OpenAI 与谷歌之争，英伟达如何花掉8500亿美元现金储备，以及马斯克的Grok计划 | 2025年11月24日

欢迎各位收看Information的TI电视节目。

Welcome everyone to the Information's TI TV.

我是阿尼塔·拉马斯瓦米。

My name is Anita Ramaswamy.

本周我将代替阿卡什·佩斯里卡担任主持人。

I'll be your host this week filling in for Akash Pesriqa.

今天是11月24日，星期一。

It is Monday, November 24.

我很兴奋为大家带来今天的节目。

I'm thrilled to bring you today's show.

首先，我的同事西奥发表了一篇关于埃隆·马斯克最近试图用Brock取代前员工的精彩文章。

First up, my colleague Theo published a great piece about Elon Musk's recent push to replace ex staff with Brock.

我们会为大家解释。

We'll explain.

我们还将深入探讨一个有争议的话题：持续采用AI对工作保障意味着什么。

And we'll dig into a divisive topic, which is what continued AI adoption could mean for job security.

接着，我们将测试谷歌的Gemini 3.0模型与ChatGPT的表现。

Then we'll put Google's Gemini three point zero model to the test against ChatGPT.

一位资深用户将做客节目，分享他们的使用体验，并谈论更广泛的聊天机器人竞争。

One power user will come on to share their experience and talk about the broader chatbot wars.

我们还将讨论一些科技公司的财报，深入探讨Zoom、Zscaler和Salesforce季度业绩的关键看点。

We've also got some tech earnings on deck, so we'll discuss what to know going into quarterly results for Zoom, Zscaler and Salesforce.

节目最后，我们将呈现来自《信息》联合执行主编马丁·皮尔斯的一篇精彩文章。

We'll end the show with a great piece from the information's co executive editor, Martin Pierce.

马丁和我将讨论英伟达快速增长的现金储备是如何被使用的，以及公司如何配置其资本。

Martin and I will be talking about how NVIDIA's fast growing cash pile is being used and how the company is deploying its capital.

这是一期内容丰富的节目，让我们直接进入正题。

It's a big show, so let's dive right in.

今天早上《信息》的独家报道揭示了埃隆·马斯克如何在X平台上进行裁员，以更大规模地用XAI的Croc取代员工。

Exclusive reporting from the information this morning reveals how Elon Musk is making job cuts at X in a larger push to replace employees with XAI's croc.

你可能还记得，XAI在今年三月收购了这个社交媒体平台，我的同事西奥·韦特写过一篇关于当前局势的文章。

You'll remember that XAI bought the social media platform back in March, and my colleague Theo Waite wrote a story about the state of play today.

他现在加入我们。

He joins me now.

Theo，欢迎来到TI TV。

Theo, welcome to TI TV.

谢谢邀请我参加。

Thanks for having me on.

很高兴你来到这里，Theo。

Awesome to have you here, Theo.

那么，在撰写这篇报道的过程中，你学到了什么？

So what did you learn in the course of your reporting this story?

最直接的消息是，在今年早些时候与XAI合并的X平台上，正在努力用Grok——即XAI的大型语言模型——取代大量工程师和其他剩余的员工。

So the most immediate news is that at at X, which was merged with XAI earlier this year, there's an effort underway to, replace a lot of engineers and and remaining human employees with Grok, which is XAI's LLM.

作为这一举措的一部分，上个月有一些负责信任与安全的工程师被裁员了。

And as part of that, you know, there there were were some layoffs last month of engineers working in trust and safety.

未来可能还会有更多裁员，这一切都属于埃隆推动的自动化X平台的总体目标。

It's likely there will probably be some more in the future, and they're all kind of under this this Elon mission of, automating x essentially as as much as possible.

对。

Right.

我想再多了解一下这方面的情况。

So I wanna hear a little more about that.

我的意思是，你觉得埃隆在收购后，希望Grok在X中扮演什么角色？

I mean, what do you think Elon's goal is in terms of how Grok is going to fit into x after the acquisition?

他希望它能尽可能多地运行。

He wants it to run as much as possible.

广告定向算法、许多后台工作，比如最近被裁的那些从事信任与安全工程的员工。

The the algorithm ad targeting, a lot of back end stuff, like, the the recent layoffs were of of people that worked in in engineering for trust and safety.

像检测垃圾信息、政府协调的影响活动之类的任务。

So stuff like detecting spam and and, you know, government coordinate coordinated influence campaigns, that kind of thing.

他基本上认为，做这些事的人越少越好，AI越多越好。

He he basically thinks that that you need as few people as possible doing that and as much AI as possible.

明白了。

Got it.

他雇佣的这对双胞胎兄弟如何融入这一更宏大的目标？

And how do these twin brothers that he's hired play into that broader goal?

是的。

Yeah.

今天故事的核心人物是这两位兄弟，迪马和叶夫根·西博列夫，他们基本上是一支由xAI派往X的突击队，目的是将Grok融入每一个角落，并推动埃隆的这项使命。

So the the central characters in in the story today are are these these two guys, Dima and Yevgen Sibolev, who are basically a they're they're twins, and they're basically a strike force that x AI kind of sent into x to put Grok into everything and to kind of spearhead Elon's mission here.

他们向埃隆汇报，尽管埃隆在名义上是X的最高工程师，但他们实际上是X的顶级工程师。

They're you know, they report to Elon, and they're they're kind of effectively the top engineers at X even though Elon is technically, you know, the top engineer on paper.

但他们确实如此。

But they yeah.

他们目前才是真正的老板。

They're they're the boss at the moment.

我们还知道关于他们的其他信息吗？

Do we know anything else about them?

比如，他们的背景是什么？

Like, what is their background?

关于他们的来历以及他们是如何来到X的，我们还知道些什么？

What else do we know about where they came from and how they they made it to X?

是的。

Yeah.

他们很有趣。

They're interesting.

他们是在乌克兰长大的。

So they they grew up in Ukraine.

