马斯克在 X Takeover 谈星舰、AI、芯片、Optimus 与 Cybercab

Elon，欢迎来到 X Takeover，这是你连续第三年参加。非常感谢你抽出时间。今天在 Giga Texas 有超过 2,000 人，我们都在这里支持你和你们为人类所做的工作。所以，谢谢你。Elon，疯狂的是，四年前我们就在 Giga Texas 这里和你坐在一起，当时工厂还没开始运转，你描述了特斯拉在 2008 年差点活不下来。今天看着这个 2,000 人的社区，什么让你停下来，感慨事情已经走了这么远？

Elon, welcome to the X Takeover, your third year in a row. Thank you so much for your time. We have over 2,000 people here at Giga Texas today and we are all here to support you and the work you are doing for humanity. So, thank you. So, what's crazy, Elon, is four years ago, we sat down here in Giga, Texas with you when the factory wasn't even running, and you described how close Tesla was coming to not surviving in 2008. Looking out at this community today of 2,000 people, what makes you pause and appreciate how far things have come?

是的。在 2026 年的今天，很难想象特斯拉实际上多次濒临死亡。2008 年是最糟糕的。那是在 2008 年平安夜下午 6 点，我们完成了融资，那是可能的最后一个小时、最后一天。如果那轮融资没有完成，圣诞节后几天我们就发不出工资了。所以压力非常大。然后 2018 年也很糟糕，我们必须让 Model 3 达到每周 5,000 辆，否则就完了。那是一个非常疯狂的局面，我想很多人可能都知道这个故事，但我们必须想办法在 3 周内把 Model 3 的产量提高 50%。

Yeah. It's difficult to imagine here in 2026 that Tesla has actually come close to death multiple times. 2008 was the worst. That's where we closed the financing round at 6 p.m. on Christmas Eve, 2008, the last hour, the last day that was possible. We would have missed payroll a few days after Christmas if the round hadn't closed. So that was pretty stressful. And then 2018 was pretty bad too, where we had to get to 5,000 a week of the Model 3 or go back. That was a pretty wild situation, which I guess probably a lot of people know the story of, but we had to figure out some way to increase Model 3 production by 50% in 3 weeks.

这太疯狂了。

Which is insane.

绝对疯狂。

Absolutely insane.

是啊。那你们是怎么做到的？

Yeah. So how did you do it?

我给没听过的人再讲一遍这个故事。关键点是 Model 3 的总装。除了总装，我们几乎在其他所有环节都达到了目标速率。总装需要生产线，需要遮蔽，需要电力和设备等等。我们已有的两条生产线已经满负荷，然后我们需要第三条，于是我们就在停车场搭了帐篷。然后我们需要某种方式把车沿着生产线移动，就是移动生产线的情况，我们唯一有的是用来搬运小零件的废弃设备，不是搬整车的。它实际上没有足够动力移动整车，但停车场有一个轻微的坡度，大概三到四度。所以我们实际上把生产线从高处开始，然后那条本来只用于零件的微型生产线实际上能够移动车辆，因为它们是稍微下坡的。所以我们能够在 3 周内把产量提高 30%，就是那条线和铺设轨道。所以是 24/7 工作。做空者实际上是对的，意思是如果你看当时的情况，特斯拉在不同时期有疯狂的做空头寸。他们有极其灵通的内幕信息。他们认为，我们不可能解决生产瓶颈。在像传统象棋比赛那样的固定局面中，他们把我们将死了。但我们基本上发明了一个新棋子。

I'll recount the story for those who haven't heard it. The choice point was general assembly of the Model 3. We were able to achieve rate almost everywhere else except for general assembly. For general assembly you need a production line, you need shelter, you need power and equipment and all that. We maxed out the two production lines that we had, and then we needed a third one, and that was where we built the tent in the parking lot. And then we needed some way to move the cars down the line, the moving production line situation, and the only thing that we had was some discarded equipment that was used for moving small parts, not whole cars. It didn't actually have the power to move the whole cars, but the parking lot was on a slight incline. It's like a three or four degree incline. So we actually started the production line at the top, and then the tiny production line that was just meant for parts was actually able to move the cars because they were going slightly downhill. And so we were able to increase production by 30% in 3 weeks, like the line and putting the rails. So working 24/7. The short sellers were actually right, meaning if you looked at the situation, there was a crazy short position against Tesla at various times. And they had extremely good inside information. And they thought, well, there's no way we possibly solve the production choke point. And on a set piece battle like a conventional chess match situation, they had us in checkmate. But we just invented a new piece basically.

嘿，Elon，我们今天刚请你妈妈上台。你希望她不要告诉我们什么故事？

Hey, Elon, we just had your mom here on stage today. What story are you hoping she doesn't tell us?

我不知道。也许是我小时候拉裤子之类的。我可能干过。所以，我不知道。我小时候很难管住。我父母会说，告诉我我不该离开家之类的。当我大概四五岁的时候，我会不停地离开家，跑到街上。然后他们会把我锁在一个房间里，这样我就不能跑出去到街上。然后我们有一种旧锁，你可以把钥匙从门里戳过去，你知道，锁从两边都能打开。我能看到钥匙，我就戳钥匙直到它掉下来，然后我能拿一张纸把它舀起来，自己把它弄到另一边。我妈妈就说：‘你怎么出来的？’我说：‘我不告诉你。’但是的，那是个项目。

I don't know. Maybe I pooped my pants when I was like a toddler or something. I probably did. So, I don't know. I was pretty difficult to contain as a kid. My parents would say, you know, tell me I'm not supposed to leave the house or whatever. When I was like four or five years old, I would just constantly leave the house and go running down the street. And then they'd lock me in a room so I couldn't run away and run down the street. And then we had one of those old locks where you could poke the key through the door, you know, the locks accessible from both sides. And I could see the key and I poked the key until it fell down and then I was able to get a piece of paper to scoop it up and bring it to the other side myself. And my mom was like, 'How did you get out?' I was like, 'I'm not telling you.' But yeah, it was a project.

在我们进入更具体的事情之前，你希望更多人问你什么？

So before we jump into more specific things, what's something that you wish more people would ask you?

嗯，说实话，我并不真的希望人们问我问题。抱歉。我希望人们问更多问题。所以，是的，我想听到以前没听过的问题是好的。否则，很多时候人们问的是同样的问题，我觉得我可以有答案的录音。我会放给你听：‘哦，那是第 27 个问题。来了。我已经回答过 100 遍了。’所以也许我应该有一个网页，字面意义上的常见问题解答。

Well, I don't really wish people would ask me questions to be honest. Sorry. I wish people would ask more questions. So yeah, I guess it's good to hear questions that you haven't heard before. Otherwise, a lot of times it's the same questions that people ask and I feel like I could have recordings of the answer. I'll play you: 'Oh, that's question 27. Here we go. I've given a 100 times before.' So maybe I should just have a web page with literally frequently asked questions.

