SpaceX对26.5万亿美元人工智能市场的展望

今天的播客里提到，OpenAI正在聘请一位曾处理克林顿危机的专家，来重塑人工智能的公众形象。

On the podcast today, OpenAI is hiring a Clinton crisis veteran to rebuild AI's public image.

我们想聊聊这背后意味着什么，为什么我觉得AI的公众形象这么差，我们能做点什么，真相到底在哪，又有哪些地方被夸大了。

We wanna talk about what that means, why I think we have this bad public image, and what we can do about it, what the truth to it is, and what some of the overblown areas of it are.

特朗普还取消了那项关于AI测试的行政命令。

Trump is also canceling the AI testing executive order.

不少首席执行官都拒绝出席白宫的这场活动。

A bunch of CEOs declined to come to an Oval Office event.

那边确实闹得挺热闹，但我们报道的是，他只是推迟了这项命令。

There's a bunch of drama there, but this is something we are reporting on that he delayed it.

现在看来，这命令实际上是被取消了。

Now it feels like it's actually canceled.

SpaceX称，AI市场的规模高达26.5万亿美元，而就在这个时候，Grok在用户使用量上已经落后于OpenAI和Anthropic。

SpaceX is saying that there is a $26,500,000,000,000 AI market, and this is all happening as Grok is falling behind OpenAI and Anthropic in usage.

谷歌现在也在调整他们的AI科研方向，更多地转向智能体系统的研究。

Google is now shifting their AI science strategy, moving it more towards agentic research systems.

我们来聊聊这对科学意味着什么。

We'll talk about what that means for science.

文学界对AI生成的小说几乎毫无准备。

And the, literary world has been pretty much unprepared for AI generated fiction.

之前闹出过一场大奖丑闻，暴露出不少AI内容检测的漏洞。

There was a huge prize scandal, and it exposed a bunch of the kind of AI generated detection gaps.

我对这些其实有个不太一样的看法，不过咱们先切入今天的主题。

I actually have probably a contrarian take on all of that, but let's get into the podcast today.

首先我想谈的是OpenAI眼下这场公关危机，他们刚聘了克里斯·莱恩，这位来自克林顿时代的专家。

The first thing I wanna cover is the OpenAI, kind of AI crisis, public image scandal thing that's going on right now, where essentially they've hired Chris Lane, who is a Clinton era.

他被称为危机处理专家。

He's a he's called the crisis fixer.

他的主要职责是负责全球事务和公关沟通。

But basically what he does is he leads their global affairs and their communications.

目前这家公司正和多个机构对峙，还牵扯上多起诉讼。

Right now, the company is fighting with a whole bunch of different organizations, and there's lawsuits.

现在情况很复杂，尤其是他们正面对公众对人工智能的负面情绪。

There's a lot going on, especially in the fact that they're they're kind of fighting with this negative public sentiment towards AI.

我听说，民调显示有六七成的普通人对AI持负面看法。

I think it's something like 60% or 70% of people have a negative view of AI when they poll the general public.

所以，李翰现在的目标就是重建信任。

So, Lee Han, his goal right now is to basically rebuild trust.

他还得游说各州通过不会限制OpenAI发展的AI法规，这平衡起来特别难，对OpenAI是这样，对任何人都是，他得花大量精力处理公关，还得谈失业问题和AI安全。

He also has to lobby the states to pass AI laws that aren't gonna constrain OpenAI's growth, and this is a balancing act that is pretty complicated, for OpenAI, but honestly for anyone, and he has a lot of work on on messaging and, you know, he's gotta talk about job loss and AI safety.

这方面的事情真不少。

There's a lot going on there.

我觉得特别重要的一点是，他去年夏天还帮着推动成立了‘引领未来’超级政治行动委员会，成功拉到了OpenAI联合创始人格雷格·布罗克曼和其他几位科技界人士合计上亿美元的承诺捐款。

So the thing that I do think is really important is that he actually helped launch Leading the Future Super PAC last summer that secured over a $100,000,000 in funding commitments from OpenAI cofounder Greg Brockman and a bunch of other tech figures.

OpenAI最近还支持了伊利诺伊州一项针对AI实验室造成灾难性伤害的免责法案。

OpenAI also recently backed an Illinois liability shield bill for AI Labs causing catastrophic harm.

后来遭到大量批评，他们就赶紧撇清关系了，但那确实是他们最近在游说的内容。

They then kind of distanced themselves after there was a lot of criticism from that, but that was what they what they recently were lobbying for.

所以这又是他得去应对的另一个危机。

So that's another crisis that he's going to have to try and talk to.