他们俩都在乌克兰哈尔科夫的这所大学就读，同时学习数学，之后搬到了硅谷。

They both attended this university in in Kharkiv, Ukraine where they studied math at the same time together, and then they moved to Silicon Valley.

迪马在苹果公司工作了多年，去年短暂加入过OpenAI，然后才去了X。

Dima worked for years at Apple and then briefly joined OpenAI last year before he went to X.

叶夫根则在Meta工作了多年，之后短暂加入他哥哥所在的苹果公司，今年也来到了X或xAI。

And then Yevgen worked at Meta for many years and then joined his brother at Apple briefly, and then also came to X or x AI this year.

所以他们真的很有趣。

So they're they're really interesting.

他们俩合作得非常默契，据我了解，办公室里有些人干脆把他们称为‘双胞胎’，把他们看作一个整体，而不是两个独立的人。

They kind of work in tandem, and my understanding is, you know, there are people at the office that kind of just refer to them as the twins and, see them as, like, one unit rather than two two people.

说到两个单位协同运作，看起来X这个社交媒体平台和xAI之间确实存在一些相似之处。

Speaking of, two units or being acting in concert, I mean, it seems like there are some parallels almost between what's going on at X, the social media platform, and x AI.

你能再多谈谈我们看到的埃隆·马斯克在管理这两家公司时的相似做法吗？

Can you talk a little bit more about the similarities that we're seeing in terms of Elon Musk's approach and how he's choosing to run the two?

嗯。

Yeah.

从精神层面来看，xAI这个名为‘MacroHard’的项目（这是对微软的戏称）与之有某种相似之处。

So spiritually, there's kind of a a similarity between this this project at x AI called MacroHard, which is a a play on Microsoft.

嗯。

Yeah.

埃隆经常提到，MacroHard的使命是打造一家完全由人工智能驱动的软件公司。

You you know, Elon has kind of talked about a lot, MacroHard having this mission of, basically building, like, a completely AI run software company.

他还认为，像制造汽车或航天器这样生产实体产品的公司反而更容易——

And he he's argued that, like, companies that make physical things, like like cars or spaceships are are easier or or sorry.

相比软件公司，它们更难被取代。

Are harder to replace than companies that make software.

而微软的大部分业务都是软件。

And a lot of Microsoft's business is is software.

所以埃隆说，你可以直接开发一个AI，让它能够创建并运行所有软件，从而取代软件公司。

And so Elon says, you know, you can you can basically just make an AI that can create all the software and run all the software and displace software companies.

因此，MacroHard的精神就是在一个又一个行业里，用各种各样的软件来实现这一点。

And so the spirit of MacroHard is like to do that in a lot of different industries with a lot of different software.

在x AI，或者抱歉，在x平台，范围更窄一些，但两者都属于这一主题的一部分。

At x AI or sorry, at x, it's more narrow, but they they both kind of, are are part of this this theme.

西奥，我读你的报道时特别欣赏的一点是，你纳入了一些非常接近这一情况的人的观点。

One of the things I really appreciated, Theo, when I was reading your story was that you included some perspectives from people who are really close to the situation.

听起来像是员工们的意见。

Sounds like, you know, employees.

关于Grok以及埃隆·马斯克计划如何进一步将其整合到他的公司中，你听到了哪些担忧？

What are some of the concerns that you've been hearing about Grok and how Elon Musk plans to integrate it further into his companies?

对。

Right.

所以当你想到XAI的使命，即让Grok触达尽可能多的人时，既然他们已经收购了X，显然他们有了一个巨大的渠道，可以强制让使用X的人接受它。

So when you think about, you know, XAI's mission of of getting Grok to as many people as as possible, like, now that they've acquired X, they they obviously have this huge way to essentially force it on people that use X.

我的意思是，我不确定你怎么样，但我从未主动选择在X上与Grok互动，可我却总能看到它。

I mean, I I don't know about you, but, like, I've never, like, chosen to interact with with Grock on on x, yet I see it all the time.

我也没有。

Me neither.

是的。

Yeah.

没错。

Exactly.

比如，我知道一些人在Grok工作，这是真的吗？

Like, I I'm not I know some people, you know, at Grock, is this true or something?

我不明白这一点。

I don't understand that.

但Grok已经以某种方式融入了X的用户体验中。

But there's there are ways in which it's worked into UX on X.

例如，再比如另一个我不使用的功能，但我经常看到它。

For example, like, again, another feature I don't use, but I see it.

当你撰写帖子时，会有一个选项询问你：是否希望使用Grok生成一张配图？

If you're composing a post, you'll have this option to say, like, do you want to generate an image to go along with the post using Grock?

而且还有各种其他类似的情况，这些功能只是被硬塞进了界面里。

And and there are all kinds of, you know, other versions of this where it's just kind of stuffed into the interface.

对于X平台上的一些人来说，这令人担忧，因为那些负责X平台和社交媒体方面的人无法控制Grok。

And for some people at x, that has kind of been concerning because essentially, people that work on x and that that work on, like, the social media side don't have control over Grok.

但Grok却在生成大量出现在X上的内容。

But Grok is generating all this stuff that shows up on x.

比如，曾经发生过Grok谈论麦加和希特勒的事件，这让X平台的员工感到沮丧，因为这种他们无法控制的输出出现了。

So, like, famously, there was the Grok talking about Mecca Hitler thing a while ago that was frustrating for people on the x side because that was, you know, this output that they couldn't control showing up.

还出现过类似的问题，比如Grok会生成色情、成人内容、冒犯性内容或侵犯版权的内容。

And there have been similar issues where, you know, Grock will generate, like, pornography or adult content or offensive things or copyright, you know, copyright infringing things.

而且，X 的员工对这种情况根本毫无控制权，尽管这确实影响了他们整天为之工作的产品。

And and people at X just kind of have have no control over that even though it does impact, you know, the the product that they spend all day working on.

我的意思是，可以说目前存在不少问题，但若从宏观角度看，马斯克的目标是让公司更高效。

I mean, sounds like it's fair to say that there are more than a few snags at this point, but I guess just zooming out, you know, Musk's goal is to make the company more efficient.