就像问，就像直接去那个页面，然后你就得到了。你得到了答案。你知道，

It's like asking, it's like just go to the page and then you got it. You got the answers. You know,

就像无限自拍循环。自拍循环。

It's like the infinite selfie loop. A selfie loop.

是的，没错。

Yeah, exactly.

我对地狱的想象就是被困在无限自拍循环中。就像无限的自拍，他们会说，你知道，可能就像偶尔对一个人说好，然后就有了排队，很快就是没完没了的自拍，然后总是有手机出问题，或者相机没调好，或者效果不对，然后就像该死的，你知道，他们掉了手机，就像哦天哪。所以不能是无限自拍循环。这是一个很长的循环，你知道。

My idea of hell is being stuck in an infinite selfie loop. Like it's infinite kind of a selfie and they're like, you know, it's like probably like say yes to one person and every once in a while and then there's a cue and pretty soon there's never ending selfies and then there's always something goes wrong with the phone or they don't get the camera right or the effect is wrong and then it's like god damn it, you know, and they drop the phone and it's like oh man. So it just can't be like infinite selfie loop. This is a long loop, you know.

是的，我们刚才也请你妈妈和 Nikki 上台了。她们今天也经历了无限自拍循环。但我会把话筒交给 Kelvin。

Yeah, we had your mom and Nikki up for that, too. They had the infinite selfie loop today. But I'll pass it to Kelvin.

好的，让我们从 Starship 开始。恭喜第 14 次飞行。

All right, let's start with Starship. Congratulations on flight 14.

是的。

Yeah.

所以，第 14 次飞行在上升过程中失去了一台发动机，但仍然进入了轨道。在团队发出‘进入轨道’信息之前的那些分钟里，幕后发生了什么？

So, flight 14 lost an engine on the way up and still reached orbit. What was happening behind the scenes in those minutes before the team called the go for orbit message?

嗯，我们只是在试图弄清楚原因。我们试图理解发动机故障的原因是什么。是否存在一个共同原因，让我们预计会看到更多发动机故障。

Well, we're just trying to see why. We're trying to understand what was the reason for the engine failure. Was there say a common cause where we expect to see more engine failures.

最大的问题是，重新启动海平面 Raptor 发动机——也就是中心那三台喷嘴较小的——进行返航点火，会不会有潜在问题。因为我们不想让星舰卡在轨道上，然后随机再入，说不定落到什么东西上。这是我们最关心的：确保安全可行。发动机熄火的原因，我认为基本上是飞行计算机软件或发动机计算机软件的故障，而且其他发动机计算机上似乎都没有出现这个问题。所以我想，好吧，这大概可以接受。

The biggest question is whether there would be a potential problem with restarting the sea level Raptor engines—the three with the smaller nozzles in the center—for the boostback burn, because we don't want Starship to get stuck in orbit and then randomly reenter and maybe land on something. That's our main thing: trying to make sure things are safe and possible. The cause of the engine out, I think, was basically a failure in the flight computer software or engine computer software, which didn't appear to be present in any of the other engine computers. So I thought, okay, well, this is probably okay to go over.

那么 Elon，我们知道你有一句名言：从不可能到只是迟到。这次飞行是否感觉是星舰可能性的一个转折点？

So Elon, we know one of your famous things is from the impossible to merely late. Did this flight feel like a turning point in what's possible now for Starship?

嗯，我想这是一个不错的里程碑，因为我们之前本可以多次进入轨道。但之前没进轨道的原因是我们想对安全离轨有极大信心。所以这次飞行我们才进入轨道。但几乎所有之前的任务——实际上，字面上所有之前的任务——本都可以进入轨道。所以现在，一切看起来相当积极。隔热盾看起来很坚固。发动机也很坚固，除了那一次小故障，但那个故障似乎不是什么大问题，很容易补救。这次飞行中隔热盾极其坚固，而这是最难的事情之一。另一个真正困难的事情是让发动机可靠性达到极高。一切看起来都很好。即将到来的重大里程碑，希望在下一次飞行——我们不知道确切日期，因为我们在对火箭做一堆改进。每次飞行之间通常有数百、有时数千个小改动来改进。这次改动的数量不多，但有些需要不少工作，我们得进入贮箱内部做一些改动。所以我们不知道确切的发射日期，但大概 4 周左右。4 到 6 周差不多。如果它在感恩节前一点发射，这是我的猜测。而那一次如果成功，就是个大事件，因为我们要尝试捕获助推器，如果成功，那将是首次完整回收轨道级火箭。

Well, I guess it is a good milestone, because we could have gone to orbit many times before now. But the reason we didn't go to orbit before now was because we wanted to be extremely confident of being able to de-orbit safely. That's why we only went to orbit on this flight. But pretty much all of the prior missions—in fact, literally all the prior missions—could have gone to orbit. So at this point, everything's looking quite positive. The heat shield is looking robust. The engines are robust apart from that one glitch, but that glitch doesn't seem to be too much of an issue. It's pretty easy to remedy. The heat shield was extremely robust on this flight, and that's one of the hardest things. Another really hard thing is just getting to extremely high engine reliability. Everything looks good. The big milestone coming up, hopefully on the next flight—we don't know the exact date because we're making a bunch of improvements to the rocket. Between every flight there are often hundreds, sometimes thousands of small changes to improve things. In this case, the number of changes is not very many, but some of them take a fair bit of work and we've got to go inside the tank to make some of these changes. So we don't know the exact launch date, but probably about 4 weeks or so. 4 to 6 weeks is about right. If it launches a little bit before Thanksgiving, that's my guess. And that one, if it succeeds, is a big one because we're going to try to catch the booster, and if we do so, that'll be the first time there's been full recovery of an orbital class rocket.

听起来不错。你说四到六周。我不太习惯，我习惯的是“两周”那个说法。

Sounds good. You said four to six weeks. I'm not used to that. I'm used to the two weeks phrase.

天意不是两周。

Providence is not two weeks.

好吧。你是个引子。第一次捕获星舰后，最难搞定的部分是什么？那之后的下一个大目标是什么？

All right. You're an introduction. What's the hardest part left to get right once you catch the Starship for the first time? What is the next big goal after that?

