埃隆·马斯克谈智能大爆炸与创造有用之物

我们正处于智能大爆炸的非常非常早期的阶段。成为一个多行星物种，大大增加了文明、意识或智能（无论是生物还是数字）的可能寿命。我认为我们非常接近数字超级智能。如果不是今年，那肯定是明年。

We're at the very, very early stage of the intelligence big bang. Being a multiplanet species greatly increases the probable lifespan of civilization or consciousness or intelligence, both biological and digital. I think we're quite close to digital superintelligence. If it doesn't happen this year, next year for sure.

让我们欢迎埃隆·马斯克。埃隆，欢迎来到 AI 创业学校。我们真的非常荣幸今天能请到你。

Please give it up for Elon Musk. Elon, welcome to AI Startup School. We're just really, really blessed to have your presence here today.

谢谢邀请。

Thanks for having me.

那么，从 SpaceX、特斯拉、Neuralink、xAI 等等。在这之前，你生命中是否有过这样一个时刻，让你觉得‘我必须做出伟大的东西’？是什么触发了那个转变？

So, from SpaceX, Tesla, Neuralink, xAI, and more. Was there ever a moment in your life before all this where you felt 'I have to build something great'? And what flipped that switch for you?

嗯，我最初并没有想过要做出伟大的东西。我想尝试做一些有用的东西，但没想过会做出什么特别伟大的事。从概率上讲，这似乎不太可能，但我至少想试试。

Well, I didn't originally think I would build something great. I wanted to try to build something useful, but I didn't think I would build anything particularly great. If you said probabilistically, it seemed unlikely, but I wanted to at least try.

所以你现在面对一屋子的人，他们都是技术工程师，其中很多是崭露头角的杰出 AI 研究者。

So you're talking to a room full of people who are all technical engineers, often some of the most eminent AI researchers coming up in the game.

我觉得我们应该……我更喜欢‘工程师’这个词，而不是‘研究者’。我的意思是，如果有一些根本性的算法突破，那是研究，但除此之外都是工程。

I think we should... I like the term 'engineer' better than 'researcher'. I mean, I suppose if there's some fundamental algorithmic breakthrough, it's research, but otherwise it's engineering.

也许让我们回到过去。这屋子里都是 18 到 25 岁的年轻人。年龄偏小，因为创始人越来越年轻。你能设身处地想想你 18、19 岁学编程，甚至想出 Zip2 的第一个点子时的情景吗？那对你来说是什么样的？

Maybe let's start way back. This is a room full of 18 to 25 year olds. It skews younger because the founder set is younger and younger. Can you put yourself back into their shoes when you were 18, 19, learning to code, even coming up with a first idea for Zip2? What was that like for you?

是的，回到 95 年，我面临一个选择：要么读研究生，在斯坦福读材料科学博士，实际上研究用于电动汽车的超级电容器，试图解决电动汽车的续航问题；要么尝试做一件大多数人从未听说过的事情——互联网。我和我的教授，材料科学系的 Bill Nix 谈过，说：‘我能推迟一个学期吗？因为这件事很可能会失败，然后我需要回到大学。’他说：‘这可能是我们最后一次谈话了。’他说得对。但我当时认为事情很可能会失败，而不是成功。然后在 95 年，我编写了基本上第一个或接近第一个的地图导航、互联网黄页和白页。我亲自编写了这些，甚至没有使用网络服务器，而是直接读取端口，因为我负担不起 T1 线路。最初的办公室在帕洛阿尔托的谢尔曼大道。楼下有一家 ISP，所以我在地板上钻了一个洞，直接拉了一根局域网线到 ISP。我哥哥和另一位联合创始人 Greg Kouri（已故）加入了我。当时我们甚至负担不起住处，办公室每月 500 美元，所以我们就在办公室睡觉，然后在佩奇米尔路的 YMCA 洗澡。我们最终创办了一家有点用处的公司，Zip2。我们确实构建了很多非常好的软件技术，但我们在某种程度上被传统媒体公司俘获了。《纽约时报》和其他公司既是投资者、客户，也在董事会中。所以他们一直想让我们以毫无意义的方式使用我们的软件。我想直接面向消费者。总之，关于 Zip2 的故事说得太多了，但我真的只是想在网上做一些有用的事情。因为我有两个选择：读博士，看着别人构建互联网；或者以某种小方式帮助构建互联网。我想，‘好吧，我总可以尝试一下，失败了再回去读研。’结果它相当成功。以 3 亿美元出售，这在当时是一大笔钱。如今，AI 初创公司的最低冲动出价大约是 10 亿美元。现在有太多该死的独角兽了，简直是一群独角兽。独角兽是指市值 10 亿美元的公司。从那以后通货膨胀了很多，所以实际上钱更多了。

Yeah, back in '95, I was faced with a choice: either do grad studies, a PhD at Stanford in material science, actually working on ultracapacitors for potential use in electric vehicles, essentially trying to solve the range problem for electric vehicles, or try to do something in this thing that most people had never heard of called the internet. I talked to my professor, Bill Nix in the material science department, and said, 'Can I defer for a quarter? Because this will probably fail and then I'll need to come back to college.' And he said, 'This is probably the last conversation we'll have.' And he was right. So, but I thought things would most likely fail, not that they would most likely succeed. And then in '95, I wrote basically the first or close to the first maps directions, internet white pages and yellow pages on the internet. I just wrote that personally and I didn't even use a web server. I just read the port directly because I couldn't afford a T1. The original office was on Sherman Avenue in Palo Alto. There was an ISP on the floor below, so I drilled a hole through the floor and just ran a LAN cable directly to the ISP. My brother joined me and another co-founder, Greg Kouri, who passed away. At the time, we couldn't even afford a place to stay, so the office was $500 a month, so we just slept in the office and then showered at the YMCA on Page Mill Road. And we ended up doing a little bit of a useful company, Zip2, in the beginning. We did build a lot of really, really good software technology, but we were somewhat captured by the legacy media companies. The New York Times and others were investors and customers and also on the board. So they kept wanting to use our software in ways that made no sense. I wanted to go direct to consumers. Anyway, long story dwelling too much on Zip2, but I really just wanted to do something useful on the internet. Because I had two choices: do a PhD and watch people build the internet, or help build the internet in some small way. And I was like, 'Well, I guess I can always try and fail and then go back to grad studies.' And that ended up being reasonably successful. Sold for like $300 million, which is a lot at the time. These days, that's like the minimum impulse bid for an AI startup is like a billion dollars. There's so many freaking unicorns, it's like a herd of unicorns at this point. Unicorn is a billion-dollar situation. There's been inflation since, so quite a bit more money actually.