另一个他得面对的问题是，OpenAI经济研究团队的不少前成员辞职了，他们都表示，这个团队已经变成了一个游说机构，一直在淡化OpenAI对经济造成的负面影响，只要一碰到不利的情况，管理层就直接说：‘AI一切都好’。

And also another crisis he's gonna be dealing with is that a bunch of former members of OpenAI's economic research they quit and they all said that they were quitting because, that particular team became an advocacy arm and they were saying that it was basically downplaying a lot of OpenAI's negative economic impacts, and that, you know, like, it was kind of inconvenient to the leadership, then they would just say, like, look, everything about AI is good.

特朗普已经正式取消了AI测试的行政命令。

Trump is officially canceling the AI testing executive order.

很多首席执行官都拒绝参加这场原本在椭圆形办公室举行的活动。

A whole bunch of CEOs declined to come to a an off or, kind of an Oval Office event.

这项命令原本要求AI模型在公开发布前有最多90天的时间提交给政府进行安全审查，但现在这项规定被取消了。

This is getting scrapped before it was gonna be you know, AI models had up to ninety days that they had to give, you know, their model to the government to go test, make sure it was safe before public release.

不少顶尖AI实验室的CEO都推掉了这场临时安排的椭圆形办公室会议，他们认为这项命令实际上会拖慢美国AI产品的发布进度。

A lot of the top labs, AI CEOs declined this kind of last minute Oval Office event about this, and the order, essentially, they were saying was gonna slow down AI launches in The US.

中国显然正在加速自己的AI治理框架，很多人说，如果我们得因为政府审查而推迟90天发布模型，那中国可能只比我们落后30到60天，而说实话，政府的审查结果大概率和各AI实验室内部的红队测试没什么区别。

China obviously is accelerating its own AI governance framework, and a lot of people were saying, look, China might be thirty to sixty days behind us with a lot of their models if we're gonna have to, you know, delay all of our model releases by ninety days for the government to do a review, which honestly probably isn't gonna come up with a verdict that's any different than the red teaming offices at all of the AI labs.

我觉得Anthropic就挺负责任的，他们明确表示：我们的Mythos模型存在不少问题。

I mean, I think we saw Anthropic pretty responsibly say, look, our Mythos model has a lot of issues.

他们把这任务交给了不少大机构去修复漏洞。

They gave it to a lot of big organizations to go fix bugs.

还交给了政府，让他们处理代码漏洞和其他问题。

They gave it to the government to fix code vulnerabilities and other things.

所以感觉这些顶尖实验室已经在认真推进所有测试和安全工作了，再加九十天，反而可能让中国趁机加速，让他们的模型赶超我们。

And so it feels like these you know, the top labs are doing a good job of trying to get all of the all of the testing done, all the safety stuff done, and so adding another 90 on top of that felt like it might, it might basically push us so China could perhaps accelerate and have their models be better than us.

这正是很多举措被取消的主要原因。

That was, kind of the reason for why a lot of this was canceled.

OpenAI 原本是支持特朗普这项指令的。

OpenAI supported this particular order that Trump was working on.

马斯克和扎克伯格都劝特朗普取消它。

Elon Musk and Mark Zuckerberg both were telling Trump to cancel it.

有意思的是，原本安排了一场 CEOs 集体讨论会，但很多人临时抽不开身，最后特朗普干脆直接宣布取消整个行政令——可当时已经有不少高管在飞往白宫的路上，接到消息后又掉头返航了。

Something that's interesting is there was this whole event where all of these CEOs were supposed to talk about it, a bunch of them weren't gonna be able to make it, and ultimately Trump decided to cancel it and just say, look, I'm canceling this whole executive order, but apparently there's a bunch of executives that were in the air flying to the White House when the event got canceled, decided to, you know, turn around.

所以不少 AI 实验室本来想说，嘿。

So a bunch of the AI labs were trying to say, hey.

你看。

Look.

给我们一个十四天的测试期，别搞那九十天的，这样才更有竞争力，能赶超中国。

Give us, like, a fourteen day testing window instead of the original proposed ninety day testing window, and that was just going to be more competitive against China.

所以像马斯克和扎克伯格这样的人都说，这可是AI行业的重大胜利，但马斯克和他的公司可不止这一条新闻。

So people like Elon Musk and Mark Zuckerberg are both saying this is a big win for the AI industry, but this isn't the only news we're getting from Elon Musk and his companies.

SpaceX现在在他们的IPO文件里宣称，AI市场将有26.5万亿美元的潜在机会。

SpaceX right now is currently pitching that there is going to be a $26,500,000,000,000 AI market opportunity.