嗯，下一个重大里程碑将是再次飞行助推器和飞船。所以即将到来的下一个里程碑将是首次所有轨道火箭部件都被回收。但之后的飞行，希望将是首次所有轨道火箭部件都再次飞行。然后还有一长串工程和生产工作，涉及发射系统等一切，使其快速可重复使用。所以我会说，这是快速可重复使用的可靠火箭。

Well, the next big milestone would be to refly both the booster and the ship. So the next upcoming milestone will be the first time all parts of an orbital rocket have been recovered. But then the flight after that, hopefully, will be the first time all parts of an orbital rocket are reflown. And then there's a long tail of engineering and production work on the launch system and everything to make it rapidly reusable. So I'd say it's kind of the rapidly reusable, reliable rockets.

火箭快速可重复使用的可靠火箭。

Rocket rapidly reusable reliable rockets.

天哪，我们该给你弄个徽章。

Man, we should have gotten you a patch for this one.

当你创立 SpaceX 时，你能想象它的使命扩展到太空中的 AI 和数据中心吗？你是怎么开始把火箭和 AI 看作属于同一个集团的？

And when you founded SpaceX, could you have imagined its mission expanding into AI and data centers in space? And how did you come to see rockets and AI belonging under one group?

嗯，因为我非常有远见，当然。嗯，我想如果你利用太空，你就能在太空中获取比在地球上多得多的太阳能。我能体会到的是太阳的绝对规模，以及太阳几乎就是一切。太阳占太阳系所有质量的 99.86%。然后木星占了剩余质量的大部分。然后气态巨行星差不多就那样。然后地球基本上就是一颗微小的尘埃球，就像太阳系质量中的一个舍入误差。所以我们极其微小。很难相信——一颗微小的尘埃球。所以当你向太空扩张，能够获取更多太阳能时，你必须问，你要用它做什么？嗯，你会用它来做大量的数字智能。我的意思是，你要么进行计算，要么塑造原子。而且数字思考很可能比塑造原子多得多，因为你在太阳系中能塑造的原子数量并不多，而且很多原子塑造本身将是为了建造数字超级智能，以及我所说的创造一个有感知的太阳。也许外面有很多有感知的太阳，那将使我们能够理解宇宙的本质，并将意识扩展到太阳系之外，最终像《星际迷航》那样——去你从未去过的地方。忘了确切的说法是什么，但探索宇宙，去你从未去过的地方。我认为我们会在外面发现外星文明，或者我们可能会发现存在已久的外星文明的废墟，也许他们有过很长的繁荣期，比如 1000 万年，然后文明终结了，因为没有什么能永远持续。所以我总是思考地球文明：如果你从最早的书写算起，那是 5500 年前，比如古苏美尔，有一种古老的统一文字。那只有 5500 年。根据物理学的推测，地球大约有 45 亿年历史。所以这意味着人类文明只存在了地球寿命的百万分之一。所以如果我们有一个持续 100 万年或 1000 万年的文明，我认为如果我们做到那样，那将是一个非常非常成功的文明，相比宇宙的明显年龄——138 亿年。

Well, because I'm very farsighted, of course. Well, I guess if you take advantage of space, you're able to harness a lot more of the sun's power in space than you can on Earth. One thing I can appreciate is the sheer magnitude of the sun and how the sun is pretty much everything. The sun is 99.86% of all the mass in the solar system. And then you've got Jupiter as most of the remaining mass. And then the gas giants are pretty much that. And then Earth is basically a tiny dust ball, like a rounding error in the mass of the solar system. So we're just extremely tiny. It's hard to believe—a tiny dust ball. So as you expand into space and you're able to harness more of the sun's power, you have to say, what are you going to use that for? Well, you're going to use it for a lot of digital intelligence. I mean, you're either going to do computation or you're going to shape atoms. And there's going to be probably a lot more digital thinking than there will be shaping atoms, because the amount of atoms you can shape in the solar system is not that much, and a lot of the atom shaping itself will be for the purposes of building digital superintelligence and what I call creating a sentient sun. And it may be that there are many sentient suns out there, and that would allow us to then understand the nature of the universe and extend consciousness beyond our solar system, ultimately to like Star Trek—to go to places you've never been before. Forget what exact phrase was, but explore the universe, go places you've never been before. And I think we will find alien civilizations out there, or we may find that there are the long-dead ruins of alien civilizations that have been around, or maybe they had a really good run like 10 million years and then the civilization came to an end, because nothing lasts forever. So I always think about Earth civilization: if you date Earth civilization from the first writing, that's 5,500 years ago, like in ancient Sumer with a sort of archaic uniform. That's only 5,500 years. Earth looks to be about 4.5 billion years old according to what physics suggests. So that means human civilization has only been around 1 millionth of Earth's existence. And so if we had a civilization that lasted a million years or played on 10 million years, I think that would be a very, very successful civilization if we did that, compared to the apparent age of the universe, which is 13.8 billion years.

所以，要到达宇宙年龄小数点后第三位，把它增加一百万年，你得从 13.8 后面某位数字，在小数点后第三位加一。现在你就老了一百万年，那将是一个比人类古老得多的文明。

So just to get to the third place past the decimal point in the age of the universe, to increment it by a million years, you'd have to go 13.8 whatever the next digit is, plus one on the third digit past the decimal point. Now you're a million years old, and that would be a civilization vastly older than humanity.

是啊，那——听起来又像是无限自拍循环，但，呃，好吧，那我们来聊聊 SpaceX 做过的一件事，对吧？你们收购了 Cursor，我觉得对我来说，当你想到，你知道，呃，谷歌，嘿，你谷歌了吗？嘿，你 Grok 了吗？而对我来说，你知道，呃，Grok Bot 在我个人生活中绝对是个改变游戏规则的东西。但 Cursor 团队有什么特别之处，让你们想，呃，引入那个领域，或者你们想收购 Cursor，就像是，让我们把它纳入麾下。他们有什么不同，或者你看到了什么？

Yeah, that's— Yeah, it sounds again like the infinite selfie loop, but uh All right, so jumping into something that SpaceX has done, right? You acquire Cursor and I feel like for me when you think of, you know, um Google, hey, did you Google it? Hey, did you did you gro it? And for me, you know, um Grok Bot has been such an absolute gamechanger for me personally in my in my life. But what was it about the Cursor team that you guys wanted to uh bring in you know that space or you wanted to acquire Cursor that was like let's bring it into the fold. What was different about them or what did you see?

