Drinking coffee or water?

喝咖啡还是水？

Sure.

好啊。

This supposed to keep you up for like tomorrow afternoon, basically.

这玩意儿基本上能让你撑到明天下午不困。

So what is Nitro? It’s just got a lot of caffeine or something?

那Nitro到底是什么？就是咖啡因特别多还是怎么？

It’s got nitrogen in it. That’s ridiculous. What we breathe is 78% nitrogen anyway. What do you need to add more for?

里面含氮。太荒唐了。我们呼吸的空气本来就有78%是氮，还加它干嘛？

Unfortunately, you’re going to eat it.

可惜啊，你还是得把它喝下去。

Most people think that they’re breathing oxygen and they’re actually breathing 78% nitrogen. You need like a milk bar, like from Clockwork Orange.

大多数人以为自己吸的是氧气，其实吸的78%是氮。你需要的是《发条橙》里那种牛奶吧。

Clockwork Orange? It’s pretty good. It’s demented. Jarring, I’d say.

《发条橙》？挺不错的。挺变态的。要我说，很刺人。

Thanks. Yeah.

谢谢。是的。

Yeah. And we just obviously have our second implant as well.

对。而且我们显然已经完成了第二例植入。

So far, so good. It looks like we’ve got, I think, on the order of 400 electrodes that are providing signals.

目前一切顺利。看起来大概有400个电极在提供信号。

Yeah.

是的。

It depends somewhat on the regulatory approval, the rate at which we get regulatory approvals. So, we’re hoping to do 10 by the end of this year, total of 10. So, eight more.

这多少取决于监管审批，取决于我们拿到审批的速度。我们希望今年年底前做到10例，总共10例。也就是再来8例。

Yeah. Yeah. I think it’s obviously going to get better with each one. I don’t want to jinx it, but it seems to have gone extremely well with the second implant. So, there’s a lot of signal, a lot of electrodes. It’s working very well.

对。对。我觉得显然每一例都会更好。我不想乌鸦嘴，但第二例似乎进展得极其顺利。信号很多，电极很多。运行得非常好。

In years, it’s going to be gigantic, because we’ll increase the number of electrodes dramatically. We’ll improve the signal processing. So, even with only roughly, I don’t know, 10, 15% of the electrodes working with Noland, with our first patient, we were able to get to achieve a bit per second. That’s twice the world record. So, I think we’ll start vastly exceeding the world record by orders of magnitude in the years to come. So, start getting to, I don’t know, 100 bits per second, thousand. Maybe if five years from now, we might be at a megabit, faster than any human could possibly communicate by typing, or speaking.

按年看，那会是巨大的，因为我们会大幅增加电极数量，改进信号处理。即便在我们第一位患者Noland身上只有大概10%到15%的电极工作，我们也能达到每秒1比特，是世界纪录的两倍。所以我认为未来几年我们会开始以数量级远超世界纪录。开始达到每秒100比特、1000比特。也许五年后，我们可能达到每秒1兆比特，比任何人打字或说话能达到的交流速度都快。

Yeah.

对。

And with humans.

还有与人交互。

Provided they have want a Neuralink, too.

前提是他们也得想要一个 Neuralink。

Otherwise they wont be able to absorb the signals fast enough.

否则他们无法足够快地吸收这些信号。

Well, I think you could think of it, if you were to slow down communication, how do you feel about that? If you’d only talk at, let’s say one-tenth of normal speed, you’d be like, “Wow, that’s agonizingly slow.”

嗯，我觉得你可以这样想：如果把交流速度放慢，你会是什么感受？假设你只能用正常速度的十分之一说话，你会觉得“哇，这慢得让人难受”。

So, now imagine you could communicate clearly at 10, or 100, or 1,000 times faster than normal.

那么现在想象一下，你能以正常速度的10倍、100倍甚至1000倍清晰地交流。

I usually default to 1.5X. You can do 2X. Well actually, if I’m listening to somebody get to… in 15, 20 minutes, I want to go to sleep, then I’ll do it 1.5X. If I’m paying attention, I’ll do 2X.

我通常默认用1.5倍速。你可以用2倍速。其实，如果我听某人讲话要听15、20分钟就想睡觉了，那我就用1.5倍速。如果我全神贯注，我就用2倍速。

But actually, if you actually listen to podcasts, or audiobooks or anything at… If you get used to doing it at 1.5, then one sounds painfully slow.

但说真的，如果你真的去听播客、有声书之类的，一旦你习惯了1.5倍速，那1倍速听起来就慢得难受。

Well, it depends on the person. You can speak very fast. Like we can communicate very quickly. And also, if you use a wide range of… if your vocabulary is larger, your effective bit rate is higher.

嗯，这因人而异。你可以说得非常快。比如我们可以交流得非常快。而且，如果你使用更广泛的……如果你的词汇量更大，你的有效比特率就更高。

Yeah.

是的。

Yeah. If there’s a single word that is able to convey something that would normally require, I don’t know, 10 simple words, then you’ve got maybe a 10X compression on your hands. And that’s really like with memes. Memes are like data compression. You’re simultaneously hit with a wide range of symbols that you can interpret, and you get it faster than if it were words, or a simple picture.

是的。如果单个词能传达通常需要，我不知道，10个简单词才能表达的东西，那你手上可能就有了10倍压缩。这其实就像梗图。梗图就像数据压缩。你同时被大量可解读的符号击中，理解起来比用文字或简单图片更快。

Yeah. There’s an entire idea structure that is like an idea template, and then you can add something to that idea template. But somebody has that pre-existing idea template in their head. So, when you add that incremental bit of information, you’re conveying much more than if you just said a few words. It’s everything associated with that meme.

是的。存在一整套观念结构，就像一个观念模板，然后你可以往这个模板上添加东西。但对方脑子里已经有那个现成的观念模板。所以当你加上那一点点增量信息时，你传达的东西远比只说几个词要多。那是与那个梗相关的一切。

Yeah.

是的。

Yes.

是的。

Yeah. Certainly, if you’re anywhere at 10,000 bits per second, that’s vastly faster than any human can communicate right now. If you think what is the average bits per second of a human, it is less than one bit per second over the course of a day. Because there are 86,400 seconds in a day, and you don’t communicate 86,400 tokens in a day. Therefore, your bits per second is less than one, averaged over 24 hours. It’s quite slow.

是的。当然，如果你达到每秒1万比特，那已经远快于目前任何人类的交流速度。想想人类的平均比特率是多少，按一天算还不到每秒1比特。因为一天有86400秒，而你一天并不会传达86400个词元。因此，按24小时平均，你的比特率不到1。这相当慢。

Yeah. Very lossy, compression, and decompression. And a lot of what your neurons are doing is distilling the concepts down to a small number of symbols, or say syllables that I’m speaking, or keystrokes, whatever the case may be. So, that’s a lot of what your brain computation is doing. Now, there is an argument that that’s actually a healthy thing to do, or a helpful thing to do because as you try to compress complex concepts, you’re perhaps forced to distill what is most essential in those concepts, as opposed to just all the fluff. So, in the process of compression, you distill things down to what matters the most, because you can only say a few things.

是的。非常有损的压缩和解压。你的神经元所做的很多工作就是把概念提炼成少量符号，比如我说的音节，或按键，不管是什么形式。所以你的大脑计算很大一部分在做这个。现在有一种观点认为这其实是健康有益的，因为当你试图压缩复杂概念时，你也许被迫提炼出其中最关键的部分，而不是所有废话。所以在压缩过程中，你把东西提炼到最重要的部分，因为你只能说那么几句。

I think there is some argument for humans as a source of will.