他们就是在IPO文件里这么写的。

They're saying this in their IPO filing.

他们说这几乎跟美国的GDP规模差不多了。

They're saying this is nearly as large as The US GDP.

可就在这个时候，他们自家的Grok聊天机器人，在企业级AI用户中的份额其实还很小。

Now all of this is also happening while their own Grock chatbot has a relatively small portion of the corporate AI user kind of user base.

企业技术研究机构做了一项调查，发现企业在内部使用Claude的比例，从2025年的21%飙升到了2026年的48%。

So, there was a research done by Enterprise Technology Research, and what they found is that Claude among companies jumped from about 21% in 2025 to 48% in 2026 of, you know, people using Claude inside their organization.

Gemini 从 27% 上升到了 40%，说实话，现在几乎全世界用 Google 的人都在用 Gemini，所以它的用户基数大得惊人。

Gemini rose from 27% to 40%, and I mean, honestly, almost everyone in the world that's gonna be using Google is using Gemini at this point, so there is a massive percentage of the population, that Gemini is gonna see.

相比之下，Grok 在企业市场的使用率一年内从 4% 增长到 7%。

Grok, on the other hand, their corporate usage grew from 4% to 7% in a year.

这个绝对增长确实小得多，而且 Grok 的付费个人用户数量也非常少。

This is definitely a smaller absolute gain, and the paid consumer usage of Grok is also really small.

也就是说，在整个 AI 市场中，虽然有 7% 的用户在用 Grok，但根据 App Magic 对 26 万美国消费者的调查，今年第二季度真正为此付费的只有 0.2%。

So of that, you know, let's say 7% of the over AI overall AI market that's using Grok, only point 2% in, the second quarter of this year according to App Magic's survey of 260,000 US consumers are actually paying for it.

SpaceX 大力宣传这个市场潜力巨大，但 Grok 并不是其中的主力。

So SpaceX is making a big pitch that the market is huge, but Grok is not the biggest one.

还有一份报告指出，在政府部署的 400 多个 AI 系统中，绝大多数都是 OpenAI 的，而 Grok 仅在四个政府项目中被实际采用。

There's also another report that came out that said of the government's, like, 400 deployments of different AI systems, a vast majority of those were clawed in OpenAI, and Grok was only deployed four times in government systems, for things that are actually happening.

所以，在 SpaceX 这次大规模上市之前，我觉得有必要提醒一点：机会确实很多。

So, you know, the I think it's an important thing to note as we go into this big SpaceX IPO is that I think there's a ton of opportunity.

我们知道，Anthropic 每月支付 SpaceX 十五亿美元，用来租赁他们的 Colossus 一号数据中心。

We know that Anthropix paying $1,500,000,000, I think, a month to OpenAI or sorry, not to OpenAI, SpaceX to lease their Colossus one data centers.

他们确实从这方面赚了不少钱，但真正靠GrokAI模型赚的钱并没有多到离谱。

They're making a lot of money from kind of from that, but there's not, an insane amount of money being made from the actual Grok AI model.

明白了。

Okay.

咱们聊聊谷歌这边的情况吧。

Let's talk about what's going on with Google.

他们正在调整AI科研策略，不再专注于开发像AlphaFold这样的专用工具，而是转向能自主开展研究的通用智能系统。

They are shifting their AI science strategy from building specialized tools like AlphaFold towards general purpose agentic systems that can autonomously conduct research.

听起来有点拗口，有些人说这是战略转向，但AlphaFold的联合创始人、诺贝尔奖得主约翰·詹珀，其实已经转去做编程工作了。

Kinda sounds like a mouthful, and some people are saying this is a pivot, but the noble laureate John Jumper who co created AlphaFold, he actually just moved to coding work.

这其实挺重要的，因为这说明谷歌相信，通用推理模型完全有可能超越专门领域的工具。

This is this is kinda big because I think it's kind of showing that Google is betting that general reasoning models can actually outpace domain specific tools.

如果我们真能做出一个在科学领域特别出色的模型，那也许更好的办法不是搞专用工具，而是把通用的Gemini模型或者代码模型做得更强，这才是加速科学进步的关键。

So, you know, if we're really making a a model that's super good at doing science, well, it turns out maybe just making the general Gemini model better better or the code model better, and that's actually how we're gonna accelerate science, not a a domain specific model.

这对谷歌和Gemini这样的前沿模型来说，是个重大的转变。

This is a big shift, especially for frontier model like Google and Gemini.

但真正让人惊讶的是，AlphaFold 至今仍是全球研究人员最常用的工具。

Now what makes this pretty striking is the fact that AlphaFold is still the most popular, tool for researchers worldwide.