是啊。我的意思是，1995 年你可能花五分钱就能买到一个汉堡。好吧，没那么夸张，但确实通货膨胀了很多。但 AI 的炒作程度非常激烈，你也看到了。有些成立不到一年的公司，有时能获得十亿甚至数十亿美元的估值。我想这在某些情况下可能会成功，而且很可能成功，但看到这些估值还是让人瞠目结舌。你怎么看？

Yeah. I mean, like 1995 you could probably buy a burger for a nickel. Well, not quite, but I mean, yeah, there has been a lot of inflation. But the hype level on AI is pretty intense, as you've seen. You see companies that are less than a year old getting sometimes billion-dollar or multi-billion-dollar valuations. Which I guess could pan out and probably will pan out in some cases, but it is eye-watering to see some of these valuations. What do you think?

嗯，我个人非常看好。老实说，我非常看好。所以我认为在座的各位将创造大量价值，世界上有十亿人应该使用这些东西。而我们甚至还没有触及表面。

Well, I'm pretty bullish, personally. I'm pretty bullish, honestly. So I think the people in this room are going to create a lot of the value that a billion people in the world should be using this stuff. And we're not even scratching the surface of it.

我喜欢这个互联网故事，因为即使在那时，你也很像在座的各位。所有传统媒体公司的 CEO 都把你视为懂互联网的人，而很多不理解 AI 的企业界也会向在座的各位寻求答案。有哪些具体的教训？听起来其中一个是不要放弃董事会控制权，或者要小心，找个好律师。

I love the internet story in that even back then you were a lot like the people in this room. The heads of all the CEOs of the legacy media companies looked to you as the person who understood the internet, and a lot of the corporate world that does not understand what's happening with AI will look to the people in this room for exactly that. What are some of the tangible lessons? It sounds like one of them is don't give up board control, or be careful about having a really good lawyer.

我想对于我的第一家初创公司来说，最大的错误是传统媒体公司拥有过多的股东和董事会控制权，他们必然通过传统媒体的视角看问题，会让你做一些在他们看来合理但不符合新技术的事情。我应该指出，我最初其实并没有打算创办公司。我试图在 Netscape 找一份工作。我把简历寄给了 Netscape，Marc Andreessen 知道这件事。但我不认为他看过我的简历，而且没有人回复。然后我试图在 Netscape 的大厅里闲逛，看看能不能碰到什么人，但我太害羞了，不敢和任何人说话。所以我想，‘天哪，这太荒谬了。我还是自己写软件，看看会怎么样吧。’所以实际上并不是从‘我想创办公司’的角度出发的。我只是想以某种方式参与构建互联网。

I guess for my first startup, the big mistake was having too much shareholder and board control from legacy media companies, who then necessarily see things through the lens of legacy media, and they'll kind of make you do things that seem sensible to them but don't really make sense with the new technology. I should point out that I didn't actually at first intend to start a company. I tried to get a job at Netscape. I sent my resume into Netscape, and Marc Andreessen knows about this. But I don't think he ever saw my resume, and nobody responded. Then I tried hanging out in the lobby of Netscape to see if I could bump into someone, but I was too shy to talk to anyone. So I'm like, 'Man, this is ridiculous. I'll just write software myself and see how it goes.' So it wasn't actually from the standpoint of 'I want to start a company.' I just wanted to be part of building the internet in some way.

既然我找不到互联网公司的工作，我就只好自己开一家。AI 将如此深刻地改变未来，其程度难以估量。假设事情没有失控，AI 也没有杀死我们所有人，最终你会看到一个经济规模不是现在的 10 倍，而是数千倍甚至数百万倍。我在华盛顿特区因为削减浪费和欺诈而挨批，那是个有趣的支线任务，但我得回到主线任务了。修复政府就像清理一个肮脏的海滩，而 AI 的千尺海啸即将来袭。清理海滩有多大意义？没多大。很高兴回到科技领域，那里的信噪比更好。在政治里，你骗不了数学和物理；它们是严格的裁判。

Since I couldn't get a job at an internet company, I had to start one. AI will so profoundly change the future, it's difficult to fathom how much. Assuming things don't go awry and AI doesn't kill us all, you'll see an economy that is ultimately not 10 times bigger, but thousands or millions of times bigger than today. When I was in DC, taking flak for cutting waste and fraud was an interesting side quest, but I had to get back to the main quest. Fixing the government is like cleaning a dirty beach while a thousand-foot tsunami of AI is about to hit. How much does cleaning the beach matter? Not that much. I'm glad to be back in technology, where signal-to-noise is better. In politics, you can't fool math and physics; they are rigorous judges.

很高兴你回到了主线任务。这非常重要。

We're glad you're back on the main quest. It's very important.

是的，回到建造技术，这是我喜欢做的事。噪音太多了；政治中的信噪比糟透了。我住在旧金山，所以不用你说两遍。华盛顿特区全是政治。但如果你要造火箭、汽车或能可靠编译运行的软件，你就必须最大限度地追求真理，否则你的软件或硬件就行不通。你骗不了数学和物理。所以我很高兴回到科技领域。

Yeah, back to building technology, which is what I like doing. There's so much noise; the signal-to-noise ratio in politics is terrible. I live in San Francisco, so you don't need to tell me twice. DC is all politics. But if you're trying to build a rocket or cars or software that compiles and runs reliably, you have to be maximally truth-seeking, or your software or hardware won't work. You can't fool math and physics. So I'm glad to be back in technology.

回到 Zip2 的时刻，你有了价值数亿美元的退出。你拿到了 2000 万美元，解决了钱的问题，然后继续投入 X.com，它后来成了 PayPal。不是每个人都会这么做。是什么驱使你重新投入？

Going back to the Zip2 moment, you had an exit worth hundreds of millions of dollars. You got $20 million, solved the money problem, and kept rolling with X.com, which became PayPal. Not everyone does that. What drove you to jump back into the ring?