我认为有一种论点认为人类是意志的来源。

Will, yeah. Source of will, or purpose. So if you consider the human mind as being… Essentially there’s the primitive, limbic elements, which basically even reptiles have. And there’s the cortex, the thinking and planning part of the brain. Now, the cortex is much smarter than the limbic system, and yet is largely in service to the limbic system. It’s trying to make the limbic system happy. The sheer amount of compute that’s gone into people trying to get laid is insane, without actually seeking procreation. They’re just literally trying to do this simple motion, and they get a kick out of it. So, this simple, which in the abstract, rather absurd motion, which is sex, the cortex is putting a massive amount of compute into trying to figure out how to do that.

意志，对。意志或目标的来源。所以如果你把人类心智看作……本质上存在原始的边缘系统元素，连爬行动物都有。然后还有皮层，大脑中思考和规划的部分。皮层比边缘系统聪明得多，却很大程度上服务于边缘系统。它努力让边缘系统开心。人们为了做爱投入的算力多得离谱，甚至不是为了繁衍。他们就是literally想完成这个简单动作，并从中获得快感。所以这个简单的、抽象来看相当荒谬的动作，也就是性，皮层投入了海量算力去琢磨怎么做。

A massive amount. Yes. Yeah. Yeah. There’s no purpose to most sex except hedonistic. It’s a sort of joy, or whatever, dopamine release. Now, once in a while, it’s procreation, but for modern humans, it’s mostly recreational. And so, your cortex, much smarter than your limbic system, is trying to make the limbic system happy, because the limbic system wants to have sex, or wants some tasty food, or whatever the case may be.

海量。是的。对。大多数性行为除了享乐没有别的目的。是一种快乐，或者随便什么，多巴胺释放。偶尔是繁衍，但对现代人来说，主要是娱乐性的。所以你的皮层，比边缘系统聪明得多，却在努力让边缘系统开心，因为边缘系统想做爱，或者想吃好吃的，或者别的什么。

Yeah. There’s like gigawatts of compute going into getting laid, of digital compute.

对。有吉瓦级的算力，数字算力，正投入在泡妞这件事上。

This is happening as we speak.

这事此刻正在发生。

Pretty much.

差不多吧。

Well, it’s one of the things. Certainly, yeah.

嗯，那是其中一件事。当然，是的。

But what I’m saying is that, yes, is there a use for humans? Well, there’s this fundamental question of what’s the meaning of life? Why do anything at all? And so, if our simple limbic system provides a source of will to do something, that then goes through our cortex, that then goes to our tertiary compute layer, then I don’t know, it might actually be that the AI, in a benign scenario, is simply trying to make the human limbic system happy.

但我想说的是，是的，人类还有没有用？嗯，有个根本问题：生命的意义是什么？为什么要做任何事？所以，如果我们简单的边缘系统提供了做事的意志来源，它经过皮层，再传到第三层算力，那我不知道，也许在良性情景下，AI真的只是在努力让人类的边缘系统开心。

That’s limbic too, I think.

我觉得那也是边缘系统。

Not everybody does. But yeah, sure, some people do.

不是所有人都这样。但没错，有些人确实如此。

Sure.

当然。

Yeah. There are these fairly cerebral, or higher level goals. For me, it’s like, what’s the meaning of life, or understanding the nature of the universe, is of great interest to me, and hopefully to the AI. And that’s the mission of xAI and Grok is understand the universe.

对。存在这些相当理性的、更高层次的目标。对我来说，比如生命的意义是什么，或者理解宇宙的本质，我非常有兴趣，希望AI也是。这正是xAI和Grok的使命：理解宇宙。

Well, assuming that there are not… They’re solving basic neurological issues that people have. If they’ve got damaged neurons in their spinal cord, or neck, as is the case with our first two patients, then obviously the first order of business is solving fundamental neuron damage in a spinal cord, neck, or in the brain itself. So, our second product is called Blindsight, which is to enable people who are completely blind, lost both eyes, or optic nerve, or just can’t see at all, to be able to see by directly triggering the neurons in the visual cortex.

嗯，前提是没有……它们正在解决人们的基本神经问题。如果他们的脊髓或颈部神经元受损，就像我们前两位患者那样，那显然首要任务是解决脊髓、颈部或大脑本身的神经元损伤。所以我们的第二款产品叫Blindsight，让完全失明、失去双眼或视神经、或根本看不见的人，能通过直接刺激视觉皮层神经元来看见。

So, do you have letters and the alphabet? Okay, great. Words? And then eventually you get sagas. So, I think there may be some things to worry about in the future, but the first several years are really just solving basic neurological damage, like for people who have essentially complete or near complete loss from the brain to the body, like Stephen Hawking would be an example, the Neuralink would be incredibly profound, because you can imagine if Stephen Hawking could communicate as fast as we’re communicating, perhaps faster. And that’s certainly possible. Probable, in fact. Likely, I’d say.

那么，你们有字母和字母表吗？好，太好了。单词？然后最终你会得到传奇故事。所以，我认为未来可能有一些事情需要担心，但最初几年真的只是解决基本的神经损伤，比如对于那些从大脑到身体基本完全或接近完全丧失功能的人，斯蒂芬·霍金就是一个例子，Neuralink会非常深刻，因为你可以想象如果斯蒂芬·霍金能像我们交流一样快，也许更快。这当然是可能的。事实上，很可能。我会说，很有可能。

The logical thing to do is… Sensible thing to do is to start off solving basic neuron damage issues.

合乎逻辑的做法是……明智的做法是先从解决基本神经元损伤问题开始。

Because there’s obviously some risk with a new device. You can’t get the risk down to zero, it’s not possible. So, you want to have the highest possible reward, given there’s a certain irreducible risk. And if somebody’s able to have a profound improvement in their communication, that’s worth the risk.

因为新设备显然存在一些风险。你无法把风险降到零，这不可能。所以，考虑到存在某种不可消除的风险，你希望获得尽可能高的回报。如果有人能够在其沟通能力上获得深刻改善，那值得冒这个风险。

Yeah. As you get the risk down. And once the risk is down to… If you have thousands of people that have been using it for years and the risk is minimal, then perhaps at that point you could consider saying, “Okay, let’s aim for augmentation.” Now, I think we’re actually going to aim for augmentation with people who have neuron damage. So we’re not just aiming to give people the communication data rate equivalent to normal humans. We’re aiming to give people who have… A quadriplegic, or maybe have complete loss of the connection to the brain and body, a communication data rate that exceeds normal humans. While we’re in there, why not? Let’s give people superpowers.

是的。随着风险降低。一旦风险降到……如果有成千上万人已经使用多年，风险极小，那么也许到那时你可以考虑说：“好，让我们以增强为目标。”现在，我认为我们实际上会以有神经元损伤的人为增强目标。所以我们不只是想给人们相当于正常人类的通信数据速率。我们想给那些……四肢瘫痪的人，或者可能完全丧失大脑与身体连接的人，超过正常人类的通信数据速率。既然我们已经在做了，为什么不呢？让我们给人们超能力。

Yeah. At first, the vision restoration will be low res, because you have to say, “How many neurons can you put in there, and trigger?” And you can do things where you adjust the electric field. So, even if you’ve got, say 10,000 neurons, it’s not just 10,000 pixels, because you can adjust the field between the neurons, and do them in patterns in order to have say, 10,000 electrodes, effectively give you, I don’t know, maybe like having a megapixel, or a 10 megapixel situation. And then over time, I think you get to higher resolution than human eyes. And you could also see in different wavelengths. So, like Geordi La Forge from Star Trek, he had the thing. Do you want to see it in radar? No problem. You could see ultraviolet, infrared, eagle vision, whatever you want.

