他们开玩笑说，定义世界模型的方式有多少种。但当我们考虑世界模型时，我们通常是在仿真的背景下考虑它，对吧？某种实际上——

So, first off, world models means about 100 different things, and we had a team at CVPR recently, and I was joking with them about just how many different ways you can define what a world model is. But when we're thinking about a world model, we're typically thinking about it in the context of a simulation, right? Something that is effectively—

最初是一家仿真公司。嗯。

Started as a sim company. Yeah.

嗯。某种能够充分代表真实世界，并且在某种意义上具有反应性，你可以让一个自主智能体在这个世界中行动，而世界模型会适当地响应那个自主智能体。

Yeah. Something that is like sufficiently able to represent the real world and is reactive in a sense where you can actually have, let's say, an autonomous agent that's acting in this world, and the world model is behaving appropriately in response to that autonomous agent.

也许，Peter，我认为这里值得非常明确。我们往下看一层。模拟器中的确定性，那种仿真到现实的差距，基于物理的，你知道，渲染一直到这种生成的世界。

Maybe, Peter, I think it's worth being super explicit here. We just go one level lower. The determinism in simulators, kind of the sim-to-real gap, physics-based, you know, rendering all the way to like this generated world.

嗯。

Yeah.

我们适合在哪里，或者哪里——你知道——嗯。描述一下这个格局，我想也许。

Where do we fit on it, or where—you know—yeah. Describe the landscape, I think maybe.

嗯。所以这就像，比如说，广义的仿真，对吧？做仿真的方式有很多种。更经典的仿真方法非常基于物理，你可以用各种方式分解物理，在不同抽象层次上，你可以带传感器或不带传感器仿真。是仅仅身体仿真，还是我们实际上在仿真，例如，环境中的光线，或者几乎像 CGI 的做法那样。

Yeah. So this is like, let's say, simulation broadly, right? There's so many different ways of doing simulation. And so the more classical approaches of simulation are very physics-based, and you can decompose physics in all different ways and all different levels of abstraction, and you can simulate with sensors or without sensors. And is it just a body simulation, or are we actually simulating, for example, the light in the environment, or almost think about like the way CGI is done.

如果我们真的有技术美术师——我们确实有技术美术师——他们会创建资产放入模拟器中，模拟真实的道路标志，具备你在现实世界中看到的反射率和材质属性。但正如你们所知，好莱坞正在经历自身的根本性变革，现在你们已经生成了……同样的事情也正在我们的领域发生。

If we literally had technical artists—and we do have technical artists—they would create assets that would go into the simulator, mimicking real road signs with the reflectivity and material properties you'd see in the real world. But as you guys know, Hollywood is going through its own fundamental change, and now you've generated the same thing is happening in our universe as well.

是的。是的。

Yeah. Yeah.

所以那是在基于物理的模拟的最远端，而另一端是纯粹的神经模拟。但在这个频谱之间，有很多不同的事情可以做，每件都有其自身的价值。其中之一就是基于高斯的模拟，你实际上拥有一个真实世界的表示，具有 3D 表示。那个 3D 表示是一致的，意味着如果你有一个参考点，比如一个相机，并且那个相机在那个 3D 世界中移动，因为高斯实际上表示那个世界的 3D 几何，你会得到非常高质量的输出。那有很多价值，那是一种世界模型。但当你沿着频谱进一步进入神经模拟，你会进入生成视频流的领域。你可以想象一个神经网络实际上输出视频作为神经元的输出，并且那可以是反应式的，这给你一些非常有趣的特性。

So that's sort of on the far end of physics-based simulation, and the opposite end is purely neural simulation. But within that spectrum, there are many different things you can do that are each useful in their own right. One of those things is a Gaussian-based simulation, where you have effectively a representation of the real world that has a 3D representation. That 3D representation is consistent, meaning that if you have some reference point, say a camera, and that camera moves within that 3D world, because the Gaussian is actually representing the 3D geometry of that world, you'll get very high quality output. There's a lot of value in that, and that's one type of world model. But when you go further on that spectrum into neural simulation, you get into generating the video feeds. You can think of a neural network that's actually outputting a video as what's coming out of the neurons, and that can be reactive, which gives you some very interesting properties.

