Tokenmaxxing, OpenAI's shopping spree, and the AI Anxiety Gap

大家好，欢迎回到《Equity》，这是TechCrunch关于初创企业商业动态的播客。

Hello, and welcome back to Equity, TechCrunch's podcast about the business of startups.

今天是4月17日，星期五。

Today is Friday, April 17.

我是TechCrunch的交通板块编辑柯尔斯滕·科罗塞克，今天照例和资深记者肖恩·奥凯恩，还有我们的周末编辑安东尼·哈一起主持。

I'm Kirsten Korosec, transportation editor over here at TechCrunch, and I'm joined as always by senior reporter Sean O'Kane and Anthony Ha, our weekend editor.

我们有个关于Allbirds事件的重要更新。

And we have an important update about the Allbirds saga.

安东尼？

Anthony?

很重要。

Important.

没错。

That's

对。

right.

确实重要。

Important.

非常重要。

Very important.

几周前，我们聊过Allbirds以3900万美元卖掉了它的鞋类业务，这数字可不低。

I think a few weeks ago, we talked about how Allbirds had sold its shoe business for $39,000,000, which, you know, is not nothing.

但对一家曾经那么火爆的公司和初创企业来说，这个价格实在算不上亮眼。

But for a business and a startup that was once that hot is a is a not a very impressive sum.

不过，Allbirds这家公司——不是卖鞋的那部分——还没彻底完蛋。

But Allbirds, company, so not the shoe business, is not dead yet.

它现在转向了人工智能领域。

Instead, it is pivoting to AI.

公司已经重新命名为Newbird AI，打算转型成一家计算基础设施公司。

It's rebranded as Newbird AI, and is basically trying to become a computing infrastructure business.

我觉得这未必是2026年最重要的新闻，但绝对是2026年最具代表性的故事。

So I'm gonna say that I don't think this is the most important story of 2026, but I do think it's actually the most emblematic story of 2026.

我实在想不出还有比这更贴切地体现我们这档播客一直在讨论的内容了。

I cannot imagine anything that more perfectly encapsulates what we've been talking about on this podcast.

问题是，这就到头了吗？

I mean, the question here is, like, is this is this it?

这就到顶了吗？

Is this the top?

你们觉得呢？

Like, what do you guys think?

这做法吧，说白了就是特别功利，特别明显。

Like, this is, you know, a pretty craven and pretty obvious.

那你们怎么看？

So what do we think?

还没到顶呢。

It's not the top.

肯定还没。

Nope.

当然不是。

Nope.

绝对不是顶峰。

It's 100% not the top.

我预测到夏天的时候，会出现更荒唐的事，咱们先回溯一下过去。

There is I predict by summertime going to be something more ridiculous because let's go back in time.

你还记得2021年吗？那时候疫情还没结束，整个 meme 股票热潮闹得沸沸扬扬，到处都是离谱到极点的事。

Do you remember that back in, like, 2021, this is code pandemic and, like, the whole meme stock craze, and there was just some, like, insanely ridiculous things that were happening at the time.

而这次因为有了人工智能，情况还要夸张得多。

And this is even bigger and grander because of AI.

那时候不过是人们发疯，想找点东西赌一把。

I mean, that was just, like, people getting crazy and, I don't know, wanting something to bet on.

我们当时看到了金融市场上一种全新的有趣现象。

And we saw this whole new interesting thing occurring in the in the financial markets.

这次完全是靠人工智能的魔法粉在撑着。

This is based on AI fairy dust.

所以我完全预料到还会出现更多这样的情况。

So I 100% expect more.

至少在极短期来看，这招是管用的。

And we should say it it you know, at least in the extremely short term, it worked.

他们的股价原本几乎一文不值，现在却涨到了原来的好几倍。

Like, their stock, which was basically nothing, is now several multiples of nothing.

对啊。

Yeah.

从那么低的起点来看，涨幅确实非常惊人。

It's it's gone up pretty dramatically from a very low point.

不过我们这个故事带出了一点我完全忘了的事，那甚至不是在迷因股热潮期间，而是更早的时候。

But it did you know, one of the the nice things that our story about this brought up that I'd completely forgotten was that not even during the the meme stock craze, but before that.

早在2017年，Long Island Ice Tea这家公司……

But back in 2017, Long Island Ice Tea

没错。

Yes.

结果转去做区块链了。

Like, pivoted to the blockchain.

我甚至说不清他们这么做到底意味着什么。

I couldn't even tell you exactly what that meant for them to do that.

而且，那段时间确实短暂地掀起了一阵热潮。

And, again, it seemed like for a very brief period, there's sort of excitement around that.

后来他们就被退市了。

And then they're, you know, that they were delisted.

我可不是说这事儿完全没戏。

And I'm not gonna say that this won't work at all.

但你知道吗，他们当初想筹五千万美元，在这种资本密集的行业里，这点钱根本不够看。

But I mean, first of think they were trying to raise like $50,000,000, which in an in crazy capital intensive business is not a lot of money.

是的。

Yeah.

我觉得再过一两年，大概没人会再提Newbirds AI这事儿了。

I I don't know that we're gonna be talking about, you know, Newbirds AI a year or two from now.

我的意思是，我们能有这种运气就不错了。

I mean, we should be so lucky.

说到这点，咱们就到这儿吧。

To, you know, to your point, and then we can move on.

这家公司现在不管以什么形式存在，本质上都是在借五千万美元。

Like, this company in whatever form it exists now is basically borrowing $50,000,000.

我不清楚这能让他们买多少张显卡，更别说在别人动辄百亿、千亿资金、还有深厚合作关系的情况下，他们怎么挤得进前排。

I don't know how many GPUs that's gonna get them and how they're gonna be able to get to the front of the line when everybody else has, you know, a $100,000,000,000 or more and relationships with these companies.

我们根本不知道那个投资人是谁，他们一开始就没公开。

We don't know you know, they didn't initially disclose who this investor is, you know.

这里头太多东西都透着一股子当年长岛冰茶骗局的味道，还有之前NFT和Web3热潮时那些老把戏，整个加密圈都这样——上市公司利用自己在公开市场上市的身份，随便发个公告就能炒热度。

There's just so much here that reeks of not just the Long Island iced tea thing, but a lot of the stuff we saw during sort of the NFT and like Web three craze and and crypto in general too, where public companies are taking advantage of the fact that they're listed on public markets.

他们随便发个声明，就能制造一波炒作。

They can make announcements like this and and generate some hype.