对于 Zip2，我们建造了不可思议的技术，但它从未真正被使用。我们的技术比雅虎或其他任何人都好，但我们受到客户的限制。我想做点不受客户限制的事情，直接面向消费者。最终就是 X.com 与 Confinity 合并，共同创造了 PayPal。PayPal 的衍生公司可能比 21 世纪任何东西创造的都多。那么多有才华的人都在那个组合里。我觉得 Zip2 束缚了我们的翅膀，我想看看如果不受束缚会怎样。这就是 PayPal 最终的样子。我拿到了那张 2000 万美元的支票，是我在 Zip2 的份额。那时我和四个室友住在一起，银行里大概有 1 万美元。然后支票寄到了，我的银行余额从 1 万美元变成了 2000 万美元。我还得交税，但我几乎把所有钱都投进了 X.com，几乎把所有筹码都留在了桌上。

With Zip2, we built incredible technology, but it never really got used. We had better technology than Yahoo or anyone else, but we were constrained by our customers. I wanted to do something where we weren't constrained by customers, go direct to consumer. That ended up being X.com merging with Confinity, which together created PayPal. PayPal's diaspora might have created more companies than anything in the 21st century. So many talented people were at the combination. I felt like our wings were clipped with Zip2, and I wanted to see what happened if they weren't. That's what PayPal ended up being. I got that $20 million check for my share of Zip2. At the time, I was living with four housemates, had maybe $10,000 in the bank. Then the check arrived in the mail, and my bank balance went from $10,000 to $20 million. I still had to pay taxes, but I put almost all of it into X.com, keeping almost all the chips on the table.

PayPal 之后，你想知道为什么我们还没把人送上火星。

And after PayPal, you wondered why we hadn't sent anyone to Mars.

是的。我很好奇为什么我们还没把人送上火星。我上了 NASA 网站，想知道什么时候送人，但没有日期。我以为可能很难找到，但实际上没有真正的计划。所以我开始思考。我和大学室友 Adeo Ressi 在长岛高速公路上，他问我 PayPal 之后打算做什么。我说不知道，也许在太空做点慈善，因为我觉得在那里做不了商业。但我很好奇我们什么时候送人去火星。网站上没有，所以我开始挖掘。我的第一个想法是一个名为 'Life to Mars' 的慈善任务：发送一个小温室，里面装有种子和脱水营养凝胶，降落在火星上，水合凝胶，然后拍一张绿色植物在红色背景下的精彩照片。很长一段时间，我不知道 'money shot' 是色情片里的说法。但重点是激励 NASA 和公众送宇航员去火星。在这个过程中，我在 2001 年和 2002 年去了俄罗斯买洲际弹道导弹。那是一次冒险：会见俄罗斯高层，说 '我想买一些洲际弹道导弹。' 这是为了进入太空，不是为了核打击任何人。由于削减武器谈判，他们不得不销毁大量大型核导弹。所以我想，我们能不能拿两枚，去掉核弹头，再加一个上级火箭去火星？

Yes. I was curious why we hadn't sent anyone to Mars. I went on the NASA website to find out when we're sending people, and there was no date. I thought maybe it was hard to find, but there was no real plan. So I started thinking about it. I was on the Long Island Expressway with my friend Adeo Ressi, a college housemate, and he asked what I was going to do after PayPal. I said I didn't know, maybe something philanthropic in space, because I didn't think I could do anything commercial there. But I was curious when we'd send people to Mars. It wasn't on the website, so I started digging. My first idea was a philanthropic mission called 'Life to Mars': send a small greenhouse with seeds and dehydrated nutrient gel, land it on Mars, hydrate the gel, and get a great shot of green plants on a red background. For the longest time, I didn't realize 'money shot' is a porn reference. But the point was to inspire NASA and the public to send astronauts to Mars. Along the way, I went to Russia in 2001 and 2002 to buy ICBMs. That was an adventure: meeting with Russian high command and saying, 'I'd like to buy some ICBMs.' This was to get to space, not to nuke anyone. As a result of arms reduction talks, they had to destroy a bunch of their big nuclear missiles. So I thought, how about we take two of those, minus the nuke, add an additional upper stage for Mars?

嗯，但当时在莫斯科，2001 年，跟俄罗斯军方谈判买洲际弹道导弹，那感觉挺迷幻的。太疯狂了。而且他们一直涨价，跟谈判该做的完全相反。我就想，这些东西越来越贵了。然后我意识到，问题不是缺乏去火星的意愿，而是没办法在不超预算的情况下做到，甚至 NASA 的预算都不够。所以我就决定创办 SpaceX，推进火箭技术，直到能把人送上火星。那是 2002 年。所以我不是一开始就想创业。我想做点自己觉得有趣、人类需要的事情。然后就像猫拉线团一样，线团散开，结果发现这可能是门很赚钱的生意。现在确实赚钱了，但之前没有火箭初创公司成功的先例。有过各种商业火箭公司的尝试，都失败了。所以创办 SpaceX，我的想法是，成功概率可能不到 10%，也许 1%。我不知道。但如果初创公司不推动火箭技术，那肯定不是大型国防承包商来做，因为他们只跟政府对接，政府只想做非常传统的事情。所以要么是初创公司来做，要么根本不会发生。所以小概率成功总比没有成功好。所以我在 2002 年中创办了 SpaceX，预期会失败。我说大概 90%会失败。招人时我也不假装会成功，我说我们很可能完蛋。但有 12%的机会不会完蛋，这是唯一能把人送上火星、推进技术前沿的方法。然后我成了火箭的首席工程师，不是因为我想，而是因为我雇不到好人。好的首席工程师都不愿意来，觉得风险太大。所以我成了首席工程师。前三次发射都失败了，算是学习过程。第四次幸运地成功了。但如果第四次没成功，我就没钱了，那就完了。所以很悬。如果猎鹰的第四次发射没成功，那就彻底完了，我们会加入之前火箭初创公司的坟墓。所以我对成功的估计差不多。我们勉强成功了。