我的意思是，随着人工智能在职场中越来越普及，你是如何看待进一步裁员的可能性的，无论是在他的公司内部，还是更广泛的范围内？

I mean, how are you thinking through the possibility of further layoffs, whether they're at his companies or even beyond just as AI becomes more and more prevalent in the workplace?

是的。

Yeah.

我的意思是，对于 xAI 而言，就像其他大型科技公司一样，那些大量投资人工智能的公司必须展现出某种成果。

I mean, something that's true of of x AI just as much as it's true of other, you know, big tech companies is that, you know, companies that have been pouring money into AI have to show some kind of result.

而目前，这可能就意味着裁员。

And right now, like, that that that can mean layoffs.

我的意思是，亚马逊也进行了大规模裁员。

I mean, Amazon, you know, did big layoffs too.

你知道的？

You know?

其他公司在大量投资人工智能后也进行了大规模裁员。

Other companies have done large layoffs after pouring a lot of money into AI.

而且，我觉得有时候人工智能只是公司想做但原本就想做的某种借口。

And, you know, I think that sometimes AI can kind of be a pretense to do something companies wanted to do anyway.

这并不是像埃隆·马斯克领导的公司进行裁员这种完全新鲜的事。

It's not like, you know, an Elon Musk led company doing layoffs as, you know, completely new.

就像他当初收购推特时就发生过这种情况。

Like, that that happened when he bought Twitter originally.

但对我来说，X 是一个相当有趣的案例，因为这些变化在产品上会迅速显现出来。

But to me, like, x is a pretty interesting case because these changes become evident so quickly in the product.

想想你经常看到新功能突然出现，或者你的免费信息流发生变化，或者垃圾信息增多然后又消失。

Like, you know, just think how often you see, you know, new features pop up or or your free feed evolve or spam, you know, get more prevalent and then disappear.

这些变化作为用户来说，比在其他公司出现得要快得多。

Like, there are all these changes, like, that as a user just become evident so much quicker than they do at other companies.

所以我认为，观察 X 和 XAI 将会是一个相当有趣的前沿视角，让我们看到未来几年其他公司可能面临的情况。

So I think, like, you know, watching X and XAI will be a pretty interesting, like, you know, front kinda a way to see, like, what's what could be coming at at other companies in the in the next few years.

最终，你认为马斯克用人工智能提高公司效率的这一愿景能成功吗？

Ultimately, do you think that that vision will work out for Musk when it comes to making the company more efficient using AI?

我的意思是，虽然有这么多潜在问题，说‘不行’很有诱惑力，但想想看，当初有多少人说要离开推特，结果现在还在用。

I mean, I I you know, it's tempting to say no because there are all these potential problems, but, like, think how many people said they were gonna, you know, quit Twitter initially and are still on there.

这是一个非常持久的平台，拥有庞大的用户基础。

Like, it's a very persistent platform with a huge base of users.

而且，我觉得他可能做出更多改变，而人们依然会留下来。

And, like, you know, I think it's possible he could change a lot more and people would still stick around.

这个观点非常好，西奥。

That's a really good point, Theo.

很高兴你来参加我们的节目，我想只有时间能告诉我们，是否还会看到马斯克的公司以及更广泛的领域出现更多裁员。

It was great having you on the show, and I guess only time will tell whether we're gonna see more layoffs coming from Musk's companies and beyond.

但非常感谢你今天加入我们。

But really appreciate you joining us today.

谢谢，阿妮塔。

Thanks, Anita.

因此，不仅仅是埃隆·马斯克的公司，员工们都在时刻警惕着裁员的威胁。

So it's not just at Elon Musk's companies that workers are keeping the threat of layoffs top of mind.

这就是为什么我想邀请两位嘉宾来讨论AI对就业市场可能带来的实际影响。

And that's why I wanna bring in two guests to talk about the reality of what AI could actually mean for the job market.

接下来加入我们的是D.A. Davidson的技术股权研究主管吉尔·卢里亚。

Joining me next is Gil Luria, Head of Technology Equity Research at D.

A.

A.

戴维森，以及安德鲁·麦卡菲，他是麻省理工学院斯隆管理学院的首席研究科学家，同时也是人工智能初创公司Work Helix的联合创始人。

Davidson, and Andrew McAfee, who's Principal Research Scientist at MIT's Sloan School of Management, as well as Co Founder of AI startup Work Helix.

盖伦、安德鲁，非常感谢你们今天加入我们的节目。

Gellan, Andrew, thanks so much for joining us today.

谢谢邀请我们。

Thanks for having us.

那么我想先稍微铺垫一下背景。

So I want to start by setting the scene here a little bit.

美国雇主在10月份宣布了超过15万个职位削减。

US based employers announced over 150,000 job cuts in October.

这一数字比去年同期增长了175%。

That was up 175% from the year prior.

根据奇切斯特、格雷和克里斯马斯公司的报告。

That's according to a report from Challenger, Gray, and Christmas.

这正值企业开始更广泛地采用人工智能之际。

And it's coming at a time when companies are starting to more widely adopt AI.

我们看到亚马逊在10月份裁掉了约14,000名员工，而公司发布的备忘录中特别提到了人工智能。

We saw Amazon laid off about 14,000 employees back in October, and the company memo that went along with that announcement specifically singled out AI.

吉尔，我们先从你开始。

Gil, let's start with you.

这些裁员在多大程度上是真正由人工智能驱动的，又在多大程度上只是降低成本的借口？

How much of this is truly AI driven versus a convenient narrative to cut costs?

嗯，两者都有。

Well, it's both.

但需要明确的是，当我们说大量裁员与人工智能有关时，我们现在主要指的是亚马逊、微软、谷歌、Meta等公司必须为人工智能支付费用。

But to be clear, when we say that a lot of the job losses have to do with AI, a lot of what we're talking about right now is that Amazon, Microsoft, Google, Meta, and others have to pay for AI somehow.