目前大约有三百万名研究人员在使用它。

There's about 3,000,000 researchers that are using it.

Isomorphic Labs，也就是谷歌的药物研发部门，刚刚在 B 轮融资中筹集了二十亿美元，专门用于智能药物发现。

Isomorphic Labs, which is Google's drug, development unit, they actually raised $2,000,000,000 in a series b funding for agentic drug discovery.

我还想说，谷歌并不是唯一在做这件事的公司。

I'll also say that Google isn't the only one really chasing this.

OpenAI 的通用 GPT-5.5 模型最近在没有经过任何专业训练的情况下，就推翻了一个数学猜想。

OpenAI had their general purpose GPT 5.5 model, and it recently disproved a mathematics conjuncture without any sort of specialized training.

所以你看，现在这些前沿模型在科学、数学和其他领域都做出了令人惊叹的成果。

So think there's a lot of these frontier models that are doing some really incredible work in science, math, and in other areas.

随着模型越来越强大，我们能从它们那里得到的结果也越来越好。

As the models get better, we're able to just get better results out of them.

不过，谷歌确实不是唯一一家在全力推进这件事的公司。

But, yeah, Google isn't the only one that's really going hard here.

OpenAI 也在积极推动这方面的发展。

OpenAI is also trying to push this.

接下来，我想聊聊文学界闹出的一桩丑闻。

Next up, I wanna talk about the scandal in the literary world.

据说有个大型文学比赛，根本没料到会有AI生成的小说参赛。

Apparently, there was a big contest and they were not prepared for AI generated fiction.

结果，格兰特提交的一篇获得英联邦文学奖的短篇小说，被一款叫Pangram的软件检测出有百分之百的概率是AI生成的。

What happened is that Grant actually published one of the Commonwealth prize winning short stories, and then a software run called Pangram detected that there was about a 100%, chance of that short story being AI generated.

英联邦文学奖基金会表示，他们完全依赖作者的原创文字，绝不使用AI。

The Prize Foundation, which is a Commonwealth Prize, said that they rely 100% on writers' words, that they didn't use AI.

一旦发现用了AI，直接取消资格。

It disqualifies you if you use AI for this.

目前没有技术核查机制，参赛者只需提交作品就行。

There's no technical verification process, so people just submit their stories.

主办方觉得不错，就直接发表。

They publish them if they like them.

因为这场风波，很多人开始问，这些文学机构到底该怎么可靠地识别出AI生成的作品呢？

And then because of this whole scandal, a lot of people are just asking, like, how do these literary institutes reliably catch AI submissions?

当然，你可以直接问作者：你用AI了吗？

Of course, you can ask the writers, like, did you use AI?

但有意思的是，现在像《格兰塔》这样的机构，居然用Claude来检测AI内容，等于说让AI去抓AI。

But they they're also people are also pointing out that it's sort of ironic that right now in order to detect AI, Granta, for example, they use Claude to detect AI, and then it's like AI detects AI.

说到底，这事儿还挺滑稽的。

Anyways, it's kind of funny.

在我看来，这根本算不上什么大问题。

In my opinion, I actually don't think this is a big deal.

我觉得作者用AI写作，然后拿奖，完全没问题。

I think you should be allowed to use AI to write work and win prizes for it.

你完全可以是个极具创造力的人。

I think you can be an incredibly creative person.

你依然可以请AI帮忙润色句子、找找更贴切的形容词，诸如此类。

You could still ask AI to help you, tighten up a sentence or think of a new adjective or etcetera, etcetera.

AI在创意领域的用法太多了，根本说不清什么时候算用得过头了。

There's a lot of ways you can use AI in creativity, so it's really hard to know when using AI is too much AI.

就像电影里用CGI一样，格莱美和奥斯卡都说用了AI就取消资格，可到底什么才算AI呢？

It's like using CGI in a movie, you know, because the Grammys and Oscars are saying like if you use AI disqualifies your film for certain things, and so it's, you know, like, what is AI?

什么才算电脑生成的？

What is computer generated?

这些艺术机构到底想在哪儿划一条线呢？

What where is the line that a lot of these art institutions are are trying to draw?

我觉得吧，只要作品够美、能打动人，就该算数。

And I think personally, if you make something beautiful that resonates with a lot of it should still qualify.

不过说起来，他们这次是怎么发现这个问题的，还挺有意思。

It's actually interesting how they caught this in particular though.