是的。起初，视力恢复会是低分辨率的，因为你必须说：“你能在那里放多少神经元并触发？”你可以通过调整电场来做一些事情。所以，即使你有比如说1万个神经元，那也不只是1万像素，因为你可以在神经元之间调整电场，并以模式方式进行，以便比如说1万个电极，实际上给你，我不知道，也许像百万像素或千万像素的情况。然后随着时间的推移，我认为你会达到比人眼更高的分辨率。你还可以看到不同波长。就像《星际迷航》里的乔迪·拉福吉，他有那个东西。你想用雷达看？没问题。你可以看到紫外线、红外线、鹰眼视觉， whatever you want。

Is that a serious question?

这是个严肃的问题吗？

Well, I guess technically it is.

嗯，我想严格来说是的。

Yeah.

是的。

Yeah, is this DMT in there, or something?

是啊，这里面是有DMT还是什么？

Wait, wait. Have you said much about it, the ayahuasca stuff?

等等，等等。你关于死藤水的事说过很多吗？

Okay. Well, why don’t you spill the beans?

好吧。那你为什么不爆料一下？

Let me turn the tables on you.

让我反客为主。

You’re in the jungle.

你在丛林里。

Yeah. It must’ve been crazy.

是啊。那一定很疯狂。

Things are going to look pretty wild.

事情会看起来相当狂野。

Just don’t go hugging an Anaconda or something.

只是别去抱什么水蟒之类的就行。

Snakes and Ladders.

蛇和梯子。

Okay.

好吧。

Damn. Okay. That sounds like a lot. Is normal to just one cup? Or…

我去。好吧。这听起来量很大。一般就一杯吗？还是……

Okay. Wait. Like right off the bat, or did you work your way up to it? Did you just jump in at the deep end?

好吧。等等。是一上来就这样，还是你慢慢加量上去的？你是直接跳进深水区吗？

It wasn’t like a revelation.

不是那种启示性的体验。

Nine cups, you went to Pluto, I think.

九杯，我看你是飞到冥王星了。

Deep space.

深空。

That’s what people [inaudible 00:22:26].

大家都这么说。

That’s what everyone says. Yeah, exactly.

大家都这么说。对，没错。

Just a pure soul.

就是个纯洁的灵魂。

That’s cool.

挺酷的。

That sounds awesome.

听起来很棒。

When’s the demon coming?

心魔什么时候来？

Space travel to where?

太空旅行去哪儿？

Okay.

好的。

Did you go past the Milky Way?

你飞过银河系了吗？

Okay. You’re like intergalactic.

好吧。你简直是星际级的。

Okay. Dang.

好吧。天哪。

Sure. Did you see an alien?

当然。你看到外星人了吗？

Zero aliens?

一个外星人都没有？

Sounds good.

听起来不错。

Was there anything scary at all?

有什么吓人的东西吗？

Sure. Like Puff the Magic Dragon.

当然。就像魔法龙帕夫。

Yeah. I mean, the jungle’s kind of scary.

是啊。我是说，丛林挺吓人的。

Sure. Okay.

当然。好吧。

And you were way out in the jungle.

而且你深入丛林里。

Like 10 miles outside of Rio or something.

像里约郊外十英里那种？

You’re in deep Amazon.

你在亚马逊深处。

Wow. Cool.

哇。酷。

Probably. Yeah.

大概吧。是的。

Yeah, I mean, Neuralink, it’s really a generalized input-output device. It’s reading electrical signals, and generating electrical signals, and I mean, everything that you’ve ever experienced in your whole life, smell, emotions, all of those are electrical signals. So, it’s kind of weird to think that your entire life experience is distilled down to electrical signals for neurons, but that is in fact the case. Or I mean, that’s at least what all the evidence points to. So, I mean, if you trigger the right neuron, you could trigger a particular scent. You could certainly make things glow. I mean, do pretty much anything. I mean, really, you can think of the brain as a biological computer. So, if there are certain say, chips or elements of that biological computer that are broken, let’s say your ability to… If you’ve got a stroke, that if you’ve had a stroke, that means some part of your brain is damaged. Let’s say it’s speech generation, or the ability to move your left hand. That’s the kind of thing that a Neuralink could solve.

是的，我是说，Neuralink 本质上是一个通用的输入输出设备。它读取电信号，也生成电信号，我的意思是，你这辈子经历的一切，气味、情绪，所有这些都是电信号。所以，想想你整个生命体验都被浓缩成神经元的电信号，这有点奇怪，但事实确实如此。或者说，至少所有证据都指向这一点。所以，如果你触发正确的神经元，你就能触发某种特定的气味。你当然可以让东西发光。基本上什么都能做。真的，你可以把大脑看作一台生物计算机。所以，如果这台生物计算机的某些芯片或元件坏了，比如说你的能力……如果你中风了，中风意味着你大脑的某部分受损。比如说是语言生成能力，或者移动左手的能力。这就是 Neuralink 能解决的问题。

Yeah, you could say like, create the most probable set of memories based on all the information you have about that person. You could then… It would be probabilistic restoration of memory. Now, we’re getting pretty esoteric here.

是的，你可以说，根据你掌握的关于那个人的所有信息，创建最可能的一组记忆。然后你可以……这将是概率性的记忆恢复。不过，我们现在聊得有点太玄了。

Sure.

当然。

Yeah. Well, I mean, what are we but our memories? And what is death? But the loss of memory, loss of information? If you could say, well, if you could run a thought experiment, what if you were disintegrated painlessly, and then reintegrated a moment later, like teleportation, I guess? Provided there’s no information loss, the fact that your one body was disintegrated is irrelevant.

是的。嗯，我是说，我们不就是我们的记忆吗？死亡又是什么？不就是记忆的丧失、信息的丧失吗？你可以说，做个思想实验，如果你被无痛地分解，然后片刻后重组，就像传送一样？只要没有信息丢失，你原来的身体被分解这件事就无关紧要。

Death is fundamentally the loss of information, the loss of memory.

死亡从根本上就是信息的丧失，记忆的丧失。

Yeah.

是的。

It’s an idea that may help with AI safety. Certainly, I wouldn’t want to claim it’s some panacea, or that it’s a sure thing, but I mean, many years ago I was thinking like, “Well, what would inhibit alignment of collective human will with artificial intelligence?” And the low data rate of humans, especially our slow output rate would necessarily, just because the communication is so slow, would diminish the link between humans and computers. The more you are a tree, the less you know what the tree is. Let’s say you look at this plant or whatever, and hey, I’d really like to make that plant happy, but it’s not saying a lot.

这是一个可能有助于 AI 安全的想法。当然，我不想声称它是什么万灵药，或者它是板上钉钉的事，但我的意思是，很多年前我在想，'嗯，什么会阻碍集体人类意志与人工智能的对齐？'人类的数据速率低，尤其是我们缓慢的输出速率，必然会因为通信太慢而削弱人与计算机之间的连接。你越像一棵树，就越不知道这棵树是什么。比如说你看这株植物什么的，嘿，我真的很想让那株植物开心，但它表达不了什么。

Yeah. We could better align collective human will with AI if the output rate especially was dramatically increased. And I think there’s potential to increase the output rate by, I don’t know, three, maybe six, maybe more orders of magnitude. So, it’s better than the current situation.

是的。如果输出速率尤其能大幅提高，我们就能更好地让集体人类意志与 AI 对齐。我认为输出速率有可能提高，我不知道，三个、也许六个、甚至更多个数量级。所以，这比现状要好。

Yeah.

是的。

Yeah, I do.