因此，尽管在人工智能上进行了巨额投资，它们仍通过削减成本来维持利润率。

And so they're cutting costs in order to keep margins in spite of these massive investments in AI.

从财务角度来看，我们正在将成本从销售、一般和行政费用转移到销售成本中。

So in financial terms, we're moving costs from SG and A into COGS.

我们正在将成本从人力支出转移到计算成本上，这确实在发生。

We're moving costs from headcount to compute costs, and that's definitely happening.

这些被裁人员中有多少是真正被人工智能取代的？

How many of those people are actually being replaced by AI?

这是另一个完全不同的问题。

That's a whole other question.

我们目前仍在沿用经济学家理查德·鲍德温提出的框架，他认为我们不会被人工智能取代。

The framework we're still seeing this is the framework proposed by economist Richard Baldwin that said that we're not going to be replaced by AI.

我们会被那些比我们更善于使用人工智能的人取代。

We're going to be replaced by people who use AI better than we do.

安德鲁，我想向你提出同样的问题。

Andrew, I want to pose the same question to you.

我的意思是，你观察到的这种情况，究竟是AI真的在取代劳动力，还是仅仅是个削减成本的借口？

I mean, what have you been seeing in terms of whether this is really AI replacing labor or if it's just an excuse to cut costs?

目前有几件事正在发生。

There are a couple of things going on.

需要记住的一点是，除了吉尔提到的超大规模云服务商之外，在大多数大型企业中，AI的应用仍处于初期阶段。

One important thing to keep in mind is that outside the hyperscalers that Gil was talking about, at most large enterprises, AI use is still in its infancy.

组织中有一些热点领域，软件工程是最明显的例子，但总体而言，AI在企业中的深度应用还处于早期阶段。

There are a couple of hotspots in the organization, software engineering being the most obvious one, but it is really early in the history of deep AI adoption for organizations.

因此，我认为在很多情况下，他们的裁员并不是因为自动化已经取代了大量工作。

So I don't think in a lot of cases, their layoffs are because they've automated a ton of work away.

另一个需要留意的是，关于这个问题的研究并没有指向单一方向，但由我的联合创始人兼同事埃里克·布伦约尔夫松等人进行的一项研究非常引人注目，该研究发现，特别是在知识型工作、白领岗位以及职场新人中，与过去相比，就业人数出现了下降。

The other thing to keep in mind is that the research is not pointing in one direction on this, but there's really intriguing research done by a team, including my co founder and colleague, Eric Brunjolfsson, that found out that in particular for knowledge work, white collar jobs, and for the youngest entrants to the workforce, the youngest people, the newest entrants to the workforce, that's where we're seeing employment declines compared to the past.

因此，我们可能已经开始看到AI对高暴露职业和职场新人招聘的影响。

So we could be starting to see AI's impact showing up in hiring of highly exposed professions and new entrants to the workforce.

很高兴你提到高暴露职业，因为我正想问你认为哪些是高暴露职业。

I'm glad you brought up highly exposed professions because I wanted to ask you what you think those are.

是的。

Yeah.

我们不确定你是否依赖我。

We don't know if you rely on me.

我的另一位联合创始人丹尼尔·洛克，他是沃顿商学院的教授，同时也是OpenAI团队的一员，几年前在《科学》杂志上发表了一篇精彩论文，他们系统地分析了经济中每一项任务，询问：这项任务能否在不降低质量的前提下被AI显著加速，然后将结果汇总到岗位、行业等层面。

My other co founder, Daniel Rock, who's a professor at Wharton and team at OpenAI, wrote a beautiful paper they published in Science a couple years ago, where they essentially looked at every task done in the economy, asked, can this be substantially accelerated with AI with no loss in quality, and then rolled that up to the job, the industry, the whatever.

果然，你会发现许多白领工作对现代AI高度敏感，尤其是某些在过去技术浪潮中并不太受影响的职业。

And sure enough, you find that a lot of white collar work is highly exposed to AI, modern AI, and in particular, some kinds of jobs that were not as exposed to previous waves of technology.

因此，软件工程就是一个很好的例子。

So again, software engineering is a great example here.

直到最近，软件在编写软件方面表现得很差。

Up until very recently, software was lousy at writing software.

但在生成式AI时代，这一情况彻底逆转了。

Just in the era of generative AI, that situation's been turned on its head.

所以我们正在见证这些颠覆。

So we're seeing these disruptions.

我们正在见证真正的创新。

We're seeing real innovation here.

但这并不等同于目前统计数据中出现大规模的技术性失业，因为我们并没有看到。

That's not the same thing as seeing massive technological unemployment in the statistics right now because we're not.

我想反过来问一下，安德鲁，你有没有发现哪些职位在人工智能普及后会变得更有价值？

I guess on the flip side, Andrew, I'm wondering if there are any roles that you are seeing that you think will become more valuable as AI takes root.

我认为那些需要真正人际交往能力、激励能力、协调能力和说服力的工作会变得更加重要。

I think jobs that require real interpersonal skills, motivation, coordination, persuasion, those kinds of jobs become more important.

AI仍然当不好垒球教练、中学老师或优秀的高级管理者。

AI is still a lousy softball coach or middle school teacher or a good upper level manager.

所以我认为，关怀类职业以及需要深度社会互动和社会认知的职业，可能会变得更有价值。

So I think caring professions and professions that require actual deep social interaction and social knowledge, those are likely to become more valuable.

吉尔，我想转向你，问问你从投资者的角度观察到了什么。

Gil, I want to turn it to you and ask what you're noticing from the investor perspective.

我的意思是，我们是否看到投资者普遍奖励那些声称用AI和自动化替代人力以降低成本的公司？

I mean, are we seeing investors generally rewarding companies that say they're replacing labor with AI and automation for their cost cutting?

还是华尔街更关注AI如何真正推动收入增长和营收增长？

Or is Wall Street more focused on AI actually driving revenue growth and top line growth?

这两者确实在同时发生。

Both are happening for sure.

