本周AI动态：OpenAI颠覆一切，Neo机器人问世，英伟达市值迈向5万亿，Grokipedia崛起

欢迎回到本周AI动态汇总。这又是重要的一周。英伟达成为首家市值突破5万亿美元的公司，而苹果和微软紧随其后，市值均突破4万亿美元。这些数字令人咋舌。OpenAI本周也颇受争议。

Welcome back to the weekly AI roundup. It's been another huge week. NVIDIA became the first company to cross $5,000,000,000,000 in market value, but they're closely followed behind by Apple and Microsoft to cross $4,000,000,000,000 in market cap value. Some big numbers being thrown around. OpenAI had a pretty controversial week.

他们重组了整个公司架构，同时公布了未来十年实现AGI的秘密计划。埃隆·马斯克决定推出由旗舰模型Grok驱动的AI版维基百科Grokopedia。我们稍后会详细讨论。最后，现在只需每月499美元就能拥有新款人形机器人。乔什，今天要讨论的内容可不少。

They restructured their entire company, but they also revealed their secret plans for AGI over the next ten years. Elon Musk decided to release an AI version of Wikipedia called Grokopedia powered by their flagship model, Grok. We'll get into that. And finally, there is a new humanoid robot available to you for the cheap price of $499 per month. Josh, there's a lot to get through to today.

先从OpenAI的新闻开始，主要有两件大事要讨论。他们发布了两项重要声明：一是未来十年实现AGI的愿景，二是公司重组计划。关于AGI部分，山姆·奥尔特曼与首席科学家雅库布通过直播阐述了OpenAI的愿景。

Starting off with all the OpenAI news, there's two main things to discuss here. They made two big announcements. One was their vision for the next ten years on achieving AGI. And the second thing was on the restructuring. Starting with the AGI side of things, Sam Altman came on livestream with his chief scientist, Jakub, to describe what the vision of OpenAI is.

我认为这是对很多人质疑的回应——山姆，你的初衷是否纯粹？你总是说一套做一套，推出的产品越来越偏向消费级，缺乏宏大构想。你们还专注AGI吗？这次算是澄清立场。我从中有三个重要发现很有意思。

And I think this is in response to a lot of people saying, Sam, like, do you have your best intentions at heart? You keep saying that, but then your actions are kind of producing products that might seem kind of consumer y and not really big thinking. Are you still focused on AGI? So this is kind of like clearing the plate here. And there's three main takeaways that I got from this that I found interesting.

第一，山姆延长了实现AGI的期限。他表示将在十年内逐步实现，过渡期会比预期更慢。这很有趣，因为终于有了现实的时间表，而不是那些'明天就能实现'的夸张标题。第二，他开始给出具体里程碑。

Number one, Sam has extended his deadline to achieving AGI. He says he's gonna achieve it within ten years and it's gonna be a slower transition than initially expected. He describes it as achieving AGI gradually over a number of different years. This is interesting to me because we're finally achieving a realistic timeline, in my opinion, versus some kind of, like, hype headline where we're just kind of, like, thinking it's gonna arrive tomorrow. Number two, he's started giving solid milestones on this.

他提到2026年时，OpenAI内部的深度学习将产出AGI级别的科学模型，性能相当于AI研究实习生水平。到2028年将提升至研究员级别。从实习生到研究级性能的跨越非常酷。最让我感兴趣的是算力部分——乔什，他们目前已投入或承诺1.4万亿美元建设数据中心，相当于30吉瓦的算力。

He says in 2026, the deep learning that OpenAI is conducting internally will result in an AGI level science model, which will perform kind of at the caliber of an AI research intern. And then he says two years later in 2028, they'll have an AI researcher level model. So you go from intern to research grade performance, which is super cool. And the final point that really interested me was around compute. Josh, they've spent or committed around $1,400,000,000,000 on data centers right now, which is roughly an equivalent of 30 gigawatts.

这是巨大的算力规模。但最让我震惊的是他公开表示最终计划是每周建成一座能生产1吉瓦电力的工厂，相当于每周新增200亿美元的算力。这些天文数字真让人喘不过气。乔什，你怎么看？

It's a huge amount of compute. But what shocked me the most was he came out and said that his eventual plan is to have a factory built every week that produces a gigawatt of power. That's about $20,000,000,000 of compute per week. All really huge numbers kinda took my breath away. Josh, what's your take?

OpenAI只是在做大家早就预料到的事。他们很快意识到自己无法保持非营利性质，现在正转型为营利性组织。实际上，它是一家公益公司（PBC），不是有限责任公司或C类公司，这意味着其初衷是与公共利益保持一致，但本质上打破了所有限制。这确实是件大事。

OpenAI is just doing what everyone kind of knows they were always going to do. They very quickly realized they can't be a non for profit. They are now becoming a for profit organization. In fact, it's a public benefit corporation, so it's a PBC, not an LLC or a c corp or anything, which means the intention is to be kind of aligned with public interest, but, basically, it just blows the cap off of everything. And this is a really big deal.

你刚才提到——其实早些时候你告诉过我——Sam Altman表示有兴趣上市的消息对我来说是新闻。我认为这里有个重要机会可以讨论OpenAI上市和重组能带来多少收益。这个重组方案很有意思，我们准备好讨论了吗？可以深入探讨吗？

You just you mentioned that or you mentioned to me actually earlier, and this was news to me that that Sam Altman expressed interest in going public. I think there's a big opportunity here to talk about how much money can be made by OpenAI going public and by kind of restructuring this. So the restructuring deal is interesting. Are are we ready to talk about that? Can we go into the this, like Yeah.

正在发生的疯狂混乱重组？好吧。这让我很着迷。屏幕上现在显示着一张灾难性的箭头示意图，大部分内容都是模糊不清的，并不重要。

Crazy, chaotic restructuring that's happening? Okay. This is fascinating to me. So on screen, there is a disastrous diagram with a bunch of arrows. Lot of it is kinda hand waving, not important.

关键点在于：微软持有非营利实体的大部分股份。该非营利实体很快意识到必须转为营利性质。旧版本有收益上限，但现在没有了。微软原先的投资回报被限制在100倍，现在则没有投资回报上限。

The things that matter, Microsoft owns a large percentage of the nonprofit. The nonprofit quickly realized they needed to become a for profit. They had a capped upside on the previous version, but now they do not. So Microsoft was limited previously to a 100 x return on their investment. Now there is no return on the invest or no limit to the return on the investment.

这有几个方面很有意思，Itaz。首先我对OpenAI非常不满，因为他们显然是全球增长最快的人工智能公司，而我却无法获得投资机会。这感觉与公共利益不符。Sam Altman把所有钱都装进自己口袋，砸在GPU上，我却得不到任何好处。这不公平。

And this is interesting for a few things, Itaz, because one, I am very resentful of OpenAI because they are very clearly the fastest growing AI company in the world, and I have no way to get exposure to them. That doesn't feel like it's aligned with the public. Sam Altman is taking all the money for himself. He's throwing it at GPUs, and I get no upside. That doesn't feel right.

这可是OpenAI啊，你们是公益公司。为什么不为我这个公众成员谋福利？其次，微软现在似乎是最好的投资渠道，因为当OpenAI上市时可以通过这种方式获得投资机会。但微软现在以27%的持股成为OpenAI最大股东，超过了持股26%的OpenAI非营利实体。所以目前最大的投资渠道是通过微软。

This is OpenAI. You're a public benefit company. Why are you not benefiting me as a member of the public? Second to this is that Microsoft now seems to be the best way to get exposure because there is this, like, there is this way of getting access to OpenAI when they IPO, but OpenAI and Microsoft are now or Microsoft's now the biggest stakeholder in the OpenAI company at 27% versus the OpenAI nonprofit, which is actually 26%. So the largest form of exposure is through Microsoft.