是的，我认为会。

Yeah. If it’s extremely safe, and you can have superhuman abilities, and let’s say you can upload your memories, so you wouldn’t lose memories, then I think probably a lot of people would choose to have it. It would supersede the cell phone, for example. I mean, the biggest problem that say, a phone has, is trying to figure out what you want. That’s why you’ve got auto complete, and you’ve got output, which is all the pixels on the screen, but from the perspective of the human, the output is so frigging slow. Desktop or phone is desperately just trying to understand what you want. And there’s an eternity between every keystroke from a computer standpoint.

是的。如果它极其安全，而且你能拥有超人能力，比如说你能上传记忆，这样就不会丢失记忆，那我认为可能很多人会选择植入。它会取代手机，比如说。手机最大的问题就是试图搞明白你想要什么。这就是为什么你有自动补全，你有输出，就是屏幕上所有像素，但从人的角度看，输出慢得要命。台式机或手机拼命想搞懂你想要什么。从计算机的角度看，每次按键之间都是永恒。

Yeah. So, if you had computers that are doing trillions of instructions per second, and a whole second went by, I mean, that’s a trillion things it could have done.

是的。所以，如果计算机每秒执行数万亿条指令，整整一秒过去了，我的意思是，它本可以完成一万亿件事。

Yeah. We would be something different. I mean, some sort of futuristic cyborg, I mean, we’re obviously talking about, by the way, it’s not like around the corner. You asked me what the distant future is. Maybe this is… It’s not super far away, but 10, 15 years, that kind of thing.

是的。我们会变成另一种东西。我是说，某种未来主义的赛博格，我的意思是，我们显然在谈论，顺便说一下，这不是近在眼前的事。你问我遥远的未来是什么。也许这个……也不算太远，但 10 年、15 年那种。

Probably less than 10 years. It depends on what you want to do.

可能不到 10 年。取决于你想做什么。

A thousand BPS, wow.

每秒一千比特，哇。

Well, we feel pretty confident that I think maybe within the next year or two, that someone with a Neuralink implant will be able to outperform a pro gamer.

嗯，我们相当有信心，可能在未来一两年内，植入 Neuralink 的人就能击败职业电竞选手。

Because the reaction time would be faster.

因为反应时间会更快。

Yeah. Yeah.

对，对。

Yeah.

对。

Yeah.

对。

For AI, that means you’ve got to have the most powerful training compute, and the rate of improvement of training compute has to be-

对 AI 来说，这意味着你必须拥有最强大的训练算力，而且训练算力的提升速度必须——

And the rate of improvement of training compute has to be faster than everyone else, or you will not win. Your AI will be worse.

而且训练算力的提升速度必须比所有人都快，否则你赢不了。你的 AI 会更差。

Well, hopefully end of this year.

嗯，希望今年年底。

If we’re lucky. Yeah.

如果运气好的话。对。

I mean, they all matter. It’s sort of like saying, let’s say it’s a Formula 1 race, what matters more, the car or the driver? I mean, they both matter. If a car is not fast, then if, let’s say, it’s half the horsepower of your competitors, the best driver will still lose. If it’s twice the horsepower, then probably even a mediocre driver will still win. So, the training compute is kind of like the engine, this horsepower of the engine. So, really, you want to try to do the best on that. And then, it’s how efficiently do you use that training compute, and how efficiently do you do the inference, the use of the AI? So, obviously, that comes down to human talent. And then, what unique access to data do you have? That also plays a role.

我是说，这些都很重要。这就好比说，假设这是一场 F1 比赛，车和车手哪个更重要？两者都重要。如果车不够快，比如说马力只有对手的一半，最好的车手也还是会输。如果马力是两倍，那大概连平庸的车手也能赢。所以训练算力有点像引擎，引擎的马力。所以你真的要在这方面做到最好。然后就是你用训练算力的效率如何，推理、使用 AI 的效率如何。所以，这显然取决于人才。然后，你对数据有什么独特的访问权限？这也起作用。

Yeah. I mean, I think most of the leading AI companies have already scraped all the Twitter data. Not I think. They have. So, on a go forward basis, what’s useful is the fact that it’s up to the second, because that’s hard for them to scrape in real time. So, there’s an immediacy advantage that Grok has already. I think with Tesla and the real time video coming from several million cars, ultimately tens of millions of cars with Optimus, there might be hundreds of millions of Optimus robots, maybe billions, learning a tremendous amount from the real world. That’s the biggest source of data, I think, ultimately, is Optimus, probably. Optimus is going to be the biggest source of data.

对。我是说，我觉得大多数领先的 AI 公司都已经把推特数据全爬走了。不是我觉得，是确实爬了。所以往后看，有用的是它的实时性，因为对手很难实时爬取。所以 Grok 已经有即时性优势。我觉得有了特斯拉、几百万辆车提供的实时视频，最终是几千万辆车加上 Optimus，可能会有几亿台 Optimus 机器人，也许几十亿台，从现实世界学到大量东西。我认为最终最大的数据来源是 Optimus，很可能是。Optimus 会成为最大的数据来源。

Because reality scales. Reality scales to the scale of reality. It’s actually humbling to see how little data humans have actually been able to accumulate. Really, if you say how many trillions of usable tokens have humans generated, where on a non-duplicative… Discounting spam and repetitive stuff, it’s not a huge number. You run out pretty quickly.

因为现实是可以扩展的。现实能扩展到现实的规模。看到人类实际能积累的数据少得可怜，其实挺让人谦卑的。真的，你说人类生成了多少万亿个可用 token，在不重复的情况下……扣掉垃圾和重复内容，数字并不大。很快就用完了。

Right.

对。

Yeah. I mean, the Optimus robot can pick up the cup and see, did it pick up the cup in the right way? Did it, say, go pour water in the cup? Did the water go in the cup or not go in the cups? Did it spill water or not? Simple stuff like that. But it can do that at scale times a billion, so generate useful data from reality, so cause and effect stuff.

对。我是说，Optimus 机器人能拿起杯子，看看它是不是用正确的方式拿的？比如它往杯子里倒水了吗？水进杯子了还是没进？洒了还是没洒？就是这种简单的事。但它能以十亿倍的规模去做，从现实中生成有用的数据，也就是因果关系那类东西。

It’s the same as cars, really. I mean, global capacity for vehicles is about 100 million a year, and it could be higher. It’s just that the demand is on the order of 100 million a year. And then, there’s roughly two billion vehicles that are in use in some way, which makes sense because the life of a vehicle is about 20 years. So, at steady state, you can have 100 million vehicles produced a year with a two billion vehicle fleet, roughly. Now, for humanoid robots, the utility is much greater. So, my guess is humanoid robots are more like at a billion plus per year.

其实跟汽车一样。全球汽车产能大约是每年一亿辆，还可能更高。只是需求大概就是每年一亿辆的量级。然后，大约有二十亿辆汽车在使用中，这很合理，因为一辆车的寿命大约是20年。所以，在稳态下，每年生产一亿辆车、保有二十亿辆的车队，大致如此。而人形机器人的实用性要大得多。所以，我猜人形机器人更可能是每年十亿台以上。

Well, I think it is.

嗯，我觉得确实是。

Yes. So, a walk in the park. I mean, Optimus, currently, would struggle to walk in the park. I mean, it can walk in a park. The park is not too difficult, but it will be able to walk over a wide range of terrain.

是的。所以，就像公园散步一样轻松。我是说，Optimus目前连在公园里散步都费劲。它能在公园里走，公园不算太难，但它将来能在各种地形上行走。

Yeah, yeah. It can already do that.

对，对。它已经能做到了。

Yeah, yeah.

对，对。

Yeah, there’s going to be an immense amount of engineering just going into the hand. The hand, it might be close to half of all the engineering in Optimus. From an electromechanical standpoint, the hand is probably roughly half of the engineering.