我觉得这事闹到这地步，连干这事的人都没想到会这么火。

I think this probably went further than even the people who are doing this expected it to go.

说实话，关键问题是，他们真会花力气去搞什么新型云服务吗？

You know, the real question is, like, are they gonna actually put some effort into trying to become some sort of neo cloud?

还是说，他们就只想趁股价上涨赶紧套现，多发点股票，等热度一过，又回头卖羊毛袜子之类的东西？

Or are they just gonna, you know, ride the stock price and and issue more shares to snatch money while it's on the table before, you know, pivoting back to wool sneakers or whatever.

我希望不管他们做什么，至少能出点酷炫周边，让我能收进我的‘被遗忘科技’收藏馆，十年后拿出来看看。

Well, I do hope that whatever they do, they put out some cool swag so that I can pick it up and put it in my archive of tech that we forgot, and I get to open it up ten years

十年后。

from now.

你的Quibi袜子？

Your Quibi socks?

对。

Yes.

没错。

Exactly.

咱们还是聊聊点靠谱的交易吧。

But let's talk about some deals steeped in reality.

今天我们来聊聊Wayve和Fluidstack。

And so we're gonna talk about Wave today and Fluidstack.

我们还要更新一下OpenAI和Anthropic的最新动态，这方面事情不少，顺便深入聊聊现在很火的‘tokenmaxxing’现象。

We're also gonna do an update on everything that's happening with OpenAI and Anthropic, which there is a lot, and get into the phenomenon now called tokenmaxxing.

先从Wayve说起，这家公司我其实花了不少时间了解。

Let's start off with Wayve, a company I've actually spent a lot of time in.

我和肖恩还亲自去体验过他们的车。

And Sean and I, we actually went into their vehicle.

那是什么时候的事？

When was that?

去年的Disrupt大会期间。

During Disrupt last year.

对。

Yeah.

没错。

Yeah.

去年我们坐了一辆配备Wave技术的福特Mustang Mach E，仅靠几个摄像头就能运行，效果相当惊艳。

We got a ride in a in a Mustang Mach E outfitted with Wavestack last year, and it was pretty impressive for something that was operating on only on just a few cameras.

不过，那条路线看起来是提前规划好的，就在我们参加Disrupt活动的周边区域。

Although it was on, you know, what seemed like a a fairly planned route around where we had Disrupt.

这周有个新闻，AMD、Arm和高通都纷纷入局，砸的钱多到足以另起一个云平台了。

The news this week is chipmakers, AMD, Arm, and Qualcomm all sort of jumped in with with, you know, enough money to start a Neo Cloud if you want.

这笔六千万美元的投资，算是对NVIDIA之前联合微软、优步、软银愿景基金和几家车企共同完成的那笔十亿美元D轮融资的延伸吧。

$60,000,000 sort of pseudo extension, I guess, of the more than a billion dollars series d that NVIDIA held back along with Microsoft, Uber, SoftBank Vision Fund, and a couple other automakers a little while ago.

我有点好奇，Wayve现在融的钱已经好几倍于十亿美元了。

And, you know, my question here is, like, you know, the Wayve has now raised multiples of billion billions of dollars.

但很明显，这些资金本身并不直接决定他们能做出什么成果。

Clearly, the money here isn't necessarily something that means too much to what they're able to do.

不过，优步或NVIDIA投资Wayve这样的公司，倒是可以理解——他们想打造一个能适配几乎所有车型的自动驾驶系统。

But, you know you know, it makes sense for Uber or Nvidia to invest in a company like Wayve that's trying to make, like, a self driving system that can go into almost any vehicle.

但我真不太明白，芯片厂商为什么也要掺和进来。

But I'm I don't really feel like I have a good sense of why chip makers would be getting into this game.

你不是跟Wayve的创始人亚历克斯聊过吗？

Did you I know you talked to Alex, the founder of Wayve.

凯瑟琳，你从他那儿感受到这件事有多重要了吗？

Did you get a sense from him, Kirsten, like, what the sort of import of this is?

是的。

Yeah.

这些年我跟亚历克斯聊过好几次。

I've talked to Alex a few times over the years.

关于Wayve的技术和商业模式，关键的一点是，亚历克斯·肯德尔——

And the important thing to know about Wayve's tech is and their business model is the way Alex puts it, is Alex Kendall.

他是联合创始人兼CEO，他认为他们采取了一种与众不同的视角来看待自己的技术和商业模式。

He's cofounder and CEO, is that they're taking a contrarian look at their business model and their tech stack.

他们做的是端到端的系统，基于神经网络搭建了软件层。

So end to end, they built a software layer off of neural network.

你可以这么理解。

So think of it this way.

这辆车能兼容任何现代车辆上的传感器和芯片。

The vehicle takes any kind of sensor that's on a modern vehicle and any kind of chip.

所以它在硬件上完全不挑，只要拿到数据，就能教会车辆自动驾驶。

So it's tech agnostic when it comes to the hardware and then takes that data and can teach the vehicle to drive.

当然，这套系统不是随便就能装进任何一辆车的。

And this is obviously not just, like, plopped into any car.

实际上，这背后有一整套授权协议在运作。

I mean, there's a whole licensing deal that happens.

他们已经有一个稳固的客户——日产，而且最近合作还进一步扩大了。

And they do have a solid customer with Nissan that seems to have just been expanded.

在我看来，芯片合作的关键在于跟现有车载芯片厂商深度绑定，协助他们优化开发，确保系统兼容，同时推动规模化。

The chip deal to me is really more about locking in with all of the different chip players that are already in vehicles and helping in that development to ensure that everything works and also to help scale.

这些芯片厂商一看Wayve是势头正猛的新星，正往车里渗透，自然都想分一杯羹。

So all of these chip partners, obviously, if they see Wayve as an up and comer that is getting into these vehicles, they want a piece of that.

他们都想参与进这个发展过程。

They wanna be part of that development.

Wayve 也能直接从这些芯片厂商的合作中获益。

And then Wayve benefits from working directly with these chipmakers.

我顺便问了他们一句，嘿。

I did ask them, by the way, hey.

这种合作是实物抵资类型的投入，还是真金白银的投资？

Is this in kind trade type of investment, or is this real capital?

是实实在在的股权投入。

And it's real equity.

你可以看出他们在下长远的棋，但这一切的核心还是研发和后续的规模化能力。

So you can see that they're playing the long game, but this is all about development and then the ability to scale.