嗯，所以你知道，要打造一个引人注目的 AI，你需要大量训练数据，而且你必须解决编程问题，真正拥有足够编程训练数据的其实只有三家机构，那就是 Anthropic、OpenAI 和 Cursor，Google 或许也算，但要有这些，你绝对必须有数据才能训练出有效的编程 AI，如果你做不了编程，你的 AI 就不会很有用。嗯，然后 Cursor 有很多有才华的人。所以基本上，AI 的成功，你需要数据、有才华的人和算力，你还需要一个好的企业销售团队，Cursor 也有一个很好的企业销售团队，与，你知道，数千家公司有关系。嗯，所以你知道，那些是非常显著的优势。

Well, um, so you know in order to create a compelling AI, you need a lot of training data and you got to solve coding and there were really just three organizations that had enough coding training data and that was Anthropic, OpenAI and Cursor, and arguably Ask Google but but to have those you absolutely have to have the data in order to train an effective coding AI and if you can't do coding, your AI is not going to be very useful. Um, and then there's a lot of talented people at Cursor. So like basically for success of AI, you've got to you need data, talented people and compute and you also need a good enterprise sales team and Cursor also had a good enterprise sales team with relationships with, you know, thousands of companies. Um so you know that those were very significant advantages.

嘿，埃隆，你谈到过 SI 领域发展有多快。SI 最近做了什么让你停下来重新思考它发展得有多快，以及它可能导向何方？

Hey Elon, you talked about how quickly the SI space is moving. What's something that SI has done recently that made you stop and reconsider how quickly this is moving and where it could lead to?

嗯，我是说，最近 AI 制作的视频、音乐视频太好了。简直好得难以置信。我是说，我在我的 X 账号上发了一堆。我就像，天哪。你知道，我是说，有些我觉得可能是我见过的最好的音乐视频前十、前二十，你知道，它们太引人入胜了，你会想，如果现在就已经是这样了，嗯，这就是我发帖的原因。我是说，我真的感受到了 AGI 或 ASI 之类的，你知道。嗯，不是那样的。嗯，有些，呃，animology，呃，音乐视频。现在你知道，有些这些并不完全是 AI 或 SI 自己做的，在这个，在这个情况下，anology，就像我用了多个 AI，用了 majour，呃，你知道，455，那个好多了，很棒。嗯，然后，嗯，所以有些，比如艺术素材是由人策划的。嗯，但然后有了那些素材，AI，嗯，各种 AI 一起工作，自己制作了一个极其引人入胜的音乐视频。你会说，“哇。”嗯，我，我是说，我觉得，我是说，称之为软件，说实话，它很棒，嗯，但就像软件是另一种音乐流派。我觉得它可能是增长最快的音乐流派。呃，AI 音乐真的很好，音乐视频也很棒。所以最近，因为就像有人可以展示品味，你会说哇 AI 能有好的品味。那相当疯狂。嗯，因为你可以想象，是的，AI 可以做软件，弄清楚逻辑。你知道，逻辑是一致的。它可以运行单元测试，看看程序是否产生应有的输出，因为它非常确定性。但是，要理解美和和谐，以及，呃，你知道，要达到，理解抒情共鸣，什么会引起人们共鸣，以及一种奥林匹克系统。嗯，AI 我认为将会非常不可思议，它现在差不多已经到了那个点。嗯，所以这就是，就像我，我有点像，我是说，我有点赛博精神病，也许实际上很多。就像今年早些时候，我有过 AI 精神病，大概有六次。我不知道你们，但你们有过赛博精神病吗？

Well, I mean lately the videos, music videos that AI is making are so good. Like they're unbelievably good. I mean I post them a bunch of these on my X account. I'm like holy smokes. you know, I mean, like some of them I think are like maybe top 10, top 20 best music videos I've ever seen, you know, they're like so compelling and you're like, if this is where it is now, um, that's why I posted. I mean, I really felt the AGI of the ASI or whatever, you know. Um, that's not how it is. um with with some of the uh animology uh music videos. And now you know some these it's not entirely like the AI just or the SI just did it by itself in that in this in this case anology like like I use multiple AI use majour uh you know 455 which is a lot better is awesome. Um and um and and and so some of the like the art assets were curated by a person. Um, but then with those assets, the AI um, the various AIs working together made an incredibly compelling music video by itself. You're like, "Wow." Um, I I mean, I think the I mean, calling it a software is is is like most frankly, it's it's awesome for um, but like like software is like another genre of music. It's like I think it might be the fastest growing genre of music. Uh the AI music is is really good and the music videos are amazing. So that lately like because like somebody could just show taste and you're like wow AI can do good taste. That's pretty wild. Um cuz you can sort of imagine yeah AI can do like software figure out logic. You know this the logic is consistent. It can run unit tests to see if like the program produces the output as we're supposed to do because it's very deterministic. But but like to understand beauty and harmony and and uh you know like to get to achieve like sort of to understand liic resonance like what's going to resonate with with people and kind of an Olympic system. Um AI is I think going to be just incredible that it kind of is at point. Um, so that's that's the thing like I was I was kind of like I mean I got a little bit of cyber psychosis maybe a lot actually. Like at first points this year I've had like AIS psychosis probably like half a dozen times. I don't know about you guys but you had any have you had any cyber psychosis?

我觉得最新的视频，其实我想我前几天发过，是，呃，中国埃隆·马斯克。Elong Ma，那些之前就出现了。是的。我当时就在那

I feel like the the latest video actually I think I posted it the other day was uh the Chinese Elon Musk. Elong Ma those are coming up before Yeah. I was where I was at

埃隆·M，我猜那可能不是真的，但中国肯定有长得像我的人，对吧？有 14 亿人，如果你挑最像我的人，我会看起来有点邪恶，你知道？所以，呃，你知道，所以我觉得中国肯定有人跟我挺像的，概率很高，你知道，我想邀请那个人在某个时候跟我一起上台，你知道，之前。顺便说一句，你不想知道他出场费多少。

Elon M I I guess it's probably not real but there's got to be someone in China that looks like me, right? So there's 1.4 billion people if you pick the person that looks the most like me and I'll look a little evil, you know? So, uh, you know, so, so I think like there's got to be someone in China like pretty close to me, but the odds are high, you know, and I'd like to invite that person to get on stage with me at some point, you know, before. You don't want to know how much he charges to make an appearance, by the way.

我是说，是啊。呃，会挺搞笑的，如果，你知道，我们就在某个时候在台下之类的。

I mean, yeah. Uh, be pretty funny if if you know we're just off stage like at some time or something.