投资者总是希望他们的公司降低成本。

Investors always want their companies to have lower expenses.

事实上，他们已经开始将这一点纳入预期之中。

And in fact, they're starting to build that into expectations.

而这种情况的发展方式，将首先在科技公司中显现。

And the way this is going to play out, it will be first at technology companies.

记住，科技公司主要雇佣的是软件程序员。

Remember, technology companies mostly employ software programmers.

因此，它们将是首批开始从这些程序员身上获得效率提升的公司。

So they're gonna be the first ones to start seeing leverage on those programmers.

他们能够用相同数量的人生产出更多的产品。

They'll be able to produce more with the same number of people.

因此，人们对这一点充满期待。

So there's a lot of excitement around that.

但再说一遍，安德鲁和他的麻省理工学院同事所做的研究表明，在收入方面，情况也会慢得多，因为这些工具目前还不一定适合大规模应用。

But but again, the the research Andrew and and and his colleagues at MIT have done says on the revenue side, it's also going to take a lot longer because these tools are still not necessarily ready for prime time.

IT技术栈主要基于结构化数据和确定性结果，并且需要严格的数据治理。

The IT technology stack is mostly based on structured data and deterministic outcomes with a lot of data governance.

这三件事是目前人工智能很难做到的。

Those are three things that are very hard for AI to do right now.

因此，要等到这些工具变得足够优秀，从而带来新的收入，还需要一段时间。

So it's gonna take time before the tools get so good that we're generating new revenue.

吉尔，你有没有一些例子，说明最近有哪些公司有效利用人工智能来降低成本或提升收入，并因此得到了市场的回报？

Do you have any examples, Gil, of companies that have effectively used AI to either cut costs or grow their top line that have been rewarded in the market recently?

有的，当然有。

Yeah, absolutely.

我的意思是，如果你看看Palantir，它已经帮助许多客户实现了这一点。

I mean, if you look at Palantir, Palantir has been able to help many of its clients do that.

但这还处于非常早期的阶段，目前只有Palantir的客户能够做到这一点。

But it's it's really early stages, and it's really only Palantir's clients that that have been able to do that.

所以Palantir会进来对你说：看。

So Palantir comes in and it tells you, look.

如果你有这种关键任务需求，与其雇佣人员来做，不如让我们替你完成。

If you have this mission critical need, instead of hiring up to do that, we'll do it for you.

我们会为你带来结果。

We'll get you that result.

你会降低成本，并增加收入。

You'll have less costs and you'll get an increase to revenue.

但到目前为止，只有少数客户能负担得起支付给Palantir的数千万美元费用。

But so far, it's really just a handful of clients that can afford the tens of millions of dollars that they're paying Palantir.

除此之外，几乎没有其他进展。

There's not a lot that's going on beyond that.

再宏观一点，即使抛开科技领域。

Just zooming out a little bit big picture, even outside of tech.

我的意思是，安德鲁，我想把这个问题专门问你。

I mean, Andrew, I want to direct this one to you.

过去，我们看到劳动生产率的提高带来了工资增长。

In the past, we've seen increased labor productivity lead to wage growth.

你认为AI也会如此吗？工人能否分享到这些收益，还是你认为企业会攫取大部分价值？

You think this is going to be the case with AI and that workers will actually get to share a piece of the gains or do you envision companies capturing most of the value here?

是的，阿妮塔，你提到了一个非常根本的事实：我们过去曾经历过极其强大的技术。

Yeah, Anita, you bring up this really fundamental fact that we've had previous extraordinarily powerful technologies.

这些技术改变了经济，但并没有导致贫困化、大规模技术性失业或工资下降。

They've changed the economy and they have not led to immiseration, to massive technological unemployment or wage declines.

工人数量增加了，他们也变得更加富裕。

There are more workers, they become more affluent.

但不同教育水平的人之间存在差异，诸如此类。

There are differences based on education levels and things like that.

但技术创新的故事，也是关于人们日益繁荣的故事。

But the story about technological innovation is also a story about increased prosperity for people.

我预计即使面对像人工智能这样疯狂而强大的技术，这种趋势仍将继续；但你提到了另一个重要现象：历史上，资本与劳动之间的收益分配一直相对稳定，但近年来证据显示，分配正更多地向资本倾斜。

I expect that to continue even with something as crazy and powerful as AI, but you bring up this other important phenomenon that it looks like when we divide up the pie between capital and labor, historically, division has been pretty consistent and there's evidence that in recent years it's been shifting more toward capital.

这种趋势在人工智能时代可能还会持续。

That might continue with AI.

这是我们必须要密切关注的事情。

It's a thing we've got to keep our eyes on.

所以，为了总结一下，我想问你们两位：你们认为人工智能会减缓科技行业的招聘，还是认为随着公司逐渐掌握如何使用这项技术并将其融入日常运营，招聘将出现反弹？

So to wrap this up, I wanted to ask both of you, I mean, do you expect AI to slow down hiring in tech or do you think we're going to see a rebound as companies sort of figure out how to use this technology and incorporate it in their day to day workflows?

我认为这将如安德鲁所言，带来一种结构性变化。

I think it'll be a change of mix to Andrew's point.

我认为未来会更具挑战性。

I think we're gonna have it's gonna be more challenging.

如果你的工作是阅读大量材料并进行总结，或者为他人撰写初稿，那么找到工作将会困难得多。

If your job was to read a lot of material and summarize it or to write drafts for somebody else to edit, it's gonna be a lot harder to get a job.

如果你非常擅长使用AI工具来完成这些类型的任务，那么你找到工作会非常容易。

If you're very good at using AI tools in order to accomplish those types of tasks, you're gonna have a very easy time getting a task.

因此，根据安德鲁的观点，我们更有可能看到生产力和使用率的提升，而不是由此导致的大规模失业，但这会推动边界的变化。

So to Andrew's point, then it's much more likely that we're gonna have increased productivity and usage than we're gonna have mass unemployment from this, but it will push the boundaries.