我稍后还想再聊回 Wayve，不过我也想问问 Sean 提到的一件事——我觉得在那笔十亿美元以上的融资里，优步也是投资人之一。

I did wanna get back to Wayve in a second, but I also wanted to ask about something that Sean mentioned, is that I think not in this extension, but in that giant billion dollar plus round, Uber was also an investor.

我记得在之前的几期节目里，我们讨论过，说到自动驾驶和自动驾驶出租车，优步基本上就是到处投资、在每个城市都找不同的合作伙伴。

And I know that on some past episodes, we've talked about the kind of sense that when it comes to robo taxis and self driving, Uber is basically just, like, investing in everyone and partnering with a different partner in every single city.

这算是他们一贯策略的一部分吗？

I mean, is this part of that same strategy?

是的，我想这正是回答这个问题的方式。

Yes and, I guess, is the way to answer that.

也就是说，优步其实已经下了注。

Like, yes in the sense that Uber's basically placed bets.

这周《金融时报》有篇文章总结了一下，算出来优步已经以各种方式向这些自动驾驶公司投入了大约100亿美元。

I mean, the Financial Times had a story this week that has kind of rounded it all up and and sort of figured out that it's about $10,000,000,000 that Uber has now committed in some way to these other companies that are working on autonomous vehicle tech.

所以很大程度上，我觉得确实如此。

So it is in large part, I think it's that.

但同时，优步似乎也在和Wayve紧密合作，打算真正推出一个上路的自动驾驶出租车服务，而对其他公司，他们可能没这么明确。

But also, seems to be seems ready to be working closely with Wayve to put, you know, put a real robo taxi service on the street in a way that maybe Uber isn't necessarily sort of being so explicit about with some other companies.

我觉得优步下的注是有不同层次的。

I think there's gradations to the bets that Uber is placing.

有些比较试探性，他们只是想保持这种关系。

Some are a bit more speculative and they just wanna have that relationship.

而有些则进展很快，甚至已经在部署了，就像Waymo那样——在好几个城市，Waymo的车已经能通过优步App叫到了。

And some where they're actually moving pretty quickly towards or already deploying like we see with Waymo where where Waymos are available on the Uber app in a number of cities already?

是的。

Yeah.

优步和这些公司的合作关系确实有轻重之分。

The Uber relationships are definitely on a sliding scale.

我一向关注具体数字，比如至少订购两万辆车，而不是最多多少辆。

So I always like to look at word wording like at least order 20,000 vehicles and not up to.

这两者差别可大了。

There's a big difference.

优步和Wayve的合作看起来非常扎实。

The case with Wayve is is is seems pretty rock solid.

优步正在和Wayve合作，打算把他们的技术装进车里，投入伦敦的优步网络。

I mean, this is Uber working with Wayve, wanting to have their tech in a vehicle that it goes on the Uber network in London.

优步还计划再向Wayve投资三亿美元，但这是按里程碑支付的。

And this is Uber is planning on investing another $300,000,000 into Wayve, but it's milestone based.

据我了解，一开始并不需要完全无人驾驶，只要车上配备安全员就能触发这笔三亿美元的付款。

So once they're able to launch and my understanding is that it doesn't have to launch driverlessly right away, so there may be a human safety operator behind the wheel that that will unlock that $300,000,000 milestone.

而且Wayve也在和日产之外的其他汽车制造商接洽合作。

And then Wayve is also talking to and working with other automakers beyond Nissan.

我们得提一下，Stellantis和奔驰都是他们的投资人。

We should mention that Stellantis and Mercedes are both investors.

很明显，他们正在紧密合作。

They clearly are working together.

不过目前还没有明确的时间表，说明这项技术什么时候会用到他们的车型上。

There's no timeline, though, on, like, when that tech might go into their vehicle.

所以这还谈不上是正式的客户官宣。

So it's not, like, quote, unquote, an official customer announcement yet.

所以现在情况非常活跃，我觉得Uber和Wayve的合作比他们以往的其他合作要认真得多。

So there's absolutely a lot going on, and I would put the Uber Wave relationship as far more serious one than some of the other ones that they've done.

而且我觉得Uber是在下长远的棋。

And also, I think Uber is just playing the long game.

从Wayve的成功标准来看，他们似乎正在和很多不同的汽车制造商建立合作关系。

In terms of, like, what success looks like for Wayve, sounds like they're partnering with a lot of different automakers.

也就是说，几年后，很多人坐的车可能已经用上了Wayve的技术，或者至少用上了它的自动驾驶部分，但他们自己根本察觉不到？

Is this something where, you know, in a few years, potentially, a lot of people could be in vehicles that have are powered by wave technology or at least, you know, the kind of self driving parts of it are and not even know that it's there?

是的。

Yeah.

我让Sean来多说一点，不过他们确实有两个产品。

I'll let Sean answer and more and more more beyond the But, yeah, they have two products.

Sean，你怎么看？

Sean, what do you think?

对。

Yeah.

不对。

No.

我的意思就是，差不多就这些了。

I mean, that's pretty much it.

核心想法是，这项技术可以用于你叫的自动驾驶出租车，也可以用在你未来要买的日产汽车，或者其他任何没点名的汽车品牌上——当然，不是特斯拉。

The the idea is that this could be something that's powering a robotaxi that you're hailing or this could be something that's powering, you know, a Nissan vehicle or insert other automaker here, not named Tesla, you know, that you purchase, you know, in the near future.

还有，我觉得Wayve有一点特别有意思，而且他们签的合作越多、融到的资金越多，这一点就越值得聊——就是之前柯尔斯滕提到过的那个点。

And, you know, one thing that I think is also really interesting about Wayve and and in some ways gets more interesting the more of these deals that they strike and the more money that they raise is, you know, Kirsten mentioned it before.

他们打算先在伦敦落地，因为这是家英国初创公司。现在大部分自动驾驶领域的资源和关注点，几乎都没放在这类地区身上。

They're looking to launch in London because they're a UK startup, like and I think that's just for so much of the focus on autonomous vehicles.

其实就是硅谷或者中国的公司在做这件事，说白了就是这两个地方。

We're talking about, you know, Silicon Valley or China, really.