据说贾米尔·纳赛尔那篇叫《林中之蛇》的文章，‘stubborn’这个词用了六倍于人类作者的次数，还反复用‘as if it had’这个表达。

Apparently, Jamir Nasser's essay or article, which was called The Serpent in the Grove, it used the word stubborn six times more than human writers, and it also used the phrase as if it had.

这种句式AI特别爱用，他用了五次之多，统计上明显指向是大模型的输出，但其实有些写得差的人类文章里也会出现。

That's something that AI uses a lot, and it had said that five times more, which statistically tells that cluster, in the LLL, that there you know, this is an output from an LLM model, but it also appears in weak human writing.

所以这真的很难，因为他们评判的标准就是这种所谓的‘弱人类写作’。

So it's really hard because and quote unquote weak human writing, this is how they grade it anyways.

对吧？

Right?

因为现在AI检测太难了，有人可能就是随便说一句‘as if it had’，

Because it's these this AI detection is so hard, and someone could write something that technically like, someone could just, you know, say as if it had.

说得多了，

They could say that a lot.

或者就是特别喜欢用‘stubborn’这个词，动不动就往里塞。

Or they could just love the word stubborn and put it in all the time.

结果AI检测一看，哎，有情况。

And, you know, this AI detection is like, oh, look.

从统计上看，这人写东西跟普通人不一样。

Statistically speaking, they're not writing like normal people.

但也许这根本就不是AI写的。

Well, maybe it's not an AI model.

也许这就是那个人的写作风格。

Maybe that's just how that person writes.

所以要分辨出来还真不容易。

So it's kind of hard to to pull that.

但Pangram确实扫描了所有英联邦文学奖的得主。

But Pangram actually scanned all of the Commonwealth Prize winners.

他们发现了2026年两位获奖者和2025年的一位获奖者。

They found two of the twenty twenty six awardees and the twenty twenty five winner.

他们说，所有这些人都明显使用了AI生成的内容，而基金会的负责人也确认了这一点——但所有提交作品的作家都声称自己没用过AI。

They say all of them appeared to use AI generated content, and the the director of the foundation confirmed that, any writers that were submitting said that they didn't use AI.

他们根本没去核实过这些说法。

There was no way they verified this.

在这种背景下，诺贝尔奖得主奥尔加·托卡丘克也公开承认，她确实使用AI，还提到自己会用类似‘亲爱的，我们怎么才能把这段写得更优美、更有深度呢？’这样的提示语。

With all of this going on, Olga Tokatskuk, who is a Nobel Nobel laureate, she also admitted, she posted online, said that she uses AI, she uses prompts like, she says, quote, Darling, how could we beautifully elaborate this for developing ideas?

这番话引发了大量批评，她不得不发布了一份长达三点的公开说明，解释自己的创作过程，说实话，我觉得这事儿真挺让人无语的。

That apparently triggered a ton of backlash, She and had to make this big huge three point public clarification about her creative process, which I just think it's so I don't know.

这真有意思。

It's so funny.

没人希望看到AI随便生成的东西被当成原创，但与此同时，AI在创作过程中其实有太多好用的地方。

Like, no one, I guess, wants to have one shot AI slop being written for stuff, but at the same time, there's so many great ways to use AI to help in your creative process.

所以有时候，这事儿确实有点荒唐。

So I I think it's, it can be a little bit ridiculous sometimes.

如果你想每月花8.99美元就能使用80多种AI模型，不妨去aibox.ai看看。

If you wanna get access to over 80 different AI models for $8.99 a month, Go check out aibox.ai.

这是我自己的创业项目，你可以在一个聊天窗口里直接使用所有主流的文本、图像、音频和视频模型。

That's my own startup, and I give you access to all of the top text, image, audio, and video models in one place, in one chat thread.

所有工具你都能用上。

You're gonna get access to everything.

我做这个就是为了帮大家省钱，同时也把所有登录入口都集中到一个地方。

I built this so that it could save people money, but also so you can consolidate all of your logins into one place.

想了解的话，简介里有aibox.ai的链接。

If you wanna check that out, there's a link in the description to aibox.ai.

另外也去逛逛 aichatdaily.com。

And also go check out aichatdaily.com.

那是我的网站，和这个播客是配套的。

That's my website that is kind of partnered or or, like, aligned with this podcast.

它会针对我每期节目的内容写深度文章，还会每天发一封邮件，把所有内容直接送到你的邮箱里。

It generates deep dive articles on everything I cover in the show, and it also has a newsletter where I'll send you all of this into your inbox every day.

你可以去看看。

So you can go check that out.

描述里也有链接。

There's also a link in the description.

非常感谢大家收听，我们下期再见。

Thanks so much everyone for tuning in, and I'll catch you in the next episode.