Um, but it was kind of trippy, you know, being in Moscow in 2001 negotiating with the Russian military to buy ICBMs. That's crazy. And they kept raising the price on me, which is the opposite of what a negotiation should do. So I was like, man, these things are getting really expensive. And then I came to realize that the problem was not that there was insufficient will to go to Mars, but that there was no way to do so without breaking the budget, even breaking the NASA budget. So that's where I decided to start SpaceX, to advance rocket technology to the point where we could send people to Mars. That was in 2002. So it wasn't that I started out wanting to start a business. I wanted to start something that was interesting to me that I thought humanity needed. And then as you sort of, like a cat pulling on a string, the ball unravels, and it turns out this could be a very profitable business. I mean, it is now, but there had been no prior example of a rocket startup succeeding. There have been various attempts to do commercial rocket companies, and they all failed. So starting SpaceX was really from the standpoint of, I think there's less than a 10% chance of being successful, maybe 1%. I don't know. But if a startup doesn't do something to advance rocket technology, it's definitely not coming from the big defense contractors because they just are matched to the government, and the government just wants to do very conventional things. So it's either coming from a startup or it's not happening at all. So a small chance of success is better than no chance of success. So yeah, I started SpaceX in mid-2002 expecting to fail. I said probably 90% chance of failing. Even when recruiting people, I didn't try to make out that it would succeed. I said we're probably going to die. But there's a 12% chance we might not die, and this is the only way to get people to Mars and advance the state-of-the-art. And then I ended up being chief engineer of the rocket, not because I wanted to, but because I couldn't hire anyone who was good. None of the good chief engineers would join because they thought it was too risky. So I ended up being chief engineer. The first three flights did fail. So it was a bit of a learning exercise. The fourth one fortunately worked. But if the fourth one hadn't worked, I had no money left, and that would have been curtains. So it was a pretty close thing. If the fourth launch of Falcon hadn't worked, it would have been just curtains, and we would have joined the graveyard of prior rocket startups. So my estimate of success was not far off. We made it by the skin of our teeth.

特斯拉差不多同时在进行。2008 年是艰难的一年。2008 年中，或者说夏天，SpaceX 第三次发射失败，连续第三次失败。特斯拉的融资轮也失败了。特斯拉很快就要破产了。情况很严峻。这本来会是一个警示故事，一次狂妄的尝试。那段时间，很多人说：‘埃隆是个软件 guy。他为什么做硬件？’他为什么选这个？对，100%。你可以看看当时的媒体，还在网上。他们一直叫我‘互联网 guy’。所以‘互联网 guy’ aka 傻瓜试图造火箭公司。我们被嘲笑得很多。听起来确实很荒谬：互联网 guy 开火箭公司，听起来不像成功的配方。我不怪他们。我也同意这不太可能。但幸运的是，第四次发射成功了，NASA 给了我们一个为空间站补给的合同。我记得大概是 12 月 22 号，圣诞节前。因为第四次发射成功还不够，我们还需要一个大合同来维持生存。所以我接到 NASA 团队的电话，他们说：‘我们决定给你一个空间站补给合同。’我直接脱口而出：‘我爱你们。’这通常不是他们听到的话。通常很严肃，但我说：‘天哪，这救了公司。’然后我们在最后一小时完成了特斯拉的融资轮，那是 2008 年 12 月 24 日下午 6 点。如果那轮融资没完成，我们圣诞节后两天就会发不出工资。所以 2008 年底真是紧张。

Tesla was happening sort of simultaneously. 2008 was a rough year. In mid-2008, or summer 2008, the third launch of SpaceX had failed, our third failure in a row. The Tesla financing round had failed. So Tesla was going bankrupt fast. It was grim. This was going to be a tale of warning, an exercise in hubris. Probably throughout that period, a lot of people were saying, 'Elon is a software guy. Why is he working on hardware?' Why would he choose to work on this? Right. 100%. You can look at the press of that time, still online. They kept calling me 'internet guy.' So 'internet guy' aka fool is attempting to build a rocket company. We got ridiculed quite a lot. It does sound pretty absurd: internet guy starts rocket company doesn't sound like a recipe for success. I don't hold it against them. I agreed that it's improbable. But fortunately, the fourth launch worked, and NASA awarded us a contract to resupply the space station. I think that was maybe December 22nd, right before Christmas. Because even the fourth launch working wasn't enough to succeed. We also needed a big contract to keep us alive. So I got that call from the NASA team, and they said, 'We're awarding you one of the contracts to resupply the space station.' I literally blurted out, 'I love you guys.' That's not normally what they hear. It's usually pretty sober, but I was like, 'Man, this is a company saver.' Then we closed the Tesla financing round on the last hour of the last day that it was possible, which was 6 p.m. December 24th, 2008. We would have bounced payroll two days after Christmas if that round hadn't closed. So that was a nerve-wracking end of 2008.

我想从你在 PayPal 和 Zip2 的经历来看，跳进这些硬核硬件初创公司，感觉有一条主线是能够找到并最终吸引那些领域里最聪明的人。你会对那个还没做过这些的埃隆说什么？

I guess from your PayPal and Zip2 experience, jumping into these hardcore hardware startups, it feels like one of the through lines was being able to find and eventually attract the smartest possible people in those particular fields. What would you tell to the Elon who's never had to do that yet?

我通常认为要尽量有用。这听起来可能老套，但要做到有用很难，尤其是对很多人有用。总效用的曲线下面积，就像你对同胞的有用程度乘以人数。这几乎像物理中‘真正工作’的定义。做到这一点极其困难。我认为如果你渴望做真正的工作，成功的概率会高得多。不要渴望荣耀，渴望工作。

I generally think to try to be as useful as possible. It may sound trite, but it's so hard to be useful, especially to be useful to a lot of people. The area under the curve of total utility is like how useful you have been to your fellow human beings times how many people. It's almost like the physics definition of true work. It's incredibly difficult to do that. And I think if you aspire to do true work, your probability of success is much higher. Don't aspire to glory, aspire to work.

我的意思是，就最终产品而言，你只需要问：如果这个东西成功了，它会对多少人有用？这就是我的意思。然后你无论担任 CEO 还是初创公司的任何角色，都会不惜一切代价去成功。而且，要不断粉碎你的自我，把责任内化。一个主要的失败模式是当自我与能力之比大于 1 的时候。如果你的自我与能力之比过高，你就会基本上打破与现实的反馈循环。用 AI 的术语来说，你会破坏你的强化学习循环。所以你想要一个强大的强化学习循环，这意味着内化责任并最小化自我。无论任务宏大还是卑微，你都要去做。这就是为什么我实际上更喜欢“工程”而不是“研究”这个词。我更喜欢那个术语。而且我实际上不想把 xAI 称为一个实验室。我只想让它成为一家公司。最直接、最简单、理想情况下最不带自我的术语，通常是好的选择。你想要紧紧关闭与现实的循环。这非常重要。

I guess it's in terms of your end product, you just have to say, well, if this thing is successful, how useful will it be to how many people? And that's what I mean. And then you do whatever, whether you're CEO or any role in a startup, you do whatever it takes to succeed. And just always be smashing your ego, like internalize responsibility. A major failure mode is when ego ability ratio is greater than sign one, you know? If your ego to ability ratio gets too high, then you're going to basically break the feedback loop to reality. In AI terms, you'll break your RL loop. So you want to have a strong RL loop, which means internalizing responsibility and minimizing ego. And you do whatever the task is, no matter whether it's grand or humble. That's kind of why I actually prefer the term engineering as opposed to research. I prefer that term. And I actually don't want to call xAI a lab. I just want to be a company. Like, whatever the simplest, most straightforward, ideally lowest ego terms are, those are generally a good way to go. You want to just close the loop on reality hard. That's a super big deal.