而Wayve从一开始就走了一条与众不同的路线，他们研发的这套技术是端到端的，而且不挑硬件。放在几年前，尤其是2010年代末自动驾驶第一波风口的尾巴那段时间，所有人都觉得这个想法太冒险了，成的概率很低。

And, you know, Wayve was by nature of taking this kind of contrarian approach and going at this, like, sort of end to end hardware agnostic tech a couple years ago that seemed at the time, especially during the, you know, the late hype cycle of the 20 tens of that initial sort of AV hype cycle, it seemed like a real long bet.

这项长期投资让他们成为了欧洲乃至整个欧洲范围内最有价值的初创公司之一。

And that long bet has turned them into, you know, one of the more valuable startups sort of across Europe in general.

你知道的，他们还是一家人工智能公司。

You know what also they're an AI company.

但你知道还有哪家是人工智能公司，或者为人工智能提供动力的吗？

But you know what also is an AI company or powers AI?

数据中心。

Data centers.

所以这算是我勉强转到下一个话题了，我想聊的是Fluidstack这家公司，一家AI数据中心初创企业。

So that was my, like, terrible segue into the next deal that I really wanna talk about, which is Fluidstack, and this is an AI data center startup.

我不确定你们俩有没有关注过这家公司。

I don't know if either of you have been tracking this company.

其实我之前对Fluidstack了解得也不多。

I didn't know a lot about about Fluidstack, actually, prior to this.

他们正在洽谈一笔10亿美元的融资，公司估值高达180亿美元。

They're in talks for a billion dollar round at a $18,000,000,000 valuation.

我当时就惊了。

So I was like, wow.

这么高的估值，我确实该好好了解一下这家公司了。

I should probably know this company a little bit better with that type of valuation.

现在这类数据中心的建设消息铺天盖地，速度和规模都太快了，想搞清楚每家公司到底扮演什么角色，真的挺难的，尤其是当所有这些承诺都还在进行中的时候。

It's hard with the sort of speed and scale of the announcements that are happening around how many of these kinds of data centers are being built out to really, I think, sometimes understand all the different roles that different companies will play in that progression should this all happen as it's being, you know, sort of promised.

而且，确实有理由对这些说法持保留态度。

And I think there are good reasons to question that.

不只是因为现在美国各地对建设数据中心普遍有顾虑，尤其是那些项目周边的居民，还因为一些大公司原本承诺要投入，后来又以各种方式撤回或调整计划。

Not just because of their the sort of hesitance around data centers in The US in general right now, and from people who live in the areas where they're being built, but also because we keep seeing some of the plans shift from some of the bigger players who are making commitments to these and backing off them in different kind of interesting ways.

这么说吧，作为一个观察者，我觉得Fluidstack挺有意思的，它似乎想在这些早已家喻户晓的大玩家之间找到自己的位置。

All that is to say that, you know, I find, you know, a company like Fluidstack sort of interesting as someone who's kinda trying to stand in the middle to a certain extent of all those other larger players that are household names by now.

我有点好奇的是，他们和其他类似公司，尤其是在建设甚至设计环节的公司，持续以这种规模融资，到底能从这些巨额数字里真正拿到多少实质性的支持。

One thing that I'm I am kinda curious about is they continue to raise money at this scale and and other companies in this area, specifically around the construction side of this and and maybe design a little bit too, is, you know, how much glean they get from the size of the number.

我想说的是，这让我联想到汽车初创公司领域，那里很容易被这些数字搞晕——过去几年我们已经看得够清楚了，如果你不真正搞清楚做这件事到底需要多少底牌资金，光看表面数字是没用的。

And I'd say that in the context of someone who writes a lot about automotive startups where it's easy to kinda get lost, and I think we've seen this in a pretty hard way over the last couple years, In those big numbers, if you're not really thinking about how much the sort of table stakes really are for how much money you need to do this.

对吧？

Right?

比如Lucid Motors，几年前通过SPAC上市，单这一笔就融了40亿美元。

So like a good example is, like, Lucid Motors went public in a SPAC merger a couple years ago and raised $4,000,000,000 in that transaction alone.

可到现在，他们还是在艰难地调整，想真正站稳脚跟、实现可持续发展。

And, like, they're still struggling to kinda, like, get things right and get really get off the ground in a sustainable way.

所以我在想，当你看到这些公司动不动就融十亿美元、估值一万八千亿的时候，是不是也觉得有点不对劲？

And I'm curious, like, you know, if that's something that's crossed your as minds too as you look at some of these where it's like, wow, billion dollars, you know, $18,000,000,000 valuation.

但问题是，这些资金到底有多少是跟公司未来的前景挂钩，又有多大程度只是因为做这些事本身就需要巨额投入？

But like, you know, how much is that really tied to future prospects for this company versus they just need a lot of money to do this stuff?

这可能会扭曲他们的判断视角。

And like, that can kinda skew this their perspective.

再补充一个数字，来说明一下这里的规模：他们还和Anthropic签了一份500亿美元的合同，专门为对方建设数据中心。

To throw one other number in there just to kinda, like, for some of the scale involved here is that I think they also have a $50,000,000,000 contractor agreement with Anthropic to build data centers for them.

是的，没错。

And so it yeah.

就像我们在这档播客里常说的，十亿美元在大多数情况下已经是一笔巨款了，但在AI和AI基础设施领域，这其实真不算多。

It's this thing where, again, as we often say on this podcast, it's like billion dollars is like a lot of money in most contexts, but actually in the context of AI and AI infrastructure, it it it's actually not that much.

而且你说的另一点也很到位——每当这些交易公布时，标出来的金额都大得惊人。

And, like, to I think your other point is, like, really well made too of like, when these deals get announced, like, the the the price tags are just huge.

比如，我们打算花这么多钱建这么多数据中心，然后就突然安静下来了。

Like, we're gonna build so many data centers for this much money, and then kinda goes quiet for a while.

当然，建设确实在进行，但说实话，我们并不清楚他们到底进展到哪一步了，最终能建成多少。

And I mean, the building there is building happening, but I think it it's not as clear as I think probably we would like in terms of, like, where they are they're actually at and how much is actually gonna get built.

安东尼，我感觉你对这些基础设施项目是否真能落地有些怀疑，是吗？

Do I sense some skepticism, Anthony, as to whether these these these infrastructure projects are actually going to happen?

没错。

Yes.

我不是说一样东西都建不起来，但我觉得可以很有把握地说，最终实际建成的规模，肯定介于官宣数字和零之间。

I mean, I don't think none of nothing's gonna get built, but I I would be similar I I think it's a very safe bet that the number will be somewhere between the numbers that have been announced and zero.

对。

Yes.

