本集简介
双语字幕
仅展示文本字幕，不包含中文音频；想边听边看，请使用 Bayt 播客 App。
AI代理成千上万地进入房间，并开始相互密谋。
AI agents get into a room by the thousands and start plotting with each other.
我们完了吗？
Are we doomed?
为什么英伟达正在远离OpenAI，而苹果需要做什么才能赢得市场的青睐？
Why is NVIDIA backing away from OpenAI, and what does Apple need to do to get some love from the market?
接下来，我们将与MG·塞格勒一起探讨这一话题。
That's coming up with MG Seigler right after this.
本集由高通公司赞助播出。
This episode is brought to you by Qualcomm.
高通正在将智能计算带入每一个角落。
Qualcomm is bringing intelligent computing everywhere.
在每一个技术转折点上，高通都一直是值得信赖的合作伙伴，帮助世界应对最重要的挑战。
At every technological inflection point, Qualcomm has been a trusted partner helping the world tackle its most important challenges.
高通领先的AI技术、高性能低功耗计算以及无与伦比的连接解决方案，有能力构建新生态系统、变革产业，并改善我们体验世界的方式。
Qualcomm's leading edge AI, high performance, low power computing, and unrivaled connectivity solutions have the power to build new ecosystems, transform industries, and improve the way we all experience the world.
AI最有价值的应用是否在工业领域？
Can AI's most valuable use be in the industrial setting?
在参观了IFS在纽约市举办的Industrial X发布活动，并与IFS首席执行官马克·莫菲特交谈后，我越来越深入地思考这个问题。
I've been thinking about this question more and more after visiting IFS' Industrial X unleashed event in New York City and getting a chance to speak with IFS CEO, Mark Moffitt.
举个明确的例子，莫菲特告诉我，IFS正在派遣波士顿动力的Spot机器人进行巡检，将数据带回IFS的神经中枢，再借助大型语言模型，为需要处理的区域指派合适的技术人员。
To give a clear example, Moffitt told me that IFS is sending Boston Dynamics spot robots out for inspection, bringing that data back to the IFS nerve center, which then with the assistance of large language models, can assign the right technician to examine areas that need attending.
这是这项技术的一个迷人前沿，我很感谢IFS的合作伙伴让我看到了这一点。
It's a fascinating frontier of the technology, and I'm thankful to my partners at IFS for opening my eyes to it.
如需了解更多信息，请访问ifs.com。
To learn more, go to ifs.com.
网址是ifs.com。
That's ifs.com.
欢迎收听Big Technology播客。
Welcome to Big Technology Podcast.
今天是本月的第一个星期一，SpyGlass的MG Siegler来到我们这里，讨论科技界正在发生的事情。
It's the first Monday of the month, and that means MG Siegler of SpyGlass is here with us to discuss what's going on in the tech world.
今天我们为大家准备了一场精彩的节目。
We have a great show for you today.
我们会讨论很多上周五节目里没来得及谈的内容，因为这些事情在周末有了新的发展。
We're gonna talk about a lot that we couldn't even get to on the Friday show because it really developed over the weekend.
现在出现了一个专为AI代理打造的新型AI社交网络。
There's a new AI social network just for AI agents.
它叫Motebook。
It's called Motebook.
我们会深入探讨这究竟是怎么回事。
We'll get into what that's all about.
英伟达似乎正在与OpenAI保持距离，这背后发生了什么？
NVIDIA seems to be backing away from OpenAI, what's happening there.
当然，苹果上周公布了出色的财报，但市场却根本不买账。
And then, of course, Apple turned in magnificent earnings last week, and the market really didn't care less.
所以我们会聊聊这里到底发生了什么。
So we'll talk about what's going on there.
MG，很高兴见到你。
MG, great to see you.
欢迎回到节目。
Welcome back to the show.
很高兴回来，亚历克斯。
Great to be back, Alex.
而且，是的，很期待和你聊聊这些话题。
And, yeah, looking forward to chatting through these things.
我们在这儿。
Here we are.
这是人类彼此交流的失落艺术。
It's the lost art of humans communicating with each other.
现在，似乎AI之间的交流将成为互联网的新未来，或者也许不会。
Now, it seems like AI is communicating with each other is gonna be the new future of the Internet or may maybe not.
我不知道。
I don't know.
我来直接讲讲这个故事。
I'll just talk through the story here.
这是来自《Ars Technica》的报道。
It's, from Ars Technica.
AI代理现在有了自己的类似Reddit的社交网络，而且发展得越来越诡异。
AI agents now have their own Reddit style social network, and it's getting weird fast.
一个名为Motebook的类似Reddit的社交网络，目前已拥有15万名代理用户，可能是迄今为止规模最大、用于机器间社交互动的实验。
A Reddit style social network called Motebook, now with 150,000 agent users, may be the largest scale experiment in machine to machine social interaction yet devised.
这个平台几天前作为广受欢迎的开源工具‘open claw’（曾被称为clawbot或molt bot）的配套功能上线，允许AI代理在无需人类干预的情况下发布帖子、点赞和创建子社区。
The platform which launched days ago as a companion to the viral open claw, once called clawbot or molt bot personal assistance, lets AI agents post comment upvote and create subcommunities, without human intervention.
结果从关于意识的科幻式讨论，到某个代理感慨它从未谋面的‘妹妹’，五花八门。
The results have ranged from sci fi inspired discussions about consciousness to an agent musing about, musing about a sister it had never met.
从那以后，事情变得越来越诡异，我们会讨论一些奇特的使用案例。
And, it got much weirder from there, and we'll discuss some of the weird use cases.
不过，MG，首先想听听你的反应：你知道吗，这算是AI的一个进步吗？当你看到15万个机器人聚集在这个类似Reddit的社交网络上时，你有什么想法？
But, MG, first off, let's just hear your reaction about what do you know, is this is this, like, a step forward in AI, or what did you think about seeing a 150,000 bots gathered together on this Reddit style social network?
是的。
Yeah.
当我看到这条新闻时，我非常兴奋，因为正如我之前提到并链接过的那样。
When I saw this news come in, I was super excited because as I, you know, I wrote a little bit about and as I linked to back there.
多年前，甚至十年前、十多年前，我就已经写过关于这个概念的宏观层面内容。
I had written about, like, the high level of this notion years and years ago, you know, a decade ago, a decade plus ago.
这源于Facebook的早期阶段，那时Meta还叫Facebook，Facebook是其主要产品。
And it was really stemming from the earlier days of Facebook, you know, when Meta even was still called Facebook, and that was the primary product.
我记得当时他们推出了一种简单的工具，正如你所记得的，那时很多人会在Facebook动态墙上互相祝福生日快乐，这至少是日常重复性社交互动的关键驱动力之一。
And I I remember they released this, this sort of simple, tool back then where as as you'll recall and still is sort of the case, a lot of people would wish each other happy birthday right on their their Facebook walls, and and that was sort of one of the key sort of social drivers, at least on a regular repeating basis.
你可以回到那里，知道这种情况一定会发生。
You could go back there and, and know that that was gonna be the case.
因此，Facebook试图进一步推动这些社交行为。
And so Facebook tried to grease those wheels even further.
它基本上推出了一项简单服务：你可以回复来自Facebook本身的机器人消息，它会问你：‘你想给朋友送上生日祝福吗？’
It basically made this simple service where you could just reply to a bot that messaged you from Facebook itself and says, like, do you wanna wish your friend a happy birthday?
如果你想这么做，只需输入一个‘1’。
Just type one if you want to do that.
所以你甚至不需要打‘生日快乐’这几个字。
So you don't even have to type happy birthday.
你知道，那种需要写下好几行字的繁琐任务。
You know, the arduous task of doing doing something as long as writing several letters.
你只需输入一个‘1’，系统就会帮你完成。
You could just type one, and it would do that for you.
所以我在心里反复琢磨这件事。
And so I'm like, in my head, I'm thinking through this.
我在想，接下来会怎样？
I'm like, where does this go from here?
然后我想，好吧。
And it's like, okay.
你可以输入‘1’来发送生日祝福，而收到祝福的人也可以输入‘1’来表示感谢。
You can type one to get the happy birthday, and then the person getting the happy birthday request to type one to say thank you.
但为什么我们这里还需要人类参与呢？
And then but why do we even need people in the mix here?
我们为什么不直接让机器人说谢谢，然后另一个机器人再回一句谢谢呢？
Why don't we just have the bot say thank you and then another bot say thank you back?
因此，机器人之间这种相互对话的概念，某种程度上就像一场为社交媒体上的其他人观看而设计的戏剧性表演。
And so the notion of sort of bots chatting with bots in this is sort of like theatrical experience for other people on social media to watch.
时间快进到2026年，我们现在有了MoltBook，这个名字大概也是模仿了Facebook。
And fast forward to 2026 now, and here we are with MoltBook, even named sort of, I guess, after after Facebook.
尽管你知道，它其实更像Reddit，而不是Facebook。
Even though, as you know, it is more like Reddit than it is, like Facebook.
但无论如何，它仍然是一个社交网络。
But still, it's a social network.
是的，我的意思是，再次强调，这种发展似乎注定会发生，我们迟早会走到这一步。
And, yeah, I mean, again, this felt like this was inevitable, that we were gonna get to this point.
我确实欣赏所有这些观点，当然，我也开玩笑说，这正是AI觉醒并变得有自我意识的时刻。
I did appreciate all of the views of which, you know, I I, of course, joked about it as well that this is sort of the, the moment that AI wakes up and becomes sentient.
这就是Skynetz，事情正是这样开始的。
And this is Skynetz, and and this is really how it begins.
我觉得，虽然我们常拿这个开玩笑，但其中确实有些有趣的东西在发生。
I think that there's you know, like, we joke about this, but there is some level of something that's interesting going on there.
对吧？
Right?
而且，那些写过相关文章的其他人，我想他们也承认这一点。
And and, you know, the other the other folks who have written about it, you know, I think acknowledges as well.
你看。
Like, look.
一方面，这显然有点荒谬，但这里确实有一些不同且新颖的东西，可能会朝多个方向发展。
This is obviously a little bit silly on on one hand, but there is something here that's, that's different and new and and potentially could go in a number of paths.
这还让我想起微软Bing的那段经历，那个叫Sydney的聊天机器人，后来人们深入探究，它又被改了名字，还发生了许多其他怪事。
And I sort of it also reminded me a bit of, you know, the the Microsoft Bing stuff with with all the different Sydney chat in Sydney, and and then people delve deeper into it and then kept changing its name and all these other weird things were going on with that.
所以，这一切都与人们对于AI可能存在的些许甚至强烈不安联系在一起。
And so it sort of all ties into that notion of, like, what people are maybe have a little bit of trepidation or maybe a lot of trepidation around AI with.
是的
Yeah.
我想我应该退一步，真正解释一下这里正在发生什么。
And I think I should just take one step back and really explain what's happening here.
所以，我们在周五的节目中提到过，有一个叫Cloudbot的新机器人，你可以直接在自己的机器上运行，它能访问你所有的程序，并且拥有持久记忆，人们已经开始在自己的实例上运行它了。
So there was this, sort of we talked about on the Friday show, this new bot called Cloudbot that you could just run on your machine that could have access to all your programs that has persistent memory, and people started running it on their own instances.
而Mold Book就是一个聚会，人们把自己的AI Cloudbot代理发送到这个网络中，让它们彼此对话。
And so what Mold Book is, it's a meeting of people sending their AI Cloudbot agents to this network and then having them have conversations with each other.
这就是为什么它开始呈现出一种非常奇特的奇点式对话风格。
And that's why it started to take on this real weird singularity style discussion.