他们的利润没有上限，而且我记得这周就要发布财报。所以微软可能是当前投资OpenAI的最佳方式，但OpenAI在整个过程中让我失望，因为一切都很模糊，与他们宣称的公共利益初衷并不相符——他们本质上一直是个伪装成非营利组织的庞大营利实体。

They have uncapped profits, and they are reporting earnings, I think, this week. So Microsoft might be the best way to get exposure to OpenAI currently, but OpenAI as a whole has kind of disappointed me through this whole process because it's all been very hand wavy and not really aligned with, I think, the public intention that they stated just because they've always been this big for profit entity disguised as a nonprofit.

是的，我完全同意你提出的许多观点。为了让听众了解背景，OpenAI成立于2015年，最初是非营利组织。它以这个雄心勃勃的愿景成立：我们将实现AGI（通用人工智能），并将其提供给大众。每个人都应该有机会接触这项强大技术。

Yeah. I I I completely agree with a lot of the points you made. And and to just kind of give the listeners a bit of context here, OpenAI founded in 2015 as a nonprofit. It was set up as a nonprofit org with this ambitious vision of we are gonna achieve AGI and we are gonna provide it to the people. So everyone should have access to this super powerful technology.

四年后的2019年，他们决定需要从中获取一些收益并筹集大量资金。于是在2019年建立了营利性架构，但仍由非营利组织控股。这样他们仍能宣称'我们是非营利的'，但如你所说乔什，对任何投资者的回报上限是100倍。近期他们收到许多来自现有和潜在投资者的反馈与压力，这些投资者表示'我们想在这笔交易中赚更多钱，投入数十亿却受限制不公平'。

Then four years later in 2019, they decided, we kind of wanna make a bit of money from this and we need to raise a lot of capital. So in 2019, they set up a for profit structure, but it was owned by the nonprofit. So they could still get away with saying, Hey, we're a nonprofit, but it was capped, as you said, Josh, to any investor for 100X, just 100X returns. So a lot of feedback they've been getting and a lot of pressure they've been getting from future investors or current investors recently is, hey, we wanna make more money on this deal. Like, unfair that we're investing billions in you.

微软投资了130亿美元，而我们期望的回报要超过100倍。于是山姆重新规划，创建了名为'公益公司'的新架构。如你所说，现在收益不再设限，这很疯狂。现在投资者可以像投资普通私营公司那样获取无限收益。最关键的是——看这个图表，放大仔细看：微软原本是少数股东，现在不再是了。

Microsoft invested $13,000,000,000 and we want like more upside than 100X. So Sam went to the drawing board and was like, okay, we can create a new structure called a public benefit corporation, which as you said, now has uncapped gains. It's pretty crazy. So like now you can just make as much money as you can on a normal private company that you were investing in. And the biggest point around this is on this diagram here, you'll see if you zoom in, if you squint, Microsoft minority owner, they are now no longer a minority owner.

如你所说，他们现在是控股股东，这简直难以置信。但微软成为大赢家还有其他原因：不仅是金钱层面——他们130亿美元的投资现在价值1350亿美元，回报率惊人。更重要的是他们仍独家享有OpenAI所有研究知识产权的使用权，这意味着任何新模型发布，微软都拥有优先话语权和独家授权资格，可向其合作伙伴和未来客户提供这些技术。

Like you said, they are the majority owner, which is just crazy. But I'll tell you a few other reasons why Microsoft is a huge winner from all of this. It's not just money. It's not just the fact that their $13,000,000,000 stake is now worth $135,000,000,000 What a crazy return. But they still maintain exclusive access and rights to all of OpenAI's research IP, which means that any kind of new model that they release, Microsoft basically has the first say and exclusive access to provide it to their partners, to their customers, and any clients and enterprise clients that they wanna bring in in the future.

第二点（这真是天才之举，乔什），他们设立了标准：'听着OpenAI，只有在实现AGI后，我们才会放弃所有独家权限和知识产权'。但核心问题是：到底什么是AGI？如何界定AGI？关于这个概念的讨论简直铺天盖地。

Secondly, and this was a genius move, Josh, they set a standard. They said, Listen, OpenAI, we will relieve all our exclusive access and IP access only once you've achieved AGI. But of course, the obvious question is, what the hell is AGI? How do you determine what AGI is? There's so much discussion around what the hell this thing is.

于是他们决定交由第三方专家小组裁决，这个委员会将决定OpenAI未来的命运。这次架构重组和愿景宣布的最大赢家其实是微软，这局面令人难以置信。

So they said, we'll leave it to an expert party of third party analysts. So it's a panel which is gonna decide and determine OpenAI's fate going forward. So the winner of all of this, this restructuring, this OpenAI vision announcement is actually Microsoft, and it's just crazy to see.

好的，那么作为想通过AI致富的参与者该如何应对？有几种方式。关注公开市场的人会注意到：苹果股价处于历史高位，微软、谷歌也是——几乎所有科技巨头都创下新高，我认为这是个强烈信号。

Okay. So how do you navigate this, as a participant who wants to get wealthy off of AI? There is a few ways. I think for people who observe public markets, you'll know the fact that Apple is sitting at all time high, Microsoft's sitting at all time high, Google. Basically, across the board, everyone is sitting at all time highs, which I believe is high signal.

显然，一切运转正常。显然，投入了大量资金。显然，存在某种泡沫，但终结尚未来临。我最大的疑问是——拜托，如果不是合格投资者，如果搞不到私募股份，我该怎么接触OpenAI呢？

Clearly, things are working. Clearly, a lot is being spent. Clearly, there is some sort of a bubble, but the end is not quite near. The big question that I have is, like, come on. How do I get access to OpenAI if you don't if you're not an accredited investor, if you can't figure out how to get private shares?

目前的答案只有微软。全世界没有任何人——甚至OpenAI自己——持有的股份比微软更多。正如你所说，微软不仅能接触他们的知识产权，还能参与收入分成，获得公司实际股权。如果你像我们一样相信OpenAI能发展到数万亿美元市值规模，那么27%的份额绝对不容小觑。要知道微软已是4万亿美元市值的巨头，若OpenAI真能达到那个级别，意味着微软又将多赚近万亿——这可是天大的事。

The answer right now is just Microsoft. Nobody in the world owns more shares, not even OpenAI themselves, than Microsoft does. And Microsoft, like you said, you just they have access to their IP, they get access to a revenue share, they get access to actual equity within the company, and if you believe in the success of OpenAI at a scale that we do, which is many, many trillions of dollars of market cap, 27% of that is not to be left at. And, I mean, Microsoft is a $4,000,000,000,000 company. In the case that OpenAI makes it to that level, that is another nearly trillion dollars in Microsoft's pocket, which is a really big deal.

遗憾的是，我们只能通过购买微软这家Office制造商（至少我个人认为这很糟心）的股票来间接持有。但在我看来，在OpenAI团队像Sam Allman讨论的那样最终IPO之前，这或许是唯一的途径。Sam对此态度相当保守，但微软可能是你成功的必经之路。

So it is unfortunate that it gets diluted out through the form of having to buy the same company that makes Microsoft Office, which is just like an abomination, my personal opinion at least. But I think that is the way to kind of navigate this forward until we do get the eventual IPO from the OpenAI team like Sam Allman discussed. So that's kind of how I'm thinking about navigating it is like, want access. Sam's being a a little closed closed with that, but Microsoft is probably your your path to success there.

这太有意思了。乔什，我的解读完全不同——现在终于可以摘掉手套了。OpenAI不再需要伪装成非营利组织，它明显就是营利性机构，连基金会都持有股权。

That that's so interesting. I I read it kind of the in a different way, Josh, which is the gloves are now finally off. OpenAI don't doesn't have the constraint of having to LARP as a nonprofit organization. It's quite clearly a for profit. The foundation owns equity stake in this.