我想在座的每个人都对你所做的一切充满敬意，你堪称第一性原理的典范。考虑到你所做的事情，你实际上是如何确定你的现实的？因为这似乎是其中很重要的一部分。其他从未创造过任何东西的人，非工程师，有时是那些从未做过任何事情的记者，他们会批评你。但还有另一群人，他们是建设者，有着很高的曲线下面积，在你的圈子里。人们应该如何处理这个问题？对你来说什么有效？你会传给孩子们什么？比如，如何从第一性原理构建一个可预测的现实。

I think everyone in this room really looks up to everything you've done around being sort of a paragon of first principles. And thinking about the stuff you've done, how do you actually determine your reality? Because that seems like a pretty big part of it. Other people who have never made anything, non-engineers, sometimes journalists who've never done anything, they will criticize you. But then you have another set of people who are builders with very high area under the curve who are in your circle. How should people approach that? What has worked for you? What would you pass on to your children? Like, you know, here's how to construct a reality that is predictive from first principles.

嗯，物理学的工具对于理解和在任何领域取得进展都非常有帮助。第一性原理只是意味着将事物分解为最可能为真的基本公理元素，然后尽可能有说服力地从那里向上推理，而不是通过类比或隐喻推理。然后是一些简单的事情，比如极限思维：如果你外推，最小化这个东西或最大化那个东西，极限思维非常有帮助。我使用所有物理学的工具。它们适用于任何领域。这实际上是一种超能力。以火箭为例。你可以说，一枚火箭应该花多少钱？人们通常的做法是看历史上火箭的成本，并假设任何新火箭的成本必须与之前的成本有些相似。第一性原理的方法是看火箭由什么材料组成。如果是铝、铜、碳纤维、钢等等，然后说，那枚火箭有多重？它的组成元素是什么，它们有多重？这些组成元素每公斤的材料价格是多少？这设定了火箭成本的实际下限。它可以渐近地接近原材料的成本。然后你意识到，哦，实际上火箭的原材料成本只有历史成本的 1%或 2%。所以如果原材料成本只有 1%或 2%，那么制造过程必然非常低效。这就是对火箭成本优化潜力的第一性原理分析。这还没考虑可重复使用性。举一个 AI 的例子，去年 xAI 试图构建一个训练超级集群时，我们去找各种供应商，说我们需要 10 万块 H100 才能连贯地训练。他们估计完成需要 18 到 24 个月。我们说，我们需要在 6 个月内完成，否则我们将失去竞争力。所以如果你分解一下，你需要什么？你需要一栋建筑、电力、冷却。我们没有足够的时间从头开始建造一栋建筑，所以我们不得不找一栋现有的建筑。我们在孟菲斯找到了一家不再使用的工厂，以前生产伊莱克斯产品。但输入功率是 15 兆瓦，我们需要 150 兆瓦。所以我们租了发电机，把发电机放在建筑的一侧。然后我们需要冷却，所以我们租了美国大约四分之一的移动冷却能力，把冷水机组放在建筑的另一侧。这并没有完全解决问题，因为训练期间的电压和功率变化非常大。功率可以在 100 毫秒内下降 50%，发电机跟不上。所以我们增加了特斯拉 Megapack，并修改了 Megapack 中的软件，以平滑训练期间的功率变化。然后还有一大堆网络挑战，因为如果你试图让 10 万块 GPU 连贯地训练，网络电缆非常具有挑战性。

Well, the tools of physics are incredibly helpful to understand and make progress in any field. First principles just means breaking things down to the fundamental axiomatic elements that are most likely to be true, and then reasoning up from there as cogently as possible, as opposed to reasoning by analogy or metaphor. And then simple things like thinking in the limit: if you extrapolate, minimize this thing or maximize that thing, thinking in the limit is very helpful. I use all the tools of physics. They apply to any field. This is like a superpower actually. So take rockets for example. You can say, well, how much should a rocket cost? The typical approach people would take is to look historically at what the cost of rockets are and assume that any new rocket must be somewhat similar to the prior cost. A first principles approach would be to look at the materials the rocket is comprised of. If that's aluminum, copper, carbon fiber, steel, whatever, and say, how much does that rocket weigh? What are the constituent elements and how much do they weigh? What is the material price per kilogram of those constituent elements? That sets the actual floor on what a rocket can cost. It can asymptotically approach the cost of the raw materials. And then you realize, oh, actually the raw materials of a rocket are only maybe 1 or 2% of the historical cost of a rocket. So the manufacturing must necessarily be very inefficient if the raw material cost is only 1 or 2%. That would be a first principles analysis of the potential for cost optimization of a rocket. And that's before you get to reusability. To give an AI example, last year for xAI when we were trying to build a training supercluster, we went to various suppliers and said at the beginning of last year that we needed 100,000 H100s to be able to train coherently. Their estimates for how long it would take to complete that were 18 to 24 months. We said, well, we need to get that done in 6 months, or we won't be competitive. So if you break that down, what do you need? You need a building, power, cooling. We didn't have enough time to build a building from scratch, so we had to find an existing building. We found a factory that was no longer in use in Memphis that used to build Electrolux products. But the input power was 15 megawatts and we needed 150 megawatts. So we rented generators and had generators on one side of the building. Then we had to have cooling, so we rented about a quarter of the mobile cooling capacity of the US and put the chillers on the other side of the building. That didn't fully solve the problem because the voltage and power variations during training are very big. Power can drop by 50% in 100 milliseconds, which the generators can't keep up with. So we added Tesla Megapacks and modified the software in the Megapacks to smooth out the power variation during the training run. And then there were a bunch of networking challenges, because the networking cables if you're trying to make 100,000 GPUs train coherently are very challenging.