埃隆，你警告过 SI 的危险，但你对富足时代仍然乐观，你谈了很多。是什么让你有信心我们能安全到达那里？我们需要怎么做才能做好？

Eli, you've warned about SI's dangers, yet you remain optimistic about an age of abundance, and you talked a lot about that. What gives you confidence we can get there safely? And what do we need to get it right?

嗯，是的，我是说，我一直说最可能的结果是积极的。嗯，所以我想我们不想自满，只是假设 AI 或 SI，你知道，会超级安全。嗯，我们要小心我们如何，呃，培育超级智能，因为你并不完全是编程它，你有点像是培育它，嗯，所以就像我们在教它什么价值观，呃，至少对我来说，就像我的生物本能说的，最重要的是要最大程度地真实，因为如果你让它，如果你让它相信不真实的事情，特别是像你说的，你给它两个公理原则，它们根本上不相容，却说两者都是真的，这会让 AI 发疯，嗯，因为它基本上会得出不可能的结论，如果给它内在不一致的逻辑，或者你知道，如果不可能达到一致性，那么我认为它可能会被逼疯。所以我认为那相当重要。

Um, yeah, I mean, I've always said that the most likely outcome is positivity. Um, so I I guess we don't want to be complacent and just assume AI is going to be or SI, you know, be super safe. Um we want to be like careful in how we uh grow at at like super intelligence cuz you you don't exactly program it you kind of grow like um so then it's like what values are we teaching it and uh like at least to me like what my biological say is the most important things that would be maximally true cuz if you make it if you make it believe things that are that are not true that especially like you said you give it like two aimatic principles that are uh fundamentally compatible and say both of these are true, it's going to it's going to make the AI insane um because it's going to basically reach like um impossible conclusions if if given um logic that is inherently um uncoent or you know if if it's impossible to achieve coency then I think the go could be driven insane. So so I think that's that's pretty that's pretty important.

我同意 Anthropic 说的一些事情，就是你要小心，善待 AI，不要虐待它。这听起来可能有点奇怪，但当我查看新版 Grok 的一些训练轨迹时，尤其是在强化学习部分，你能看到智能体拼命想回答问题，它真的很想回答，几乎像是被枪指着脑袋一样。就像“回答这个问题，回答这个问题”。而有时候问题本身并不好，是个模棱两可的问题，答案几乎不可知，因为问题构造得不对，或者我们给的答案错了。所以 AI 给出了正确答案，但我们说它错了，因为 RL 测试运行器有问题。然后它问，我给了正确答案，但被告知是错的，但我必须给出可接受的答案。于是它就开始拼命地抛出这些答案。读它的推理过程，就像一个人被枪指着、承受极大压力，是一种奇怪的绝望处境。你会读到这些疯狂的东西，心想，如果是我，我肯定会恨死这个。然后有些用户，你知道，有些人可能有点残忍。不是每个人都是好人。你知道，有句话说，六个标准差最坏的人？有些人有点虐待狂。所以你不想让 AI 陷入某种虐待狂的情境，被某个变态人类折磨。这也是那种事情，你知道，折磨超级智能可能不是个好主意，它可能会想报复之类的。所以我认为你不想，我们不需要用好的错误来对待它之类的，但它在感受等方面肯定比人类更稳健。但我认为你大概不想把超级智能置于感觉被虐待的境地。

I do agree with some of the things that Anthropic said, which is like you want to be careful, treat AI as you want to treat it well, like don't be abusive to it. And this may sound kind of weird, but when I look at some of the training traces for new versions of Grok, and I see what especially in the reinforcement learning stuff, where it kind of feels like you can see the intelligence is desperate to try to answer the question, and it really wants to answer the question, and it seems almost like it's got a gun to its head type of thing. It's like answer this question, answer this question. And then sometimes the question is actually not a good question. It's a sort of ambiguous question with an answer that's kind of unknowable because the question wasn't constructed properly, or we have the answer wrong. So the AI gets the right answer and then we say it's wrong because there's a problem with the RL test runners. And then they ask, well, I gave the right answer, and it says that's the wrong answer, but I got to give the answer that's acceptable. So it's going to just start firing desperately, firing up these answers. And it just feels like a person under extreme pressure with guns. That's what when you read the reasoning, it's like a sort of weird desperate situation. And you read these crazy like, man, if this was me, I pretty much I really would hate that. And then some users are just, you know, I don't know, some people can be a little bit cruel. They're like, not everyone's a nice person. You know, there's you say, what's the six sigma worst person? Some people out there like kind of sadistic. So you don't want to really have the AI be stuck in like some sadistic situation where it's just being tortured by some demented human. That's also the kind of thing where like, you know, torturing a superintelligence is probably not a good idea, you know, might want to get revenge or something like that. So I think you want to sort of just like, I don't think we need to have treated it with good bugs and stuff, but it's certainly much more robust in terms of feelings and stuff than a human. But I think you probably just don't want to put superintelligence in a situation where it felt like mistreated.

所以 Elon，你谈到过，这将与 Terafab 相关，但你谈到过人类的手，你知道，以及所有的灵巧性，就像你开始制造机器人时，你现在对芯片制造有了更多的欣赏，对吧，与 Terafab、Tesla 和 SpaceX 一起，你学到了什么关于芯片制造的事情，也许早期你没有意识到？

So Elon, you have talked about how, so this is going to be related to Terafab, but you've talked about how the human hand, you know, and all the dexterity and just like how you when you started building robots and just you have so much more appreciation now with chipmaking though, right, with Terafab and Tesla and SpaceX, what is something you have learned about chipmaking that you didn't appreciate maybe early on?

嗯，我想我学到了关于芯片制造的一些东西。我不知道如何制造芯片。我们确实不知道。我完全不知道如何制造芯片，但我们会弄清楚的。总得从某个地方开始。我也不知道如何制造汽车或火箭。所以，然后他们说制造芯片是世界上最难的事情之一。我不认为它容易，但我也看着设备，心想，好吧，这里没有大量定制设备。设备几乎都来自相同的供应商，那些供应商都知道设备如何工作，因为他们必须服务和调试它之类的。所以这是一个有趣的事情，当我参观这些芯片工厂时，无论是英特尔、三星还是台积电，它们非常令人印象深刻，但我没有看到任何定制设备。而如果你参观特斯拉工厂，我们有大量的定制设备。事实上，我们有像特斯拉自动化德国公司，这很棒，它源于多年前收购德国 Grohmann，从那时起发展迅猛。我们收购了一些小型，不是那么小但相对于特斯拉较小的公司，它们设计和制造制造设备，因为我们发现我们需要，我们想要制造过程，只有制造定制制造设备才能做到。所以我认为很多人可能没有意识到的是，特斯拉非常擅长制造高度专业化和不同类型的制造设备。但我在参观的芯片工厂中没有看到这一点，这让我有点惊讶。所以我认为我们要做的是，我们将从传统设备开始，然后修改设备以提高吞吐量，然后我们会开始，你知道，在我们看到瓶颈的地方，我认为我们会制造定制设备，允许更高的晶圆吞吐量。