反应式是指自我在环境中做了某事，其他智能体对自我做出响应。

Reactive is in the ego does something in the environment and the other agents respond to the ego.

完全正确。然而，在这种反应性中，你并不能保证它是准确的。现在的问题是如何将这个模拟、这个世界模型与真实世界以及真实世界实际反应的方式对齐。如果你在真实世界和世界模型之间实现了完美对齐，我想你基本上就解决了宇宙的问题，对吧？那是一个极其困难的问题。但随着我们在这方面取得进展，训练物理 AI 模型会变得容易得多，因为你可以在模拟中做更多的事情。但最困难的部分是我们总是在谈论性能。实验室很容易，因为他们可以制造数万亿参数的模型，那些模型可以非常慢，那没问题。但我们在物理 AI 中没有那种奢侈。我们处理的是实时，就像实际的时钟实时，所以我们在必须做某事之前只有那么多毫秒。那些性能约束在很多方面限制了问题。所以我们可以有非常大的模型，而且我们确实在离车环境中使用非常大的模型，但一旦你上车，所有这些约束都是非常真实的。现在我们需要训练一个更小的模型，它必须满足这些安全约束、这些确定性约束。这就是物理 AI 的困难之处。这也是护城河。它使我们的工具和我们的能力有价值，因为在一个物理系统中满足所有这些约束确实非常困难。

Exactly. However, you're not guaranteed in that reactivity that it's accurate. And now it's a question of how can I align this simulation, this world model, with the real world and the way the real world would actually react. If you have perfect alignment between the real world and the world model, I think you've just solved the universe roughly, right? That's an impossibly difficult problem. But as we make progress towards that, it makes training physical AI models much easier because you can do more of that in simulation. But the hardest part is we're always talking about performance. The labs have it easy because they can make models that are trillions of parameters, and those models can be super slow, and that's fine. But we don't have that luxury in physical AI. We deal in real time, like the actual clock real time, so we have so many milliseconds before we have to do something. Those performance constraints actually constrain the problem in a lot of ways. So we can have very large models, and we do have very large models used in the offboard environment, but once you go onboard, all of those constraints are very real. Now we need to train a much smaller model that has these safety constraints, these determinism constraints. And that's the hard part about physical AI. It's also the moat. It's what makes our tooling and our competencies valuable, because it's just really hard to meet all of these constraints in a physical system.

我们什么时候会先得到：一个完美模拟的真实世界环境用于训练自主设备，还是《侠盗猎车手 6》？

When will we get first: a perfectly simulated real world environment for training autonomous devices, or Grand Theft Auto 6?

你知道，只要他们不断推出精彩的预告片，我是说，我看着，我觉得我不花一分钱就能得到娱乐，还因为……重新认识了汤姆·佩蒂。

You know, as long as they keep putting out great trailers, I mean, I watch, I feel like I'm getting entertained without paying a dollar, and reintroduced to Tom Petty because of...

你能给我们一些时间线吗？

Will you give us some timelines?

这个先放一放。

So this for a second.

是的。说吧。说吧。

Yeah. Go for it. Go for it.

嗯，不，你看，我的意思是，《侠盗猎车手》最大的创新是开放世界、开放世界沙盒游戏。所以它是一个模拟城市，至少在理论上是这样。

Well, no, look, I mean, the whole thing with Grand Theft Auto is the big innovation was open world, open world sandbox gaming. So it's a simulated city, at least in theory.

在那个频谱上。我是说，我们从这个频谱上从视频游戏世界雇了很多人。这绝对是真实的。

On that spectrum. I mean, we hire so many people out of the video game world on that spectrum. It's absolutely real.