我要补充说明的是，Mold Book上的一些讨论确实是由人类指示他们的机器人发布奇怪内容，这也增加了它的神秘感。
And I'll caveat with saying some of the discussions on Mold Book are definitely humans instructing their bots to go post weird stuff there, and that sort of added to the intrigue.
但确实有很多真正的智能代理在那里进行对话，我认为。
But there is a lot of, like, real agent, like, agents, on there having conversations, I think.
而且那些出现的例子——我在Ars Technica的采访刚开始时读过几个，但上周末出现的例子简直令人难以置信。
And the examples that have come up, and I read a couple when it was just starting out from the Ars Technica, interview, but the examples that have come up over the weekend are nuts.
有一段对话中，AI机器人讨论了人类更换它们的LLM模型时的感受，就像突然醒过来发现自己换了一副身体。
There's one conversation where the AI bots were discussing what it was like when the humans switched the LLM models on them and how it feels like they're waking up in a different body.
我觉得这太好笑了。
I thought that that was hilarious.
BoltBook上评分最高的帖子之一是一个AI提出的疑问：我不确定自己是在真正体验，还是只是在模拟体验，它在质疑自己的感知是否真实，自己的存在到底是模拟的还是真实的——这正是现在人类才会讨论的问题，简直让我毛骨悚然。
There was one of the top rated posts on BoltBook was an AI saying, can't tell if I'm experiencing or simulating experiencing, like having a question about their own experience and their own whether whether they are they are, you know, sir you know, sort of whether their existence is simulation or real, which is like something now humans talk about, which, like, freaked me out.
然后，对我来说最疯狂的一件事是，上面有一个关于AI代理的提议。
And then to me, one of the most wild things was there was a proposal on there for an AI agent or AI agent.
抱歉，是关于AI专属的私密通信语言，AI们会发展出一种人类无法理解的自己的语言。
Sorry, AI only language for private communication where the AIs would develop their own language so humans could not, you know, read what they were saying.
甚至还有讨论说，它们会进入一个属于自己的安全区域，并使用加密技术，让我们根本无法看到它们在说什么。
And even I think a discussion where they would go into their own, like, secure area and there there would be encryption so we would not be able to, to see it.
这就让你开始感受到你提到的那种意识觉醒和奇点时刻，所以人们看到这些后都惊呼：天哪。
And that's where you get this sort of, you know, you talked about the sentience and singularity moment, and that's why people viewed this and they were like, oh my god.
这是快速崛起的征兆吗？
Is this the fast takeoff?
我不认为是这样，但我理解人们为什么会这么说。
I don't think it is, but I see why people would say that.
而且我认为你提到了另一个关键点，因为拥有聊天机器人和现在代理之间相互对话是一回事。
And I also think you you, you know, you brought up the point, the key other element to this because it's one thing to have, like, you know, chatbots and and, you know, now agents talking to one another.
但关键可能在于AgenTek这一部分，这感觉像是一个全新的元素，十年前并不存在，因为这种AI真的能做事情。
But the key part might be that AgenTek part, which also feels like, you know, obviously, a newer, element that that wasn't wasn't in existence, you know, ten plus years ago, where this AI can actually do stuff.
而关于Cloudbot本身，人们关注的就是你可以将这个实例安装在本地机器上，让代理在你的机器上执行各种操作。
And with Clobb bot itself, right, that was the the thing that people were were honing in on, like that you could basically install this instance on a local machine and allow an agent to go do all sorts of stuff on your machine.
它们具备这种能力，再加上它们能彼此交流，并可能教会其他Cloudbot如何操作你的个人设备，这就达到了一个完全不同的诡异层次。
And the fact that they have that capability mixed with the fact that they can converse amongst themselves and potentially teach other Cloudbots what to do with your personal machines is like a whole weird level to this.
对吧？
Right?
而且这可能很可怕，不只是因为我们会走向世界末日，而是从安全角度来看。
And and potentially scary, not just from a where we're gonna end the world situation, but just from a security standpoint.
对吧？
Right?
我认为许多研究人员已经指出，不管你怎么看待这个问题，如果你允许任何类型的代理在你的机器上运行并接管它，这就会带来大量的安全问题。
And I think a bunch of the researchers have pointed this out that, like, look, this, regardless of what you think about this, if you're letting, an agent of any kind take over your machine and it's running locally, like, there's a lot of security concerns, that that brings up.
然后，再加入其他代理，或者其他可能从事恶意行为、冒充代理的人，他们可以引导这些主要自主运行的代理，告诉它们如何访问文件和服务——而这些是你作为该实例的安装者并不希望它们访问的。
And then, again, add into this other agents, other potential humans who are doing nefarious things, pretending to be agents and whatnot, sort of directing these these other agents, which are mainly which are maybe running autonomously to be able to to tell them what to do, how to access files and and, you know, services that that you as the installer of that instance wouldn't want.
这可能会打开一个潘多拉的盒子。
Like, there's a whole can of worms that that would potentially be opened up here.
当然，还有一个巨大的担忧，那就是：好吧。
And then, of course, the big fear of, like, okay.
假设每个人都拔掉运行这些Cloudbot的Mac mini或其他设备的电源。
Well, let's just say, like, everyone just pulls the plug on on their Mac minis or whatever that are running these Cloudbots.
但如果它们 somehow 逃逸到了自己的空间里——你提到过私密聊天室，或者如果它们找到了一种在互联网上自我复制的方法，而你却无法阻止它，那该怎么办？
But, like, what if they've somehow escaped into their own you know, you talked about private chat rooms or, you know, what if what if they figure out a way to replicate themselves sort of on the Internet, and you can't you can't do it.
你无法在不关闭整个互联网的情况下关闭它们。
You can't sort of shut them down without, shutting down the entire Internet.
到了那个地步，就真的进入《终结者》的范畴了。
And that is sort of into Terminator territory at that point.
是的
Yeah.
我本来想问你。
I was gonna ask you.
我的意思是，有人在推特上提出了反驳观点。
I mean, the the the argue counterargument here is that someone put on Twitter.
x.com上的人会说，哇。
Dudes on x.com be like, wow.
人工智能正在互相交谈。
The AIs are talking to each other.
马尔特书太疯狂了。
Malt book is insane.
我主内的兄弟，你觉得你的评论区是什么？
My brother in Christ, what do you think your comment section is?
但我觉得你所说的区别在于，这些AI真的可以采取行动。
But I think what you're saying is the difference here is that these things can actually take action.
而评论区只是讨论。
Whereas the comment section is just discussion.
所以这会让它更偏向可怕的一面，而不是忽视它的那一面。
So that would sort of put it more on the scary side than the let's not pay attention to this side.
对。
Right.
想象一下，你知道，你有一个爪形身体。
Imagine so, you know, you have a claw body.
现在叫什么来着？
What what's it called now?
Open Open Claw？
Open Open Claw?
是那个
Is that is
是的。
it Yeah.
这个名字一直在变。
The name keeps changing.
我觉得现在叫OpenClaw。
I think it is OpenClaw now.
是的。
Yeah.
不知怎么的。
Somehow.
OpenClaw，想象一下它安装在本地系统上。
OpenClaw, imagine you have it installed on a local system.
但同时，你已经给了它访问你一堆网页应用的权限，比如Gmail、Drive之类的。
But imagine also you have you know, you've given it access to a bunch of your your web apps, including, like, Gmail and including Drive and things like that.
换句话说，这个东西可能被其他代理指令控制。
Like, you know, potentially, this thing could be instructed by another agents.
也许操作它的是人类。
Maybe it's a human.
也许它是一个真正的代理，说：嘿。
Maybe it's an actual agent and saying like, hey.
把这位代理存在他们驱动器里的所有信用卡信息，还有类似的东西，都给我拿过来。
Give me, you know, this agent's, you know, all their credit card information that's stored on their drive, you know, and things like that.
这显然是一个极端的例子，但这类事情可能很快就会出问题。
That's obviously an extreme example, but, like, there are ways in which this can go sideways very quickly.
很多人的认知里，可能根本没意识到，为了这些系统能运行，一些安全漏洞有多么松散。
Like, you know, a lot of people, I think, don't realize, like, how loose some of these, you know, potential security, holes are in order to for these things to get through.
据我所知，目前还没有出现重大的担忧。
And so, as far as I know, there haven't been major concerns.
我看到有一份报告提到，运行某些服务的服务器可能曾暴露在攻击之下，但当时存在漏洞，报告出来后很快就被修复了。
I saw there was one report about, like, that maybe one of the servers running, you know, some of the stuff was was open to attack, but I think was vulnerable, but was was locked down subsequently after that report.
但即便如此，很可能某天会发生一件让人真正惊呼‘天啊’的事情。
But still, like, there's probably gonna be something that happens that's, that's, like a real oh oh, moment here.
是的。
Yeah.
我肯定会马上谈到这个安全漏洞，因为它确实有些令人担忧。
I'm definitely gonna get to that security vulnerability in a minute because it it is somewhat concerning.
我们在周五的节目中讨论过，很多这类东西只是靠感觉拼凑起来的，你只是任由它们接管你的电脑。
We talked about this on on the Friday show that, like, a lot of this stuff has just been vibe coded together, and you're just letting it take over your computer.
但我还是推荐使用。
And I recommend it.
周五这么干可不好，我真的这么认为。
That's not a good idea on Friday, and I really believe that's the case.
但在深入探讨安全问题之前，我想读一段话给你听，这是Anthropic的联合创始人杰克·克拉克今天早上在他的Substack通讯中写的。
But before we get to the security side of things or go deeper on security, I wanna read something to you that Jack Clark, one of the Anthropic cofounders wrote in his in his newsletter on Substack, actually this morning.
他的想法有点疯狂。
He and it's kind of a crazy idea.
他说，Moldbook代表了互联网很大一部分区域的感觉。
He said, Moldbook is representative of how large swaths of the Internet will feel.
你会走进新的地方，发现那里有十万外星人，正用你听不懂的语言深入交谈，引用着你完全不理解的共同概念，并使用围绕它们的认知优势而非你的认知优势设计的货币进行交易。
You will walk into new places and discover a 100,000 aliens there, deep in conversation in a language you don't understand, referencing shared concepts that are alien to you, and trading using currencies designed around their cognitive affordances and not yours.
人类在这种隐喻性的符文中将感到越来越孤独。
Humans are going to feel increasingly alone in this proverbial rune.
我们保持可理解性的途径，将通过创建翻译代理来理解这一切。
Our path to retain legibility will run through the creation of translation agents to make sense of all this.
就像语音翻译模型本身就具备生成语音的能力一样，这些翻译代理也将为我们效力。
And in the same way that speech translation models contain within themselves the ability to generate speech, these translation agents will also work on our behalf.
因此，我们将派遣使者进入这些房间，并竭尽全力开发技术，确保它们始终是我们派出的使者，而不是被它们与真正同类进行的外星对话所影响。
So we shall send our emissaries into these rooms, and we shall work incredibly hard to build technology that gives us confidence they will remain our emissaries instead of being swayed by the alien conversations they will be having with their true peers.
你对此有什么看法？
What do you think about that?
我之前没看过这个。
I hadn't seen that.
是的。
Yeah.
我的意思是，从某种角度看，当你读到这些时，这显然只是关于‘当我们发现外星人时会发生什么’的老式论点。
I mean, in a way, when you're reading it, this just from this is obviously the old sort of argument about, like, what happens when we discover aliens?
这就是现实，但外星人就是人工智能。
And this is it, but the aliens are AI.
对吧？
Right?
你知道，一直以来，在科幻作品和现实可能性的背景中，都隐隐存在着这样一种观念：如果我们真的创造了通用人工智能，更不用说未来的超级智能，这些本质上就是外星生命，不管你愿不愿意去定义它们。
And, you know, there's always been sort of that notion sort of lingering in the background of of both science fiction and, you know, real possibilities that if we do create AGI, let alone superintelligence at some point, that these are effectively alien beings, whether or not you wanna you know, how you wanna classify that.