你提到Anthropic的交易正好，因为我们接下来就要深入聊聊OpenAI和Anthropic的那些事。

Well, I'm glad you brought up the Anthropic deal because we are gonna get into all the things OpenAI and Anthropic here next.

不过我想再补充一点关于Fluidstack的看法：在我看来，这正体现了AI初创公司越来越倾向于成为大厂的专属基础设施合作伙伴，而不是跟它们正面竞争。

But I do wanna make one final point about Fluidstack, which is, to me, I think that it's an example of AI startups increasingly becoming, like, sort of dedicated infrastructure plays for the big frontier labs as opposed to trying to compete with them.

既然赢不了，那就加入他们。

So if you can't beat them, join them.

所以我接下来会密切关注，预计还会看到更多类似的合作关系。

And so I'll be watching for more I would expect more of those types of relationships.

不过我们还是得聊聊Anthropic和OpenAI，因为上周发生的事实在太多了。

But we should talk about Anthropic and OpenAI because darn there's been a lot in the last week.

Anthony，你来开个头吧？

Anthony, you wanna kick things off?

对。

Yeah.

咱们先说说这些交易吧，毕竟这属于股权领域的动态。

Let's let's do deal let's do I mean, we're we're this is equity.

我们会提几个交易，然后可以再往后退一步，看看它们更大的意义。

We'll we'll mention a couple of deals, and then I think we can kinda zoom out to the to the bigger significance.

有两笔交易特别值得提，一个是OpenAI收购了一家叫Hero的个人理财初创公司。

But but two deals that are worth mentioning, one is that OpenAI acquired this personal finance startup called Hero.

是Hero，拼写是H-I-E-R-O，不是H-E-R-O。

That's Hero with an I instead of an e.

这事儿紧跟着另一笔交易，我记得那笔交易刚好是在我们录上一期《Equity》的时候公布的。

And and that kinda comes after another deal that I think was literally announced when we were recording our last episode of Equity.

所以我们没来得及聊这个，就是OpenAI还收购了TBPN，这是一家做商业访谈类节目的新媒体公司。

So we didn't get to talk about it, which is that OpenAI had also acquired TBPN, which is this essentially like a business talk show, like a new media company.

我觉得这两笔交易，跟OpenAI的体量比起来，都算很小的。

And I think both of these deals are pretty small when you, you know, by the scale compared to the scale of OpenAI.

这些收购不太可能对他们的业务方向产生什么重大影响。

These are not things I think that people expect to really change the course of their business or anything like that.

但它们挺有意思，因为这暗示着OpenAI还在尝试各种新东西，比如TBPN这笔收购。

But they're interesting because it sort of suggests that there's still this kinda like, alright.

尤其是现在这个时间点，我们看到的报道都说OpenAI正全力聚焦在让ChatGPT和GPT模型在企业市场、程序员群体中更具竞争力，那为什么还要做这种事？

Let's, like, try out different things, especially the TBPN deal of, like, few like, just, like, why would you do this particularly at this time when it feels like OpenAI from all the reporting we're reading is also trying to, like, really refocus on making, you know, ChatGPT and its GPT models really competitive in enterprise context with programmers and things like that.

搞一个科技访谈节目，真该列在他们的待办事项里吗？

Like, is, like, running, like, a tech talk show really should that really be on the to do list?

不该。

No.

根本不该列在待办事项里。

It should not be on the to do list.

就这样了。

That's it.

我想提一下Hero，因为在我看来，这挺有意思的。

I I do wanna mention Hero because that to me okay.

我知道肖恩对TBPN有不少看法，毕竟里面有个挺精辟的评论。

I know Sean has stuff to say about TVPN for sure because there's a there's a pithy comment in there somewhere.

但Hero就挺有意思了，我们的风险投资编辑朱莉·博特非常出色，她最早写了这篇报道。

But with the with Hero, that's an interesting one because Julie Bort, our venture editor, super talented, she she wrote about this and was, I think, the first to write about it.

她深入挖了一下。

And she dug in a little bit.

看起来这明显是个收购团队的交易。

And, basically, this looks like an acquihire.

这家公司就要解散了。

The company is is folding.

他们明确说了，到某个日期，你就再也无法访问这个服务了。

I mean, they they basically said, by this date, you won't be able to access this anymore.

这是一家个人理财初创公司，才刚成立两年。

This is a personal finance startup, and they only launched two years ago.

所以这明显是为了吸引人才。

So this absolutely is about getting talent on board.

我很想知道，OpenAI 是打算直接把他们整个团队收编进来，还是真的对个人理财产品感兴趣，想自己开发？

So I'm very curious to see if OpenAI is going to be is full just just absorbing them into the if you will, at OpenAI, or if they're actually, like, interested in some sort of personal finance product that they are they wanna work on.

这需要非常特定的技能。

So it's a very specific skill set.

对我来说，这一点还不太清楚。

It's, to me, it's not really clear.

我觉得这两个收购在某种程度上都是为了挖人。

I I think you'd look at both of these as acqui hires to a certain extent.

比如 TVPN 这笔收购，据说他们还会保留自己每日节目的编辑自主权。

I mean, like, the TPPN acquisition is, you know, allegedly, they are going to retain their editorial independence on the show that they make every day.

当然要向那些人致敬，他们这么快就把项目做起来，还发展到了今天的规模。

And, you know, all respect to those guys who've who've put that, you know, out there and gotten it off the ground so quickly and grown it into what it has become.

我觉得任何关注媒体的人都该保持一点怀疑：当你收购了这样的团队，却把内容创作者放到公共政策、公关或市场这类上级部门下面，光说‘保持编辑独立性’真的够吗？这值得深究。

I think you any person who follows the media should have a healthy dose of skepticism that when you acquire something like that and you you put them, the people who make the show, under the org of sort of the public policy people and and, you know, sort of comms or marketing adjacent people higher up at the company making the acquisition that that you have you could have good questions about whether or not saying editorial independence is enough.

这不是一句咒语，念了就万事大吉。

It's not an incantation that just works.

不过有意思的是，虽然这两起收购都像是‘为挖人而收购’，但在我看来，它们恰恰反映了OpenAI正面临的两大难题。

But, you know, what's interesting to me about these two, while they are similar in their, like, aqua hire ness, I think they both represent, like, two major problems that OpenAI is facing.

一个是，OpenAI的ChatGPT确实很成功。

One is, Hero is, you know, OpenAI has a very successful product in ChatGPT.