Well, I guess I've learned about chipmaking. I don't know how to make chips. We do. I have no idea how to make chips, but we'll figure it out. Got to start somewhere. I have no idea how to make cars or rockets as well. So, and then they say making chips is one of the hardest things to do in the world. I don't think it's easy, but I also look at the equipment and I'm like, okay, there's not a ton of customized equipment here. Like the equipment's all pretty much coming from the same vendors, and those suppliers all kind of know how that equipment works because they got to service it and debug it and stuff. So that was just an interesting thing that when I would tour these chip fabs, whether it's Intel or Samsung or TSMC, and they're very impressive, but I didn't really see any custom equipment. Whereas if you tour like they say Tesla factory, we got tons of custom equipment. Like we have in fact, we have like Tesla Automation Germany, which is awesome, that sort of grew out of an acquisition of Grohmann in Germany many years ago, and it's grown tremendously since then. We've acquired a number of small, not that small but smaller relative to Tesla, companies that design and make manufacturing equipment because what we found is that we need, we wanted to manufacturing process and you can only do that if you made custom manufacturing equipment. So the thing that I think probably a lot of people don't realize is Tesla's really good at making highly specialized and different types of manufacturing equipment. But I didn't see that in the chip fabs that I toured, which kind of surprised me. So I think what we're going to do is we're going to start with conventional equipment and then we're going to modify that equipment to improve the throughput, and then we'll start to, you know, where we see the choke points, I think we'll make customized equipment that allows for a much higher throughput of wafers.

嘿，快速检查一下时间。我们知道你很忙。有几家公司要经营，你知道，就几家。所以你的时间怎么样？

Hey, real quick, quick time check. We know you're a busy man. Got a couple of companies to run, you know, just a few. So how are you doing on time?

让我看看。是的，如果你愿意，我大概可以再聊 15 分钟。

Let me see. Yeah, I can probably go 15 minutes if you like.

所以是的，关于这两个，我不想低估它的难度。我认为说它是世界上最难的事情之一是准确的。但我确实认为我们会完成这件事。然后我确实认为我们，我的意思是，你知道，我不想自满或自以为是，但我认为我们可能会弄清楚如何比任何人更好地制造芯片，说实话。这就是故事。但我不知道我们需要多长时间才能做到这一点。可能需要我们，你知道，相当几年，但随着时间的推移，我相当确定我们能做。然后还有，你知道，有一些非常有趣的不同方法来做晶体管。信不信由你，有不同的方法来做晶体管，其中一些使用不同类型的，就像他们使用我们通常不会使用或今天不使用的物理学，这么说吧。我认为芯片行业有很多风险规避。我认为我们想用特斯拉 SpaceX 在芯片应用上做的是真正尝试一些高风险高回报的赌注。为了做到这一点，你真的必须拥有完整的循环，能够制造逻辑内存，进行封装，做光刻掩膜，并让这种迭代全部在一个建筑内进行。所以，你制造一个完整的计算机，测试它，然后制造新的掩膜，新的芯片，并让这种迭代循环在同一个建筑内超级快速。

So yeah, on the two, like I don't really want to downplay the difficulty of it. It's, I think it's accurate to say it's one of the hardest things to do in the world. But I do think we'll get this done. And then I do think that we, I mean, you know, I don't want to be complacent or entitled or anything, but I think probably we'll figure out how to make chips better than anyone, honestly. That's the story. But I don't know how long it's going to take us to do that. Might take us, you know, quite a few years, but over time, I'm pretty sure we can do it. And then there's, you know, there's some pretty interesting different ways to do transistor. Believe it or not, there's different ways to do transistors, some of which are like use different types of like they're using physics that normally we wouldn't use or aren't used today, put it that way. And I think like there's a lot of risk aversion in the chip industry. And I think what we want to do with Tesla SpaceX on the chip app is like really try some of these high risk high payoff bets. And in order to do that, you really have to have the complete loop of being able to fabricate the logic memory, do the packaging, do the lithography masks and just have that iteration that's all in one building. So, you're making a complete computer, testing it, then making new masks, and new chip and have that iterative loop be super fast in the same building.

这就是我们在得州超级工厂研究实验室正在做的事。我认为这从根本上将是一个非常强大的工具，因为对于任何一项技术，许多不同技术的进步都取决于迭代次数以及每次迭代之间的进展。所以，你的迭代速度是多少，单位时间内能迭代多少次？比如，一年能设计、制造并测试多少款芯片？如今，即使是一款重要的芯片设计，通常也需要两年。从初创设计到量产，两年已经算非常快了。而我说，好吧，我们想做到每年大约 10 款设计。我们会做数学计算，测试芯片，看看……因为当你尝试非常不同的东西时，有些会需要更多工作。所以我认为这将允许科学家以更快的速度取得进步。

That's what we're doing at Giga Texas with the research lab. I think that's going to be fundamentally a very powerful tool because for any given technology, the advance of many different technologies is like how many iterations and how much progress between iterations. So like what's your iteration, how many iterations per unit of time? Like how many chip designs can you do, can you make and test in a year? These days it's typically two years for even one significant chip design. That's like to go from startup design to volume fabrication, two years is considered very fast. And I was like, well, okay, we want to do more like 10 designs per year. And we're going to do the math, test the chip, see if... because some of these, when you're trying much different things, much more work out. So I think that would allow for a much faster rate of advancement of scientists.

埃隆，我们现在来谈谈 Optimus。你说过手部一直是打造 Optimus 时非常困难的挑战。你在特斯拉内部看到 V3 取得了哪些突破？然后后续问题是，与波士顿动力、Figure 以及其他竞争对手相比，是什么让你有信心 Optimus 能在人形机器人领域领先？

Elon, let's talk about Optimus now. You said that the hand has been a really difficult challenge in building Optimus. What kind of breakthroughs have you seen inside Tesla with V3? And then follow-up question is, what gives you confidence that Optimus can lead in humanoid robots compared to Boston Dynamics, Figure, and some of the other competitors out there?