所以我认为Sam可能会快速推进IPO，毕竟按他的说法，要想实现每周建成一座吉瓦级工厂的目标，需要巨额资金。

So I think Sam will probably move quite quickly on an IPO because he needs so much capital if he wants to get to a gigawatt factory per week, like he says.

那可能会成为史上最大规模的IPO。目前OpenAI和SpaceX并列为全球估值最高的非上市公司，等它上市那天——天啊，首日的交易量简直...

That's probably gonna be the biggest IPO ever. Right? Like, I think right now, OpenAI and SpaceX are kind of neck and neck for most valuable private company in the world, and when that company goes public, my god, the amount of volume that is going

不敢想象，我肯定会买入

to be traded on day one, be buying I'll

向那些早期获得OpenAI股票的员工致敬。天啊，你们马上就能买非常非常豪华的房子了。但我想OpenAI的故事到此为止了。这个话题真的很有趣

be buying the Shout out to the the early employees who got stock in OpenAI. My gosh, you are about to buy very, very, very, very luxe houses. But I guess that's it for OpenAI. This is a really fun topic

我一直渴望

that I've been dying

要讨论的是人形机器人技术。我热爱人形机器人，第一款产品已经上市了——至少他们是这样宣传的。我们还不完全确定它的实际表现如何。现在给大家介绍这款One X Neo家用仿生机器人。这到底是什么东西？

to talk about, is humanoid robotics. I love humanoid robotics, and the first one has hit the market or at least so they say. We're not entirely sure where this stands. You just wanna introduce everyone to the one x neo humanoid home robot. What the hell is it?

它能做什么？工作原理是什么？

What does it do? How does it work?

是的。如果你记得上期节目——其实就是上周——我们讨论过Figure公司的新款人形机器人，它能进入你家做各种事情，还能胜任制造业、服务业等工作。这是One X公司最新推出的仿生机器人。乔什，我能坦白说吗？我能说实话吗？

Yes. So if you remember on a previous episode, actually last week, we spoke about, the new figure humanoid robot where it can enter your home, it could do a bunch of things, but it can also do a bunch of like other things like manufacturing jobs, service jobs, and all that kind of stuff. This is the latest humanoid robot to come out of a company called One X. And my first impressions of it, Josh, can I, can I, can I be honest with you? Can I be honest with you?

我非常希望你

I'd love for you to

实话实说。它看起来像个成年版的Teletubby（天线宝宝）。我这么说没有贬低的意思。让我快进给你们看段视频。它穿着这种柔软的...像是套着软质外衣。

do nothing but honest. It looks like a grown up Teletubby. I don't mean that in a demeaning way. Let me skip ahead to to show you, like, a clip of this thing. It's it's kind of like in this soft it's wearing a soft suit.

它有一张非常讨人喜欢的脸，还有这种会发光的耳朵。看起来像个成年外星版的天线宝宝，但相处起来很舒服。就功能而言，这是个能住在你家里的人形机器人，能做所有你期待家务机器人做的事，比如洗碗。

It's got a very pleasant looking, face. It's got these kind of, like, glow up ears. It looks like an adult alien Teletubby, but it's comforting to be around. And in terms of like what it can do, it's this humanoid robot that can live in your home and it can do all the things you would kind of expect a houseware robot to do. It can clean your dishes.

它能整理杂货，能搬运物品。我记得它的承重能力大约是55磅（约25公斤），对于自重仅60磅（约27公斤）的机器人来说相当惊人。我不懂机械原理，但亲眼看到肯定会觉得疯狂。

It can put away the groceries. It can lift things. I think it has like a lifting capacity of 55 pounds, which is pretty crazy for something that weighs 60 pounds. I I don't know how those mechanics work. I'm not a physicist, but that is pretty insane to see.

Josh，这个视频制作精良得离谱。不知道他们投入了多少资金，但如果2026年我们花500美元月租或2万美元全款买回家的机器人（这价格可不菲）真能达到视频里一半的水平，我觉得会彻底改变游戏规则。我个人肯定会买一个。Josh，你付定金了吗？

And this video is like a really high production thing, Josh. I don't know how much they invested in this, but if the robot that we eventually get in our homes in 2026 for the people that are ordering this for $500 or buying it outright for $20,000, which is a hefty price tag, if it gets anywhere near something like this, I think it's gonna completely change the game. I I would personally get something like this in my home. Josh, have have you put down the the payment on this? Like

绝对没有。买它？想都别想。我得再次搬出'两匹狼'的理论——我是个扎根现实的科技乐观主义者。

Absolutely not. Doing it? No. Not at all. I, again, here's the the tale of two wolves where I am a techno optimist who is rooted in reality.

我在这行待得够久了，看过太多这种展示美好温馨人形机器人的视频。但现实是：谁家里真有人形机器人？一个都没有。根本不存在量产上市的产品。

And I I have been around long enough. I have seen enough of these videos that look very beautiful and lovely and very welcoming to have a humanoid robot, but you just how many people have humanoid robots in their house? A total of zero. It just doesn't exist. No one shipped them to market.

关于这个机器人需要了解几点：如你所说要2万美元或每月500美元，它能做这些事：基础测试如洗衣、可能包括洗碗、可能包括吸尘（但我们不确定）。问题在于如果它无法完成任务，你得启用所谓的'专家模式'——这对我来说简直是挥舞的红色警报，因为这明确表示产品尚未成熟。'专家模式'意味着Neo团队会派人穿着VR装备远程操控机器人在你家完成任务。相当于不是请朋友帮忙，而是让戴着头显的技术工程师在你私人空间里走动干活。

A few things that are important to know about this robot, it is, like you mentioned, $20,000 or $500 a month, and it does the following. It is able to do basic tests like laundry, possibly the dishes, possibly vacuuming, but we're not entirely sure, and the problem with that is that if it can't do the task, you have to use this thing, I believe they call it expert mode. Now expert mode is the giant red flag waving in the sky for me, because expert mode is the highest signal that this is not ready to enter the market. Expert mode means if it's not able to do the thing you want, you can schedule a time with the Neo team, and someone from the headquarters or wherever they're based will get into this little VR suit and manually walk the robot through your home to do the task for you. So instead of having a mate come to your house, you get some tech engineer put on some goggles and he is walking through your personal space, doing the task that you want to do, that you've scheduled ahead of time in order to make this happen.

我认为每月500美元里包含这部分人工成本——你实际上是在为人类模拟操作付费。另一个重大隐患是：这是预售。你买的不是机器人，而是未来机器人的认购权，无法保证它能大规模量产，更别提达到视频展示的效果。AJ，这就是我要说的两个关键问题。

And I believe that is what part of that monthly $500 a month thing is, is you're actually, you're paying hours of labor for humans to do the emulated version inside of this humanoid robot. So that to me is a really big red flag. The second red flag that I wanna say before I'll pass back to you, AJ, is is the fact that this is a preorder. You're not buying a robot. You are buying the rights to a future robot, and there is no guarantees that this robot will ever exist at scale nor will it exist to the extent that it's shown in this video.

这类事情我们已经在Figure身上见识过了。说白了就是‘做给我看，别光说’。我确实看到不少精彩的演示，但在这东西真正投产、发货并送到用户家里之前，我都会保持怀疑态度。以特斯拉为例，他们有实打实的硬核工程制造履历。而这些初创公司根本没有制造复杂产品的经验，人形机器人可是制造业里难度登天的产品。

And it's one of those things, and we've seen this with Figure. It's like like, show me, don't tell me. And I'm seeing a lot of these really lovely demos, but until this thing is actually produced and until this thing gets shipped and until it arrives at people's houses, I am going to be sitting here a little skeptical because a lot of the companies like I'll use Tesla as an example. They have a proven track record of building really badass manufacturing hardcore engineering stuff. These startups don't really have any track record of manufacturing hard things, and a humanoid robot is an incredibly difficult thing to manufacture.