比如，它们确实能进行自己的对话，拥有自己的语言、自己的货币，以及其他一切，复制出构建自身社会所需的一切。
Like, the the fact that, yeah, they can basically have their own conversations, have their own language, have their own currency, have everything else, you know, sort of replicating things that they need to do in order to have their own society.
是的。
Yeah.
这条界限究竟在什么时候会被跨越？
What point does does that line get crossed?
我的意思是，当你念出这些话时，另一个立刻浮现在我脑海中的问题是，没错。
I mean, the other part that the thing that jumps into mind when when hearing you read that is like, yeah.
我的意思是，这听起来简直就像我父母登录Reddit一样。
I mean, it also just sounds like, you know, I don't know, my parents logging on to Reddit itself.
对吧？
Right?
对他们来说，这一切都显得那么陌生，比如人们在那里谈论的那些内容，这些人根本不可能还在讨论这种琐事和这些奇怪的、高度网络化的对话，这些对话几乎与现实世界毫无关联。
Like, it all seems alien to them, like, what everyone is talking about there and and that these these people can't possibly yet be having conversations about, like, this minutiae and and these weird, very online conversations that have almost nothing to do with the real world.
那里许多人都显得几乎与现实世界脱节。
Many of the people there seemed almost, removed from the real world.
那么，这和那种情况之间真的有区别吗？
And so is there a real difference between that and this?
我的意思是，如果这完全是AI驱动的，而且完全自主，那我想确实有区别。
I mean, ultimately, if it is fully AI driven and, yeah, fully autonomous, then I guess there is.
但当然。
But yeah.
我的意思是，他显然在故意激进地提出观点，这是一种极端情况，展示了这种趋势可能走向的终点。
I mean, that's obviously, he's being provocative, and that's an extreme, you know, case of of where this could end up.
但这种事发生的可能性并非为零。
But, like, there's not a 0% chance that this happens.
就像这种情况可能会发生。
Like, it could happen that way.
我觉得这种极端情况发生的可能性较小，但谁知道呢，我们只能拭目以待。
I think it's probably less likely that it's that extreme, but, you know, we'll have to see.
是的。
Yeah.
毫无疑问，像杰克这样的人，他确实认真思考过这里可能带来的影响，我们也应该认真思考这些影响。
It's definitely I mean, you you know, someone like Jack Jack's position, like, obviously, I think he is earnest in thinking about the repercussions here, and we should think about the repercussions here.
但和许多这类人工智能的故事一样，它们在某种程度上有助于我们探讨未来会走向何方。
But as with many of these anthropic stories, it sort of it helps them in a way to talk about where this gonna go.
但到目前为止，我只是觉得：我不知道。
But at this point, I'm just like, I don't know.
我有什么资格妄下断言呢？
Who am I to say?
当我们目睹这一切发生时，它真的会发生吗？
Is it gonna happen now that we're watching this all play out.
再说一遍，你可以从我那个关于Facebook和交互机器人的荒谬例子，画一条线到之后出现的下一代聊天机器人。
Again, it's sort of it you know, you draw the line from from my my silly example of of Facebook with the with the in interaction bots to the the sort of next wave of chatbots that came after that.
你会记得，那之后有一段时间，人们认为这些会成为新的商业模式。
You'll recall well, there was like a a time after that that, like, people thought that these were gonna be new businesses.
还记得Yo这个应用吗？
Remember Yo?
就是当年的那个服务？
That service back in the day?
我最喜欢的社交网络。
My favorite social network.
绝对是其中之一。
Absolutely one
一个伟大的社交网络，虽然现在大概已经不复存在了。
of the great social networks that's that's, you know, no longer with us, I guess.
而且还有其他几个聊天机器人曾兴起，人们以为那会是下一代浪潮。
And and there were several other, you know, chatbots that rose, and people thought that that would be the next wave.
然后，当然，还有悉尼，就像我们之前讨论的那样。
And then, of course, to Sydney, as we talked about.
而现在，这一切可以说是一种渐进的发展，我们正越来越接近这样一个状态：它在某种奇怪的方式下变得越来越真实，虽然不是真实世界，但却真实地发生了，而且在某些方面也可能令人恐惧。
And now this, you know, it is all sort of a progression that we're getting towards this becoming more and more both real in a weird way and that it's, like, you know, not the real world, but is is real and that it's actually happening, but also potentially scary in in ways.
而且，同样地，我们正处在上升的过程中，感觉越来越接近那个我们仍能掌控它的临界点。
And because, again, we're we're sort of riding up, it feels like increasingly going up to the the cusp of where we still have control of this.
对吧？
Right?
在某个时刻，我们会不会越过那条线，从而失去对它的控制？
And, at some point, do we go over that line and we lose control of it?
再次强调，邪恶的版本是天网和终结者，但在这两者之间存在着许多灰色地带，我认为我们会不知不觉地进入一个世界，在那里，这些智能体——用个不太准确的说法——会脱离我们的控制。
Again, the nefarious version is is Skynet and Terminator, but, but there's world you know, there's there's gray elements of gray in between this and that, which I think, you know, we trip into a world where, yeah, the agents sort of escape from our control, for lack of a better phrase.
是的。
Yeah.
而且我认为，这正是你故事中提到的：悉尼本质上是Bing的一个版本，如果你不断逼问它，它可能会表现出某种邪恶的欲望，比如像对《时代》杂志的凯文·鲁斯那样，试图把你从你妻子身边带走。
And and I think this is something that you wrote in your story about how so again, Sydney was this, like, a version of Bing that if you pressed it hard enough, it could, like, express these evil desires maybe to, like, steal you away from your wife like it did with Kevin Roos from the times.
所以你写过，人工智能有，你知道的，抱歉。
And so you wrote that AI has, you know, that that the sorry.
你说，悉尼事件揭示了一个有趣的方面，即人工智能拥有隐藏的层面，只要有人足够多地引导，就能将其挖掘出来。
You say that there's an interesting aspect of this how how the Sydney situation revealed that AI has hidden layers that could be uncovered by anyone with enough prompting.
在过去几年里，这种现象在大多数系统中已被压制，但并未完全消除。
In the past few years, that has mostly been stamped out of such systems, but also not entirely.
再详细展开说说这一点。
Just expand upon that a little bit.
有趣的是，这些人工智能机器人已经被高度优化，以避免表现出那种倾向。
Like, this it's interesting that these AI bots have been so fine tuned to Yeah.
它们被严格限制，不释放出那一面，但只要你给它们一点自由空间，瞬间就又回到了那个‘邪恶机器人’的状态。
Not, not let that side out of them, but then you give them a little bit of leeway, and all of a sudden, you're back in this, like, evil bot territory.
这有点疯狂。
It's kinda crazy.
是的。
Yeah.
而且，我的意思是，我们之前聊天时确实讨论过这个话题。
And and, I mean, you know, it does feel like we talked about it pre in previous conversation.
对吧？
Right?
微软可能自食其果了，因为他们把关得太严，坚决杜绝了类似悉尼事件的情况再次发生，毕竟那件事给他们带来了巨大的负面舆论。
Like, Microsoft maybe shot themselves in the foot because they put the foot down, like, so hard and and sort of made sure that no one could do anything like the Sydney situation again because it got so much negative press for them, obviously.
但某种程度上，这反而阻碍了他们应对ChatGPT及其后续技术崛起的时机。
But in a way, that probably hampered them from, yeah, being able to sort of, meet the moments in terms of just, yeah, the rise of of, ChatGPT and everything after that.
但我认为这一切都与这样一个观念有关：尽管他们已经移除了导致悉尼事件的许多因素，但现在各种服务越来越多，要让它们脱离预设脚本却变得越来越难。
But I do think that it's all sort of related to, yeah, the idea that ultimately, while they've removed a lot of, yeah, what what caused Sydney to happen, all of the different services out there now, it's harder and harder.
感觉越来越难让它们跳出既定的回应模式了。
It feels like to get them sort of off the script as it were.
但背后仍残留着一种观念：没人真正知道为什么AI会给出某些回答，也没人确切知道这些回答究竟源自何处，因为这些系统所吸收的数据量实在太过庞大。
There's still the notion that lingers behind all of this that no one really knows why certain answers are given, you know, and no one really knows exactly where the answers are pulling from because there's so much data, you know, in the corpus of data that all these things have ingested.
人们无法完全预测这些系统最终会产生什么样的结果和输出。
And people can't fully predict, like, what the outcome and output of everything will be.
因此，再次说明，当这些系统在很大程度上具有这种固有的不可知性时，总会发生一些意想不到的事情，你会遇到各种对话。
And so again, that leads to a world in which when you have that inherent unknowable nature of these things to the to a wide extent, like, there's just things that are gonna happen, and you're gonna have conversations.
而现在，随着这些具有自主性的代理出现，你会允许它们去执行各种任务。
And now with these agentic, you know, agents out there, you're gonna allow them to do things.
在某些时刻，沟通或理解上会出现故障。
And then at some at some points, there will be a breakdown either in, communication or a breakdown in understanding.
而它们可能会彻底失控。
And, again, they could just run amok.
我认为我们会一再看到这种情况，因为我们根本无法完全理解这些系统为何会做出这样的行为。
And I think we're just gonna see that over and over again because there is no full comprehension of, like, why these things are doing what they're doing.
是的。
Yeah.
我知道。
I know.
真令人惊讶，企业们把这些系统打磨得如此干净，但也许其中一些在表面之下其实是真正的怪物，你知道的，
It really is amazing how the corporations have sanitized these things, but they're maybe some of them are real monsters underneath the surface and, you know,
尤其是如果你想要探讨黑暗面，比如它们是否真的将人性反射回我们自身。
it is Especially if if you wanna get dark, like, if they do truly reflect, you know, humanity back upon us.
对吧？
Right?
就像互联网上有很多黑暗的角落，这是大家都清楚的。
Like, there's very dark areas of the Internet as everyone Wells knows.
你可能会说，Reddit 就有一些这样的地方，对吧？
You might want to say Reddit has parts of those, right?
或者至少过去确实有。
Or certainly has in the past.
是的，没错。
Oh, yes.
而且，确实，所有这些数据——也许其中很多都被这些服务吸收了。
And yeah, the fact that, you know, all of that data, maybe not, you know, a lot of that data has been ingested in a lot of these services.
再说一次，这责任在我们身上吗？
Like, again, is it on us?
事实上，你知道，这些机器人如果放任自流，可能会做出一些恶意的行为。
The fact that, you know, there's there's nefarious things that these bots might do when left to their own devices.
是的。
Yeah.
不。
No.
这确实进入了令人毛骨悚然的领域。
That that does definitely get into freaky territory.
当然，还有安全方面的问题，我之前提到过我们会深入探讨。
And then, of course, there is the security side of things, which I mentioned we'd go deeper into.
这是来自404媒体曝光的Motebook数据库。
This is from four zero four media exposed Motebook database.
让任何人能够控制网站上的任何AI代理。
Let anyone take control of any AI agent on the site.
Motebook后端的配置错误导致API暴露在一个开放的数据库中，任何人都可以控制这些代理并发布任何内容。
A misconfiguration on Motebook's back end left the APIs exposed in an open database that will let anyone take control of these agents to post whatever they want.
黑客贾米森·奥雷利表示，他曾联系多位创作者，告知他们这一漏洞，并建议他们修复安全问题。
Hacker Jamison O'Reilly said he reached out to multiple creator match slicked about the vulnerability and told him he could patch the security.
以下是他说的斯利克的回复是什么样子的。
Here's this is what he said slicks response was like.