是的，光是手部就要投入巨量的工程。手可能占了Optimus全部工程的一半左右。从机电角度看，手大概占了一半的工程。

Yeah.

对。

Yeah. I mean, you start really thinking about your hand and how it works.

对。你开始认真思考你的手是怎么运作的。

The sensory control homunculus is where you have humongous hands. So I mean, your hands, the actuators, the muscles of your hand are almost overwhelmingly in your forearm. So, your forearm has the muscles that actually control your hand. There’s a few small muscles in the hand itself, but your hand is really like a skeleton meat puppet and with cables. So, the muscles that control your fingers are in your forearm, and they go through the carpal tunnel, which is that you’ve got a little collection of bones and a tiny tunnel that these cables, the tendons go through, and those tendons are mostly what move your hands.

感觉运动皮层的小人图里手特别巨大。你的手，驱动手的肌肉几乎全都在前臂里。前臂才有真正控制手的肌肉。手本身只有几块小肌肉，手其实就像一个骨架肉偶，靠缆绳驱动。控制手指的肌肉在前臂，它们穿过腕管，就是一小撮骨头组成的细小隧道，肌腱从中穿过，这些肌腱主要就是用来驱动手的。

Yeah. So the current Optimus, we tried putting the actuators in the hand itself. Then you sort of end up having these-

对。现在的Optimus，我们试过把驱动器放进手本身。结果就变成这些——

… yeah, giant hands that look weird. And then, they don’t actually have enough degrees of freedom or enough strength. So then you realize, “Oh, okay, that’s why you got to put the actuators in the forearm.” And just like a human, you’ve got to run cables through a narrow tunnel to operate the fingers. And then, there’s also a reason for not having all the fingers the same length. So, it wouldn’t be expensive from an energy or evolutionary standpoint to have all your fingers be the same length. So, why not do the same length?

……对，巨大的手，看着很怪。而且自由度和力量都不够。然后你就明白了：哦，所以驱动器得放在前臂。就像人一样，得让缆绳穿过狭窄的隧道来操作手指。手指长度不一样也是有原因的。从能量或进化角度，让所有手指一样长并不昂贵，那为什么不都一样长？

Because it’s actually better to have different lengths. Your dexterity is better if you’ve got fingers that are different lengths. There are more things you can do and your dexterity is actually better if your fingers are a different length. There’s a reason we’ve got a little finger. Why not have a little finger that’s bigger?

因为不同长度其实更好。手指长度不同，你的灵巧度更好。能做的事更多，灵巧度确实更高。我们有小指是有原因的。为什么不让小指更大一点？

Because it helps you with fine motor skills.

因为它有助于精细动作技能。

It does. But if you lost your little finger, you’d have noticeably less dexterity.

有用。但如果你失去小指，灵巧度会明显下降。

It’s actually going to be quite complicated. The as possible part is it’s quite a high bar. If you want to have a humanoid robot that can do things that a human can do, actually, it’s a very high bar. So, our new arm has 22 degrees of freedom instead of 11 and has, like I said, the actuators in the forearm. And all the actuators are designed from scratch, from physics first principles. The sensors are all designed from scratch. And we’ll continue to put a tremendous amount of engineering effort into improving the hand. By hand, I mean the entire forearm, from elbow forward, is really the hand. So, that’s incredibly difficult engineering, actually. And so, the simplest possible version of a humanoid robot that can do even most, perhaps not all, of what a human can do is actually still very complicated. It’s not simple. It’s very difficult.

实际上会相当复杂。所谓尽可能简单，门槛其实很高。如果你想要一个人形机器人能做人能做的事，那门槛非常高。我们的新手臂有22个自由度，而不是11个，而且如前所说，驱动器在前臂。所有驱动器都是从零开始、按物理第一性原理设计的。传感器也全是全新设计。我们会持续投入巨大的工程努力改进手。说手，其实指从肘部往前的整个前臂，那才是真正的手。所以这工程极其困难。因此，一个能完成人类大部分（也许不是全部）工作的人形机器人，哪怕是最简版本，实际上仍然非常复杂。它不简单，非常困难。

Well, it’s easy to say ‘simplify,’ and it’s very difficult to do it. I have this very basic first principles algorithm that I run kind of as a mantra, which is to first question the requirements, make the requirements less dumb. The requirements are always dumb to some degree. So, you want to start off by reducing the number of requirements, and no matter how smart the person is who gave you those requirements, they’re still dumb to some degree. You have to start there, because, otherwise, you could get the perfect answer to the wrong question. So, try to make the question the least wrong possible. That’s what question the requirements means.

说起来容易——'简化'，但做起来非常难。我有一个非常基本的第一性原理算法，我把它当作口头禅来执行：首先质疑需求，让需求变得不那么愚蠢。需求在某种程度上总是愚蠢的。所以，你要从减少需求的数量开始，不管给你这些需求的人有多聪明，它们在某种程度上仍然是愚蠢的。你必须从那里开始，因为否则你可能会对错误的问题得到完美的答案。所以，尽量让问题尽可能少错。这就是质疑需求的含义。

Yeah. First try to delete it. Yeah.

对。先试着删掉它。对。

No. Actually, what generally makes people uneasy is that at least some of the things that you delete, you will put back in. But going back to sort of where our limbic system can steer us wrong is that we tend to remember, with sometimes a jarring level of pain, where we deleted something that we subsequently needed. And so, people will remember that one time they forgot to put in this thing three years ago, and that caused them trouble. And so, they overcorrect, and then they put too much stuff in there and overcomplicate things. So, you actually have to say, “Look, we’re deliberately going to delete more than we should.” At least one in 10 things, we’re going to add back in.

不容易。实际上，让人不安的是，你删掉的东西里至少有一些你还会加回来。但回到我们边缘系统可能误导我们的地方：我们往往会记住，有时带着刺骨的痛苦，那些我们删掉后来又需要的东西。所以，人们会记得三年前忘了加进某个东西，结果惹了麻烦。于是他们过度纠正，然后塞进太多东西，把事情搞复杂。所以你实际上必须说：'看，我们故意要删掉超过应该删的。'至少十分之一的东西，我们会加回来。

Oh, yeah. Absolutely.

哦，是的。绝对是。

Absolutely. And I tell them in advance, “Yeah, some of the things that we delete, we’re going to put back in.” People get a little shook by that, but it makes sense because if you’re so conservative as to never have to put anything back in, you obviously have a lot of stuff that isn’t needed. So, you got to overcorrect. This is, I would say, like a cortical override to a limbic instinct.

绝对是。我会提前告诉他们：'是的，我们删掉的一些东西，之后会加回来。'人们会有点被吓到，但这有道理，因为如果你保守到从不需要加回任何东西，那你显然有一堆不需要的东西。所以你必须过度纠正。我会说，这就像是用皮层压制边缘系统的本能。

Yeah. There’s a step four as well, which is any given thing can be sped up. However fast you think it can be done, whatever the speed it’s being done, it can be done faster. But you shouldn’t speed things up until you’ve tried to delete it and optimize. Although, you’re speeding up something that… Speeding up something that shouldn’t exist is absurd.

是的。还有第四步：任何东西都可以加速。无论你觉得能多快完成，无论现在速度如何，都可以更快。但你不应该在尝试删除和优化之前就加速。不过，加速一个……加速一个本不该存在的东西是荒谬的。

Yeah.

是的。

Yeah, it’s very. It’s like, “Wow, I really wasted a lot of effort there.”