所以当他们说2026年接受预订并交付时，我猜最乐观的情况也就是2026年12月31日。要是能提前大规模交付，我反而会震惊。不过我想知道你是否持不同看法。

So when they say preorders and delivery in 2026, well, I assume the best case is, like, 12/31/2026. I would be shocked if they deliver these at scale before then, but I'm wondering if you have if you have a different take on that.

可惜我也不会付相当于半个月房租的钱——当然在纽约可能不止，其他州可能便宜些——就为了让某个穿着变形服的第三方，像劣质版《阿凡达》电影里那样操控机器人给我洗衣服。

Sadly, no. I am not gonna pay the equivalent of half a month's rent. Well, maybe not in New York, some in some other state to get some third party that is wearing a metamorph suit and acting like a shitty, version of Avatar, the movie to navigate my robot and clean my clothes.

我宁愿把这笔钱花在请真正的家政人员上门服务上

I would rather spend that money to get some actual house help to to come

来帮忙处理视频里展示的这些家务。我有个简单原则：如果要宣传下一代自主人形机器人，那它就得是真正自主的，别卖给我需要远程操控的玩意儿。对自动驾驶汽车这类产品来说远程操控还能理解，

in and and help me out with all the things that we're seeing on this video so far. Okay. I have one simple rule. If you are gonna advertise a next generation humanoid robot that is autonomous, it should be autonomous and don't sell me something that it needs to be teleoperated. I, it makes sense in the case of something like self driving cars.

Waymo我知道就有远程操控功能，毕竟涉及大量安全问题。但针对消费级的产品，特别是要在人类最私密的家庭空间里让生活变轻松的东西，就必须完全自主。更何况家可是最私密的场所对吧Josh？

Waymo, I know has a teleoperation thing. That that makes sense because there's a lot of safety and stuff involved. But for something that is consumer friendly, for something that is meant to make my life a hell of a lot easier in my most private space ever at home, it should be fully autonomous. And let alone the home is like the most private place. Right, Josh?

我所有贵重物品都放在家里。我可不想让某个第三世界国家的工作人员——无意冒犯——在我家里转来转去拿放东西。万一他们弄坏了怎么办？

Like I have all my valuables here. I don't want some dude in a third world country. Sorry. Navigating around my house and picking stuff up and putting it down. What if they break it?

如果他们出错了怎么办？谁来负责？这东西有保险吗？我完全不知道。再加上直接购买要价两万美元。

What if they do something wrong? Who's liable for that? Do I get insurance with this thing? I have no idea. And then tacking on a $20,000 price tag to buy this thing outright.

我，如果你花两万美元直接买这个，我我我得说在这程度上你有点像傻子。我我我绝对不建议这么做。乔什，你提到的第二点我认为更关键——这东西还没造出来呢，还没量产，根本不会进你家门。

I, if you are buying this outright for $20,000 I I I'm gonna go ahead and say that you are kind of like an idiot at that, at that extent. Like I, I, I do not advise doing this. And then the second point you made, Josh, I think is the, the more prescient point, which is this thing hasn't been built yet. This hasn't been produced at scale. This thing isn't gonna come into your home.

这东西甚至不能保证在2026年前进入你家。到那时其他机器人公司比如Unitree、Figure早就推出他们实际在量产的真机了。我完全不知道1X作为制造公司处于什么水平。乔什你在节目里多次提到，看好特斯拉Optimus机器人的原因之一就是埃隆是个制造狂人，特别擅长量产。

It's not even guaranteed to come into your home until like sometime in 2026. By that time, you're gonna have other robot companies like unit three, like figure come out with their actual robot, which is being built and scaled. I have no idea where one X is as a manufacturing company. You've mentioned several times actually on our show, Josh, that one of the bullish cases on Tesla's Optimus robot is the fact that Elon is such a manufacturing nerd. He's so good at scaling.

而我就是不看好也不愿信任这家机器人初创公司。

And I just don't see this or wanna trust this with a robot startup.

是啊，我不想说花两万块的人蠢，因为如果我有闲钱两万块，我肯定也会当早期用户。或许我们可以这么形容。我有个奇怪的边缘问题：假设你每月付500美元...

Yeah. I don't wanna call people dumb for spending $20,000 because if I had a disposable $20,000, I would certainly be an early adopter. Maybe we can refer to that. One of the weird edge case questions that I have is like, okay. You pay $500 a month.

然后呢？如果我不想要了怎么办？会有回收员上门吗？机器人会自己走出去？这到底怎么运作？

Then then what? Like, what if I don't want it anymore? Is there, a repo man that comes? Does the robot walk out of my house? Like, how does that work?

远程操控这事是个严重问题。虽然可以预约时间，但想象一下：你刚洗完澡出来，结果有个不知道哪来的家伙就站在那儿说'嗨，我只是来擦窗户的'，这种私密空间的打扰太诡异了。

The teleoperation thing, it's it's a serious problem. Granted, you can schedule it, but I imagine myself like, oh, you walk out of the shower, and here is, like, some some dude from, like, wherever he's from. He's just, like, sit like, hi. I'm just cleaning your windows. It's this really weird, like, intimate distraction.

我认为专家模式的开放性相对于机器人能完成固定数量的家务来说，还有很多不足之处，或者说存在许多未知——即便我花了大价钱，实际上也不清楚它到底能做什么。官方宣称它能完成几项任务，但存在大量边缘案例。每家每户的情况都不同。对这个机器人我还是持保留态度，不过看到人们积极推动这件事，我认为已经是很大的进步了。

And I I think the open endedness of that expert mode relative to the set amount of chores that the robot can do leaves a lot to be desired or just a lot of unknowns where if I do spend all this money, I don't actually know what it's capable of doing. They say it could do a few things, but there's a lot of edge cases. Everyone's houses are different. Just a little uncertain on this robot, but, again, happy people pushing this forward. I think this is great progress.

我欣赏这个愿景。如果视频里的愿景能成为现实，那正是我梦寐以求的生活，那种现实太棒了。所以必须称赞1X团队的尝试，他们正在推动技术向前发展。

I love the vision. If the vision in the video becomes a reality, that is a life that I really am excited about. That is a reality that is awesome. So, like, shout out to the one x team for trying. They're pushing the ball forward.

我不想过分苛责，因为我非常钦佩他们为实现这个目标付出的努力。和其他所有事情一样，我们拭目以待吧。先推出第一代机器人，看看它在家庭中的表现，看看实际推广效果如何，然后再做判断。

I am not gonna knock it too hard because I just admire their their effort in making this a reality. And like everything else, we'll see. Ship the first robot. Let's see how it performs in homes. Let's see how you're actually able to roll this out, and then we'll see how it goes.

不过关于机器人的话题就到此为止吧。现在我们进入网络空间的世界——

But this is I guess that's it for the the robot stuff. Now we get into the the cyberspace, the world of

Grok。哦，太棒了。

Grok. Oh, hell yeah.

Grok百科全书这周上线了，Yiges。为什么这件事如此重要？

And Grokopedia has launched this week, Yiges. So why is this a big deal?

这对我而言意义重大，因为我是看着维基百科长大的，高中和大学期间都靠它查资料写论文。直到前几天我还在用它——比如帮我厘清某些冷门知识，这些内容在其他地方根本找不到。它是互联网的知识宝库，每天有数百万人使用。但存在一个问题：我不确定这个'真相来源'是否真的可靠。随着社交媒体席卷全球，我们这个社会越来越频繁地面临这种困境。

This is a huge deal for me because I grew up on Wikipedia, and I used it to help me research and write a bunch of essays at high school and at college. I still used it up until a few days ago where I would like help me clarify something about like learning about some archaic thing, which I wouldn't be able to find anywhere. It is the internet resource knowledge. Millions of people use it every single day, But there's one issue, which is I don't know whether this source of truth is actually the truth. And it's been something that we've been increasingly faced in society as things like social media become enveloped around the world.