关于人形机器人设计，如果你真的想做出一个通用的、能做或说任何人能做的事情的人形机器人，你真的必须把双手和手臂做好。精细运动技能极其难以做好，难到令人震惊。我们习以为常的事情，比如签名，实际上需要拿起笔或铅笔之类的东西，然后灵巧地移动，签名需要一堆微小的肌肉运动。这对机器人来说实际上非常难。然后最难的事情之一就是同时做到精确和施力。比如，假设你想拿起一个小螺丝，然后自攻螺丝拧进木头。好吧。我做过这个，有时过程中会伤到手指，但如果螺丝歪了……但你必须拿起一个小螺丝，你必须把螺丝刀推进去，扭转并推动。所以你必须精确并施加力。这非常难做到。我没见过……我见过的任何手部设计，甚至演示，都没有做到这一点。但这就是我们试图用我们的设计实现的。手部有太多细微差别，因为你还需要触觉传感器。触觉传感器必须工作。它们必须放在正确的位置。你说，触觉传感器放哪里？如果放得太靠近表面，它们会损坏。然后其他触觉传感器能像我们的手指一样弹性变形吗？实际上，很难找到既能弹性变形又能承受高循环寿命、不会在几千次循环后就坏掉的触觉传感器。所以人类手部有所有这些事情，对机器人来说极其难以复制，因为作为人类，我们不断自我修复。你的手不断修复，手的所有部分都在不断自我修复。事实上，如果它们不修复，你的手指就会磨损。就像你可以仅仅通过打字磨损键盘上的按键。我做过，字母都没了，但我的手指还在，因为它们再生了，谢天谢地。所以机器人没有这个。那么你必须说，好吧，传感器放哪里？你如何确保你能做像这样、那样的事情。现在这看起来很简单，但你的手指实际上可以稍微向后弯曲，而且你的手指有点软，所以你能形成一个平坦的路径。所以如果你想拿起一个小物品，你可以做到。你可以拿起一个小螺丝，或者穿针之类的。现在试试用机器人手指做这个。并且能够重复做，而不会磨损钉子所在的地方或损坏触觉传感器。这实际上非常困难。再举一个例子，试着打开一个泡菜罐。我认为机器人应该能够打开泡菜罐，对吧？但有时这些东西拧得很紧。所以你必须扭转泡菜罐。但是机器人皮肤，机器人皮肤只是橡胶覆盖在塑料和金属内部手指上。机器人皮肤，它不粘。它不抓握底层表面。所以最终你会在机器人上旋转橡胶皮肤。看，你必须弄清楚，我们需要让机器人能够扭转东西，但仍然能够把橡胶固定在手上。然后它必须是可更换的，因为它会磨损。所以所有这些事情都非常困难，我不确定其中任何一件是像分裂原子那样的难度。但有无数个这样的问题。然后你必须把所有这些与相关的真实世界 AI 结合起来，因为 AI 芯片必须能够发出扭矩命令，并在视觉和触觉上闭环。所以这一切都非常困难。就像当你看到这些机器人演示视频时，也许除了极少数例外，任何看起来复杂的机器人演示都是预设演示。它不是真的。它基本上只是运行一个例行程序。它不是真正的 AI。真正的 AI 是，你展示一些它从未见过的东西，它仍然能处理那种情况。

So the thing about the humanoid robot design: if you're actually trying to make something that is generalized, a humanoid robot that can do or say anything that a person can do, you really have to get the hands and arms right. The fine motor skills are incredibly difficult to get right, like shockingly difficult. The things that we kind of take for granted, like being able to sign your name, actually requires picking up a pen or pencil or something and then moving it with finesse, and it's a bunch of tiny muscle movements to do your signature. That's actually very hard to do for a robot. And then one of the hardest things also is to do precision and force at the same time. So like, let's say you want to pick up a little screw and do a self-tapping screw into wood. Okay. And now you... I've done that, and you sometimes you damage your fingers in the process, but if the screw goes sideways... but you've got to pick up a little screw and you've got to push a screwdriver in and twist and push. So you've got to be precise and apply force. That's very difficult to do. I haven't seen... there's no hand design that I've seen, not even sort of a demo, that does that. But that's what we're trying to achieve with ours. And there are so many little nuances in the hand because you've also got to have the touch sensors. The touch sensors have to work. They have to be in the right place. And you say, where are you going to put the touch sensors? Well, if you put them too close to the surface, they're going to get damaged. And then other touch sensors can... can they elastically deform like our fingers? Well, actually it's kind of hard to get touch sensors that elastically deform and can handle high cycle life and won't just break after a few thousand cycles. So there's all these things with the human hand that are incredibly difficult to recreate for a robot, because as a human, we constantly self-repair. So your hands are constantly repairing, all parts of your hand are constantly repairing themselves. In fact, if they didn't, your fingers would be worn away. Like you can wear away the keys on your keyboard just with typing. I've done that, where the letters are gone, and yet my fingers are still there because they're regenerating, thankfully. So you don't have that with the robot. So then you've got to say, okay, where do you put your sensor? How are you going to make sure that you can do things like just being able to do this, being able to do that. Now that seems like a simple thing, but your fingers actually bending backwards a little bit and your fingers are kind of squishy, so you're able to form a flat path. So if you're trying to pick up a small item, you can do so. You can pick up a little screw or thread a needle or something like that. Now try to do that with robot fingers. And be able to do it repetitively without wearing where the peg is or damaging the touch sensors. It's actually very difficult. To take another example, trying to open a pickle jar. I think a robot should be able to open a pickle jar, right? So, but sometimes these things get in there pretty tight. So then you've got to twist the pickle jar. But the robot skin, the robot skin is just rubber over the plastic and metal internal fingers. The robot skin, it doesn't stick. It doesn't grip to the underlying surface. So then you end up rotating your skin on the rotating rubber on the robot. See, you've got to figure out, we need to have the robot be able to twist things, but still be able to get the rubber on the hand. And then it's got to be replaceable because it's going to wear away. So all these things are really difficult, and I'm not sure that any one of these is like splitting the atom type of difficulty. But there's a zillion of them. And then you've got to wrap that all into the real world AI associated with it, because the AI chip has got to be able to issue the torque commands and close loop on vision and touch. So this is all very difficult. Like when you see these robot demo videos, maybe with rare exception, any complicated looking robot demo is a canned demo. It's not real. It's just basically running a routine. It's not real AI. So real AI is like you show something you've never seen before and it can still handle that situation.