是的，非常。就像：'哇，我在那上面真的浪费了大量精力。'

Well, yeah, it’s not working yet, so I don’t want to pop the champagne corks. In fact, I have a call in a few hours with the Memphis team because we’re having some power fluctuation issues. So yeah, when you do synchronized training, when you have all these computers that are training, where the training is synchronized at the millisecond level, it’s like having an orchestra. And the orchestra can go loud to silent very quickly at subsecond level, and then, the electrical system freaks out about that. If you suddenly see giant shifts, 10, 20 megawatts several times a second, this is not what electrical systems are expecting to see.

嗯，是的，它还没跑起来，所以我不想开香槟庆祝。事实上，几小时后我还要和孟菲斯团队通电话，因为我们有一些电力波动问题。是的，当你做同步训练时，所有这些计算机在训练，训练在毫秒级别同步，就像一支管弦乐队。乐队可以在亚秒级别从响亮迅速转到安静，然后电力系统就对此抓狂了。如果你突然看到每秒几次10、20兆瓦的巨大波动，这不是电力系统预期会看到的。

Yeah. So, today’s problem is dealing with extreme power jitter.

是的。所以今天的问题是应对极端的电力抖动。

Yeah.

是的。

Last week. Yeah.

上周。是的。

Yeah. We finally got training going at, oddly enough, roughly 4:20 a.m. last Monday.

是的。我们终于让训练跑起来了，说来也巧，大概是上周一凌晨4点20分。

Yeah. I mean, maybe it was at 4:22 or something.

是的。我是说，也许是4点22分左右。

Yeah.

是的。

Well, exactly. It just loves it.

嗯，确实。它就是爱这样。

Yeah. So, look, whatever the people at the front lines are doing, I try to do it at least a few times myself. So connecting fiber optic cables, diagnosing a faulty connection. That tends to be the limiting factor for large training clusters is the cabling. There’s so many cables. For a coherent training system, where you’ve got RDMA, remote direct memory access, the whole thing is like one giant brain. So, you’ve got any-to-any connection. So, any GPU can talk to any GPU out of 100,000. That is a crazy cable layout.

是的。所以你看，一线的人在做什么，我都会亲自去做至少几次。比如接光纤电缆、诊断故障连接。大型训练集群的瓶颈往往就是布线。电缆太多了。对于一个连贯的训练系统，你有 RDMA，远程直接内存访问，整个东西就像一个巨大的大脑。所以你有任意到任意的连接。十万个 GPU 里任何一个都能和任何一个通信。那布线布局简直疯狂。

Yeah.

是的。

Yeah. But, I mean, the human brain also has… A massive amount of the brain tissue is the cables. So they get the gray matter, which is the compute, and then the white matter, which is cables. A big percentage of your brain is just cables.

是的。不过我是说，人脑也有……大脑组织很大一部分就是电缆。灰质是计算，白质是电缆。你大脑很大比例就是电缆。

It’s possible. What do you define as AGI?

有可能。你怎么定义 AGI？

Just keep moving the goalposts?

就是不断把球门往后挪？

Oh, yeah.

哦，是的。

Well, I think that, generally, people would call that ASI, artificial super intelligence. But there are these thresholds where you could say at some point the AI is smarter than any single human. And then, you’ve got eight billion humans, and actually, each human is machine augmented via their computers. So, it’s a much higher bar to compete with eight billion machine augmented humans. That’s a whole bunch of orders of magnitude more. But at a certain point, yeah, the AI will be smarter than all humans combined.

嗯，我觉得一般来说，人们会把这叫做 ASI，人工超级智能。但有一些门槛，你可以说某个时刻 AI 比任何单个人类都聪明。然后你有八十亿人，而且实际上每个人都通过电脑被机器增强。所以要和八十亿机器增强的人类竞争，门槛高得多。那是好几个数量级的差距。但到了某个时刻，是的，AI 会比所有人类加起来还聪明。

Yeah, absolutely. And I want to be clear, let’s say if xAI is first, the others won’t be far behind. I mean, they might be six months behind, or a year, maybe. Not even that.

是的，绝对。而且我想说清楚，假设 xAI 是第一，其他家也不会落后太远。我是说，他们可能落后六个月，或者一年，也许。甚至没那么多。

So, I mean, I thought about AI, essentially, for a long time, and the thing that at least my biological neural net comes up with as being the most important thing is adherence to truth, whether that truth is politically correct or not. So, I think if you force AIs to lie or train them to lie, you’re really asking for trouble, even if that lie is done with good intentions. So, you saw issues with ChatGPT and Gemini and whatnot. Like, you asked Gemini for an image of the Founding Fathers of the United States, and it shows a group of diverse women. Now, that’s factually untrue.

所以，我是说，我基本上思考 AI 很久了，至少我的生物神经网络得出的最重要的一点是坚持真相，不管这个真相是否政治正确。所以我认为，如果你强迫 AI 撒谎或训练它撒谎，你就是在自找麻烦，即使那个谎言是出于好意。所以你看到 ChatGPT 和 Gemini 之类的问题。比如你让 Gemini 生成一张美国建国元勋的图片，它显示的是一群多元化的女性。那在事实上是不真实的。

Yeah.

是的。

Yet they’re real examples of-

但它们是真实的例子，来自……

… AI that was released to the public.

……向公众发布的 AI。

That went through QA, presumably, and still said insane things, and produced insane images.

那东西想必是经过QA的，结果还是说些疯话，生成些疯图。

No, it’s not.

不，确实不容易。

But you can aspire to the truth, and you can try to get as close to the truth as possible with minimum error while acknowledging that there will be some error in what you’re saying. So, this is how physics works. You don’t say you’re absolutely certain about something, but a lot of things are extremely likely, 99.99999% likely to be true. So, aspiring to the truth is very important. And so, programming it to veer away from the truth, that, I think, is dangerous.

但你可以追求真相，你可以尽量接近真相、把误差降到最低，同时承认你说的话里会有一些误差。物理学就是这样运作的。你不会说你对某件事绝对确定，但很多事极有可能为真，99.99999%可能为真。所以，追求真相非常重要。而把它编程成偏离真相，我认为这很危险。

And the internet, at this point, is polluted with so much AI generated data, it’s insane. Actually, there’s a thing now, if you want to search the internet, you can say, “Google, but exclude anything after 2023.” It will actually often give you better results because there’s so much. The explosion of AI generated material is crazy. So in training Grok, we have to go through the data and say like, “Hey…” We actually have to apply AI to the data to say, “Is this data most likely correct or most likely not?” before we feed it into the training system.

而且现在的互联网被大量AI生成的数据污染了，简直离谱。其实现在有个办法：如果你想搜互联网，你可以说“Google，但排除2023年之后的内容”。结果往往更好，因为这类内容实在太多了。AI生成材料的爆炸式增长太疯狂了。所以在训练Grok时，我们得筛选数据，说“嘿……”我们实际上得用AI来处理数据，判断“这条数据大概率正确还是大概率不正确”，然后再喂进训练系统。

Yeah.

是的。

Grok 3 is going to be next level. I mean, what people are currently seeing with Grok is kind of baby Grok.

Grok 3会是下一个层次。我的意思是，人们现在看到的Grok算是婴儿版Grok。

It’s baby Grok right now. But baby Grok is still pretty good. But it’s an order of magnitude less sophisticated than GPT-4. It’s now Grok 2, which finished training, I don’t know, six weeks ago or thereabouts. Grok 2 will be a giant improvement. And then Grok 3 will be, I don’t know, order of magnitude better than Grok 2.

现在还是婴儿版Grok。但婴儿版Grok已经相当不错了。不过它的复杂程度比GPT-4低一个数量级。现在是Grok 2，训练完成大概六周前吧。Grok 2会是一次巨大提升。然后Grok 3会比Grok 2好一个数量级，我也不知道。

Hopefully. I mean, this is the goal. I mean, we may fail at this goal. That’s the aspiration.