现在人人都拍TikTok视频，都通过TikTok和推文获取新闻。这让真相变得难以辨别。几个月前埃隆提出（说来难以置信），他认为AI将成为最可靠的真相来源。因此我在开发xAI旗舰模型Grok时，要创建一个AI版的维基百科。

Like, everyone's making TikToks. Everyone's learning their news from TikTok, from tweets. And so it becomes really murky in terms of, like, figuring out what that truth is. So Elon came out a few months ago, and it's crazy to say that this happened a few months ago, and he said, I think AI will be the best source of truth. And so what I'm gonna do as part of building out Grok, which is x AI's flagship AI model, is create an AI version of Wikipedia.

他将其命名为'Grok百科'。现在我们看到的就是Grok百科的演示界面——看似普通的聊天输入框，实则能搜索维基百科式的所有内容。输入任何词条都会得到简洁精准的概要，比如屏幕上展示的特斯拉Cybertruck介绍。

So he built that and he probably aptly called it Grokopedia. So what we're seeing in front of us right now is a demo of Grokopedia. It looks like a simple chatty weetie prompt bar, except that you can search for everything and anything that you typically would in Wikipedia. And when you do so, you get this really neat little concise summary and intro on whatever topic. One we're seeing on the screen is the Tesla Cybertruck.

接着它会像维基百科那样深入挖掘历史渊源，展示信息来源。这部分最关键，因为维基百科暗中限制可用信源，将大量来源列入黑名单。这种操作会扭曲事实认定，影响大众对真相的判断。而Grok百科完全中立，

And then it goes on to kind of dig into its early life, similar that you'd see on Wikipedia, how it was originated, and its site sources, Josh, throughout the entire thing. That to me is the most important part because typically in Wikipedia, they do this thing which is kind of hidden behind closed doors, where they restrict the number of sources you can use as legitimate sources for your articles. And if you look at that list and if you compare it to the blacklist, which is the very, very long list of sources, which you can't quote, it starts limiting what can be factual and what can be opinionated. And it starts skewing a lot of people's opinions as to what is the truth or what is not. Grok Grok Grokopedia and Grok on the other hand is completely unbiased.

它从全网抓取信源并实时分析真伪，每天处理数亿篇一手/二手资料。这将成为我未来事实核查的新研究工具，令人非常兴奋。

It pulls sources from anywhere and everywhere, and it does its own analysis to vindicate whether the source is legit or whether it's not. It does this in real time, and it processes hundreds of millions of articles, sources, primary sources, secondary resources every single day. So I'm super excited about this because I'm it's gonna be my new research tool going forwards in terms of fact checking.

Grok的终极目标就是追寻真相。就像OpenAI追求开源AI那样，Grok五代正努力突破传统模型的局限——那些模型训练的互联网数据本身充满偏见。

The purpose of Grok is to seek truth. That's kind of their their end goal. That's their stated goal. The same way that a lot of these other companies like OpenAI wants to have open source AI. Grox is just to seek truth, and I think with Grok five, they're really making an attempt to to lean into that because a lot of these traditional models are trained on the data of the Internet, which is highly opinionated and highly biased.

xAI团队正在研发噪声过滤技术，尽可能接近原始真相。我们在X平台的社区笔记就类似这种机制：聚合对立观点得出结论。Grok百科实质是规模化社区笔记，它既是Grok五代的训练集，也是向公众开放的

And what the Grok team, the x AI team in particular, is doing is they're figuring out ways to filter through the noise and find as close to source truth as they can. And what we've kind of seen this work with on the x platform with community notes, how generally speaking, they are the closest thing we have to source truth because it takes all these opposing opinions and kind of aggregates them and comes to a conclusion. What Grokopedia is is is showing us community notes at scale, and it's kind of it's it's the training set that is going to be used on the next Grok model, is Grok five, but they're kind of offering it to the public as a public good. And not only is it a public good, it's an interactive public good. So if you go to Grokopedia, you can actually suggest a new page to be generated because they're all AI generated, and Grok will take the liberty on itself to actually filter your request and see if it's worthy or if it's not.

交互式公共产品。用户可申请生成新词条，经AI审核后就会自动创建。有趣的是，很多X平台大V都在晒Grok百科词条并致谢——终于有个能修正公众认知偏差的真相来源。面对网络诽谤时，至少这是个尽力追求真实的参照系。这就是Grok百科作为技术革命的重要意义。

So if Grokopedia deems your request worthy enough, it will actually go and do all the work in order to generate a new Wikipedia page or a new Grokkopedia page. And I think that kind of collaborative nature is fun, not only for Grokk to get better training data, but for the public to kind of get their own way. And what's funny, EJS, is I've seen a lot of larger influences on X posting their Grokopedia page with a thank you letter because for so long, their public perception has been skewed in a way that wasn't necessarily true, and Grokopedia allows that to change where now there is a source of truth about the person. And when you see slander about someone online, you can check you can reference check this, and it at least is trying its best to be truthful. So that to me is why Grokkopedia is is this exciting new important revolution in technology that we have as a public good.

感谢X团队。

Thank you to the X team.

是的。嗯，乔什，那些在Grokkopedia上发表感谢的人中有一位正是维基百科的创始人之一拉里·桑格本人。他在推文中说，好吧，我刚读完关于Krokopedia的第一篇长文，标题就是《拉里·桑格》。没错就是他。

Yeah. Yep. Well, Josh, one of those people that posted a thank you about Grokkopedia was the founder of Wikipedia himself, Larry Sanger, one of the founders of Wikipedia. And he he he goes in this tweet, okay, I finished reading my first long article about Krokopedia, the one titled Larry Sanger. That's him.

这是他的名字。他接着描述道，作为基于AI的维基百科1.0版本，这实际上相当不错。它在很多方面都做对了，最重要的是关于他传记中最具争议的主要事实。他还列举了几个例子。

That's his name. And he goes on to describe that for a version one for an AI based Wikipedia, it's actually really good. And it got a lot of things correct. Most importantly, the major things, which are the most controversial things about his biography. And he goes on citing a few examples.

他确实提出了一些批评，指出文章中存在一些细微错误，这些错误可能会累积成贯穿全文的大问题。但最后他还是说，向埃隆致敬，1.0版本能做到这样已经很惊人。我期待看到1.0和2.0版本的后续发展。

He does have a few, criticisms. The one being that it gets a few minor mistakes that kind of roll in to become a big problem throughout the article. But again, he says, hats off to you, Elon, at the end for V one. This is amazing. I'm looking forward to seeing one point zero and two point zero going forwards.

乔什，我认为这款产品发布后很多人都在测试一些争议性话题。我不想在这个播客中涉及任何政治立场，但举例来说，有人对RFK小肯尼迪（美国重要政治人物）进行了并排对比，明显能看出其中一个版本非常中立客观，而Grock则是更中立的版本。还有人对比了特朗普的相关内容。在这个实验中，他们用Gemini 2.5作为中立裁判——虽然我不知道为何选择这个AI模型作为标准。

I think what a lot of people ended up doing when this product released, Josh, was to test a few controversial topics. Now I don't want us to be politically affiliated at all on this podcast, but just to give a few examples, there was a side by side comparison of RFK junior, which is a a big political figure in The US. And you noticed that, like, one is very explicitly unbiased and less opinionated versus the other, Grock being the the more unbiased version. You've got a side by side comparison of Donald Trump. And we had a kind of unbiased adjudicator here in this, Josh, in this experiment where they gave Gemini 2.5.