嗯，跟我们说说吧。是的。频谱是什么？所以这里，这是猜测，但我认为《侠盗猎车手 6》可能是在传统计算机图形工具的传统时代仍然真正开发的最后一款主要的现实世界视频游戏。

Well, tell us about that. Yeah. What's the spectrum? So here, this is speculation, but I think Grand Theft Auto 6 will be perhaps the last major real-world video game that's still really developed in the legacy era of traditional computer graphics tooling.

技术美术师，是的。

Technical artists and yeah.

就像我认为《侠盗猎车手 7》更有可能是一个基于世界模型的视频游戏，你可以想象随着 AI 技术的发展，你有这个视频游戏世界模型的概念。有某种基线数据存储，以某种方式表示真实世界，然后你有一个翻译层，实际上将那个数据存储转化为你可以看到并在其中奔跑的东西。

Like I think that Grand Theft Auto 7 will much more likely be a world model based video game, where you could imagine as AI tech evolves here, you have this concept of this video game world model. There's some sort of baseline data store that represents the real world in some way, and then you have a translation layer that's actually turning that data store into something that you can see and run around in.

因此，游戏可能是真实世界，对吧？正如你所说，你可以在游戏中完全重现真实世界。这在飞行模拟器中已经发生了，不是吗？最新的飞行模拟器不是真的精确渲染了整个星球吗，至少从空中看？是这样吗？

The game as a consequence could be the real world, right? As you said, you could have a complete recreation of the real world in the game. This has kind of happened with flight simulators, hasn't it? Isn't the most recent flight simulators literally the entire planet rendered accurately, at least from the air? Is that right?

是的。是的。我的意思是，你真的是，那是我们的主要收入来源。当我们创业时，我们从微软飞行模拟器组织雇了很多人。

Yeah. Yeah. And I mean, you're really, that's where our bread and butter is. When we started the business, we hired so many people out of the Microsoft Flight Simulator organization.

但当你飞过，你知道，不管怎样，纽约，或者现在在飞行模拟器中飞越杜斯时，它是真实的城市，对吧？

But when you fly over, you know, whatever, New York, or when you fly over to Duth in the flight simulator now, it is the real city, right?

完全正确。但他们在那里玩了一些把戏，其中很多是保真度。真实世界，你放大得越多，它保持一定的保真度。你在那里玩的把戏是你基本上非常激进地下采样，然后当你靠近时，它变得更高保真。但真实世界不是那样的。如果你试图以这种保真度重建世界，那将消耗宇宙的所有能量。这相当复杂。顺便说一句，这可能是反对我们生活在模拟中的最好论据。

Exactly. But there are some tricks they play there, and a lot of that is fidelity. The real world, the more you zoom in, it stays a certain level of fidelity. The tricks you play there is you basically are downsampling very aggressively, and then as you get closer, it becomes more high fidelity. But the real world isn't like that. If you were to try to rebuild the world with this level of fidelity, it would take all the energy of the universe. It's quite complex. And that's probably, by the way, the best argument against us living in a simulation.

但当然，你会说，嗯，我们所在的模拟器不遵循我们当时所处的物理定律。我们怎么知道我们所在的模拟器渲染了我们看不到的所有东西？

But of course, you would say, well, the simulator we're in doesn't follow the laws of physics that we're at that point. How do we know that the simulator we're in is rendering all the stuff that we can't see?

是的。是的。是的。

Yeah. Yeah. Yeah.

据我所知，这个房间外发生的一切甚至都不存在。

As far as I know, everything happening outside this room doesn't even exist.

是的。我的意思是，你知道，佛教相信这是一个不同类型的播客。

Yeah. I mean, you know, Buddhism believes this is a different type of podcast.

就像你睁开眼睛时，世界被渲染出来，然后你闭上眼睛，世界就消失了，这简直是宗教性的。

It's like you know when you open your eyes the world is rendered and then you close your eyes the world that's literally religious.