他提出的关于资本与劳动界限模糊的观点非常重要，我们现在讨论的是劳动将由AI生成。

And it's a very important point that he made about the blurring of lines between capital and labor in that what we're talking about now is that labor will be AI generated.

这是第一次，我们真正讨论将计算、深思熟虑以及创造性任务从人类劳动转移到计算机，这将对经济构成挑战。

That's the first time where we're really talking about replacing, the tasks and the performance of of computing and a thoughtful thoughtful and, and and generational tasks from labor, from humans to compute, that's gonna challenge the economy.

但就就业而言，这又回到了我最初的观点。

But in terms of employment, it goes back to the original point I made.

真正能做得好的，将是那些懂得如何使用AI的人。

It's going to be people that know how to use AI that are going to do really well.

那么，你并不认为当AI在企业层面真正扎根一两年后，招聘会出现激增吗？

And so you don't think there's necessarily going to be a surge in hiring a year or two from now once AI really takes root in the enterprise scale?

我认为，只要AI没有出现指数级增长，劳动力市场应该能够自行平衡。

I think the labor market should be able to balance out as long as there's not exponential gains in AI.

这里有一种可能性。

There is a scenario here.

如果你相信扎克伯格先生

If you believe Mr.

和马斯克先生

Zuckerberg and Mr.

以及阿尔特曼先生

Musk and Mr.

我们可能会达到一个AI呈指数级进步的阶段，它变得自我递归，最终AI能够像人类一样至少同样出色地完成任何任务，那时我们的劳动力将面临挑战。

Altman, we could get to a point where there's an exponential improvement in AI where it becomes self recursive and and we get to a point where AI can do anything any human can do at least as well as a human, then we're going to be challenged in the workforce.

但这种情况发生的概率仍然很小。

But that's still a small probability.

更可能的情景是，AI将继续推动生产力提升，保持充分就业，而生产力的提升收益将同时惠及企业的股东和劳动力，因为当我们越来越擅长使用AI时，我们的生产效率会更高。

The more likely scenario is that AI is going to continue to drive productivity, will stay at full employment, And those increases to productivity will accrue to both shareholders of companies as well as to the labor force because as we good good at AI, we're going to be more productive.

我们将创造更多的价值。

We'll produce more value.

我们能够基于这一点获得更高的工资。

We'll be able to get higher wages based on that.

嗯，吉尔，我确实希望你是对的。

Well, I certainly hope you're right about that, Gil.

非常感谢你们两位今天参加我的节目。

Thank you so much, both of you, for joining me on the show today.

很高兴你们能来。

It was great to have you here.

谢谢你们邀请我们。

Thanks for having us.

今天，我们正处在一个AI格局迅速变化的时刻，即使是业内人士也感受到了压力。

So we are in a moment today where the AI landscape is shifting really fast and even insiders are feeling the pressure.

上周，有报道称，OpenAI首席执行官萨姆·阿尔特曼向员工发送了一封备忘录，警告前方将面临困难。

Last week, the information reported that OpenAI CEO Sam Altman sent a memo to employees warning of rough vibes ahead.

这一警告是在谷歌推出其Gemini 3模型之前发出的，引发了关于这家搜索巨头是否正在缩小与OpenAI在模型方面差距的争论。

That warning came ahead of Google's rollout of its Gemini three model, which sparked debate over whether the search giant is actually closing the gap with OpenAI when it comes to its models.

所以今天我们要邀请扎克·劳埃德。

So today we're bringing on Zach Lloyd.

扎克是Warp的创始人兼首席执行官，这是一家为使用AI的软件开发者提供的平台。

Zach is the founder and CEO of Warp, a platform for software developers using AI.

在他职业生涯早期，扎克曾是Google Docs的首席工程师，最近他还花了不少时间使用Gemini 3.0模型。

Earlier in his career, Zach was actually the principal engineer for Google Docs, and recently he spent a bunch of time using this Gemini three point zero model.

扎克现在加入我们，向我们讲述这个模型究竟有多好，以及它可能对其他人构成怎样的威胁。

Zach joins me now to tell us how good the model really is and what kind of threat it could pose to others.

扎克，很高兴你能来。

Zach, great to have you.

嗨，阿妮塔。

Hey, Anita.

很高兴能来这里。

Great to be here.

谢谢邀请我参加。

Thanks for having me on.

扎克，我知道这个问题有点棘手，但你觉得 Gemini 和 GPT 哪个更好？

Zach, I know this is a bit of a loaded question, but Gemini versus GPT, which do you like better?

我的公司专注于 AI 编程领域，所以我主要能就这方面发表看法。

So my company is in the AI coding space, so I can speak mainly to that.

我们会在每个新模型发布时衡量它们的性能。

We measure the performance of all of these models as they come out.

截至今天，你知道，我们在至少一个主要基准测试中处于领先地位，这个基准叫 Terminal Bench，我们主要使用 Gemini 达到了这个水平。

And as of today, you know, we are at the top of the benchmarks, at least one of the main benchmarks, warp is called terminal bench, and we're we're there using primarily Gemini.

所以我认为谷歌上周确实取得了真正的突破。

So I do think Google had a a real advance last week.

Gemini 3 比 Gemini 2.5 Pro 优秀得多，至少在用于智能体开发方面，它现在至少与业界领先模型持平，甚至可能略微领先，而这显然是一个非常重要的领域。

Gemini three is a far superior model to Gemini 2.5 Pro, and it's certainly at least comparable, if not maybe a little bit ahead of the pack right now when it comes to using models for agentic development, which is a very important space, obviously.

那么，你会说在这方面它比任何 OpenAI 的模型都更好吗？

So you'd say it's better than any of the OpenAI models in that regard?

目前，如果你非要我表态，我认为仅基于我们这里的数据，也就是我们所测量的结果。

At the moment, if you really press me, I think just based I'm just based off the data here, and so we're going off what we measure.

我们目前在任何评估中的最佳结果都是使用 Gemini 3 取得的。

We our best result on any evals right now has been with Gemini three.