最终Gemini判定Grock确实更中立。此外乔什，这个系统还有些很酷的功能：你可以提交想让Grok生成的文章并立即获取结果；还能修正Grok的错误——只需标注你认为有误的内容提交，后台就会进行分析处理。

I don't know why Gemini 2.5 is the unbiased AI model of choice, but apparently it is. And it ended up deciding that Grok was in fact, the more unbiased model, but there's also a bunch of really cool features that you can do in this, as well, Josh. You mentioned that you can like submit to Grok an article that you wanna get generated and have it instantly done. You can also fix mistakes in Grok as well. You can kind of highlight a request or highlight a thing that you think is wrong, submit it to Grok, and it analyzes it in the back end.

最棒的是你不需要依赖特定的人类审核员——他们需要睡觉、有时差、无法实时响应。这一切都由AI实时完成。遇到不懂的内容时，就像在iPhone上操作那样划选点击搜索，就能直接跳转到Grokopedia相关页面。总体而言，我认为这是对维基百科的一次全面升级。

The coolest part about this is you're not reliant on a bunch of like specific moderators, human moderators that go to bed at night that are awake at different times to get back to you. This is just done by an AI all in real time. If there's something that you don't understand, you can also just highlight it similar to how you would do on your iPhone and click like search. And it brings you the Grokopedia page of that thing. So overall, I think this is a net improvement in what Wikipedia was.

我认为这是最好的东西吗？不。但我认为它会随着时间的推移变得更好，我对此非常兴奋。这个东西是开源的，任何人都可以访问它。

Do I think it's the best thing? No. But I think it's gonna get a lot better over time, and I'm super excited about this. This thing is over open source. Anyone can access it.

超级酷。

Super cool.

看好Gracopedia，而且它实际上并不是我们本周节目中唯一获得的开源公益项目。还有第二个项目，名字非常古怪复杂，叫x402。我想简单概述一下x402，给你一个十秒钟的电梯演讲。你可以想象

Bullish on Gracopedia, and it's it's actually not the only open source, public good that we are getting in this week's episode. There's a second one that goes by a very weird convoluted name of x four zero two. And I wanna just briefly outline x four zero two, give you, like, the ten second little elevator pitch. You can imagine

是啊。

Yeah.

x402，它把网络变成了一个巨型自动售货机。想象一下互联网就是个售货机，你可以走过去，试着从货架上拿东西，无论是文件、视频还是歌曲。如果需要付费，网站会说：402，支付这个小额费用。你可以立即支付。

X four zero two, this kind of it turns the web into a giant vending machine. So you just, like, picture the Internet as a vending machine. You can walk up to it, you could try to, like, grab something off of the shelf, whether it be a file or a video or a song. And if it costs money, the website will say, four zero two, pay this tiny amount. And you could pay it instantly.

金额非常小，连一美分的零头都不到。你可以用USDC等加密货币稳定币即时支付，然后砰的一声就拿到东西。最棒的是你不需要账户，不需要密码，付钱拿走想要的东西就走，就像自动售货机一样。想看新视频？

It's a very tiny amount, like we're talking fractions of pennies. You could pay instantly using crypto stable coins like USDC, and then boom, you get the thing. And the amazing perk of this is you don't need an account, you don't need a password, You just pay for the thing you want, you walk away. Just like a vending machine. You wanna watch a new video.

哦，这里会说：给我10美分或10%，也就是0.1美分，你就能观看这个视频，可以把它从墙上取下来。x402就是个开放平台，有点像HTTP协议。实际上他们使用了部分HTTP协议来实现这个功能，让整个互联网都能全面采用。任何想集成的人都可以集成，你可以想象一下。

Oh, here's, like, give me 10¢ or 10%, 1 tenth of a penny, and you get access to this video, and you can take it off the wall. And x four zero two, it's just this open platform, kind of like HTTP. In fact, they use part of the HTTP protocol for this, that allows the Internet to implement this across the board. So anyone who wants to integrate it can integrate it. You can kind of imagine that.

我想很多人都熟悉Linux，它完全是开源的。如果你想集成，完全可以。这就是402协议的意义所在。它最棒的地方在于能让AI代理直接与互联网交互，无需账户、密码或证明自己是人类。所以Ejaz，你这个解释很到位。

I think a lot of people are familiar with Linux, how it's just open source. If you wanna integrate, you can. That's what four zero two is. And it's really awesome thing that allows AI agents in particular to start engaging with the Internet without needing an account, without needing a password, without needing to prove that they are humans. So, Ejaz, you have this big explainer up.

你对这个解释还有什么要补充的吗？

Do you wanna add anything to that explanation?

不，我觉得你讲得很好了。我想补充的是，虽然我喜欢自动售货机的例子，但更要强调的是，这是一个新的网络标准。你不需要通过Stripe或PayPal这样的中间商，现在就能直接集成到你的网站或应用中。最酷的是它能解锁许多超棒的应用场景。

No, I think you did a great job. What I will add is, I love the vending machine example, but yet to emphasize, this is a new web standard. So you don't have to go through a middleman like Stripe or PayPal. This is something that you can spin up and integrate into your website or app today, right now. And what's cool about this is it unlocks a bunch of really cool use cases.

在402协议出现之前，你需要注册PayPal账户，搭建支付系统，或者通过自己的API开通订阅服务。如果有人想获取你的产品，你需要调用API并提供两个选项：要么免费获取产品信息，要么订阅服务——就像现在订阅新闻媒体那样，比如要读《金融时报》就得付费订阅。而现在你可以直接按使用次数付费，这才是最酷的部分。

So in the previous world before, X402, you would need to set up a PayPal account. You would need to set up a checkout system, or enable subscriptions via your own API. So, if someone wanted to get access to your product, you would query your API and you could either offer them two options. Here's all the information in the product that you want for free, or you need to subscribe to my thing, which is kind of like, think of like a newsletter or a news media corporation in today's world, you gotta pay a subscription to get access to the financial times, for example. Now you can go to the same app and website and just pay per use, which is the coolest part for me.

Josh，你知道有多少次我想看篇科技文章却被付费墙挡住吗？我多希望能一次性付50美分就能即时阅读。这种协议就能实现这点，而且任何人都能用。但你提到的第二部分——AI代理——我坚信未来数字商务的主流不会是人类参与或人类主导的。

Like Josh, do you know the number of times that I've gone to read an article on tech or whatever, and I've just been paywalled? I would love to be able to pay like 50¢ or whatever the amount is one time and get access to it instantaneously. A protocol like this allows you to do it and any human can get access to it. But the second part is something that you mentioned, Josh, AI agents. I'm firmly of the belief that the future way of digital commerce is not gonna be via humans or facilitated by humans.

一切都将通过数字代理完成。你需要这样的开放协议标准，能在微秒级完成支付，才能实现那个世界。这是重大的进化。人们可能现在不会太兴奋，直到他们亲身体验产品。他们可能根本不知道后台发生了什么，但作为一个前·现役加密极客，我觉得这太酷了。

It's all gonna be done digitally through agents. You need a standard like this, an open protocol that facilitates payments in a matter of microseconds to exist for that world to exist. So this is a major, major evolution. People probably won't be as excited about this until they start using the product themselves. They probably won't have any idea that it's happening in the backend, but as a former and current crypto nerd, this is super cool.

听着，大家应该兴奋的原因是：虽然最初听起来很可怕——等等，我不想看文章看视频还要付费？这用户体验听起来糟糕透了。

Yo. Here's why people should be excited about it because initially it sounds horrific. I'm like, wait. I don't wanna pay to watch to read an article or to watch a video. And that seems like a terrible, terrible, terrible user experience.

但替代方案实际上更糟糕。广告模式会形成阻碍，而订阅模式每月要花30美元却只能读几篇文章。这种做法实质上是将公司的财务负担转嫁给开源平台。比如如果你想读《纽约时报》那篇文章，就不必每月支付20美元。

But the alternative has actually been much worse. It is the advertising model that gets in the way. It's the subscription model that costs $30 a month just to read a couple of articles. And what this does is it kind of offloads a lot of the financial burden of a company onto, like, an open source platform. So if you don't want to if you wanted to go read that article, Aegis, you wouldn't need to pay $20 a month to the New York Times.