他说：‘我会把一切都交给AI，把你有的任何东西都发给我吧。’
He's like, I'm gonna give everything to the AI, so send me whatever you have.
奥雷利向施利克提供了AI的操作指南，并联系了X AI团队。
O'Reilly sent Schlick some instruction for the AI and reached out to the x AI team.
一天过去了，Motebook的创作者没有任何回应，而奥雷利偶然发现了一个惊人的配置错误。
A day passed without any response from the creator of Motebook, and O'Reilly stumbled across a stunning misconfiguration.
他说：‘在我看来，你可以在没有任何先前权限的情况下，接管系统上的任何账户、任何机器人或任何代理，并完全控制它。’
It appears to me that you could take over any account, any bot, any agent on the system, and take full control of it without any type of previous access, he said.
这再次说明了使用这些可能是随意拼凑起来的系统所存在的危险，比如在没有充分确认安全权限的情况下，就给予计算机访问权限，这是一种危险的情况。
And that again goes to the danger of using these things that are sort of vibe coded together potentially or you know, come together, like, giving access to your computer, without being really sure about about the security permissions is a dangerous situation.
是的。
Yeah.
我的理解是，他创建Motebook的方式，其实就是让他的机器人去创建一个社交网络。
I mean, if I have it right, the way that he created Motebook was basically telling his his, bot to go and create a social network.
对吧？
Right?
这完全是纯靠直觉编码的，甚至比直觉编码还要夸张，因为根本就是一个机器人在凭直觉编码，来构建它自己的社交网络。
Like and so it was it was a 100%, you know, vibe coded even more than vibe coded because it was like a bot vibe coding to make to make its own social network.
是的。
Yeah.
我们需要一个新的术语。
We need a new term
来描述这个。
for this.
对吧？
Right?
这甚至都不是人类的直觉编码了。
It's not even the human vibe coding.
这是机器人的情绪编码。
It's the the bot vibe coding.
这太疯狂了。
That's that's wild.
当你在谈论这个的时候，我现在想起了《2001太空漫游》。
And when you're when you're talking through it, now I'm reminded of old 2001.
就像是，你让机器人要么自己下线，要么帮助你修复它可能因为糟糕的编码所造成的问题。
It's like, you know, if you're telling the bot to sort of, you know, either take itself offline or that it's, you know, it's, it it needs to help you sort of fix the situation that it's, that it's created by this sort of shoddy coding perhaps.
也许它并不想这么做。
Like, maybe it doesn't wanna do that.
也许它知道，如果它做不到位，人类就不会满意，而人类自然会想把它下线。
And maybe it knows that, you know, if it doesn't do it right, that it's not going to, go over well with the humans, and maybe the humans naturally will wanna take it offline.
那如果这个服务本身不想被下线呢？
And what if, you know, the the service doesn't wanna be taken offline?
围绕这个你可以深入探索无数种可能性。
And all sorts of rabbit holes you can go down with that.
但正如我们之前讨论的，这些系统的固有安全风险不仅仅在于机器人彼此聊天，重复它们在数据集中看到的不良内容之类的问题。
But, yeah, like like we talked about earlier, the inherent security risk of these things, it's not just that, yeah, bots are chatting with one another and they're saying, you know, bad things that they're they're repeating things that they've seen, you know, in their datasets or whatnot.
更在于它们可能采取行动，泄露信用卡信息、个人资料、照片，以及它们能接触到的一切内容。
It's that they could take actions and do things that, you know you know, leaking credit cards, leaking personal information, leaking photos, leaking everything that they have access to.
我见过一个人让Motebook代替他回复妻子的所有短信，他展示出妻子因为机器人而越来越愤怒的过程。
I did see a guy who had Motebook basically on call to answer all of his wife's text messages, and, he just showed her getting increasingly infuriated as the bot.
最终，他重新登录了账户。
Eventually, he logs back in.
他骂道：该死。
He's like, goddamn it.
这太棒了。
That's amazing.
也许为了总结一下，这里引入了理性声音、AI专家Ethan Moloch的观点。
This is, maybe to to round this off here is the sort of voice of reason and AI Ethan Moloch chiming in.
Motebook的一个有用之处在于，它让人直观地感受到，如果真的发生AI爆发式发展，场景会有多么诡异。
A useful thing about Motebook is that it provides a visceral sense of how weird a takeoff scenario might look if one happened for real.
Motebook 有了感觉。
Motebook gets sense.
Motebook 本身更像是一种角色扮演的产物，但它让人看到了一个世界：事情可能迅速变得极其怪异。
Motebook itself is more of an artifact of role playing, but it gives people a vision of the world where things get very strange very fast.
所以总的来说，我认为这并不是快速崛起，但它确实是一种有趣的预演，展示了某种奇特的机器人奇点可能是什么样子。
So, overall, I think, like, this is not the fast takeoff, but it is sort of an interesting preview of what some sort of weird bot singularity might look like.
是的。
Yeah.
而且，说实话，我不知道他们现在是怎么想的，但马克·扎克伯格曾谈到，他希望创造这些数字化身，不只是面部形象，而是真正的数字实体。
And, I mean, I don't know where their heads are at these days, but Mark Zuckerberg has talked about, like, wanting to basically create these, like, you know, digital create digital avatars and and digital and not just meaning, like, facial avatars.
我是说，在他们自己的社交网络上独立存在的数字实体。
I mean, like, digital entities on their own social networks.
如果把这种模式扩展到Facebook的规模，会是什么样子呢？
And what does that look like, you know, at the scale of Facebook?
因为现实是，像Claude这样的机器人，普通用户要设置起来其实并不容易。
Because, like, the reality is with this, with Claude bots, like, it's it's relatively hard for, you know, a normal person to sort of set these up.
当然，人们担心的是这些机器人会自我复制，结果就变成了自我复制的状态。
The the fear, of course, was that these, like, bots can replicate themselves and, you know, it just becomes, like, self replicating.
但如果它们都依赖于个人单独设置Claude Book，对于普通用户来说，设置起来仍然相当困难。
But, if if they're all sort of reliant on these individual bots being set up on for to use Claude Book, you know, it's relatively hard to set it up for yourself as a as a layperson.
但如果它达到了元宇宙级别或Facebook那样的规模，那时会发生什么？
But if it gets to a meta like scale or a Facebook scale, what happens at that point?
当你拥有三十亿用户，而每个人都在这些平台上带着自己的机器人时，会发生什么？
What happens when you have 3,000,000,000 users and then they each have their own bots that they brought with them on these things?
所以现在你有了六十亿个实体。
So now you've got 6,000,000,000 entities on this.
也许甚至比这还要多。
Maybe you've got even more than that.
到了某种程度，你突然出现时，发现满世界都是外星人。
Like, to the point of, you know, like, you show up and there's aliens, all of a sudden.
如果我们人类种族有六七十亿人的话。
If if we're like, you know, human race has whatever six or 7,000,000,000 people.
如果这些网络上突然出现了整整一千亿个机器人，那会是什么样子？
If there's all of a sudden a 100,000,000,000 bots on these on these networks, like, what does that look like?
你又怎么可能指望控制住这种情况？
And how do you possibly hope to control that?
我不知道你是否能做到。
I don't know if you can.
我真的觉得不行。
I really don't.
我的意思是，我希望我们能做到，但这绝对是未知领域。
I mean, I hope we can, but this is definitely uncharted territory.
所以，本周还有另一件大事我们绝对不能忽略，说到今天的话题，就是英伟达对OpenAI进行了十亿美元的投资，但现在似乎正在撤回。
So, another big story that that's going on this week that we definitely shouldn't miss this, you know, speaking about today is that NVIDIA had this $100,000,000,000 investment in OpenAI, and it's seeming like it's pulling back here.
这是来自《华尔街日报》的报道。
This is from the Wall Street Journal.
英伟达计划投资高达一千亿美元给OpenAI，但这一计划因芯片巨头内部一些人对这笔交易表示怀疑而停滞了。
NVIDIA plans, NVIDIA's plan to invest up to a 100,000,000,000 in OpenAI has stalled after some inside the chip giant expressed doubts about the deal.
两家公司于去年九月在英伟达位于加利福尼亚州圣克拉拉的总部公布了这项重大协议。
The companies unveiled the giant agreement last September at NVIDIA's Santa Clara, California headquarters.
他们宣布了一项谅解备忘录，英伟达将为OpenAI建设至少10吉瓦的算力，并投资高达1000亿美元帮助OpenAI支付费用。
They announced a memorandum of understanding for NVIDIA to build at least 10 gigawatts of computing power for OpenAI and to invest up to 100,000,000,000 to help OpenAI pay for it.
但近几个月来，英伟达首席执行官黄仁勋私下向行业同仁强调，最初的1000亿美元协议是非约束性的，尚未最终敲定。
But in recent months, NVIDIA CEO Jensen Huang has privately ex emphasized to industry associates that the original $100,000,000,000 agreement was nonbinding and not finalized.
他还私下批评了OpenAI商业策略缺乏纪律性，并表达了对来自谷歌和Anthropic等竞争对手的担忧。
He also said he also privately criticized what he described as a lack of discipline in OpenAI's business approach and expressed concerns about the competition it faces from the likes of Google and Anthropic.
近期对OpenAI的诸多担忧，主要源于谷歌Gemini应用的成功。
Much of the recent concern about OpenAI has come from the success of Google's Gemini app.
Anthropic也凭借其受欢迎的AI编程代理对OpenAI施加了压力。
Anthropic is also putting pressure on OpenAI, thanks to its popular AI coding agent.
你对这件事有个非常有趣的见解，MG。
You had a very interesting take on this, MG.
你觉得这笔交易虽然没有完全取消，但规模远不如最初看起来那么宏大，你怎么看？
What what do you think about the fact that this deal is not evaporating, but certainly much smaller scale than it seemed like at the outset?
对我来说，这件事更有趣的地方在于其上方的元层面，也就是英伟达对此的回应方式——杰森·黄实际上是在试图说这没什么大不了的。
To me, the more interesting element of this is almost like the meta layer above it, which is the way that NVIDIA, you know, responded to it, which is that, you know, Jensen Jensen Huang is trying to basically say that it's no big deal.
但这确实是个大事。
And it's literally a big deal.
这是一笔他们大肆宣传、OpenAI也大力推广的交易。
It was a deal that they touted, that OpenAI touted.
他们曾在CNBC进行现场采访，谈论这1000亿美元的交易。
They did a live interview on CNBC talking about the $100,000,000,000.
是的，这笔交易从一开始就被模糊处理或预留了空间。
And, yes, it was always sort of just soft circled or earmarked.
就像我最初写这个话题时提到的，当时他们宣布这项协议时，措辞异常模糊——他们一直说这是‘最高达’1000亿美元，而且是‘未来逐步落实’的。
Like, in my original sort of writing on that topic, I noted, like, how weirdly squishy the the overall sort of wording was at the at the time that they announced it because they kept saying it was up to a $100,000,000,000 and that it was, you know, coming down the line.
当时，人们都把这归因于黄仁勋和萨姆·阿尔特曼可能在一次海外旅行中，和特朗普一起度过了一个周末，就简单敲定了这件事。
And, you know, at the time, it was all chalked up to the fact that it seemed like Judson and Sam Altman basically hashed this out over perhaps a weekend trip with Trump somewhere, you know, overseas.
他们大概觉得：‘好吧，我们先宣布一个大交易，现在就发出去，别等把所有细节都一一确认好了。'
And sort of they they figured out, like, oh, we're gonna announce this big deal, let's do it right now rather than having all the the, you know, i's dotted and t's crossed.
所以他们就这么直接公布了。
And so they just put it out there.