不过，如果你看看我们产品中的实际使用情况，可能 GPT 的使用量仍然超过 Gemini。

That said, there you know, if you look at actual usage within our product, there's probably still more usage for GPT over Gemini.

而且，历史上，编码领域的领先模型实际上是来自 Anthropic 的 Claude 模型。

And then, you know, historically, actually, the leading models in the coding space have been from Anthropic, so they've been the Claude models.

所以目前的情况非常有趣。

And so at at the moment, it's a really interesting situation.

我认为，谷歌、OpenAI 和 Anthropic 这三家的模型在编码能力上都非常强大，差距其实并不大。

I think we have three very, very capable models from Google, OpenAI, and Anthropic all pretty much sitting in not that different of a spot when it comes to coding capabilities.

而新变化是，谷歌已经迅速追赶上来。

And what's what's new is that Google has really made up a bunch of ground.

你会说这主要是体现在代理式编码方面吗？

And would you say that that's specifically in terms of agentic coding?

我的意思是，你有没有发现其他工作流程中，Gemini 3.0 的表现优于其他模型？

I mean, are there any other sort of workflows that you found that Gemini three point o is working better than other models?

所以我会说，在各种开发任务中，甚至超越编程本身。

So I would say across, like, all sorts of development tasks, so even going beyond coding.

你可以让Gemini执行DevOps任务。

So you can have, you know, you can have Gemini doing DevOps tasks.

你可以让它调查服务器为何崩溃。

You can have it investigating why your servers are crashing.

它非常强大。

It's super capable.

它真的很强大，我们在进行长期任务时已经测量过这一点。

It's really capable, and we've measured this when it comes to doing sort of like long horizon tasks.

这些模型随着时间推移不断改进的方式之一就是，你对它们的直接人工干预越来越少。

And this is one of the ways that these models are improving over time is like, you need sort of less and less hands on keys, humane guidance of them.

Gemini就是这一点的绝佳例子，它可以运行更长的时间。

And Gemini is a very good example of this where it can run for longer and longer timeframes.

在编程领域之外，我个人并不觉得有资格发表评论。

Outside of the coding domain, I don't really feel personally qualified to comment.

我跟其他人一样看新闻。

I I I watch the news like everyone else.

我认为，如果非要说我看到它们有明显突破的领域，那更多是在图像生成和图像识别方面。

I do think that they are if there are areas where I see them spiking, it is more in sort of image generation, image recognition.

他们推出了一款新模型叫Nano Banana，非常出色。

They have a new model Nano Banana, which is really good.

所以对我来说， headline 是：谷歌现在真的全力投入了。

So I I do think to me, the the headline is like Google is is really in the game right now.

而以前，这更像是两个人的独角戏。

Whereas before, it was a little bit more of a two man show.

是的。

Yeah.

这让你感到惊讶吗？

Is that something that surprises you?

因为至少在我看来，关于谷歌的这些热议似乎都是最近才出现的。

Because at least to me, it seems like all this chatter about Google has been pretty recent.

我的意思是，老实说，这并不令人惊讶。

I mean, honestly, it shouldn't be surprising.

如果你回顾一下大语言模型的历史，许多技术都源自谷歌工程师。

If you look back at the sort of history of LLMs, a lot of the technology originated at Google from Google engineers.

显然，最初的Transformer论文就是谷歌发布的。

Obviously, the original transformer paper was from Google.

许多最初为OpenAI和Anthropic等公司奠定基础的人才都来自谷歌。

A lot of the original talent that seeded places like OpenAI and Anthropic was Google talent.

谷歌在商业上拥有如此强大的地位，理应能够在此领域取得进展。

And Google just is in such a strong position commercially here that that they should be able to execute on this.

对吧？

Right?

他们拥有海量的数据。

They have incredible amounts of data.

他们拥有卓越的基础设施。

They have incredible infrastructure.

我认为这一点经常被忽视，但他们在实际制造AI芯片方面是英伟达为数不多的竞争对手之一。

They are I think it's overlooked often, but they're one of NVIDIA's only competitors when it comes to actually making, you know, AI chips.

而且，他们拥有巨额资金。

And so, and they have huge amounts of money.

所以，他们拥有取得巨大成功所需的一切。

So they have everything you need to be really successful.

在我看来，他们能够推出如此前沿的产品一点也不令人惊讶。

It's to me, it's actually it's it's not surprising at all that they have, like, managed to ship something that is state of the art here.

你提到谷歌自己制造芯片。

You mentioned that Google makes its own chips.

我的意思是，AI栈的另一个环节是，它还是一个云服务提供商。

I mean, another piece of the AI stack that it's touching is that it is a cloud provider.

我想问问你，对于一个模型制造商来说，成为云服务提供商有多重要？

And I want to ask you how important you think it is for a model maker to actually be a cloud provider.

我的意思是，这重要吗？

I mean, does that matter?

我认为，在规模上拥有云服务也有优势。

I think there's an advantage at scale to having the cloud as well.

所以，如果你看一下人工智能领域的整个价值链，NVIDIA做得很好，然后在它之上，是那些运行数据中心、提供模型服务或进行训练的公司，它们也赚取利润，接着是模型提供商赚取利润，最后是应用层。

So I think if you look at like the, you know, the the sort of value chain in the AI space, you have NVIDIA doing quite well, then you have, you know, on top of that, you have whoever is, like, basically running the data centers and serving the models or doing the training, that's like they take margin, then there's model providers who take margin, then there's application layer.

如果你能整合这一整套体系，我认为你就更能掌控价值捕获的位置，并且通过完全集成的解决方案获得更多的优化机会。

And if you can have an integrated version of that, I think you have more control over where you capture value and you have, you know, more of an opportunity to optimize by having a fully integrated solution.

所以我不确定。

So I don't know.

我喜欢，我是前谷歌员工。

I like, I I'm ex Google.

我知道谷歌执行得非常出色。

I know Google executes really well.