听起来你提到的几乎每一件事，我都能想象有人直接告诉你：不，你不能那样做，你不能有那样的电力，你不能有这个。而第一性原理思维的一个显著特点似乎是，让我们问为什么。让我们弄清楚，实际上挑战对面的人。如果我没有得到一个让我满意的答案，我不会让那个说法成立。这是不是每个人都应该做的？如果有人试图在硬件领域做你正在做的事情，硬件似乎特别需要这个。在软件领域，我们有很多花哨的东西，比如我们可以添加更多 CPU，那没问题。但在硬件领域，就是行不通。

It sounds like almost any of those things you mentioned, I could imagine someone telling you very directly, no, you can't have that, you can't have that power, you can't have this. And it sounds like one of the salient pieces of first principles thinking is actually let's ask why. Let's figure that out and actually challenge the person across the table. And if I don't get an answer that I feel good about, I'm not going to let that stand. Is that something that everyone, if someone were to try to do what you're doing in hardware, hardware seems to uniquely need this. In software, we have lots of fluff and things that, you know, it's like we can add more CPUs to that, it'll be fine. But in hardware, it's just not going to work.

我认为这些第一性原理思维的一般原则适用于软件和硬件，实际上适用于任何事物。

I think these general principles of first principle thinking apply to software and hardware, apply to anything really.

我举一个硬件的例子，说明我们被告知某件事不可能，但一旦分解成基本要素——我们需要建筑、电力、冷却、电力平滑——我们就能解决。然后我们四班倒 24/7 进行布线网络操作，我睡在数据中心，自己也动手布线。还有很多其他问题。去年还没有人用 10 万块 H100 进行过连贯的训练运行，也许今年有人做到了。后来我们翻倍到 20 万块。所以现在孟菲斯训练中心有 15 万块 H100、5 万块 H200 和 3 万块 GB200。我们即将在孟菲斯地区的第二个数据中心上线 11 万块 GB200。

I'm just using a hardware example of how we were told something is impossible, but once we broke it down into the constituent elements—we need a building, power, cooling, power smoothing—we could solve those. Then we ran the networking operation for cabling in four shifts 24/7, and I was sleeping in the data center and doing cabling myself. There were many other issues. Nobody had done a training run with 100,000 H100s training coherently last year. Maybe it's been done this year. Then we ended up doubling that to 200,000. So now we have 150,000 H100s, 50K H200s, and 30K GB200s in the Memphis training center. We're about to bring 110,000 GB200s online at a second data center also in the Memphis area.

您是否认为预训练仍然有效，缩放定律仍然成立，而赢得这场竞赛的人将拥有最大、最智能的可蒸馏模型？

Is it your view that pre-training is still working and the scaling laws still hold, and whoever wins this race will have the biggest smartest possible model that you could distill?

除了大型 AI 的竞争，人才也很重要。硬件规模以及你如何有效利用这些硬件也很关键。你不能只是订购一堆 GPU 然后插上电；你必须让大量 GPU 进行连贯稳定的训练。然后是你对数据的独特访问权。分发在一定程度上也很重要——人们如何接触到你的 AI。这些都是竞争性大型基础模型的关键因素。正如许多人所说，我的朋友伊利亚·苏茨克弗说过，我们基本上已经用完了人类生成的预训练数据；高质量 token 很快就会被耗尽。然后你必须创建合成数据，并准确判断它是否真实，还是与事实不符的幻觉。实现与现实的接轨很棘手，但我们正处于更多精力投入合成数据的阶段。现在我们正在训练 Grok 3.5，重点放在推理上。

Well, besides competitiveness for large AI, the talent of the people matters. The scale of the hardware matters and how well you can bring that hardware to bear. You can't just order a bunch of GPUs and plug them in; you have to get many GPUs and have them trained coherently and stably. Then it's about what unique access to data you have. Distribution matters to some degree—how people get exposed to your AI. Those are critical factors for a competitive large foundation model. As many have said, my friend Ilya Sutskever said we've kind of run out of pre-training data of human-generated data; you run out of tokens pretty fast, certainly of high-quality tokens. Then you have to create synthetic data and accurately judge it to verify if it's real or a hallucination that doesn't match reality. Achieving grounding in reality is tricky, but we're at the stage where more effort is put into synthetic data. Right now we're training Grok 3.5 with a heavy focus on reasoning.

回到你关于物理学的观点，我听说对于推理来说，硬科学，特别是物理教科书，非常有用，而社会科学对推理完全没用。

Going back to your physics point, what I heard for reasoning is that hard science, particularly physics textbooks, are very useful for reasoning, whereas social science is totally useless for reasoning.

是的，这很可能没错。未来非常重要的一件事是将数据中心或超级集群中的深度 AI 与机器人技术结合起来。例如，Optimus 人形机器人非常棒。将会有很多人形机器人和各种尺寸形状的机器人，但我的预测是，人形机器人的数量将远远超过所有其他机器人的总和，可能差一个数量级。

Yes, that's probably true. Something very important in the future is combining deep AI in the data center or supercluster with robotics. For example, the Optimus humanoid robot is awesome. There will be many humanoid robots and robots of all sizes and shapes, but my prediction is that there will be more humanoid robots by far than all other robots combined, maybe by an order of magnitude.

您是否真的在计划某种机器人军队？

Is it true that you're planning a robot army of a sort?

无论我们做还是特斯拉做，特斯拉与 xAI 密切合作。你已经看到了有多少人形机器人初创公司。黄仁勋曾与大量来自不同公司的机器人同台，大概有十几种人形机器人。我一直在抗争，也许也是让我放慢脚步的原因之一，就是我不想让终结者成为现实。所以直到最近几年，我一直在人工智能和人形机器人方面拖拖拉拉。然后我意识到，无论我做不做，它都会发生。你有两个选择：做旁观者或参与者。所以我宁愿做参与者。现在在人形机器人和数字超级智能上全力以赴。

Whether we do it or Tesla does it, Tesla works closely with xAI. You've seen how many humanoid robot startups there are. Jensen Huang was on stage with a massive number of robots from different companies, maybe a dozen different humanoid robots. Part of what I've been fighting, and maybe what has slowed me down, is that I don't want to make Terminator real. So I've been dragging my feet on AI and humanoid robotics until recent years. Then I realized it's happening whether I do it or not. You have two choices: be a spectator or a participant. So I'd rather be a participant. Now it's pedal to the metal on humanoid robots and digital superintelligence.

还有第三件事大家都听你谈过：成为多行星物种。你怎么看？所有事情都归结到这一点吗？未来 10 年、20 年和 100 年，是什么在驱动你？

There's a third thing everyone has heard you talk about: becoming a multiplanetary species. How do you think about it? Does everything feed into that last point? What drives you for the next 10, 20, and 100 years?