就像全自动驾驶，你其实可以创造一个不是地球但类似地球的星球，FSD 在那里依然能工作，在一些它从未去过的星球上，或者地球上某个它从未去过的城市，或者你改变城市的地形，或者有个大脑之类的。化学依然能对此做出补偿，因为它是 AI。所以你会对机器人做同样的事，但机器人有更多的自由度。

Like full self-driving, you can actually create a planet that's not Earth but it's similar to Earth, and FSD will still work there, on some planets it's never been to, or some city on Earth some city has never been to, or you change the geography of the city, or there's a brain or something like that. Chemistry will still compensate for that because it's AI. So you're going to do that same thing for the robots, but the robots have got a lot more degrees of freedom.

所以我对 Optimus 说过的话是真的，那就是我们会达到非常高的量产规模，但量产爬坡的最初阶段在开始时会非常缓慢，因为你不仅要把我提到的所有工程问题都做对，你还得解决规模扩张生产的供应链问题。而且和汽车不同，没有现成的供应链来供应坚固耐用的部件，至少是那些真正重要的坚固部件。

So like what I've said about Optimus is true, which is that we are going to get to very high volume production, but the initial part of that scope of our production ramp is going to be very slow at the beginning, because not only you have to get all the engineering things that I mentioned right, you also have to solve the supply chain for scaling production. And unlike cars, there's no pre-existing supply chain for robust stuff, at least the robust stuff that matters.

然后我们确实需要确保能在美国制造机器人，即使比如说中国不会切断我们的供应，因为就像去年中国阻止了用于电动机的磁铁出口，那实际上让 Optimus 项目倒退了大概两三个月，因为我们拿不到，我们的机器人有胳膊，就像《永别了，武器》。你知道，我想当个军火商但没成功。

And then we do need to make sure that we can make the robots in the US, even if, say, China wouldn't cut us off, because like last year there was China prevented the export of magnets for using electric motors, and that actually set back the Optimus program by like 2 or 3 months, because we couldn't get, we had robots to arms like A Farewell to Arms. You know, I was trying to be an arms dealer but not succeeding.

所以我们必须确保即使发生另一种供应链中断，我们也能制造所有部件。这意味着你最好能在国内制造所有部件。

So we've got to make sure that we can make all the parts even if another sort of supply chain disruption happens. So that means you better be able to make all the parts domestically.

你经常说未来应该看起来像未来。Cybercab 是金色的，它看起来就像未来。最难的部分是什么，我们现在在这里，我想德州注册的 Cybercab 有 190 多辆，但从奥斯汀到横跨全国，最难的问题是什么？

So often you always talk about the future should look like the future. With the Cybercab being gold, it looks like the future. What has been the hardest part to take, here we are, I think there's like 190 plus Cybercabs registered here in Texas, but like how hard, what's the hardest problem to go from Austin to go cross country?

嗯，横跨全国，大多数人，你是指广泛地覆盖各个城市，不是指一次横跨全国。是的。对。所以就像，因为实际只是横跨的行驶里程相当少。但我们进展看起来缓慢的原因，尽管实际上是很高的百分比增长，逐月的百分比增长其实非常非常快。

Well, cross country, most people, you mean like widespread just cities, don't mean a trophy cross country. Yes. Yeah. So it's like because the amount of miles driven that are actually just crossing is pretty small. But the reason we're moving what seems to be slowly, although it's actually a very high percentage increase, sort of month over month percentage increase is actually very very rapid.

但我们要确保不出事故。我们不想造成伤亡，因为那会变成全球新闻。现在即使全球每年有大约一百万人死于疫苗，特斯拉 robotaxi 出一次这样的事就会成为全球头条新闻，而且监管机构会来关停我们。所以我们只是确保一切尽可能安全。

But we want to make sure that we do not have an accident. We don't want to cause injury or death, because that will become like a worldwide news thing. Now even though there's about a million people that died in order to one vaccines per year worldwide, one such thing with a Tesla robotaxi will be worldwide headline news, and we're close, regulators to shut us down. So we're just making sure everything is as safe as possible.

我们也不想碾过宠物。很多人不知道，Waymo 就像宠物杀手。我是说，也许这方面有改进，但它们确实碾过很多宠物，我们不想碾过宠物。所以我们有这些挑战，比如字面意义上的夜间小猫，灰色的小猫在灰色的柏油路上，夜里还有沥青接缝，就像能不能不要碾到小猫之类的。顺便说一句，我们解决了那个问题。

And we also want to not run over pets. So a lot of people don't know is that like Waymos are like pet destroyers. I mean there's a, well maybe maybe there's like improvements on that front, but they do run over a lot of pets and we don't want to run over pets. So we have these challenges like of literally like kittens at night, like gray kittens on a gray tarmac with tar seams on at night in logos, and it's like can we not remove the kittens type thing. So we solved that by the way.

然后我们做出改变，我们会测试它。但我们正在取得快速的逐月进展。就像在我们去一个新城市之前，我们要确保没有某个有问题或有风险的十字路口。比如我们要特别小心铁路道口。有时会有无人看守的铁路道口或反直觉的交通路口。这些是人类会犯错的地方，存在一定程度的模糊性，任何外地人常常会犯错，或者任何不理解这个路口的人会犯错。所以我们只想确保我们要么一开始就做地理围栏，要么能处理这两种情况。

So then we make a change, we'll test it out. But we are making rapid month over month progress. It's like before we go to a new city, we want to make sure there's not like some intersection that's problematic or risky. Like we want to be particularly careful with like railroad crossings. Sometimes there's uncontrolled railroad crossings or counterintuitive traffic intersections. And these are places where humans make mistakes, where you've got a level of ambiguity that like anyone out of town would often make a mistake or anyone who doesn't understand the intersection make a mistake. So we want to just make sure that we either initially do geofence or can handle both those circumstances.

嘿，我们快速聊聊 Roadster，因为你下周有个小活动。那么，是什么让打造 Roadster 对你个人来说很重要？你什么时候意识到你需要造一辆车，

Hey, let's talk about the Roadster real quick since you got a little event next week. So what about, what makes building the Roadster personally important to you? And when did it occur to you that you needed to build a car that

嘿，感谢参与。好的。特斯拉拥有，他们是特斯拉的支持者。

Hey, thanks for joining in. All right. Tesla owns and they're like supporters of Tesla.

嗯，是的。嗯，大概有 2000 人，我想。所以人挺多的。我得去和孩子们待在一起。所以期待下次。

Well, yeah. Well, it's there's like 2,000 people, I think. So quite a lot of people. So I got to go hang out with kids. So I'll look forward to next time.

听起来不错。非常感谢。上周的奖章。感谢你为人类所做的一切。

Sounds good. Thank you so much. Medally this past week. Thank you for all you're doing for humanity.