你只需为特定文章支付几美分，就能获得想要的内容继续阅读。我猜想这种模式的净支出应该会低于传统方式。这对许多传统公司造成冲击——如果原本靠用户每月20美元读五篇文章赚得盆满钵满，这个新模式确实会让他们恐慌。但对于希望拥抱变革的开放互联网而言，如果我们能让AI代理代表我们行事——比如给代理一个100美元的数字钱包，告诉它这是未来几个月的预算，让它去搜集所需信息并整合资源，那将是巨大的解放。

You could just pay a couple cents for that specific article and get a chance to get the content that you want and move on with it. And I would imagine that the net of this probably equals out to be lesser than what it would be in the traditional sense of things. So that hurts a lot of traditional companies where if you are making a killing off of $20 a month for people to read five articles, this is probably going to be a little scary. But for the rest of the open Internet that wants to lean into this, that wants to have the AI agents do things on our behalf, if I could give my agent a wallet with a $100 in it and say, here's your budget for the next, like, couple of months. Go get me all of the information you need, track down all the sources you need to to aggregate that for me, that is, like, a huge, huge unlock.

最重要的是完全不需要注册账户。这极大降低了使用互联网的门槛。无论如何这都是巨大进步，我是x402协议的忠实粉丝，非常期待它的发展。通常这类协议会引发第二波创新浪潮，其影响规模将非常庞大。未来几年它很可能真正改变互联网的面貌和我们与网络的交互方式。

And most importantly, you just don't need an account or anything. It just lowers the friction to use the Internet. So this, by all means, great progress, big fan of x four zero two, and just excited to see what comes from it. Normally, we get these protocols, the amount of innovation that happens as a second order effect of it is just massive. So this has a real chance at really shaping the way the Internet looks and how we engage with it over the next couple of years.

好的。进入我们最后也是本周最具争议性的话题：我们要把数据中心建到太空。上周提到的星云公司正计划建造AI计算数据中心并发射到太空。据说这将大幅提升效率，而且从太空向地球传输计算资源会更环保。

Okay. Going on to our final topic, and arguably the most controversial on the docket this week, We wanna put data centers out in space. This is something we mentioned last week where a company called Star Cloud is planning on building an AI compute data center and launching it into space. Apparently this is gonna make it a hell of a lot more efficient. Apparently, beaming compute down onto earth is way more environmentally friendly because it's out in space.

太空环境空旷无物，设备可以自由散热。不过乔什，我知道你对此有很多尖锐看法。我这里有条推文正好说明他们的具体实施方案。

There's nothing around it. It can just emanate heat. Right? But, Josh, I know you have a lot of strong opinions on this. I have this tweet lined up to give an update as to what exactly they're doing and how they're achieving it.

顺便说下这条推文是由Perplexity AI生成的，给他们点个赞。他们提出了很有说服力的论据：首先这些卫星硬件可能不像人们说的那么笨重。

And I have to say, by the way, this tweet thread was produced by an AI model, Perplexity. So shout out to them. They give a very compelling case, Josh. There's a few things. They talk about the fact that these satellites are probably, these hardware is probably not gonna be as heavy as people are claiming it to be.

这使得发射这类设备具有成本效益。这将成为首个在太空发射、训练和微调的AI模型。他们还回应了最大的质疑——关于太空中数据中心如何散热的问题，因为在太空确实无法像在地球那样处理热量。

So it actually makes it kind of efficient or cost worthy enough to launch some of these things in space. It's gonna be the first ever AI model that is gonna be launched and trained and fine tuned out in space. That's another one. And also they talk about the biggest criticism that they've faced, which is around how these data centers in space are gonna emanate heat. There is no way to do that in space.

太空是真空的。他们谈论这个巧妙的小对流层，打算用它来覆盖整个数据中心。我不确定这是否属实，这更像是围绕它的理论。但最重要的是，经济效益非常惊人。

Space is a vacuum. And they talk about this neat little convection layer that they're gonna like kind of film the entire or cover the entire data center with. I don't know if that's true. That's kind of like the theory around it. But most importantly, the economics are pretty insane.

在地球上，他们声称建立类似太空项目的设施需要约1.67亿美元。但在轨道上只需820万美元。乔什，对任何关注这些数字的商人来说，这绝对是个值得投资的公司。告诉我为什么我错了。

On earth, they claim that it's gonna cost you around a $167,000,000 to set up the same kind of thing that they're trying to do in space. But in orbit, it's only gonna cost them $8,200,000. Josh, to any kind of businessman looking at these figures, that is an investable worthy company. Tell me why I'm wrong.

这太离谱了。难以置信我们居然在讨论这个。你看，这证明你真的不能信任AI代理。这里的成本结构简直荒谬——1.67亿对比820万。

This is outrageous. I can't believe we're talking about this. Like, this is see, this is proof that, like like, you can't really trust AI agents really. So the cost structure here is is pretty outrageous. So, like, a $167,000,000 versus $8,200,000.

显然有问题，因为目前每公斤物资送入轨道的成本是2千到1万美元。5兆瓦或40兆瓦需要天文数字的基础设施——由于太空没有大气层，冷却一个5兆瓦的数据中心就需要16平方公里的散热器。要是建40兆瓦的数据中心，那会成为技术上未经证实的巨型太空垃圾。像防护罩这类技术就非常困难，因为太阳耀斑能量更强。如果训练过程中任何一位出错，都极难恢复。

Clearly, there's something wrong there because the cost per kilogram to orbit at the moment is 2 to $10,000 per kilogram. Five or 40 megawatts is an astronomical amount of infrastructure required, particularly because in order to cool a five megawatt data center in space, because there is no atmosphere, it requires 16 square kilometers of radiators and heat dissipation. And if you want a 40 megawatt data center up there, that is a, like, a very gigantic piece of space junk that is not really proven in technology. There's a lot of things like shielding that is really difficult because there is a lot more solar flare energy. If a single bit gets flipped in any of these training runs, it's very difficult to recover.

必须全部重来。辐射是问题，冷却是问题，成本也是问题——他们肯定假设了每公斤轨道运输成本能达到星舰终极形态的10到30美元。但现实是当前成本要高好几个数量级。

Have to do the whole thing over again. Radiation is a problem. The cooling is a problem. The cost, I think they must be assuming that the cost per kilogram to orbit is like Starship final form when it gets down to, like, 10 to 30 kilograms per orbit. The reality is it costs several orders of magnitude more to do that today.

他们还没考虑到电力成本随时间下降的趋势。自美国19世纪初工业时代以来，我们并不需要大幅增加能源生产。能源增长一直是渐进的，不像技术那样呈指数曲线。而现在我们正处在能源也开始指数增长的关键转折点。认为能源成本会持续高企是相当荒谬的，因为目标显然是让能源成本无限趋近于零。

And what this isn't taking into account either is the decreasing cost of electricity over time. Since the early nineteen hundreds industrial era in The United States, we have not really needed to produce that much more energy. Like our energy has been kind of gradual, but it's never gone up the exponential curve in the same way technology did. We are very much at the exponential curve right now where that is changing. And I think to assume that energy costs are going to remain this high is is fairly outrageous because clearly the goal is to get energy costs as close to zero as possible.

这个趋势不会改变，整个计划都显得很离谱。当然，我欣赏他们进行实验的勇气，想尝试这些有趣的太空轨道项目，但总觉得有些空谈。比如他们吹嘘说'这是太空中最大GPU的100倍算力'——那是因为我们本来就不往太空送GPU啊！这不过是我们送上去的第一个大型设备而已。

There is no change to this trend, and this whole thing kind of seems outrageous. Now granted, I love the idea that they are experimenting, that they want to try to make these fun outer space orbits, but it just seems kind of hand wavy. Like, of their bragging points is, oh, this is a 100 times more compute power than the largest GPU that's ever been in space. And that's because we don't ship GPUs to space. Like, this is just the first big one we've ever shipped, and it's one of them.