我不明白为什么我不在场时，它还需要继续渲染。

I don't see why it's necessary for it to keep rendering if I'm not there.

从第一性原理出发的佛教。

Buddhism from first principles.

对，没错。你就该这么叫。这会吸引很多点击。

Yeah. Exactly. That's what you should call this. That'll get a lot of clicks.

好吧，就继续聊时间线这个话题。你知道，我们之前给过自动驾驶汽车的时间线。如果可能的话，你想给我们什么时间线，关于其他值得关注的有趣事情，比如什么时候能实现叠衣服，或者因为人形机器人而出现的其他事情。

Well, just go on the timeline topic. You know, we gave us timelines on self-driving cars. What timelines do you want to give us if any on sort of other interesting things that were worth tracking like perhaps when we'll get laundry folded or other things that emerge because of humanoid.

我还想稍微谈谈世界模型，我们如何看待世界模型的发展，因为我认为这对我们所做的工作至关重要。

And I think also maybe just touching a little bit on world models where we see world models going because I think it's fundamental to what the work we do.

确实。

True.

是的。那么，回答第一个问题，叠衣服，坦白说，离解决并不遥远。而且有很多有趣的研究正在进行。

Yeah. So, to answer the first question, laundry folding, it's not terribly far from being solved, to be clear. And there is a lot of interesting research being done.

然后人类就可以欢欣鼓舞了。我想这出自《箴言》4 章 16 节。

And then humanity can rejoice. That's in Proverbs 4:16, I think.

嗯，这里我确实认为家政服务是物理 AI 的一个杀手级应用场景，对吧？

Well, here I do think housekeeping is a killer use case for physical AI, right?

彼得也这么认为，他在公司里总是谈论这个。一个是家政服务，另一个是娱乐。彼得看好类人机器人娱乐。

Peter thinks too, he always talks about in the company. One is housekeeping and it's entertainment. Peter's long on humanoid entertainment.

是什么样的娱乐？我 100% 同意。我认为娱乐是机器人技术的杀手级应用。我觉得没人知道。

Is it like what kind of entertainment? I would 100% agree with that. I think entertainment is robotics killer app. I don't think anybody knows.

你什么意思？

What do you mean?

我是说，有些中西部白人真的很喜欢……

I mean like some of these Midwest white guys are really into...

我只是说出我的想法。

I'm just saying what I think.

我只想知道什么时候能去西部世界。我就想要这个。

I just want to know when I go Westworld. That's all I want.

不，我其实有一个好玩的……我有一只小小的中国机器狗，就像真的……它只是到处闲逛，然后做……

No, I actually have an entertaining... I have a little tiny Chinese robot dog that's just like literally it's just like a little... it just roams around and it just like does...

你愿意花钱看机器人版的洛杉矶马戏团吗？是的。我想看功夫空中飞人。

Would you pay to see Circus LA with robots? Yes. I want to see kung fu trapeze swinging.

这话听起来像是编译器专家说的。

Spoken by like a compiler's guy here.

我知道。但是……

I know. But...

我要西部世界。我要西部世界。

I want Westworld. I want Westworld.

我的意思是，有趣的是，我刚才说，底特律郊区的人其实会接受这个。我敢打赌斯特林海茨的人真的会花钱看这个。这确实是真的。

I mean the funny thing is I was just saying like well people in like the suburbs of Detroit actually that passes the test. I bet you people in Sterling Heights would actually pay to see that. It's actually true.

我收回我的话。

I stand corrected.

但回到叠衣服的话题。如果去掉时间限制，其实离实现折叠不远了。所以玩的把戏是，如果你看最新的研究视频，他们会说以 8 倍速播放之类的，对吧？那是为了让你看得下去。

But back on laundry folding for a moment. It's actually not far from being folded if you remove the time constraint. And so the trick that's played, and if you look at the latest research videos, they'll say like play it at 8x real or whatever, right? And that's for you to make it watchable.