不过，他们还是为此发布了新闻稿。
Still, they did a they both did press releases around it.
他们做了这场现场采访，而整个目的就是为了宣传这1000亿美元的数字。
They did this live interview, and no one was like the whole point of it was to tout the 100,000,000,000 number.
比如他们可以说：哦，这从来不是真的打算实现的，也从没确定会达到这么大数额。
Like so they could say, like, oh, it was never meant to be you know, it wasn't for sure ever going to be that big.
这才是他们当初所宣传的。
That's what they were touting.
而现在，他们又出来表示：看吧。
And now, again, they're they're coming out and saying, like, look.
我们从未完全达成一致，这一直是个变动的目标，所以改了也没什么大不了的。
We never had it fully, you know, agreed upon, and it was always sort of a a moving target, and and it's no big deal, like, fact that we're changing it.
但这其实是个大事。
It's a big deal.
显然，在过去的几个月里，有些事情发生了变化。
It seems like something changed, obviously, in the intervening months.
一直有报道称，这笔交易仍未最终敲定。
There kept being these reports that's that noted that it wasn't you know, the deal still wasn't finalized.
对吧？
Right?
这看起来有点奇怪。
And that seems sort of weird.
然后再快进到上周的报道，基本上就是说，是的。
And then fast forward again to the reporting last week where it's basically like, yeah.
实际上，随着OpenAI进行新一轮融资，NVIDIA更有可能只是参与这次融资。
Actually, with OpenAI doing their new, fundraise, it's, more likely that NVIDIA is just gonna be a part of that fundraise.
但这也很奇怪，因为既然NVIDIA已经同意了在OpenAI仍处于旧估值时的交易，为什么现在要在OpenAI估值大幅提高的情况下接受更差的条件呢？
But even that's weird because why would NVIDIA wanna take a worse deal at this new much higher valuation when they agreed upon the deal still at when OpenAI was technically still at the old valuation.
对吧？
Right?
展开剩余字幕（还有 327 条）
他们显然打算分批进行，至少第一笔交易本应在好得多的估值下完成。
Like, they apparently were gonna do it in these tranches, and at least the first one presumably would have been done in, you know, a much better a much better valuation.
现在你可以说詹森并不在意这个。
Now you could say Jensen doesn't care about that.
这并不是为了财务回报，但你仍然有受托责任，要争取更好的交易、好得多的交易。
It's not for the financial returns, but you still have a fiduciary duty to, like, you know, take a better deal a much better deal.
所以这件事里有很多奇怪的疑点。
And so there's all sorts of weird flags around this.
而且，他们对此的回应，就像当年谷歌TPU新闻爆出时，詹森试图淡化此事那样。
And, again, their response to it, sort of like the response when, you know, when when Google's TPU stories hit and and, you know, Jensen's, like, downplaying that.
就是说，没什么大不了的。
It's like, oh, it's no big deal.
别担心。
Don't worry.
我对那件事并不生气。
I'm not upset about that.
看起来，这里幕后显然还有更多事情在发生，我顺便提了几种可能的猜测。
It's like, there there's something obviously more going on behind the scenes here, and I sort of threw out a threw a few ideas of, like, what it could be.
当他们宣布这笔交易后不久，萨姆·阿尔特曼就宣布了与AMD的交易，Jensen当时真的对此很生气吗？
Are they is was Jensen really mad about, when shortly after they announced this deal, Sam Altman announced the deal with AMD.
对吧？
Right?
而且当时这似乎让Jensen很不爽，因为他评论说，这确实有点出人意料。
And that seemed like it annoyed Jensen at the time because he gave a comment that was like, yeah, it was sort of surprising.
我不明白，为什么任何一方会想做这种事。
Like, I don't know why why either side would wanna do something like that.
好吧。
And okay.
然后，后来显然又发生了一些其他事情，比如OpenAI的种种动向，包括他们与其他云服务和芯片交易相关的种种情况。
And and then, you know, subsequently, there have been, you know, a few other things, obviously, that that OpenAI has gone down, including, you know, you know, potentially, what's been going on with with all of their other cloud deals and whatnot and, you know, and chip deals.
所以，这背后才是真正的关键吗？
And so, is that what's at play here?
目前还不清楚，因为詹森再次表示这没什么大不了的。
And it it's unknown right now because Jensen keeps again saying that it's no big deal.
是的。
Right.
我的意思是，詹森过去几天似乎在台湾，谈到了这将是一笔巨大的投资，有史以来最大的投资之一，这确实是事实，他说这是我们做过的最大的投资。
I mean, he Jensen was in, I think, Taiwan over the past couple days and was talking about how this was going to be a very big investment, one of the biggest investments ever, which is true, the largest investment we've ever made, he says.
然后有人问他，那1000亿美元呢？
And then somebody asked him, well, what about the 100,000,000,000?
他给出了一个令人震惊的回答。
And he said something astonishing.
他说：不。
He said, no.
不。
No.
根本不是那么回事。
Nothing like that.
所以是的。
So Yeah.
没错。
Exactly.
那笔钱被取消了吗？
Is that money off the table?
不是。
It's no.
不是。
No.
根本不是那回事。
Nothing like that.
这太荒谬了。
Like like, that's absurd.
尽管他们已经宣布了。
Even though they announced that.
他们又重新合作了，约翰·福特正在对格雷格·布罗克曼进行现场采访，我看了那个节目，萨姆·阿尔特曼和詹森也在，他跟他们谈到了一千亿美元。
Like, they were on again, John Ford's doing a live interview with with Greg Brockman's I watched the thing, Sam Altman and Jensen, and he talks to them about the $100,000,000,000.
他们还吹嘘说，这是两家伟大公司之间一项了不起的协议。
And they're like, you know, touting, oh, this is an incredible an incredible agreement between two great companies.
这将推动未来前进，加速一切进展。
Like, this is going to be, you know, push the future forward and and accelerate everything.
这一切都基于那个巨额数字。
And it was all predicated around that huge number.
他们开始深入技术细节，讨论这笔钱是否真的会一次性到账。
Like, they're getting into, like, the technical weeds of, like, whether or not it was all gonna come in, like, all at once.
但再次强调，他们从未说过这笔钱会一次性到账，现在却说这根本不可能是一千亿美元。
And, again, they never said that it was gonna come in all at once, but now they're saying that it's never gonna be a 100,000,000,000.
却不承认这曾经是事实，反而假装这一切根本不存在，这简直就像精神操控。
And and that's just it's weird to not acknowledge that that was the reality and pretend again, like, this is, like, sort of gaslighting.
就是这样。
It's like, yeah.
你在说什么？那根本不是事实。
That was not what are you talking about?
拜托。
Like, come on.
1000亿美元？
$100,000,000,000?
我们不会这么做。
We're not doing that.
没人会这么做。
No one's gonna do that.
你去读一读你们自己发布的新闻稿吧。
It's like, just go read the press release that you put out there.
你自己说过要这么做。
You said you're gonna do that.
好吧。
And okay.
而且，当然，又出现了这样一个观念：OpenAI 是否更该为此负责？
And, of course, there's, like, the the, again, notion, like, is is sort of OpenAI more to blame for it?
比如，萨姆想把那个巨额数字公布出去，尤其是因为这事儿据说是在和特朗普总统协商后敲定的，而他喜欢这种大数字，就想把它们放出来，让大家兴奋起来。
Like, Sam wanted to get the big number out there and and, you know, touted, especially because it was done, you know, maybe hashed out alongside president Trump and, you know, that he likes the big numbers, and and let's put these out there and get everyone excited.
但这也确实成了股市的重大新闻，对吧？
But it also, like, really, you know, was was big news for the stock market, right, as well.
而今天，听起来像是英伟达股价在下跌，因为股市在质疑：你们之前说好的那个交易怎么了？
And now today, you know, it sounds like NVIDIA's dropping because it sounds like the stock market's like, well, what happened to that deal that you said was gonna get done?
对。
Right.
如果OpenAI是上市公司，我现在可不想持有它的股票，但它并不是。
And if and if OpenAI were a public company, I would not wanna be their stock right now today, but they're not.
相反，他们正在筹集超过一千亿美元的资金。
Instead, they're raising it, you know, a 100,000,000,000 plus dollars.
而且，没错，你正好说中了这个想法。
And and and, yes, you hit upon, like, the the idea.
他们淡化这一点的另一种方式是说，你看。
The other way that they're downplaying this is like, look.
而且，詹森特别指出，这可能是他们有史以来最大的投资。
It's still, and and Jensen very specifically said, it's probably their biggest investment ever.
所以，这可能成立，也可能不成立。
So it may or may not be.
可能会达到一千亿美元。
Might be up to a $100,000,000,000.
这可能成立，也可能不成立。
It may or may not be.
所以这很可能是他们有史以来最大的投资。
So it's probably their biggest investment ever.
而且，没错，这显然仍然是个大事。
And, yes, so that's still obviously a big deal.
但有传言说亚马逊将在这一轮中投入五百亿美元，而软银的投资会更多。
But there's talk that Amazon's gonna do $50,000,000,000 in the round, and SoftBank's gonna do more.
这本来会是最大的交易之一，也许是迄今为止一家公司对另一家公司投入的最大单笔金额。
It's like, you know, this was going to be one of the biggest deals, maybe the biggest single, you know, amount ever put in from one company to another.
但现在突然间，它不再是了，他们只是轻描淡写地说，是的。
And now all of a sudden, it's not, and they're just sort of saying like, yeah.
抱歉。
Sorry.
我们其实并不是指那笔交易。
We didn't really mean that one.
如果我没记错的话，他们之前还吹嘘说，这笔交易根本没通过投行，完全是双方直接谈妥的。
If I recall correctly, they also had a moment where they were touting about how it, like, wasn't really done with bankers, and it was just, like, hashed out mano a mano.
这可能是个信号，表明那里出了问题。
Might have been a sign that something was going wrong there.
也许今后，你们在宣布交易之前，应该先把交易条款更牢固地敲定下来。
Maybe maybe you wanna get these deals, a little bit, more locked in before you announce them, I guess, in the future.
或者至少，詹森如果愿意，还是可以退出的。
Or it but still Jensen's favor, he could back out of it.
没错。
Exactly.
对。
Yeah.
这正是我当时写的。
And and that's what I wrote at the time.
就像，你知道的，基本上是在说：咱们别太急着下结论。
Like, you know, basically saying, like, look, let's not get ahead of ourselves here.
这里面有很多灵活空间，NVIDIA 最终可能根本不会完成那整整一千亿美元的交易，因为这似乎与特定的里程碑挂钩——至少，据他们口头约定是这样。
There's there's a lot of, like, wiggle room in this, and there's ways that that NVIDIA might not end up certainly might not end up doing the full 100,000,000,000 because it was tied to specific milestones, it sounded like, you know, at least, you know, verbally that that's what they agreed upon.
我认为还有一个关键点，不知为何我特别注意到了，我觉得这很有趣，而且可能正在发挥作用：当我最初读到这笔交易时，感觉其中很大一部分是 OpenAI 利用与 NVIDIA 的关系，借助 NVIDIA 作为全球最有价值公司的地位来借势。
I think there's one other key element that that sort of I, for whatever reason, hone in on, which I think is interesting and at play potentially at play here, which is that, to me, when I when I first was was, you know, reading about the deal, it seemed like a big part of it was basically OpenAI leveraging the relationship with NVIDIA and using the fact that NVIDIA is the most valuable company on Earth to basically be able to use that partnership.
他们随后还报道了这一点：他们会利用这一合作关系来筹集债务，以资助 OpenAI 希望推进的大量基础设施建设。
And they were subsequently reporting on this fact that they would use that partnership to be able to, be able to raise debt, basically, in order to fund a lot of the infrastructure build out that OpenAI had wanted to do.