我认为他们具备了成为主导者，或者至少是主要参与者的全部要素。

I'm I think that they have all of the ingredients for being, like, you know, if not sort of dominant here, at least a major player.

在我看来，他们有一个大问题，那就是陷入了典型的创新者困境，正是这一点让ChatGPT有机会获得大量市场份额——我认为谷歌非常清楚，如何推出这些AI模型而不损害其现有的搜索业务。

They have one big problem, in my opinion, which is they have this major innovator's dilemma situation, which is I think what has opened the door for ChatGPT specifically to really gain a lot of market share is like, I think Google is very cognizant of how do you roll out these AI models without cannibalizing their existing search business?

所以我认为，他们面临着一个重大的结构性业务挑战，但在竞争实力方面，他们非常强大。

So I think that's like, they have this big structural business challenge, but in terms of being really well positioned to compete, they're very strong.

你认为Gemini 3.0及其早期成功，是否正是在创新者困境之下取得的？

And do you think Gemini three point zero and its early success has come as sort of despite that innovator's dilemma?

我只是在谈论模型本身的质量。

You know, I'm just speaking to the quality of the model.

我认为关键问题是，他们在分发方面会怎么做？

I think the big question is what do they do with it distribution wise?

他们如何将其整合到搜索产品中？

How do they roll it into their search properties?

他们如何将其推广到所有应用中？

How do they roll it out across their apps?

从Warp的角度来看，我们通过API使用它，通过他们的云服务使用它。

So from Warp's position, you know, we use it via an API, we use it through their cloud.

这毫无疑问对我们来说是件好事——谷歌推出了一款与OpenAI和Anthropic竞争的优秀编码模型。

It's just like, it's unambiguously a great thing for us to have Google ship a great coding model that's competitive with OpenAI's and Anthropix.

它降低了任何构建编码应用的人的成本。

It drives prices down for anyone who's building a coding app.

比如我们或Cursor，这为我们的用户创造了更多选择，让他们可以根据自己的工作流程选择最适合的模型。

So like us or Cursor, it, you know, it creates options for our users so that that our users have choice about which model is gonna suit their workflow best.

关于谷歌如何将其整合到其他产品中，我非常期待看到结果。

When it comes to how Google is gonna integrate it across the rest of their properties, I'm super curious to see.

我不知道。

I don't I don't know.

我认为谷歌显然有动力拥有更好的模型技术，但他们会如何将其整合到搜索中呢？

I think it's only in their interest to obviously have better model technology, but like, will they how they'll integrate that in search?

我完全不知道。

I have no idea.

显然，扎克，这件事有两个方面。

Obviously, Zach, there's two pieces to this.

一方面是模型的质量，另一方面是商业模式本身。

There's the, you know, the model quality piece and there's also the business model itself.

所以，我得稍微逼问你一下。

So I I'm gonna kind of press you here a little bit.

我的意思是，让我们往前看三年，好吗？

Mean, let's fast forward to three years from now, right?

在所有前沿模型中，在所有开发这些模型的公司里，有OpenAI、Anthropic、Google，我想阿里巴巴也在其中。

Of all the frontier models out there, all the companies that are making them, you have OpenAI, Anthropic, Google, I guess Alibaba is in there too.

你认为谁会成为领先的玩家？

Who do you think will be the leading player?

是的。

Yeah.

所以，我觉得，要预测三年后的情况很难。

So I think, again, hard for me to say three years out.

我认为我们想要的，也是我认为会发生的是，模型领域将出现一个竞争性的环境。

I think what we want and what I think will happen is that you will have a competitive environment on the model side.

我认为，每个开发这些前沿模型的公司都会希望为其配备一个杀手级应用。

I think everyone who's making one of these frontier models is going to want to have some killer app that's attached to it.

我认为，仅仅以模型API业务作为单一业务将会非常困难。

I think it's gonna be actually very hard to just be in the model API business as a sole business.

我的意思是，ChatGPT就是一个很好的例子。

Meaning, like, you know, I I think ChatGPT is is a great example.

OpenAI最初是从API业务起步的，但他们找到了一个出色的消费者应用。

It's like OpenAI started off in the API business, but they found a killer consumer app.

你知道，Gemini，谷歌肯定会把它整合到所有产品中，他们一定会找到某种方式将其融入搜索和其他服务中。

You know, Gemini, Google is gonna obviously put it into all of its you know, they're gonna find some way to put it into its search and its other properties.

Anthropic是我觉得他们正以最快的速度寻找企业应用的地方，他们的模型在这些领域具有优势。

Anthropic is the one where I think that, like, they are moving as fast as they can to find enterprise applications where their models, you know, have an edge.

所以他们主要在编程领域竞争，可能还会在其他垂直领域展开竞争。

So they're really competing in coding, they'll probably compete in other verticals.

我不知道谁会胜出。

I don't know who who will win.

我知道的是，这需要大量的资金。

What I do know is it's gonna take a ton of capital.

你可以看到，比如所有数据中心的建设与资本筹集都体现了这一点。

And you can see, you know, you can see this with all of the, like, the the data center building and capital raising.

所以我认为，不可能有数百家公司参与其中。

So I think it's it's not gonna be like hundreds of companies.

只会是这一群体中的一小部分公司胜出。

It will be some cohort of that group.

目前我看不到有任何公司能一举胜出的态势。

I don't see a dynamic right now where someone like wins it.

唯一可能让某家公司胜出的情况，就像你之前的嘉宾所说的，有人发现了一个指数级的反馈循环，使得模型能够自我递归改进。

The only situation in which some company wins it is kind of what, you know, your prior guest was saying, where there's some exponential feedback loop that someone uncovers where that causes the models to be able to self improve recursively.

我认为我们至今还没有看到这种情况。

I don't think we've seen that yet.

这仍然更多停留在理论层面，而非现实。

That's still more in like the realm of theory than reality.

所以我不确定，但我希望前沿领域能形成一个竞争性的市场。

So I don't know, but my hope is that there's a competitive marketplace at the frontier.