所以问题是，什么时候才能真正达到人类水平的性能？那还更远。

So the question is when can you actually reach human parity of performance? That's further off.

当你把模型从硬件中解耦出来。硬件现在就能做到。那曾经是个限制。所以现在硬件非常快、非常精确，这实际上是……

When you decouple models from just the hardware. The hardware can do it now. That used to be a constraint. So the hardware is very fast and accurate now, which was actually...

还有过热问题仍在处理中，但离解决不远了。这些是可以解决的。

There's still overheating issues that are still being dealt with, but it's not terribly far off. These are solvable.

我的意思是，这离我当年做机械工程师时那种幻想很远，就像没什么能……

I mean it's far off from like you know when I was a mechie that was like fantasy, like there's nothing can...

哪部电影对未来机器人的描绘最现实？

What's the movie that has the most realistic future vision of robots?

天哪，《机器管家》。

Oh man, Bicentennial Man.

还行吗？

Is it okay?

是的，很现实。为什么是那部？我其实没看过。

Yeah, it's realistic. Why that one? I actually haven't seen it.

我喜欢那个场景，我想是《我，机器人》里威尔·史密斯跳进车里，他的同伙在车里，他把车切换到手动模式。她说：“你打算自己开这东西？”就像，这就是我们……这就是应用直觉的目标，你知道。

I like that scene, I think it's I, Robot when Will Smith jumps in the car and his accomplice is in the car and he puts the car in manual. She's like, "What are you going to drive this thing yourself?" Like, that's what we're... That's applied to intuition's goal, you know.

顺便说一句，我很久没看这部电影了。可能自从上映以来就没看过。所以我的记忆可能有点不准。

Well, by the way, I haven't seen this movie in a long time. Probably since it came out. So my recollection is probably a bit incorrect.

别担心，互联网会纠正你的。

Don't worry, the internet will correct you.

但我认为《机器管家》里有完全自动驾驶的汽车。而且还有家政机器人，好吗？由罗宾·威廉姆斯扮演，它就像友好的机器人，会打扫卫生，还会照看你的孩子之类的。而且这似乎发生在不太遥远的未来。

But I think Bicentennial Man has fully self-driving cars. And it also has the housekeeping robot, okay? Which is played by Robin Williams, and it's sort of like the friendly robot that will clean up and also babysit your kids and stuff like that. And it seems like it's in the not terribly distant future.

我有不同的答案。你们看过山姆·洛克威尔那部《月球》吗？

I got a different answer. You guys ever see that movie Sam Rockwell, Moon?

哦，看过。

Oh, yeah.

是的。这个设定。我不想……这是一部很棒的电影。别看预告片，直接看电影。前提是电影的标语是“离家 25 万英里。你找到自己是谁。”然后有一个人在一个能源采集基地工作，这个基地由应用直觉公司运营，由月球科技公司运营。我不想成为维兰德-汤谷，我不想成为《异形》系列里的那个，然后是《银翼杀手》里的泰瑞尔公司。不，我想成为《月球》系列里的月球科技公司。这个人在这里工作，基地基本上自己运行，他只是在出现错误信号时照看一下。

Yeah. The setup. I don't want to... It's a great movie. Don't watch the trailer. Just watch the movie. The premise is the tagline of the movie is "250,000 miles from home. You find who you are." And it's one guy who works in an energy harvesting base run by Applied Intuition, run by Lunar Technologies. I don't want to be Weyland-Yutani, I don't want to be that from the Alien franchise, and then Tyrell Corporation from Blade Runner. No, I want to be Lunar Technologies in the Moon franchise. This one guy works on this and the base basically runs by itself and he's just there to kind of mind it when things... some error signal.

是的。它之所以如此准确，是因为 AI 系统的最先进水平是这些系统只需要偶尔的接地。它们会突然做出疯狂的事情，然后你说不，不，别那样做。

Yeah. The reason why it's so accurate is because the state-of-the-art for AI systems is like these systems they just need the occasional grounding. They'll just go off and do something crazy and then you say no no stop doing that.