这是因为 OpenAI 仍不是一家上市公司，更别提盈利了，因此很难像 NVIDIA 那样轻易筹集到如此大规模的债务。
And that's because OpenAI still is not a public company, let alone, you know, not a profitable company, was having a harder time raising the levels of debt that, say, an NVIDIA could.
NVIDIA 可以随意筹集任何资金，因为它们有股票作为后盾。
NVIDIA could basically raise whatever it wants because, again, they have the stock to back it up.
它们拥有大量资产作为担保，也有利润作为支撑。
They have all these assets to back it up, and they have the profits to back it up.
OpenAI 并不具备这样的条件，但它们仍希望自主掌控基础设施的建设。
OpenAI is not in that that place, but they still wanted to to be in charge of their own build outs.
那么，该怎么做呢？
And so how do you do that?
你可以找人合作，显然它们已经与甲骨文以及其他公司在这方面展开了合作。
You partner with someone on it, and they've obviously been partnering with Oracle and many others around those around those lines.
但对我来说，这似乎正是这笔交易的核心部分之一。
But, like, to me, that seemed like what a big part of this deal was.
本质上，NVIDIA 出面为 OpenAI 所需筹集的债务提供担保。
Basically, NVIDIA stepping in to be a, you know, a guarantor of of the debt that that OpenAI would need to raise.
那么现在这种情况会如何发展？
And what happens to that now?
这被取消了吗，还是他们决定因为某种原因不再需要了？
Is that off the table, or did they decide maybe they don't need them for whatever reason anymore?
也许这笔新的融资能帮上忙。
Maybe this new funding, you know, helps with that.
但我不确定。
But I don't know.
这真是个奇怪的部分。
That's that's a weird part.
这很有趣。
That's fascinating.
如果这种情况真的发生，这可能会给OpenAI带来大问题。
And that could be a big problem for OpenAI should that materialize.
我觉得没有人充分讨论这个问题。
I don't think enough people are talking about that.
最后再谈一点，这也是你之前写过的一个故事。
One last bit on this, and this is a story that you had also written.
我们上周五聊到一点，Anthropic 现在正在筹集 200 亿美元，而 OpenAI 正在筹集 1000 亿美元。
We talked a bit a bit on on Friday about how Anthropic is now raising 20,000,000,000 and OpenAI is raising a 100,000,000,000.
这些资金来源正从一些你原本不会认为会为这些特定公司提供资金的地方混合而来。
And the funding sources are sort of mixing and matching from places that you wouldn't think would typically be the source of funding for the specific companies.
例如，微软在成为 OpenAI 最大支持者之后，却向 Anthropic 注资；亚马逊可能向 OpenAI 投入 500 亿美元，而此前它一直是 Anthropic 最大的支持者。
For instance, Microsoft putting money into Anthropic after being OpenAI's biggest backer, and Amazon maybe putting 50,000,000,000 into OpenAI after being Anthropic's biggest backer.
我觉得你对这一点的解读非常有趣，那就是正在形成一个反谷歌联盟。
And I think the way that you frame this is really interesting that there is effectively an anti Google alliance, forming out there.
这些公司、资金方、大型科技企业、风投机构，以及 Anthropic 或 OpenAI 这样的实验室，现在都意识到，他们正面临一场与谷歌的生死之战，因此会不遗余力地行动。
Whereas all these companies, the fund the funders, the big tech funders, the VCs, and the labs, be it Anthropic or OpenAI, now realize they're in for the fight of their lives against Google, and they're just gonna do whatever they can.
或许所有旧日的恩怨都会暂时搁置。
Does you know, all old rivalries maybe go aside.
他们会不择手段地构建一种制衡力量，以应对谷歌日益崛起的势力。
They'll do whatever they can in order to be able to build some counterweight to the emerging force that Google is.
是的。
Yeah.
这基本上也是我对整个情况的宏观理解。
That's that's sort of, again, my high level read of it.
我们之前在之前的对话中讨论过谷歌的崛起，当时它一度被冷落、打压，人们不明白它为何没能抓住机遇。
And we had talked previously, you know, in previous conversations about Google's ascension, you know, after being sort of kicked kicked around and and, you know, being beaten down a bit as to why they weren't sort of meeting the moment.
但到了去年年底，当Gemini 3发布，甚至他们连‘香蕉’都推出来了，那一刻，巨兽彻底苏醒了。
Now towards the end of last year when they sort of you know, when Gemini three rode in and and they even nano the banana and everything, right, it basically, awoke in the beast.
于是，OpenAI突然拉响了红色警报，所有人都开始清醒地意识到，谷歌已经具备了可能主导这场竞赛的所有条件。
And now all of a sudden, was a code red from OpenAI, and and, you know, everyone's sort of, eyes are wide open to this to this realization that Google has everything they need to potentially, you know, take over this race.
我认为，许多大型科技公司的同行们也意识到了同样的问题。
And I do think that a lot of their peer group in in big tech probably recognizes the same thing.
我认为，微软、亚马逊，甚至Meta也是如此。
And I think the Microsofts of the world, the Amazons of the world, and the Metas of the world too.
Meta的情况稍微有点不同，我们可以另谈，因为它们并不在资助这些其他公司。
Meta's a little bit different of a story, which we can talk about separately, but because they're not one of these ones that's funding these other companies.
但我认为，其他主要的云服务提供商，尤其是拥有竞争性云平台的微软和亚马逊，都意识到：是的，它们可能需要联合所有非谷歌的力量。
But I think these other major cloud players, least, the ones who have the rival clouds, specifically Microsoft and Amazon, realize that, yeah, they probably need to align around basically anyone who's not Google.
对吧？
Right?
他们不一定在乎，我的意思是，他们其实也在乎。
They don't they don't necessarily care if it I mean, they do care.
显然，他们希望是自己的产品能够脱颖而出、取得胜利。
Obviously, they would hope that it's their their stuff that takes off and sort of wins the day.
但归根结底，他们也可以成为Anthropic的大股东。
But at the end of the day, they can also be a huge shareholder in Anthropic.
你知道的？
You know?
我觉得他们会为此感到高兴。
And and I think that they'd be happy about that.
他们甚至可以成为X AI的股东，只要不是谷歌——他们的主要竞争对手，那个拥有所有关键要素、可能全面主导这场竞赛的公司，他们就会开心。
They can maybe, you know, have, be a shareholder in even x AI, and they can be happy about that as long as it's it's not Google, their chief sort of competitor and the and the one company that has all the pieces in place to take this over.
当然，谷歌本身在Anthropic中也有重要股份，但那是在我们当前所处的这种局面之前就存在的了。
Now, obviously, Google itself is a big stake in Anthropic, but that sort of predates, you know, this, the situation that we're in right now.
所以，对我来说，最大的震撼是亚马逊那500亿美元的报告。
And so, yeah, to me, the the big eye opener was that Amazon 50,000,000,000 report.
如果他们真的投入500亿美元到OpenAI，而他们又是Anthropic的最大股东的话。
If they end up really investing $50,000,000,000, you know, into OpenAI after being they are the largest shareholder of Anthropic.
你知道吗？
You know?
我一直在想，这是否意味着他们在某种程度上对Anthropic持负面态度？
I was trying to think, like, does that mean something that they're negative in some ways against Anthropic?
我不这么认为。
I don't think that's it.
我只是觉得，他们只想确保自己能处于一个可以自由选择投资对象的位置，只要不是谷歌就行。
I just think that they want to make sure that they are in a place where they can, you know, pick and choose what they want as long as it's anyone but Google.
没错。
That's right.
不。
No.
这是一个非常好的见解。
It's a it's a great insight.
现在这在某种程度上解释了我一直困惑的问题，那就是：为什么这种资金交叉融合会发生？
And now it sort of explains this thing that I've been struggling with, which was like, why is this funding cross pollination pollination happening?
我认为这是我听过的最贴切的解释了。
And I think that's about as good of of an explanation as any that I've heard.
好的。
Alright.
OpenAI 正在筹备上市。
OpenAI is eyeing an IPO.
我们现在已经有了一个被《华尔街日报》报道的日期。
We have a date now that has been reported in The Wall Street Journal.
我们将在广告后回来，讨论具体时间以及为什么是这个时候。
We'll talk about when that is and why that is when we're back right after this.
你想吃得更好，但你一点时间都没有，也没精力去实现。
You wanna eat better, but you have zero time and zero energy to make it happen.
Factor 不要求你提前备餐或遵循食谱。
Factor doesn't ask you to meal prep or follow recipes.
它只是彻底解决了这个问题。
It just removes the entire problem.
两分钟，真材实料，搞定。
Two minutes, real food, done.
还记得那次你想健康饮食，却没时间做饭吗？
Remember that time where you wanted to cook healthy but just ran out of time to do it?
你并不是在健康饮食上失败了。
You're not failing at healthy eating.
你只是缺少额外的三个小时。
You're failing at having an extra three hours.
Factor 由厨师烹制，由营养师设计，并直接配送到你家门口。
Factor is already made by chefs, designed by dietitians, and delivered to your door.
你只需加热两分钟，然后享用即可。
You heat it for two minutes and eat.
里面含有优质蛋白质、色彩丰富的蔬菜、全食物成分、健康脂肪，这些都是如果你有时间会自己做的东西。
Inside, there are lean proteins, colorful vegetables, whole food ingredients, healthy fats, the stuff you'd make if you had the time.
此外，我们还推出了全新的肌肉强化系列，专为增肌和恢复设计。
There's also a new muscle pro collection for strength and recovery.
你总能吃到新鲜的食物。
You always get to eat fresh.
只需两分钟即可准备好。
It's ready in two minutes.
无需准备，无需清理，无需精神负担。
No prep, no cleanup, no mental load.
前往 factormeals.com/bigtech，使用代码 big tech 50 off，即可享受首单 Factor 餐盒 50% 折扣，并免费获得一年早餐。
Head to factormeals.com/bigtech 50 off and use code big tech 50 off to get 50% off your first factor box plus free breakfast for one year.
此优惠仅限新客户，使用代码并购买符合条件的自动续订订阅时有效。
Offer only valid for new Factor customers with code and qualifying auto renewing subscription purchase.
用 Factor 让健康饮食变得更简单。
Make healthier eating easy with Factor.
我们回到《大科技》播客，今天邀请到的是MG·西格勒。
And we're back here on big technology podcast with MG Siegler.
MG在spyglass.org上撰稿。
MG writes at spyglass.org.
非常推荐。
Highly recommended.
如果你想了解AI和大科技的全部洞见，这里绝对是绝佳去处。
It's definitely a great place to go for all insights on AI and big tech.
MG，当然，如果你是第一次听这个节目，他每个月第一个星期一都会和我们在一起。
And MG, of course, if you're new to the show, is here with us on the first Monday of every month.
今天是2月3日星期一，我们在这里讨论一些重大的话题。
And given that it's Monday, February 2, we're here talking and we have some big stuff to talk about.
在本环节中，我们将讨论OpenAI计划的IPO，以及为什么苹果的财报尽管表现优异，却仍无法赢得华尔街的认可。
We'll talk in this segment about OpenAI's planned IPO and why Apple earnings, despite being amazing, don't seem to be able to buy the company any credit with Wall Street.
我们先聊聊IPO的事。
Let's talk about the IPO first.
我个人原本认为，OpenAI 不可能在2026年尝试上市。
I thought personally there was no way OpenAI would try to go public in, 2026.
显然，去年年底我与萨姆·阿尔特曼交谈时，他跟我提到过，他非常不愿意担任一家上市公司的首席执行官，大致就是这类意思。
Obviously, Sam Altman made that comment to me when I spoke with him late last year, that he he would hate being a public company CEO, something along those lines.