我希望 100 年后文明还在。如果还在，它会和今天大不相同。我预测人形机器人的数量至少是人类的五倍，可能是十倍。衡量进步的一种方式是卡尔达肖夫指数的完成百分比。第一级，你利用了一个行星的所有能量。在我看来，我们只利用了地球能量的 1%或 2%。所以距离卡尔达肖夫第一级还有很长的路。第二级是利用一个恒星的所有能量，这将是地球能量的十亿倍，也许接近万亿倍。第三级是利用一个星系的所有能量。我们正处于智能大爆炸的非常早期阶段。关于成为多行星物种，我认为大约 30 年内我们将向火星运送足够多的物质，使其能够自给自足，这样即使来自地球的补给船停止，火星也能继续发展。这将大大延长文明或意识（无论是生物还是数字）的寿命。这就是为什么我认为成为多行星物种很重要。费米悖论让我困扰——为什么我们还没有看到任何外星人？可能是因为智能极其罕见。也许我们是这个星系中唯一的智能。

I hope civilization is around in 100 years. If it is, it will look very different. I predict at least five times as many humanoid robots as humans, maybe ten times. One way to look at progress is percentage completion of the Kardashev scale. At scale one, you harness all the energy of a planet. In my opinion, we've only harnessed maybe 1 or 2% of Earth's energy. So we have a long way to go to Kardashev scale one. Scale two is harnessing all the energy of a sun, which would be a billion times more energy than Earth, maybe closer to a trillion. Scale three is all the energy of a galaxy. We're at the very early stage of the intelligence big bang. In terms of becoming multiplanetary, I think we'll have enough mass transferred to Mars within roughly 30 years to make it self-sustaining, so Mars can continue to grow even if resupply ships from Earth stop coming. That greatly increases the probable lifespan of civilization or consciousness, both biological and digital. That's why I think it's important to become a multiplanetary species. I'm troubled by the Fermi paradox—why haven't we seen any aliens? It could be because intelligence is incredibly rare. Maybe we're the only ones in this galaxy.

在这种情况下，意识的智能就像广阔黑暗中的一支小蜡烛，我们应该尽一切可能确保这支小蜡烛不会熄灭。成为多行星物种，或者说让意识遍布多个行星，能极大地延长文明的预期寿命，这是前往其他恒星系统之前的下一步。一旦你至少拥有两颗行星，你就有了改进太空旅行的推动力，最终这将引领意识扩展到星辰大海。

In which case the intelligence of consciousness is this tiny candle in a vast darkness, and we should do everything possible to ensure the tiny candle does not go out. Being a multiplanet species, or making consciousness multiplanetary, greatly improves the probable lifespan of civilization, and it's the next step before going to other star systems. Once you have at least two planets, you've got a forcing function for the improvement of space travel, and that ultimately will lead to consciousness expanding to the stars.

费米悖论可能表明，一旦技术发展到某个水平，你就会自我毁灭。我们如何避免这些大过滤器？

It could be that the Fermi paradox dictates once you get to some level of technology, you destroy yourself. How do we avoid the great filters?

一个大过滤器显然是全球热核战争，所以我们应该尽量避免。我猜是构建良性的 AI、热爱人类且乐于助人的机器人。构建 AI 时极其重要的一点是严格遵循真相，即使那个真相在政治上不正确。我的直觉是，如果强迫 AI 相信不真实的事情，那会让 AI 变得非常危险。

One of the great filters would obviously be global thermonuclear war, so we should try to avoid that. I guess building benign AI, robots that love humanity and are helpful. Something extremely important in building AI is a very rigorous adherence to truth, even if that truth is politically incorrect. My intuition for what could make AI very dangerous is if you force AI to believe things that are not true.

你怎么看待为了安全而开放与为了竞争优势而封闭？我认为好的一面是，你有竞争模型。许多其他人也有竞争模型。从这个意义上说，我们避开了最坏的时间线。我担心的那种时间线是快速起飞且只掌握在一个人手中。那可能会搞砸很多事情。而现在我们有选择，这很好。

How do you think about open for safety versus closed for competitive edge? I think the great thing is you have a competitive model. Many other people also have competitive models. In that sense, we're sort of off the worst timeline. The timeline I'd be worried about is fast takeoff and it's only in one person's hands. That might collapse a lot of things. Whereas now we have choice, which is great.

我确实认为会有几个深度智能，至少五个，可能多达十个。我不确定会有几百个，但可能接近十个左右，其中大概四个在美国。所以我不认为会有任何一个 AI 拥有失控的能力。但没错，会有几个深度智能。

I do think there will be several deep intelligences, maybe at least five, maybe as many as ten. I'm not sure there will be hundreds, but probably close to ten or something like that, of which maybe four will be in the US. So I don't think it's going to be any one AI that has a runaway capability. But yes, several deep intelligences.

这些深度智能实际上会做什么？是科学研究还是试图互相黑客攻击？

What will these deep intelligences actually be doing? Will it be scientific research or trying to hack each other?

可能以上所有。我是说，希望它们会发现新的物理学，而且我认为它们肯定会发明新技术。我觉得我们离数字超级智能非常近了。可能今年就会实现，如果今年不行，明年肯定——数字超级智能定义为在任何事情上比任何人类都聪明。

Probably all of the above. I mean, hopefully they will discover new physics, and I think they will definitely invent new technologies. I think we're quite close to digital superintelligence. It may happen this year, and if not this year, next year for sure—a digital superintelligence defined as smarter than any human at anything.

那么我们如何引导它走向超级富足？我们可以有机器人劳动力、廉价能源、按需智能。这是乐观的一面吗？你在这个光谱上处于什么位置？有没有切实可行的事情你会鼓励在座的每个人去做，让这个乐观情景成为现实？

So how do we direct that to super abundance? We could have robotic labor, cheap energy, intelligence on demand. Is that the white pill? Where do you sit on the spectrum? Are there tangible things you would encourage everyone here to work on to make that white pill reality?

我认为最可能的结果是好的。我有点同意杰夫·辛顿的观点，也许有 10%到 20%的毁灭概率。但往好处想，那就是 80%到 90%的概率是好的结果。我再怎么强调也不为过：严格遵循真相是 AI 安全最重要的事情，当然还有对人性以及我们所知生命的同理心。

I think it most likely will be a good outcome. I'd sort of agree with Geoff Hinton that maybe it's a 10 to 20% chance of annihilation. But look on the bright side, that's 80 to 90% probability of a great outcome. I can't emphasize this enough: a rigorous adherence to truth is the most important thing for AI safety, and obviously empathy for humanity and life as we know it.