大语言模型也是这样。我是说，编码机器人就是这样，对吧？

LLMs are like that too. I mean that's what coding bots are like, right?

而且我认为它相当准确的另一个原因是，现在可能有点诡异，但凯文·史派西是 AI 笑脸，他只是在扮演人类来协助，但也会像“哦，你看起来很难过，山姆”那样，但真正运行基地的是他。而且希望，我的意思是我不能说我们想成为月球科技公司，因为我不知道他们在本质上是否是积极的力量……但我认为完全自主的大型能源农场将是未来。而且我认为每个人都会对这类事情感到恐惧。但就像，伙计们，这太棒了。这意味着能源成本大幅下降。这是一个不可思议的积极事物。

And the reason other reasons I think it's quite accurate, it's maybe uncanny now, but Kevin Spacey is the AI smiley face and he's just there to kind of play the human to assist, but to also like he's like "Oh, you seem like you're sad, Sam" and like that's the... but really it's the one running the base. And hopefully, I mean I shouldn't say we want to be Lunar Technologies because I don't know that they're quite a positive force in nature in that... But I think massive energy farm that's completely autonomous, that's going to be the future. And I think everyone reacts to things like that with fear. And it's like, guys, that's amazing. That means energy costs go way down. Like that's an incredible positive thing.

那将是未来。而且我认为每个人都会对这类事情感到恐惧。但就像，伙计们，这太棒了。这意味着能源成本大幅下降。这是一个不可思议的积极事物。我想……我刚刚在我的本科毕业典礼上做了演讲。

That's going to be the future. And I think everyone reacts to things like that with fear. And it's like, guys, that's amazing. That means energy costs go way down. Like that's an incredible positive thing. I think I think the... I just did this commencement speech at my undergrad.

你被喷了吗？

Did you get destroyed?

没有。你知道吗？我……

No. You know what? I...

不像埃里克·施密特。

Unlike Eric Schmidt.

是的。听着，听着，听着。我的……这是真实的故事。我妻子开始看。她说：“我觉得你在对我吼。”

Yeah. Listen, listen, listen. My... This is the true story. My wife started watching. She said, "I feel like you're yelling at me."

我看不下去。所以，我基本上——我不会点名哪些科技领袖只是回避这个问题，说“我不谈这个”。我会谈这些。部分原因是通用汽车学院（GMI）。没人会——这些人——我不想对我们从 MIT 和斯坦福招来的人妄加评判，但我觉得 GMI 的人有点不一样。他们很务实。如果你对企业的运作只有肤浅的看法，你就不会去 GMI 这样的地方。企业不过是一群人一起做项目。顺便说一句，政府里一起做项目的人、非营利组织里一起做项目的人，都会搞砸。所以说“AI 企业很糟糕”太简单了。你也不能说另一边，就是“一切都会很好”。所以你有你的角色要扮演。这基本上就是我的信息。事实就是这样。我觉得如果你觉得——如果自动驾驶卡车、自动驾驶汽车带来的明显富足，以及人们不会死——这太棒了——但你也得到更便宜能源的效率等等。如果所有这些都不能消除你的恐惧，作为一个人，你有责任真正去了解那项技术。你不能只说“好吧，我害怕它，我的反应是关掉它”。那不是——简单来说——而且我这么说不只是为了说我们在和中国竞争，但有一种困惑。那句困惑的话是“只手遮不住太阳”。