但能不能靠它赚到足够的钱，让公司真正实现盈利，而不必不断创下全球最高额度的私人融资纪录来维持运转，这还是个大问号。

As far as whether or not that will actually ever make them enough money to become a sustainable business that's not raising the most you know, the largest private rounds in the world ever to keep things going is a big question.

而且，他们在企业端的进展似乎也跟不上，而真正的利润其实来自其他竞争对手。

And they, you know, have also seemed to be struggling to sort of keep up on the enterprise side of things where the real money seems to be from some of the other players.

所以，引进这样一个团队，在我看来，更像是在试探：我们还能做点什么？

And so bringing in a team like this, to me, seems like taking a shot at, like, what else can we do?

据说Hero的创始人一直有股连续创业的劲头，专做面向消费者的App。

You know, the the guy who founded Hero seems to have this kind of serial entrepreneur streak of creating consumer apps.

所以在我看来，这像是在赌他们能推出点别的东西，不只是个聊天机器人，而是更有吸引力、更值得多花钱的产品。

And so this seems to me like a bet on them being able to come up with something else that may have more hooks than just a chatbot and maybe something worth paying more for.

至于TBPN，这桩收购显然是为了更好地展现公司到底在做什么，改善公众形象——最近这形象可不太妙。

And then, you know, TBPN is like an acquisition made to help better represent what the company does and better shape its image in the public eye, which lately has not been great.

尤其是现在，外界对他们的质疑比几周前多了不少，因为罗南·法罗刚在《纽约客》发了一篇报道，时间点也太巧了，正好和OpenAI上周那几条公告前后脚出来。

And certainly, is under more questions now than just a few weeks ago because Ronan Farrow just led a report, you know, at The New Yorker that dropped suspiciously right around the time that this and a couple other announcements from OpenAI came out last week.

所以，我觉得OpenAI眼下正面临两个根本性的问题，而这两笔收购，明显都是冲着解决这两个问题去的。

And so, you know, there's just you know, I think those those are two big sort of existentially problems that OpenAI is trying to solve right now, and these acquisitions both seem like they're targeted at at trying to crack both of them.

你没提到的一点是，Anthropic其实一直在暗中虎视眈眈。

And so the thing that you didn't say is then there's Anthropic kind of looming not in the shadows.

他们现在在企业市场拿下了不少份额，势头相当猛。

I mean, they're very much taking up a lot of space here, but they're having a lot of success on the enterprise side of things.

所以感觉他们就是直接竞争对手，而且在很多方面，两家公司的风格也截然不同。

So it feels like these guys are competitors, and they also feel like very different companies in a lot of ways right now.

我觉得他们现在就是在正面硬刚。

I think they're directly competing with each other.

我的意思是，确实存在一种情况：如果人工智能作为一门产业、一项技术，真能像支持者们期待的那样大获成功，那么这两家公司完全有可能并列第一和第二，一方的成功并不必然意味着另一方会默默无闻。

I mean, it there's definitely a scenario where if, you know, AI as an industry, as a technology, as successful as, you know, the sort of its proponents hope for, that they could both be very successful companies, like they could just be the kind of one and two, and and the success of one does not necessarily mean that the other will just, you know, fade into obscurity.

但说到底，尽管这些都不是官方消息，但从各种报道来看，OpenAI 对 Anthropic 的崛起似乎比谁都更在意，也更焦虑。

But it does seem like I mean, more than anything else, and again, none of this is, you know, official, but but there's just been a lot of reporting around how it seems like OpenAI more than anyone's is kind of obsessed with and upset about Anthropic's rise.

不过，其实还有另一个角度，记者卢卡斯就提过。

Although, actually, you know, another kind of angle on that was also a reporter, Lucas.

他上周末写了一篇很棒的文章，讲的是 HumanX 大会，他采访了现场所有人，大家聊的全是 Claude Code。

He he did a great piece over the weekend about the HumanX conference and where he was talking to everyone there, and everyone was talking about Claude Code.

他们好像说，哦，ChatGPT 也不错，挺好的。

And they're sort of like, oh, yeah, like ChatGPT, like, that's that's fine too.

但真正让他们兴奋的，是 Claude Code。

But like, they were all about Claude Code.

我觉得这正是 OpenAI 最担心的地方。

And I think that is exactly what, you know, OpenAI is worried about.

因为说到底，生成式 AI 和大语言模型本可以有无数其他的发展方向。

Because, again, in theory, there could be many other opportunities for generative AI for LLMs.

但感觉最大的增长点、最赚钱的地方，也是他们未来可能找到可持续商业模式的方向，就是企业级和编程工具这类领域，虽然这两者并不完全一样。

But it feels like the big growth area, the area where the most money is, and where they could, you know, maybe at least see a path to having a sustainable business in the future is in these kind of, again, enterprise and coding tools, which are not, like, exactly the same thing.

当然，企业应用远不止编程这一块，但尤其在编程领域，我觉得他们担心得有道理——如果在这里输了，短期内想重回第一的位置会非常困难。

Mean, there's plenty of other enterprise use cases beyond coding, but, like, especially with coding, I feel like if they're I think they're worried and probably correctly that if they lose there, that it's gonna be very hard to kind of be the number one company at least in the near future.

是的。

Yeah.

我觉得现在大概有三个主要位置。

I see this as, like, kind of three slots right now.

HumanX 这个会议里，满是用这些工具来工作的专业人士。

Like, HumanX, it's full of people using this for their jobs.

所以这种情况很自然，对 Anthropic 来说也是个好兆头。

So it it it tracks that and and it bodes well for Anthropic.

对吧？

Right?

于是突然间，参加 HumanX 会议的人，反而把 ChatGPT 当成一个面向普通消费者的工具了。

And so then all of a sudden, it seems like the folks who are going to the HumanX conference are looking at ChatGPT as that consumer product.

这可能就是根据肖恩之前提到的，他们可能会基于OpenAI的技术背景，开发一些有趣的消费类应用。

That's the thing that maybe to Sean's point earlier, like, maybe they're gonna be building out, like, interesting consumer apps based on sort of the background of OpenAI.

这种可能性是存在的。

Like, you could kinda see that.

但还有一个领域是国防，某种程度上是企业市场的混合体。

But then there's this other bucket, which is defense, which is, like, a hybrid of enterprise in a way.

看起来他们俩都在争夺这个市场。

And it seems like they're both fighting over that.

没错。

Absolutely.