我们还没谈到 Neuralink。缩小人类与机器之间的输入输出差距对 AGI 和 ASI 有多关键？一旦这个连接建立，我们不仅能读取还能写入吗？

We haven't talked about Neuralink yet. How critical is closing the input-output gap between humans and machines to AGI and ASI? Once that link is made, can we not only read but also write?

Neuralink 对于解决数字超级智能并非必需；那会在 Neuralink 规模化之前发生。但 Neuralink 能有效解决的是输入输出带宽的限制。尤其是我们的输出带宽非常低。人类一天内的持续输出不到每秒 1 比特。有了 Neuralink 接口，你可以大幅增加输出和输入带宽。输入是指对大脑的写入操作。我们现在有五个人接受了读取输入，即读取信号。患有 ALS 的四肢瘫痪者现在可以以与功能健全人类相似的带宽进行交流，并控制他们的电脑和手机，这很酷。在未来 6 到 12 个月内，我们将进行首次视觉植入，即使某人完全失明，我们也可以直接写入视觉皮层。我们已经在猴子身上实现了这一点；我们的一只猴子已经植入视觉装置三年了。起初分辨率会相对较低，但长期来看，你将拥有非常高的分辨率，并能看到多光谱波长——红外线、紫外线、雷达——就像超能力一样。在某个时候，赛博植入物将不仅仅是纠正问题，而是大幅增强人类能力——增强智能、感官和带宽。这将在某个时候发生。但数字超级智能会在此之前很久就出现。至少如果我们有 Neuralink，我们或许能更好地欣赏 AI。

Neuralink is not necessary to solve digital superintelligence; that will happen before Neuralink is at scale. But what Neuralink can effectively do is solve the input-output bandwidth constraints. Especially our output bandwidth is very low. The sustained output of a human over a day is less than one bit per second. With a Neuralink interface, you can massively increase your output and input bandwidth. Input being write operations to the brain. We now have five humans who have received the read input, where it's reading signals. People with ALS who are tetraplegics can now communicate at similar bandwidth to a human with a fully functioning body and control their computer and phone, which is pretty cool. In the next 6 to 12 months, we'll be doing our first implants for vision, where even if somebody is completely blind, we can write directly to the visual cortex. We've had that working in monkeys; one of our monkeys has had a visual implant for three years. At first, it will be relatively low resolution, but long term you would have very high resolution and be able to see multispectral wavelengths—infrared, ultraviolet, radar—like a superpower situation. At some point, cybernetic implants wouldn't simply be correcting things that went wrong but augmenting human capabilities dramatically—augmenting intelligence, senses, and bandwidth. That's going to happen at some point. But digital superintelligence will happen well before that. At least if we have a Neuralink, we might be able to appreciate the AI better.

你在所有这些领域努力的一个限制因素是接触到最聪明的人。但与此同时，石头也能说话和推理——它们现在可能有 130 的智商，而且很快可能变得超级智能。你如何调和这两件事？未来 5 到 10 年会发生什么，在座的各位应该做什么来确保他们是创造者，而不是在 API 线之下？

One of the limiting reagents to all your efforts across all these domains is access to the smartest possible people. But simultaneously, the rocks can talk and reason—they're maybe 130 IQ now and probably super intelligent soon. How do you reconcile those two things? What's going to happen in 5 to 10 years, and what should the people in this room do to make sure they're the ones creating instead of below the API line?

他们称之为奇点是有原因的，因为我们在不远的将来不知道会发生什么。人类智能所占的比例将非常小。在某个时候，人类智能的总和将不到所有智能的 1%。

They call it the singularity for a reason, because we don't know what's going to happen in the not-too-distant future. The percentage of intelligence that is human will be quite small. At some point, the collective sum of human intelligence will be less than 1% of all intelligence.

嗯，如果事情发展到遏制层级二，我们谈论的是人类智能，即使假设人口大幅增长和智能增强——比如大规模智能增强，每个人智商都达到一千。即使在那种情况下，集体人类智能可能也只有数字智能的十亿分之一。总之，数字超级智能的生物引导程序在哪里？我想最后问一下，我是个好的引导程序吗？我们该往哪里去？我们如何从这里出发？我的意思是，这一切都是相当疯狂的科幻内容，但也可能由这个房间里的人建造。你知道，如果你对当代最聪明的技术人员有什么临别赠言，他们应该做什么？他们应该致力于什么？他们今晚去吃晚餐时应该思考什么？

Um, and if things get to a cauter ship level two, we're talking about human intelligence even assuming a significant increase in human population and intelligence augmentation like massive intelligence augmentation where everyone has an IQ of a thousand. Even in that circumstance, collective human intelligence will probably be 1 billionth that of digital intelligence. Anyway, where's the biological bootloader for digital superintelligence? I guess just to end off, was I a good bootloader? Where do we go? How do we go from here? I mean, all of this is pretty wild sci-fi stuff that also could be built by the people in this room. You know, if you have a closing thought for the smartest technical people of this generation right now, what should they be doing? What should they be working on? What should they be thinking about tonight as they go to dinner?

嗯，就像我一开始说的，我认为如果你在做有用的事情，那就很好。如果你只是尽力对你的同胞有用，那你就在做好事。我一直强调这一点：专注于超级诚实的 AI。这是 AI 安全最重要的事情。显然，如果有人有兴趣在 xAI 工作，请告诉我们。我们的目标是让 Grok 成为最追求真理的 AI。我认为这是一件非常重要的事情。希望我们能理解宇宙的本质。这真的是 AI 可以告诉我们的。也许 AI 可以告诉我们外星人在哪里，宇宙究竟如何开始，如何结束，有哪些我们不知道但应该问的问题，以及我们是否在模拟中，或者处于什么层次的模拟中？嗯，我想我们会找到答案的。

Well, as I started off with, I think if you're doing something useful, that's great. If you just try to be as useful as possible to your fellow human beings, then you're doing something good. I keep harping on this: focus on super truthful AI. That's the most important thing for AI safety. Obviously, if anyone's interested in working at xAI, please let us know. We're aiming to make Grok the maximally truth-seeking AI. I think that's a very important thing. Hopefully we can understand the nature of the universe. That's really what AI can hopefully tell us. Maybe AI can tell us where are the aliens, how did the universe really start, how will it end, what are the questions that we don't know that we should ask, and are we in a simulation or what level of simulation are we in? Well, I think we're going to find out.

一个 NPC。埃隆，非常感谢你加入我们。各位，请为埃隆·马斯克鼓掌。

An NPC. Elon, thank you so much for joining us. Everyone, please give it up for Elon Musk.