I can't watch this. So, I basically—I'm not going to say which tech leaders just avoid it by punting and saying I'm not going to talk about it. I talk about this stuff. And partly it's the General Motors Institute. No one's like—these are—I don't want to throw judgment on the people we recruit out of MIT and Stanford, but I say GMI people are a little different. They're pragmatic people. You don't go to a place like GMI if you believe a superficial view of what corporations do. Corporations are just people working on projects together. And by the way, people working on projects together in government, people working on projects together in nonprofits—they all screw up. So it's too simple to say AI corporations are terrible. You also can't say the other side, which is it'll all be great. So you have a role to play. That's basically what my message is. And that's the case. I think if you feel like—if the obvious abundance that comes from self-driving trucks, self-driving cars, and the fact that people don't die—which is amazing—but then you also get this efficiency of cheaper energy, etc. If all those things don't still satisfy your fear, you as a person, it's up to your responsibility to really learn about that technology. You can't just say, 'Well, I'm afraid of it and my reaction is shut it down.' That's not—simply it's—and I don't say this just to say that we're competing with the Chinese, but there's a confusion. The saying by confusion is no hand can block the sun.

嗯。

Mhm.

而太阳就是技术进步。如果我们作为一个社会不拥抱技术进步，我们就会被甩在后面。

And the sun is technological progress. And if we as a society don't embrace technological progress, we will be left behind.

别人会去做的。

Somebody else is going to do it.

别人会去做的。而且可能不是中国人。谁知道呢？也许是一家美国公司，或者另一个国家认识到“嘿，我的公民在受苦，我要用这项技术来消除他们的痛苦”。说实话，是因为我们生活在如此伟大的社会里，我们才能有这些——

Somebody else is going to do it. And it might not be the Chinese. Who knows? Maybe it's a US company or another country that is recognizing, 'Hey, my citizens are suffering and I'm going to use this technology to remove them.' It is honestly because we live in such a great society that we can have these—

我会说，像“还有人吃不上饭”这种愚蠢的对话。是啊。

I would say stupid conversations like there still are people who don't get food. Yeah.

如果有人和我辩论，他们会立刻打趣说：“好吧，食物很多。是资本主义制度不——”不，不，不。我们说得具体点。食物很多，但把食物送到那些人手里很难。

And someone will immediately quip if they were debating me, they would say, 'Well, there's plenty of food. It's the capitalist system that doesn't—' No, no, no. Let's be very specific. There's plenty of food, but getting that food to those people is difficult.

所以，那意味着我们应该让机器人更快地把食物送到他们那里。

So, that means we should let robots get that food to them faster.

事情就是这样。所以，我觉得我——而且我觉得我们作为技术专家，有时我们倾向于说“把这些人抛在后面”。我觉得你必须带上他们。你必须向他们解释。但我们也必须把人们当成年人对待，说“如果我解释了几次你还不懂，那你就是不懂”。所以有一个中间地带。不是每个人都是白痴，或者我们应该——技术会完美无缺。有一个中间地带。我们把对话进行到一定程度，然后我们继续前进，让社会变得更好，然后结果会证明一切。我的意思是，仍然有人惊人地相信共产主义是正确的答案。我只想说为什么——而且我是资本家。我不能承认这一点。但有 70 年的历史。那已经不是辩论了。我的意思是，我认为那可以是一场辩论。如果我们坐在 1965 年辩论，你会说“好吧，也许中央控制的系统更有效”。现在没有辩论了，伙计们。个人根据自己的利益做决策的系统实际上对社会更有利。所以这并不意味着一切都是完美的，你也不能把同样的道理外推到 AI 上。不意味着一切都会完美，但总的来说肯定会更好。这大致就是我的毕业演讲内容，只是没有酒精。他们——马克和那些人在嘘，所以他们剪掉了。埃里克在嘘。他扔东西，他们只是把它剪掉了。