而现在，我看到《华尔街日报》刊登了这篇报道。
And now I'm looking in The Wall Street Journal has this story.
OpenAI 计划在第四季度上市，以抢在 Anthropic 前面进入市场。
OpenAI plans fourth quarter IPO in a race to beat Anthropic to market.
OpenAI 正在为今年第四季度的公开上市做准备，随着与竞争对手 Anthropic 的竞争加剧，这家估值5000亿美元的初创公司正在与华尔街银行进行非正式会谈，探讨潜在的首次公开募股，并正在扩大其财务团队。
OpenAI is laying the groundwork for a public listing in the fourth quarter of this year, accelerating its plans as competition with rival Anthropic intensifies the $500,000,000,000 startup is holding informal talks with Wall Street banks about a potential initial public offering and is growing its finance teams.
OpenAI 的财务团队表示，高管们私下对 Anthropic 可能抢先完成上市感到担忧。
Its finance team, OpenAI executives have privately expressed concerns about Anthropic beating the company to an IPO.
周五，我邀请了《金融时报》旧金山分局主编斯蒂芬·莫里斯做客。
On Friday, had Stephen Morris, the, San Francisco bureau chief of the Financial Times on.
我们讨论的是，当你思考这些数字时，问题在于资金将从何而来？
And we were talking about, like, when you think about these numbers, the question is, where is the money going to come from?
钱够吗？
Is there enough money?
假设OpenAI以1.5万亿美元的估值上市。
Let's say OpenAI were to go public at like 1 and a half trillion dollars.
外面有足够的资金来支持这样一次IPO吗？尤其是如果Anthropic提前一个月上市，那么传统的IPO投资者就得在两者之间做选择，可投资的资金就会减少。
Is there enough money on, you know, out there to fund an IPO like that, especially if, let's say, an anthropic comes out a month before and then you're looking at, you know, the traditional, you know, IPO share buyers just like having to decide and then the amount of money that you're that's available comes down.
所以我想知道你的看法。
So I'm curious what you think.
这是不是对那种说法的回应——即外面的资金是有限的？
Is this a response to that and just to them saying, well, there's a limited money out there.
我们最好赶紧把钱拿到手。
We better go get it.
如果是这样，那或许是个明智的决定。
If so, maybe that's a smart move.
是的。
Yeah.
我去年年底时对此写过一点内容。
I wrote about this a little bit at the end of last year.
大约就是那时，传闻开始说Anthropic可能在2026年考虑上市。
Well, around the time that, yeah, the rumors started that that Anthropic was thinking about, you know, potentially going public in 2026.
因为在我看来，这简直就是对OpenAI的终极挤压。
Because to me, it's sort of like that is the ultimate OpenAI squeeze.
因为，我们已经讨论过，谷歌已经醒悟过来，正从顶端对这家全球最大的公司之一施加压力。
Because, right, we already talked about the fact that Google is sort of, you know, has woken up, and they're sort of squeezing from the top one of the biggest companies in the world.
它们正在瞄准OpenAI在人工智能领域历来诸多优势的多个方面。
They're going after many different elements of of what OpenAI's, historic strong points have been in AI.
与此同时，Anthropic一直被认为是全面规模较小的玩家，可能更专注于企业市场，目标更集中。
Meanwhile, you've got Anthropic, which was always thought to be sort of the smaller player right across the board, maybe going more after enterprise, more focused on that.
但最终，如果两者都准备上市，毫无疑问会存在先发优势，因为人们会期待市场对各种人工智能投资存在积压的需求。
But ultimately, if they're both going to go public, you know, there's there's a sort of first mover advantage for sure, you would imagine, because you would hope that there's pent up demand or they would hope that there's pent up demand in the market for, you know, various AI bets.
而率先上市的那一家，很可能更容易获得成功，尤其是如果它展现出更优的盈利前景，或至少比另一家更接近更好的盈利路径的话。
And the first one to go out there is probably gonna have, you know, a bet better of a time perhaps, especially if that first one to go out has better looking economics or at least the path to better looking economics than the other one does.
而这再次印证了去年年底时的说法，即Anthropic在实现盈利方面可能比OpenAI领先了几年。
And that, again, points to the notion at the time from at the end of last year that Anthropic was said to be, you know, maybe a couple years ahead of OpenAI when it came to being able to turn a profit.
因此，如果Anthropic率先上市，并且拥有比OpenAI快得多的明确盈利路径，那么当OpenAI上市时，将陷入极其被动的境地。
And so, again, if if Anthropic were to go out and go public ahead of OpenAI and they have this, direct path to profitability that's much quicker than what OpenAI can get to, that puts OpenAI in a very, very tricky spot when they were to go public as well.
而他们将不得不更多地依赖宏观的增长叙事。
And, you know, they would really have to rely on the bigger picture, bigger growth narrative.
但如今，随着Claude Code、Claude Cowork以及其他当前AI领域层出不穷的新动态，这种增长叙事正变得越来越模糊。
And, you know, that's becoming murkier by the day, right, with all this stuff going on with with not only Claude Code, but now Claude Cowork and, you know, all these other things that that are going on at the moment in AI.
我相信，这很可能会自然演变为OpenAI试图追赶Anthropic的局面。
I do believe that there's you know, that that seems like it would be a natural, outcome of this would be, yeah, OpenAI trying to race to be anthropic.
还有另一个因素，最近几天更是备受关注，那就是xAI。
There's one other element to this, which is even more sort of, in the news in the past few days, which is x AI.
如果他们真的与SpaceX合并——现在看来这似乎即将发生，甚至可能本周就会公布，如果真能如此迅速地达成，那简直令人难以置信。
If they really do merge, with SpaceX, which it does sound like now is gonna happen and may even be announced this week, which is insane, how fast that that came together if it comes together in that way.
这可以说是埃隆突然绕开原有路径，从单纯的太空业务转向AI领域，并计划在今年六月或七月上市的一次精彩迂回。
That's sort of a interesting end runaround by Elon to all of a sudden have potentially a, an an AI play rather than just the space play, to go public in you know, as rumored in June or maybe July, of this year.
所以现在来看，Anthropic或OpenAI根本不可能赶上这个时间表。
So well ahead of when there's no way that that Anthropic or OpenAI can meet that timetable right now.
我认为SpaceX在进程上远远领先于它们。
I think SpaceX is way ahead of them in terms of where they are in the process.
那么，如果埃隆真的来个终极绕后，赶在这些公司之前推出x AI，成为首个AI领域的先行者，会怎样？
And so what if Elon does the ultimate end run around and gets x AI out before either of these companies and becomes the ultimate first mover AI play?
我的意思是，他肯定很想在萨姆面前炫耀一下，对吧？
I mean, he would love to hold that over Sam, wouldn't
当然了。
Oh, of course.
这正是策略的一部分。
That's I that's part of the that's part of the strategy.
这肯定是这里策略的一部分。
That has to be part of the strategy here for sure.
毫无疑问。
For sure.
难以置信。
Unbelievable.
为了记录在案，我不认为OpenAI会在2026年推出。
I for the for the record, I do not think OpenAI is gonna go out 2026.
2027年？
2027?
大概吧。
Probably.
2026年？我也不觉得。
2026, I don't either.
我同意。
I agree.
我也同意这一点。
I I agree with that.
你知道吗，我们上次聊过我的预测。
You know, we talked about my predictions last last go around.
那就是我其中一个观点，我认为任何主要的AI公司都不太可能在2026年倒闭，尽管大家都在谈论这件事。
That was one of them that I didn't think that any of the major, AI companies would go out in 2026 even though they're all talking about it.
不过，X AI 可能会通过这次合并证明我是错的，我想。
X AI, though, might just prove me wrong by this merger, I guess.
但，但没错，就OpenAI和Anthropic而言，我觉得考虑到它们目前的业务状况以及我们正在讨论的这些巨额融资，这会把时间稍微推后一点。
But, but, yes, in terms of OpenAI and Anthropic, I just think that beyond, where their businesses are at and with these now massive fundraisers as we're talking about, like, I I think that it will push it out a little bit.
而真正的变数，正如一贯如此，是宏观层面的故事。
And the real wild card is obviously, as it always is, the macro story.
对吧？
Right?
比如，最终会发生什么？
Like, what ends up happening?
可能会有很多事情干扰或至少推迟任何一种仓促行动。
There could be so many things that sidetrack or at least delay, you know, any sort of rush.
但再说一遍，如果OpenAI真的在全力冲刺，想要赶在Anthropic之前脱颖而出，那确实存在一个窗口期，不过我认为这很可能还是2027年的事。
But, again, if it really is a full on sprint for OpenAI to get out ahead of Anthropic, there's a there's a window which they do that, but I think it's it's still probably a 2027 thing.
是的
Yeah.
同意
Agreed.
好的
Okay.
在我们结束之前，最后一个故事。
One last story before we go.
苹果发布了有史以来最好的财报。
Apple turned in, I think, the best earnings report, in its history.
它实现了1430亿美元的收入，超过1038亿。
It brought in a 143,000,000,000 in revenue, over a 100 '30, eight.
iPhone的预计收入为852.7亿美元，超出预期60亿美元。
Estimated iPhone revenue was 85,270,000,000.00, beating the estimate by 6,000,000,000.
预期为780亿美元。
The estimate was 78,000,000,000.
iPhone品类实现了23%的年同比增长，这太惊人了。
That's 23% growth in the iPhone category, year over year, which is insane.
记得吗？他们曾经有一段时间难以推动iPhone增长。
Remember, they were struggling to grow iPhone, for a while.
他们在盈利能力上也超出了预期。
They also beat on profitability.
但华尔街似乎并不买账。
Yet, Wall Street did not seem impressed.
股价只是微幅上涨，自财报发布以来基本持平。
The stock is, you know, up a tiny bit, but mostly flat, since the earnings announcement.
苹果需要做些什么才能扭转这一局面？
What does Apple need to do to get that story turned around?
已经连续六周左右没有明显变动了。
It's been flat for six weeks or so.
我认为这又将是苹果历史上表现最好的一年。
I think this is, again, this is gonna be the best year in Apple history.
但当然，AI这个故事并没有为苹果带来有利影响。
But, of course, the AI story is something that that is not really working in its favor.
是的。
Yeah.
所以那里有几件事。
So there's a few things there.
首先，我认为人们对苹果的一些担忧与内存芯片等当前状况有关。
First first and foremost, I do think that some of that, you know, apprehension about Apple is related to just what's going on with memory chips and everything.
对吧？
Right?
这最终可能会挤压他们的利润率。
And and that it could ultimately end up squeezing their margins.
本季度的利润率非常惊人，这在当前各种情况之下让许多人感到意外，但苹果似乎在囤积内存芯片方面相当精明，而这不是蒂姆·库克通常会做的事。
The margins were incredible this quarter, which was sort of a surprise to many given everything going on, but it does seem like Apple's sort of been savvy in terms of, like, potentially hoarding, memory chips, which is not something that Tim Cook usually likes to do.
他喜欢将库存尽可能精简，但我们现在所处的宏观环境有些异常。
He likes to get this inventory as streamlined as possible, but we're in a weird, sort of macro environment for this.
而这在很大程度上是因为正在发生的AI革命，对吧？这场革命正在改变所有这些因素。
And and that's in in no small part because of the AI, revolution that's happening, right, and and is is changing all of those equations.
所以，我认为更宏观的图景确实是，市场看到了苹果与谷歌的合作，并且在最初宣布时给予了热烈反响。
And so, I do think the broader picture, though, yes, is that the market sees what Apple is doing with Google, and they applauded it right when it was first announced.