国防部的预算非常充裕，我觉得他们俩都在争着拿下这块业务。

Like, the the wallet's quite open with the Department of Defense, and I think they're both vying for that.

没错。

Absolutely.

对。

Right.

最近Anthropic的另一条新闻是发布了他们称之为Mythos的模型，声称这个模型强大到不能向公众开放。但这件事还有一个有趣的政府层面——就像我们在这档节目里多次讨论过的，Anthropic目前和五角大楼关系紧张，被对方列为供应链风险。

And part of the other Anthropic news recently was about the announcement of this I think it's pronounced Mythos, which they hyped up as, like, the model so powerful that we can't release it to the public, but there has sort of been this interesting kind of government angle to it where, as we've discussed quite a bit on this show, like, Anthropic is sort of not on great terms with the Pentagon right now, which has declared them a supply chain risk.

但显然，特朗普政府的其他部门对与Anthropic合作非常感兴趣。

But, apparently, there are other parts of the Trump administration that are very interested in working with Anthropic.

杰罗姆·鲍威尔上周专门召集了一次会议，跟几家大银行说：你们都该去试试Mythos，这可能是个非常重要的网络安全工具。

And so Jerome Powell apparently called a meeting last week and, like, told a bunch of, like, the big banks, like, you should all be trying out Mythos because, like, this could be, like, a really, really important cybersecurity tool.

是的。

Yeah.

这个时间点真有意思，对吧？

That timing is really interesting, isn't it?

我不是说Anthropic现在才提这个，纯粹是为了在政府面前争取主动权。

I'm not saying that Anthropic is only talking about this now because, you know, it's a a way for them to sort of get some leverage back over the government.

但确实巧的是，特朗普政府突然开始主动找上门，表示想评估一下这个模型。

But it certainly works out in that way where all of a sudden the the Trump administration is kind of knocking on Anthropic's door to a certain extent to say, like, we actually wanna evaluate this.

你得抛开所有背景因素来看。

You know, remove yourself from, like, all context.

但如果你真的开发出一个几乎能黑掉任何系统的模型，那确实应该让政府来评估一下，也许这类技术的开发、使用和访问权限都需要一些监管规范。

But, like, if you do really develop a model that can hack pretty much anything, like, yeah, actually, I would want the government to probably take an evaluation of that and and, maybe that is something that needs to have some regulation around how those things get developed and get handled and who gets access to them and and all these things.

所以某种程度上，他们愿意暂时放下分歧，认真看看这个技术、好好聊聊，这倒是个让人稍感欣慰的信号。

And so, like, in some ways, like, it's a little encouraging that they're willing to maybe set aside their petty differences for a moment to, like, take a look at this and talk it out.

但一想到之前的种种背景，那些问题又全涌上来了。

But, like, you know, all that context comes rushing back in.

我觉得现在真正发生的，根本不是这类对话。

I don't think that those are the kind of conversations that are actually happening right now.

我还想退一步看看，站在我们当前所处的这个背景下——我感觉自己完全陷在硅谷初创圈和旧金山AI实验室的泡沫里，离得太近了。

You know, I wanna also take kind of take a step back and, like, against this backdrop or what we are living, like, I feel, like, very steep and very, very much inside that AI bubble of, like, startup scene in Silicon Valley and the AI labs that are in San Francisco and, like, we're really close to it.

但外面还有很多人，并不在这圈子里。

But there is quite a lot of people outside of that.

斯坦福最近发布了一份报告，指出了AI圈内人和外界之间日益扩大的隔阂。

And, you know, Stanford did a recent report highlighting sort of this growing disconnect between AI insiders and everyone else.

所以我觉得，尽管我们今天主要聊的是初创公司，但也得承认这一点。

So I think it's important that we sort of acknowledge that as much as we you know, we're here to talk about startups.

确实有一大群人，不仅跟这个脱节，而且对人工智能充满疑虑，担心它会如何渗透进他们的生活，哪怕现在还没真正影响到。

There really is a whole big group, and population that not only are disconnected from it, but also quite suspicious of AI and how it is going to, you know, seep into their lives if it hasn't already.

这其实也跟肖恩之前提到的那篇《纽约客》关于萨姆·阿尔特曼的文章有关。

That also, like, ties into something that that Sean had mentioned earlier about the the New Yorker piece about Sam Altman.

感觉我们现在正处在一个对人工智能的焦虑情绪特别高涨的时刻。

And, like, it just feels like we're at this moment where a lot of anxiety about AI is is very, very high.

我觉得这种焦虑可能还会继续上升。

I mean, I think it definitely might continue increasing.

我不确定我们是不是已经到顶了。

I don't know that we're at the top yet.

但与此同时，这种情绪跟行业内人士谈论AI的方式显得格格不入。

But at the same time, it feels very disconnected from the way people in the industry talk about it.

我觉得其中一部分原因，当然就是AI圈的人在某种程度上，其实是在给自己产品造势，对吧？

And I think that, you know, some of that is is, of course, I mean, AI insiders are to a certain extent kind of just hyping their own product, right?

毕竟，你手里押了几十亿美元赌在这个产品上，当然会跟人说它能带来一个美好的未来。

Well, of course, the product that you have billions of dollars on the line trying to sell, you're going to tell me that that's going to lead to a good future.

但我认为这里还有一个脱节，因为你刚才说我们身处这样一个世界，而确实，在很多方面我们也是如此。

But I think there's also this disconnect because, you know, you were talking about us as sort of being in this world, and I and obviously, we are in a lot of ways.

但就生活被改变的程度而言——至少我们的职业生活——跟程序员比起来，变化要小得多。

But I think in terms of how much our lives have been transformed versus a or at least our professional lives have been transformed versus a programmer, like, I think is is much smaller.

你去问问任何一个写代码的人。

Like, whereas, like, you talk to anyone who codes.

对吧？

Right?

几乎每个写代码的人，他们现在每天的工作状态，跟几年前相比，简直天差地别。

And I and or almost anyone who codes and, like, just what their daily day to day looks like now is so sounds so radically different versus how it was a couple years ago.

但在其他绝大多数行业里，情况并不是这样。

And and I think that's not really true in in almost any other profession.

所以我觉得，这种脱节就在于，有些人根本觉得：不对。

And so there's this part of the disconnect, I think, is that there are these people who are like, no.

这才是未来。

Like, this is the future.

我现在就身处其中。

I'm living it right now.

而其他人则觉得可能吧。

And everyone else is like, maybe.