That's just how it is. So, I think I'm—and I think we as technologists, sometimes we have an inclination just to say leave these people behind. I think you have to bring them along. You have to explain it to them. But we also have to treat folks like adults and say if you don't get it after I explain it a couple times, then you just don't get it. So there's a middle ground. It's not everyone's an idiot, or we should just—technology will just be perfect. There's a middle ground. Let's have that conversation to a point and then we just move forward and we make society better and then the results show it. I mean, there are people who still shockingly believe communism is the right answer. I just want to say why—and I'm a capitalist. I cannot admit that. But there's 70 years of history there. That's not even a debate anymore. I mean, I think it could be a debate. If we're sitting here in 1965 and having a debate, you say, 'Okay, maybe centrally controlled systems work better.' There's no debate anymore, folks. Systems where individuals make decisions on their own interest actually work better for society. And so that doesn't mean everything is perfect, and you can't extrapolate that same thing with AI. Doesn't mean everything is going to be perfect, but net net it's definitely going to be better. That's roughly what my commencement speech was without the booze. They—Mark and these guys were booing so they just cut it out. Eric was booing. He's throwing stuff and they just edited it out.

你之前提到了日本市场。你为什么不简单谈谈全球雄心，以及这些技术如何相互作用，我们在这里做什么？

You mentioned the Japan market earlier. Why don't you talk briefly about sort of the global ambitions and how these technologies interplay and what we're doing here?

所以我认为美国尤其仍然是最先进的——考虑到商业模式。第二，对于像 Applied Intuition 这样的公司，我们是一家非常全球化的公司。我们与所有人合作——除了我们在中国没有办公室——但真的和全球其他所有人合作。我们是一家横向公司。我们是技术提供商。我认为更多的硅谷公司可以借鉴我们的一点做法，那就是随着主权 AI 变得越来越现实，与当地经济非常协作地合作。我们必须建立考虑到这一点的业务。顺便说一句，我们不是第一个这样做的。如果你看美国的历史，读标准石油的历史，你会看到这就是公司的历史。你会在国际上工作。阿美石油公司（ARAMCO）不是一家随意的公司，对吧？你根据当时的真实地缘政治现实来建设。所以我认为我们处理得很好。我在日本住过，在德国住过，在迪拜住过。而且我出生在巴基斯坦，我认为这也影响了我们公司。彼得只在密歇根和这里住过，但他是德国人。所以我认为我们天生更——我们更考虑全球。而且我认为当我在谷歌和 YC 时，我总是惊讶于这些公司多么短视，总是只盯着圣何塞和旧金山之间 30 英里内的市场。实际上市场非常大。我认为物理 AI，它物理的本质，我认为我们必须是一家非常国际化的公司，而且我认为我们在国际化方面取得了很大成功。

So I think America particularly is still the most advanced in terms of—when you take account the business model. The second thing for a company like Applied Intuition, we're an extremely global company. We work with everybody—minus we don't have an office in China—but really everyone else on the globe. And we're a horizontal company. We're a technology provider. And I think more Silicon Valley companies can employ a little bit of what we do, which is work very collaboratively with the local economies as sovereign AI becomes more of a real thing. We have to build businesses that take that into account. By the way, we're not the first ones to do this. If you look at the history of America, you read the history of Standard Oil, you'll see that this was the history of companies. You'd work internationally. ARAMCO is not a random company, right? You build based on the real geopolitical realities of the time. And so I think we've navigated it quite well. I've lived in Japan, I lived in Germany, I lived in Dubai. So also being Pakistani by birth, I think that's also influenced our company. Peter's only lived in Michigan and here, but he is a German. So I think innately we're more—we think about the globe more. And I think when I was at both at Google and at YC, I was always surprised at how almost myopic the companies are, just always looking at the market that's just within the 30 miles between San Jose and San Francisco. It's like actually the market is really big. I think physical AI, the nature of it being physical, I think we have to be a very international company, and I think we've had a lot of success being very international.

是啊。酷。我觉得这是一个很好的结束点。

Yeah. Cool. I think it's a good place to wrap.

好的。Peter Casser，非常感谢你来到播客，祝贺你和 Dana 的大规模发布。

Okay. Peter Casser, thanks so much for coming on the podcast and congrats on big launch with Dana.

是的，谢谢邀请我们。

Yeah, thanks for having us.

太棒了。很高兴见到你。

Awesome. Great to see you.

很好。

Great.