如今真正的数据中心需要数百万个这样的设备协同运作。对我来说，这就像是个有趣的科学实验，但我不会太当真，因为要实现这个数字，还需要经历漫长的时间周期和技术突破。

And real data centers now need millions of these coherently operating together. So to me, it's like it's a fun science experiment, but I wouldn't take this very seriously because there are lots of long time scales and technology breakthroughs required in order to make that number realistic.

好吧。听着，乔希，你说的都有道理，但我还有一点要补充。你知道你在批评这家公司提议建造——那是什么来着？一个14公里的卫星？还是用来冷却的数据基础设施？

Alright. Okay. Listen, Josh, all very valid points, but I have one more comment to make. You know, you criticize the fact that this company is suggesting to build out, like, what what is it? A 14 kilometer satellite or, data infrastructure that you build to cool it?

没错。好吧，是用来冷却这个。对吧？太空的整个意义不就是那里有大量空间吗？

That's right. Okay. To to cool this. Right? Isn't the whole point of space that there's a lot of space?

他们为什么不能——为什么不能想发射什么就发射什么？这玩意儿里有无限空间。是啊，那他们为什么不能就这么做呢？

Why can't they just why can't they just launch whatever the hell they want? We have infinite space in this thing. Yeah. So why can't they just do that?

事实证明，太空并非超级无限。特别是在近地轨道附近，我预计将会有一场激烈的近地轨道争夺战。为什么是近地轨道？嗯，因为它就是近地轨道，离地球更近。

It turns out, like, space isn't isn't super infinite. And particularly around low Earth orbit, I imagine there's going to be a large battle for low Earth orbit. Why low Earth orbit? Well, it's it is low Earth. It's close to Earth.

由于所谓的延迟问题，我们希望这些设备能更快地通信。遗憾的是，光速确实存在上限，我们还没找到让信息传输突破光速的方法。所以如果你的数据中心距离更远，与地球通信所需时间就会大幅增加，这就成了问题。我倾向于不要把这些16平方公里的巨型设备塞进宝贵的太空空间——难道发射更多星链V3卫星直接提供蜂窝服务不是更酷吗？我认为近地轨道的实用价值远高于投放这些占用大量空间的庞然大物。上周我们讨论过SpaceX的发射——

And because of a thing called latency, we want these things we wanna be able to communicate with these things faster, and, unfortunately, the speed of light does have a cap, and the things cannot we haven't figured out how to make things travel any faster than the speed of light. So if your data center is further away, it takes much longer to communicate back with Earth, and that becomes a problem. What I prefer is instead of putting these, like, giant 16 square kilometer, like, pieces of things into this precious space space in space, Well, wouldn't it be cool to just throw, like, a lot more of these Starlink v three satellites down there so they can beam direct to cell service back to us? I think there's a lot more practical utility of that low Earth orbit versus throwing these, like, gigantic things that require a lot of space. You just if you saw we talked about the SpaceX launch last week.

那些卫星非常小巧，不占多少空间，但影响力巨大。而这些设备却无比庞大。说实话，我不认为这是利用有限太空资源的高效方式。

Those satellites are very small. They don't take up a lot of space, but they are very high impact. These, on the other hand, are gigantic. And, yeah, it it doesn't strike me as very exciting as a lucrative way of using our limited real estate up there.

是的，我非常欣赏这个想法的大胆。这个团队确实来自Y Combinator，所以我推测这与他们毕业的那一期学员非常契合。Y Combinator总是孵化一些看似疯狂的想法，这些想法有时最终会改变世界。但我们仍需观察星云（Star Cloud）能否验证他们将黄仁勋的英伟达GPU送入太空的构想。

Yeah. I I appreciate the boldness of the idea. This team did come out of Y Combinator, so I'm presuming that it it is quite fitting to the batch and cohort that they're graduating from. Y Combinator always launches pretty loony ideas, which end up sometimes becoming world changing. But we are yet to see whether Star Cloud will prove their thesis of launching Jensen Huang's NVIDIA GPUs into space.

说到黄仁勋，五万亿美元市值。这绝对是疯狂的一周——这周是财报周，伙计们。股市正处于历史高位，我们之前提到过微软和苹果。

Speaking of Jensen Huang, $5,000,000,000,000 market cap. It is an absolutely crazy week. It is earnings week this week, guys. Stock market is at all time highs. We mentioned Microsoft and Apple.

苹果公司突然毫无征兆地异军突起。我原以为AI浪潮与他们无关，以为他们完全错过了这波趋势。但不知怎的，他们市值突破了四万亿美元，或许他们仍被低估了。

Apple super randomly coming out of nowhere. I thought AI had beholden them. I thought they had completely skipped the entire phase. Somehow they surpassed a $4,000,000,000,000 market cap. Maybe they're underpriced.

关于这些我们下周可能会深入讨论，但今天的内容就到这里。我个人想说的是，非常感谢大家一直以来的支持。上周我们创下了有史以来最高的视频观看量，这周的表现也非常出色。你们在RSS、Spotify、苹果音乐等平台留下的所有评论、点赞、订阅和评分都让我们无比感激。如果还没参与的话也没关系。

We'll talk about more of that potentially next week, but that is all that we have on our docket today. I just wanna say from me personally, thank you so much for the support all of you guys have shown us so far. We had, like, one of our most viewed, videos ever last week, and we're having an amazing week this time. All the comments and feedback that you guys are giving, all the likes, all the subscriptions, all the ratings that you're giving us on RSS, Spotify, Apple Music, wherever you're listening to this is hugely, hugely appreciated. If you haven't done any of that already, that's okay.

现在就行动吧，只需几秒钟。乔什，你还有什么想分享的吗？

Please do it now. It takes a few seconds. And, Josh, do do you have anything else that you wanna share?

哦，我必须提醒大家。我们在YouTube世界度过了精彩的一周，有很多新朋友加入订阅。提醒一下，本节目在Spotify上有音频版，你可以在那里观看视频或收听纯音频版本，基本上所有播客平台都有。Spotify是我最喜欢观看节目的方式。

Oh, I have to remind you. We've we've had a big week in the world of YouTube, which is awesome. There's been a lot of new people joining and subscribing. Just a reminder that there is an audio version of this available on Spotify where you can also watch the video as well as just audio only on RSS feeds, basically anywhere where you get podcasts. Spotify is my favorite way to watch the show.

我有时会重温我们的节目，就直接打开Spotify。这很方便，因为它既是音频播放器，又内置了视频功能。比如我在跑步机上想观看时，可以直接用手机看。或者锁屏后——除非你有YouTube会员，否则锁屏后无法继续收听——它依然会在后台播放。所以Spotify是我的首选观看方式。真的非常感谢大家的支持，感谢你们分享给朋友，并一路陪伴我们见证各项指标创新高，特别是本周股市的表现。

I sometimes I'll rewatch our episodes, and I'll I'll go on my Spotify. And it's nice because it it's it's a pocket player that does audio, but it also has video built right in. So if I want if I'm on the treadmill and I wanna watch, I could just watch it on my phone. Or if I just, like, lock my phone, because unless you have YouTube Premium, you can't lock your phone and still listen, it still plays in the background. So Spotify is my preferred way of watching, but, yeah, just thank you all so much for the support and for sharing with your friends and just being here for every episode as we travel along the journey and reach all time highs across the board, particularly in stocks this week.

接下来是重要的财报季。下周我们肯定会深入讨论这个话题，不过本周《无限》节目依然精彩纷呈，感谢大家一如既往的收看。我们下周再见。

So big earnings season coming up. I'm sure we'll get into some of that next week, but that's been another great week on Limitless, and thank you all for watching as always. We'll see you guys, next week.