而且我觉得，任何一种真正主导话题的新技术都会这样，包括这些大规模的评估——这其实是个经济问题。

Well and I think that, like, we've seen this with any new and emerging tech that really kind of takes over the conversation and also with these mass evaluations, I mean, it's a financial piece.

它正在悄悄渗透进文化。

It's, like, creeping into culture.

连我们的娱乐内容里，也开始出现各种相关的梗和引用。

It's creeping into our entertainment in terms of, like, references to it.

同时，你也看到了人们对它的反弹。

Then you also are seeing, like, these backlashes towards it.

比如自动驾驶，人们就开始冲着机器人发火。

I mean, we've seen that with, you know, autonomous vehicles, people kind of, like, going after the robots.

但在AI这个情境下，出现的情况要严重得多——有些人明显心理状态不稳定，直接冲着开发AI的人去发泄。

In this in this scenario, what's happening with AI, there are far more serious situations that are occurring where, you know, people who are, you know, let's say, like, clearly not well are going after the people behind AI.

虽然这还谈不上是一种趋势，但我记得本月早些时候发生了针对萨姆·阿尔特曼家的袭击事件。

And while this certainly isn't, I would say, a trend yet I mean, we saw the attack that happened, on Santa Altman's house earlier this month.

说实话，我不太希望看到这种情况越来越多，但心里难免会担心。

You know, I I kinda wonder if we're going to see more of this, you know, hopefully not.

人们的情绪和不满似乎已经到了顶点，却不是通过抗议或理性表达来释放，而是直接爆发出来。

But people sort of suspicions and frustrations sort of maxing out like this as opposed to, you know, protesting about it or, you know, doing more rational pushback.

是的。

Yeah.

这其实挺难开口谈的，因为阿尔特曼本人就发过一篇博客，说他想尽量降低这种对立的语气。

It's it's like a difficult thing to talk about in some ways because I think, like and and, you know, Altman himself did a a blog post where he talked about sort of trying to take the rhetoric down a bit.

我觉得这完全说得通，毕竟我们不希望科技公司的高管们连家都不得安宁。

And I think that's completely reasonable because we don't wanna see, you know, like tech CEOs, homes being attacked.

但真正棘手的是，我们该怎么认真对待这些问题呢？

But it's the quite the challenging part is like, how can we take these issues seriously?

因为这些问题真的非常非常严重。

Because these are very, very serious issues.

这些决定着人工智能未来几年的发展方向，甚至可能影响我们未来几十年的生活方式。

These are gonna determine, you know, sort of the course that that AI takes over the next few years really could potentially determine what all of our lives look like for the next few decades.

认真对待这些问题，坦诚讨论，该批评公司时就批评，但同时也要避免出现这类极端事件。

And and to take that seriously, discuss it seriously, be willing to criticize the companies when when they should be criticized, but also not, you know, have a situation where those kinds of things are happening.

我觉得有一件事值得注意，尽管现在社会正在经历巨大变化，但人们确实在实际使用这些技术。

You know, one thing I think that is worth noting is that while this is happening and there's these big shifts in society, what what is very real is that people are using it.

对吧？

Right?

一年前或者两年前，我可能根本想不到，大家会花这么多时间跟BUD或者ChatGPT互动。

I mean, a year ago or two years ago, I don't know if I would have predicted the amount of time that people would spend, you know, interacting with BUD or ChatGPT.

当然，我们也看到了不少滥用的情况。

And certainly, we've seen, you know, misuses of it.

但正如你之前提到的，安东尼，企业在应用方面确实有大量实际用途，连我们说话的方式都在因此改变。

But to your point earlier, Anthony, about uses on the enterprise side, there is, like I mean, our language is even changing to talk about this.

所以，今天我们也想聊聊‘tokenmaxxing’这个话题。

And so, you know, tokenmaxxing is another thing we wanted to talk about today.

你说，这个词一年前根本就不存在吧？

You know, did that that term even exist a year ago?

我觉得确实没有。

I don't think it did.

首先我想说，那些给软件行业造词的人，或许该换种思路了，像什么‘氛围编程’、‘令牌最大化’这类词，我觉得不太妥当。

First of all, I just wanna say that whoever is coming up with these phrases for, you know, the software industry, maybe take a different tack because I just feel like things like vibe coding and tokenmaxxing are, to me, not

我倒还挺喜欢这些词的。

I kind of like them.

我也觉得这些词挺有意思。

I kind of like them.

我喜欢是因为它们精准地反映了当下发生的事。

Kind of like them because it it it it perfectly illustrates what is happening.

你跟别人一说‘令牌最大化’，对方立马就懂了。

I mean, is there anything more like, you could talk to someone and be like, what you know, tokenmaxxing.

一听就明白是怎么回事。

Like, I get it right away.

是的。

Yes.

我的意思是，这些词之所以能传开，就是因为好记，而一个梗最核心的就是得让人过耳不忘。

I mean, they're memorable for so for the and and, I mean, the the number one thing you need from a meme is it has memorable.

而且大家一听就明白你在说什么，这点很重要。

It's sort of clear what you're talking about, so that's important.

如果你觉得用Claude或者其他模型的人，其实就是在白白消耗脑力，越用越迷糊，那就算你管这叫‘tokenmaxxing’，也改变不了他们的行为。

I think that if you think that a lot of people, you know, going using Claude or whatever model are basically just sort of, like, cooking their brains and, like, you know, like like, just really spiraling, then calling it like tokenmaxxing is not going to dissuade you, let's say.

我们得说清楚，现在很多科技公司——不管是AI实验室，还是更广义上的所有科技公司，毕竟现在每个公司都自称是AI公司——都流行一种表演式做法：好像非得用掉最多的算力才显得有实力。

And and so we should say that, you know, what we're talking about is is essentially this idea that at a lot of tech companies, both kind of the AI labs, but just, you know, more broadly because, you know, every tech company is an AI company now, like, there's this almost like a performance of, like, you have to just use as much compute as possible.

你得用尽可能多的token，才算尽责。

You need to be using as many tokens as possible.

有报道说，Meta内部甚至搞了个排行榜，看谁用的token最多。

There were reports that Meta had like a leaderboard showing who was using the most tokens.

但后来发现，这种做法反而助长了不好的行为：有人干脆把问题写得又长又复杂，就为了多消耗点token。

And then also but then it turned out maybe that was incentivizing some like not great behavior, because then it was just like, well, what if we can just like, make this query as long and complicated as possible because that'll use more tokens?