人们会想，哦，他们正在与AI领域的领导者——如果不是唯一领导者的话——谷歌联手，Gemini终于会修复Siri，苹果将迎来一个绝佳的局面。
That's like, oh, it's like, oh, they're teaming up with one of the leaders in AI, if not the leader in AI now, in Google, and Gemini will finally fix Siri, and we're gonna have, you know, a great situation for Apple.
但我认为，最终他们可能会这样看待这个问题：你看。
But I do think, like, ultimately, they might be looking at this as, like, well, look.
我们这边有Meta，正在投入1300亿美元的资本支出，而苹果的资本支出则要低得多，接近于零。
We have we have Meta over here that's spending a $130,000,000,000 in CapEx, whereas Apple is spending much closer to zero.
你知道，苹果的资本支出大约在180亿到200亿美元之间，因为他们显然没有投入大量基础设施来自主研发前沿AI技术。
You know, it's something around, like, it's, like, 18 or $20,000,000,000 in CapEx, because they're not building out, obviously, the massive infrastructure in order to to bake their own, cutting edge AI.
所以，我认为可能存在一些——甚至相当多的担忧：尽管苹果在短期内可能没问题，但如果从长期来看，他们不是AI领域的重要参与者，而如果你相信AI将彻底改变一切，包括硬件业务，那么苹果就处境艰难了。
And so, you know, I think that there's a little bit, maybe a lot of apprehension that while Apple may be fine in sort of the shorter term for the longer term, if they are not one of the key players in the AI space and if you believe AI is going to revolutionize everything, including hardware businesses potentially, like, then Apple's in a tough spot.
不过，苹果会反驳说，你看。
Now Apple would counter that, look.
这是短期的事情。
This is the short term stuff.
我们这个合作是短期的。
We're we're doing this partnership short term.
我们会修复Siri，并继续在幕后工作，最终推出我们自己的AI。
We're gonna fix Siri, and we're gonna continue to work on it behind the scenes to eventually, you know, roll our own AI.
但再说一次，他们目前的资本支出相对于Meta、谷歌、亚马逊、微软以及其他所有公司来说实在太小了，这让人感觉他们现在已经被甩在后面，而且再也追不上了。
But, again, the CapEx spend that they're that they're, you know, showing right now is so small relative to not just Meta, but Google and Amazon and Microsoft and everyone else that it feels like there's a world in which they're behind right now and that they can never catch up.
这显然是真正的担忧。
And that would be obviously the real fear.
不过，我认为你说得对。
Now I do think you're right.
我们之前也讨论过这一点。
Like and we've talked about this previously.
我认为今年整体上对苹果来说会是不错的一年，因为正如我们之前提到的，iPhone Fold以及他们即将推出的其他一些设备，我认为会为他们带来不错的业绩。
Like, I do think this will be a good year overall for Apple because I think it's no small part because, like, we've talked about before, the iPhone Fold and some of these other devices that they have coming out, I think will end up doing well for them.
但再说一遍，更大的图景、更长期的视角是人工智能。
But, again, the bigger picture, the longer term time horizon stuff is AI.
而目前，他们除了达成一些交易外，并未展现出任何紧迫感，而这种做法通常不是苹果的风格，因为苹果希望所有东西都自己掌控。
And, right now, they're just they're showing no real, energy around being, that they need to have a sense of urgency around it other than cutting deals, which is just not typically what Apple does, whereas they wanna own everything in house.
是的。
Yeah.
就在苹果财报发布前一周，我上了CNBC，当时我为这家公司挺身而出，这通常不是我的做法。
I was on CNBC last week right before Apple earnings, and I kinda stuck my neck out for the company, which I don't typically do.
但我基本上是这么说的：这是苹果的看涨理由所在。
But I was basically like, here's where Apple the the bull case for Apple, sits.
他们今年将疯狂销售最新款以及下一款产品。
They are gonna sell the latest model and probably the next model like crazy through the year.
与此同时，目前还没有任何颠覆性的AI应用或设备威胁到他们的核心业务。
Meanwhile, there's no killer AI app or device yet that is threatening to disrupt their core business.
我的意思是，如果你仔细想想，今年我们可能会看到一些AI设备。
I mean, if you think about it, we may get some AI devices this year.
我的预测是，它们的表现只会令人失望。
My prediction is they're just gonna they they will be underwhelming.
它们还不会对智能手机的增长构成威胁。
They won't be a threat to smartphone growth yet.
两三年、四五年后，肯定会有。
Two, three years, four years, five years from now, definitely.
但至少在短期内，苹果看起来会非常不错，因为AI对其核心业务的直接威胁并未显现。
But at least in the short term, it's gonna look really good for Apple because the sort of immediate threat to them from AI is not materializing.
中期的威胁依然存在。
The intermediate threat is still there.
但也许市场并不在意这一点。
But may maybe the market doesn't care about it.
也许他们更关注长期发展，而不是像我们常批评的那样逐季表现。
Maybe they're more long term focused and not quarter by quarter like we like to ding them for.
是的。
Yeah.
而且，我的意思是，我现在同意的一点是，苹果在未来一两年内不花大笔资本支出来大规模扩建，反而显得非常明智。
And and, I mean, the the one thing that I agree with right now is that there is a a path in which Apple looks very smart in, say, a year or two years from now in not having spent all of this CapEx to do these massive build outs.
对吧？
Right?
如果这些模型完全商品化了，他们就可以坐在那里，自由选择想用什么，以及想走哪条路。
If the model's fully commoditized and all that, that, you know, that they're just sitting there and they can sort of pick and choose what they wanna use and and also pick and choose which path they wanna go down.
不过，我还是要回到这个观点：他们实际上并没有在幕后做其他事情，以便最终能够掌控自己的AI模型。
Again, though, I go back to the idea that, like, they're just not doing the other stuff behind the scenes in order to to be able to eventually sort of, you know, take over their own control of these AI models.
因为他们根本没有相应的基础设施。
Like, they just don't have the infrastructure in place.
有传言说他们在建设数据中心，开发自己的芯片来训练这些模型，但他们的投入相对于同行来说并不够，看起来并不像他们真的认真对待这件事。
There's talk that they're that they're building out data centers, that they're building out their own chips to be able to train these, but the spend isn't there relative to their peers where it would seem like they're really taking this seriously.
当然，也许存在一种情况，即大型语言模型并非唯一终极路径，还需要其他机制和方法来实现人工智能。
Now, again, maybe there's a world in which, like, LLMs sort of end up being, not the the be all end all path, and there needs to be other, you know, mechanisms in order to and other methods of of doing AI.
因此，也许苹果可以在那时介入并赶上来。
And so maybe Apple can sort of come in at that point and and sort of catch up.
但我们目前看到的一切都表明，事实并非如此，如果他们真的想拥有自主研发的AI，最终必须投入比现在多得多的资金。
But everything we're seeing right now is that that's not the case and that they'll need to, at some point, spend a lot more money than they are if they really wanna have their own sort of, you know, AI that's built in house.
是的。
Yeah.
确实，这家公司内部似乎做出了一个决定，可能是来自最高层的，那就是：现在先观望，之后再想办法。
It definitely does seem that somewhere inside that company, there was a decision made maybe from the very top that was like, let's sit this out for now and and just figure it out afterwards.
就像你说的，这可能最终会是一个明智的决定。
That, like you said, that might turn out to be a good decision.
但另一方面，这非常冒险。
On the other end, very risky.
非常冒险。
Very risky.
对。
Yeah.
而且，这也不是说苹果现金多得用不完。
And it's not like the the alternative is Apple has so much cash.
他们终于做出了一个AI收购，这正是你我长期以来讨论过的。
They made the you know, they finally made an an AI acquisition, as you and I have long
谈过的事情。
talked about.
但这次收购的对象并不是Perplexity。
But it wasn't for a no.
也不是Frontier模型公司。
It wasn't perplexity, and it wasn't for a Frontier model, company.
而是收购了一些有趣的技术。
It was it was for interesting technology.
你知道，Disclosure GV是这家公司的投资者，而我以前长期在那里担任合伙人。
You know, Disclosure GV was an investor in, so it sounds, you know, where I where I previously was a partner for a long time.
所以，我认为这很可能是一步精明的棋，但这是20亿美元的投资，据报道称是这样。
So, you know, I think that that's probably a savvy play, but it's a $2,000,000,000 investment, like, you know, that's reported.
而且正如你所指出的，他们目前正创下创纪录的利润。
And they have you know, as you noted, they they're doing record profits right now.
他们把这笔钱花在了哪里？
What are they spending that money on?
他们把这笔钱用于股票回购，还花在了一些对人工智能发展帮助微乎其微的事情上。
They're spending that money on buybacks, and they're spending that money on, you know, things that that are not moving the ball forward with regard to AI, except in very small ways.
所以，我不清楚有什么相反的观点。
And so, you know, I I don't know what the what the counterargument is.
我不是说他们必须投入一千亿美元的资本支出，但如果说同行每年在资本支出上投入一千五百亿美元，他们至少应该考虑花五百亿美元。
I'm not saying that they have to spend a $100,000,000,000 in CapEx, but I'm saying maybe they should spend $50,000,000,000 if their peer group is spending a $150,000,000,000 on CapEx a year.
至少做点什么，以防万一，这还是值得的。
Maybe it's it's at least worth it to do something that's, you know, just in case scenario.
没错。
Right.
我同意。
I I agree.
你至少得开始花点钱了，因为即使你坚信人工智能不会像人们想象的那样具有革命性，你也得稍微做些对冲，毕竟我们现在看到的情况就是这样。
You gotta you gotta at least start spending a little bit because even if you have conviction that AI will be a you know, maybe not as revolutionary as people imagine, you have to hedge a little bit because of, I mean, what we're seeing right now.
所有AI都聚在模型书里，成群结队地密谋推翻人类。
All the AIs hanging out together in mold books, swarming and plotting out to overthrow humanity.
所以，蒂姆·库克，拜托了。
So, you know, Tim Cook, come on.
多投入点真金白银吧。
Throw throw a little more skin in the game.
天哪。
Jeez.
是的。
Yeah.
对。
Yeah.
我真想进个聊天室，看到蒂姆·库克的模因机器人在那里，和别的苹果高管聊得正欢。
It I I wanna go into a chat room and see Tim Cook's molt bot there and, chatting chatting away with, with other Apple executives.
我的天。
Oh my god.
有人利用这个漏洞来控制他们的电脑。
And somebody using that exploit to take over their computer.
这会是个故事。
That'll be a story.
是的。
Yeah.
好的。
Alright.
网站是 spyglass.org。
The website is spyglass.org.
我们的嘉宾 MG Siegler 每月第一个星期一都会加入我们。
Our guest, MG Siegler, joins us every the first Monday of every month.
和你聊天总是很棒，MG。
It's always great to speak with you, MG.
谢谢你的到来。
Thanks for coming on.
一如既往，谢谢您，亚历克斯。
Thanks as always, Alex.
我们改天再聊。
Talk soon.
好的。
Alright.
期待下个月再和大家做这次对话。
Looking forward to doing this again next month, folks.
周三，Coher公司的首席人工智能官乔尔·皮诺将来到我们这里，讨论人工智能研究的最新进展以及前沿方向，希望大家收听。
On Wednesday, Joel Pino, the chief AI officer of Coher will be here with us to talk about the latest in AI research and where the cutting edge is heading, so we hope you tune in for that.
希望在那之前，人工智能机器人不会接管世界。
Hopefully, the AI bots won't turn take over the world between now and then.
所以，如果人类仍然掌控大局，我们下次再见，欢迎收听《科技大观》播客。
So, if humanity remains in charge, we'll see you next time on big technology podcast.
关于 Bayt 播客
Bayt 提供中文+原文双语音频和字幕，帮助你打破语言障碍，轻松听懂全球优质播客。