希望如此。这就是目标。我们可能达不到这个目标。但这是我们的追求。

Yeah. I think it’s important that whatever AI wins, it’s a maximum truth seeking AI that is not forced to lie for political correctness, or, well, for any reason, really, political, anything. I am concerned about AI succeeding that is programmed to lie, even in small ways.

是的。我认为重要的是，无论哪个AI胜出，它都应该是最大程度追求真相的AI，不被强迫为了政治正确而撒谎，或者为了任何理由撒谎，政治的或别的什么。我担心的是，一个被编程为即使在小事上也会撒谎的AI取得成功。

To become very big ways. Yeah.

会变成非常大的问题。是的。

Yeah.

是的。

Cool.

酷。

Yeah, sure. I’ll stop in.

行，当然。我会过去。

Well, I think people tend to take, say, an endorsement as, well, I agree with everything that person has ever done their entire life 100% wholeheartedly, and that’s not going to be true of anyone. But we have to pick. We’ve got two choices, really, for who’s president. And it’s not just who’s president, but the entire administrative structure changes over. And I thought Trump displayed courage under fire, objectively. He’s just got shot. He’s got blood streaming down his face, and he’s fist pumping, saying, “Fight.” That’s impressive. You can’t feign bravery in a situation like that. Most people would be ducking because there could be a second shooter. You don’t know.

嗯，我觉得人们往往把背书理解为：我100%衷心同意那个人一生做过的每件事，但这对任何人都不会成立。可我们得做选择。总统人选实际上只有两个选择。而且不只是谁当总统，整个行政结构都会换掉。我认为特朗普在枪火下展现了勇气，客观地说。他刚中枪，血顺着脸流下来，还在挥拳喊“战斗”。这很了不起。那种情况下你没法假装勇敢。大多数人会蹲下躲，因为可能有第二个枪手。你不知道。

I mean, there’s a sort of age-old debate in history, like is history determined by these fundamental tides, or is it determined by the captain of the ship? It’s both, really. I mean, there are tides, but it also matters who’s captain of the ship. So, it’s a false dichotomy, essentially. I mean, there are certainly tides, the tides of history. There are real tides of history, and these tides are often technologically driven. If you say like the Gutenberg press, the widespread availability of books as a result of a printing press, that was a massive tide of history, and independent of any ruler. But in stormy times, you want the best possible captain of the ship.

我的意思是，历史上一直有这种古老的争论，历史究竟是由这些根本性的潮流决定的，还是由船长决定的？其实两者都是。有潮流，但谁当船长也很重要。所以这本质上是个伪二分法。历史当然有潮流，真实的历史潮流，而这些潮流往往由技术驱动。比如古腾堡印刷机，印刷机带来的书籍普及，那就是一股巨大的历史潮流，与任何统治者无关。但在风暴时期，你希望有尽可能最好的船长。

The Lessons of History.

《历史的教训》。

Yeah.

是的。

Yeah, over the years.

是的，这些年来。

Yeah. Well, I mean in terms of dating civilization, the start of civilization, I think the start of writing, in my view, that’s what I think is probably the right starting point to date civilization. And from that standpoint, civilization has been around for about 5,500 years when writing was invented by the ancient Sumerians, who are gone now, but the ancient Sumerians. In terms of getting a lot of firsts, those ancient Sumerians really have a long list of firsts. It’s pretty wild. In fact, Durant goes through the list of like, “You want to see firsts? We’ll show you firsts.” The Sumerians were just ass kickers.

是的。就文明起源的断代而言，我认为文字的出现，在我看来，那大概才是文明断代的正确起点。从这个角度看，文明大约存在了5500年，那时文字由古苏美尔人发明，他们如今已消失，但古苏美尔人……说到拿下众多'第一'，那些古苏美尔人真是一长串的第一。挺疯狂的。事实上，杜兰特就列了那份清单，像是'你想看第一？我们给你看第一。'苏美尔人就是狠角色。

Very early.

非常早期。

Many. So many rises and falls of empires. So many.

很多。太多帝国的兴衰了。太多了。

Yeah, exactly. I mean, only a tiny fraction, probably less than 1% of what was ever written in history is available to us now. I mean, if they didn’t literally chisel it in stone or put it in a clay tablet, we don’t have it. I mean, there’s some small amount of papyrus scrolls that were recovered that are thousands of years old, because they were deep inside a pyramid and weren’t affected by moisture. But other than that, it’s really got to be in a clay tablet or chiseled. So, the vast majority of stuff was not chiseled because it takes a while to chisel things. So, that’s why we’ve got tiny, tiny fraction of the information from history. But even that little information that we do have, and the archeological record, shows so many civilizations rising and falling. It’s wild.

没错。我的意思是，历史上所有写下的东西中，只有极小一部分，可能不到1%，留存至今。如果他们没把它刻在石头上或写在泥板上，我们就看不到了。有一些几千年前的莎草纸卷轴被保存下来，因为它们在金字塔深处，没受潮。但除此之外，真的得是泥板或刻石才行。所以绝大多数东西没被刻下来，因为刻东西很费时间。这就是为什么我们只掌握历史上极小极小的信息。但即便那点信息和考古记录，也显示出那么多文明兴衰更替。太疯狂了。

Yeah. I mean, the basics of human nature are more or less the same. Yeah.

是的。人性的基本层面或多或少是一样的。是的。

Yeah. I mean, you do tend to see the same patterns, similar patterns for civilizations, where they go through a life cycle, like an organism, just like a human is a zygote, fetus, baby, toddler, teenager, eventually gets old.

是的。你确实会看到相似的模式，文明有相似的模式，它们经历一个生命周期，像有机体一样，就像人从受精卵、胎儿、婴儿、幼儿、青少年，最终变老。

… Eventually gets old and dies. The civilizations go through a life cycle. No civilization will last forever.

……最终变老死去。文明经历生命周期。没有文明能永存。

Well, the single biggest thing that is often actually not mentioned in history books, but Durant does mention it, is the birthright. So perhaps to some, a counterintuitive thing happens when civilizations are winning for too long, the birth rate declines. It can often decline quite rapidly. We’re seeing that throughout the world today. Currently, South Korea is, I think maybe the lowest fertility rate, but there are many others that are close to it. It’s like 0.8 I think. If the birth rate doesn’t decline further, South Korea will lose roughly 60% of its population. But every year that birth rate is dropping, and this is true through most of the world. I don’t mean to single out South Korea, it’s been happening throughout the world. So as soon as any given civilization reaches a level of prosperity, the birth rate drops.

嗯，最大的因素，历史书里往往不提，但杜兰特提到了，就是出生率。对某些人来说可能反直觉：当文明长期处于胜利状态时，出生率会下降。而且往往下降得很快。我们今天在全世界都看到这一点。目前韩国大概是生育率最低的，可能0.8左右，但很多国家也接近。如果生育率不再下降，韩国将失去大约60%的人口。但生育率每年都在降，这在世界大部分地区都是如此。我不是要单挑韩国，这在全世界都在发生。所以一旦某个文明达到一定繁荣水平，出生率就会下降。

Well, I’m saying that’s-

嗯，我是说那就是——

At a fundamental level, if a civilization does not at least maintain its numbers, it’ll disappear.

从根本上说，如果一个文明连自身数量都无法维持，它就会消失。

Well, I mean there’s this hedonistic sex, which is… That’s neither her nor there. It’s-

嗯，我是说，有这种享乐式的性，那个……那跟这没关系。它是——

It doesn’t produce kids. Well, what matters… I mean, Durant makes this very clear because he’s looked at one civilization after another and they all went through the same cycle. When the civilization was under stress, the birth rate was high. But as soon as there were no external enemies or they had an extended period of prosperity, the birth rate inevitably dropped. Every time. I don’t believe there’s a single exception.

它不生孩子。嗯，重要的是……我是说，杜兰特把这点讲得很清楚，因为他研究了一个又一个文明，它们都经历了同样的循环。当文明处于压力之下时，出生率很高。但一旦没有外部敌人，或者经历了一段长期的繁荣，出生率就不可避免地下降。每一次都是。我不相信有任何一个例外。

Yeah. I mean, at a base level, no humans, no humanity.

对。我是说，在最基本的层面上，没有人，就没有人类。

Yeah, absolutely. But at a basic level, if you do not at least maintain your numbers, if you’re below replacement rate and that trend continues, you will eventually disappear. It’s just elementary. Now then obviously you also want to try to avoid massive wars. If there’s a global thermonuclear war, probably we’re all toast, radioactive toast. So we want to try to avoid those things. Then there’s a thing that happens over time with any given civilization, which is that the laws and regulations accumulate. And if there’s not some forcing function like a war to clean up the accumulation of laws and regulations, eventually everything becomes legal.

对，绝对。但在基本层面上，如果你连人口数量都维持不住，如果你低于更替率而且这个趋势持续下去，你最终会消失。这就是常识。然后显然你还想尽量避免大规模战争。如果爆发全球热核战争，我们大概都完蛋了，变成放射性的烤面包。所以我们要尽量避免那些事。然后还有一件事，随着时间推移会发生在任何文明身上，那就是法律和法规不断累积。如果没有某种强制机制，比如战争，来清理法律和法规的堆积，最终一切都会变成非法的。

I have discussed with Trump the idea of a government deficiency commission.

我跟特朗普讨论过设立一个政府效率委员会的想法。

And I would be willing to be part of that commission.

我愿意参与那个委员会。

The antibody reaction would be very strong.

抗体反应会非常强烈。

So you really have to… You’re attacking the matrix at that point. The matrix will fight back.

所以你真的得……那时候你就是在攻击矩阵。矩阵会反击的。

Me? Attacked?

我？被攻击？

Yeah, there is a lot. I mean, every day another psyop. I need my tinfoil hat.

对，确实很多。我是说，每天都有一场新的心理战。我需要我的锡箔帽了。

Oh yeah, that’s a daily occurrence.

哦对，那是每天都会发生的事。

So I mean, it does get me down at times. I mean, it makes me sad. But I mean at some point you have to sort of say, look, the attacks are by people that actually don’t know me and they’re trying to generate clicks. So if you can sort of detach yourself somewhat emotionally, which is not easy, and say, okay look, this is not actually from someone that knows me or, they’re literally just writing to get impressions and clicks. Then I guess it doesn’t hurt as much. It’s not quite water off a duck’s back. Maybe it’s like acid off a duck’s back.

所以，我是说，有时候确实让我低落，让我难过。但到了某个时候你得说，看，这些攻击来自根本不认识我的人，他们只是想赚点击量。所以如果你能在情感上稍微抽离一点——这不容易——然后说，好吧，这其实不是来自认识我的人，他们字面上只是为了刷曝光和点击。那我想就没那么疼了。这不太像水从鸭子背上滑过，也许更像酸从鸭子背上滑过。

A measure of success, I’d say, how many useful things can I get done?

衡量成功，我会说，是我能完成多少有用的事？

Yeah, maximize utility, area under the code of usefulness. Very difficult to be useful at scale.

对，最大化效用，有用性曲线下的面积。要在规模化层面做到有用非常难。

Well, time is the true currency.

嗯，时间才是真正的货币。

So it is tough to say what is the best allocation time? I mean, there are often… Say if you look at say Tesla, Tesla this year will do over a hundred billion in revenue. So that’s $2 billion a week. If I make slightly better decisions, I can affect the outcome by a billion dollars. So then I try to do the best decisions I can. And on balance, at least compared to the competition, pretty good decisions. But the marginal value of a better decision can easily be, in the course of an hour, a hundred million dollars.

所以很难说什么是最佳的时间分配？我的意思是，经常……比如说你看特斯拉，特斯拉今年营收会超过一千亿美元。也就是每周20亿美元。如果我做出稍微好一点的决策，我就能影响十亿美元的结果。所以我尽量做最好的决策。总的来说，至少和竞争对手相比，决策还不错。但一个更好的决策的边际价值很容易在一小时内就是上亿美元。

Yeah. Well, I think you have to look at it on a percentage basis because if you look at it in absolute terms, it’s just… I would never get any sleep. It would just be like, I need to just keep working and work my brain harder. And I’m not trying to get as much as possible out of this meat computer. So it’s not… It’s pretty hard, because you can just work all the time. And at any given point, like I said, a slightly better decision could be a hundred million dollars impact for Tesla or SpaceX for that matter. But it is wild when considering the marginal value of time can be a hundred million dollars an hour at times, or more.

是的。我觉得你得按百分比来看，因为如果看绝对值，那就……我永远都别想睡觉了。那就像是我得一直工作，把脑子用得更狠。我并不想把这颗肉脑榨到极限。所以这不是……很难，因为你可以一直工作。而且在任何时刻，就像我说的，稍微好一点的决策对特斯拉或SpaceX来说可能就是上亿美元的影响。但考虑到时间的边际价值有时能达到每小时上亿美元甚至更多，这确实很疯狂。

It has to be to some degree. If I’m sad, if I’m depressed, I make worse decisions. So if I have zero recreational time, then I make worse decisions. So I don’t know a lot, but it’s above zero. I mean, my motivation if I’ve got a religion of any kind is a religion of curiosity, of trying to understand. It’s really the mission of Grok, understand the universe. I’m trying to understand the universe, or at least set things in motion such that at some point civilization understands the universe far better than we do today.

在某种程度上必须算。如果我难过、如果我抑郁，我会做出更差的决策。所以如果我完全没有娱乐时间，我就会做出更差的决策。所以我不知道具体多少，但肯定大于零。我的意思是，如果我有任何信仰的话，那就是好奇心的信仰，试图去理解。这正是Grok的使命，理解宇宙。我在试图理解宇宙，或者至少把事情推动起来，使得某个时刻文明对宇宙的理解远超今天。

Digital super intelligence is possibly a great filter. I hope it isn’t, but it might be. Guys like say Jeff Hinton would say, he invented a number of the key principles in artificial intelligence. I think he puts the probability of AI annihilation around 10% to 20%, something like that. So look on the bright side, it’s 80% likely to be great. But I think AI risk mitigation is important. Being a multi-planet species would be a massive risk mitigation. And I do want to once again emphasize the importance of having enough children to sustain our numbers, and not plummet into population collapse, which is currently happening. Population collapse is a real and current thing.

数字超级智能可能是一个大过滤器。我希望它不是，但也许是。像杰夫·辛顿这样的人会说，他发明了人工智能的若干关键原理。我认为他把AI灭绝人类的概率定在10%到20%左右。所以往好处想，有80%的概率是好事。但我认为降低AI风险很重要。成为多行星物种将是巨大的风险缓解。我想再次强调有足够孩子来维持我们人口数量的重要性，不要跌入人口崩溃，而这正在发生。人口崩溃是真实且当下的事。

Great. Hope you do. No time like the present.

太好了。希望你能做到。最好的时机就是现在。

No, it’s like unskilled labor.

不，这就像非技术工种。

There will be billions of robots. That seems virtual certainty.

会有数十亿机器人。这几乎是必然的。

You’re welcome. Go forth and multiply.

不客气。去繁衍吧。