微软与英伟达投资Anthropic，谷歌Gemini 3.0，杰夫·贝佐斯的AI初创公司 | 2025年11月18日

欢迎各位来到Information的TITV。

Welcome everyone to the Information's TITV.

我叫阿卡什·帕什拉查。

My name is Akash Pashracha.

今天是11月18日，星期三。

It is Wednesday, November 18.

我们为大家准备了一场精彩节目，邀请了一些令人兴奋的嘉宾。

We have got a great show for you with some exciting guests.

但首先，让我们来看几条重要新闻。

But first, a couple of notable headlines.

今天早上，一次云flare故障导致互联网部分服务中断。

A cloud flare outage took down certain parts of the Internet this morning.

该公司确认已修复问题，并认为此次事件现已解决。

The company confirmed a fix has been implemented, and they believe the incident is now resolved.

X、ChatGPT以及我们Information自己的平台都受到了此次中断的影响。

Sites like X, ChatGPT, and even our own platform at the information were impacted by the outage.

该公司的首席技术官在X平台上发布道歉声明，澄清此次中断并非由攻击引起，而是由一个导致系统崩溃的漏洞所致。

The company CTO posted an apology on X clarifying that the outage was not caused by an attack, but by a bug that caused systems to crash.

该公司承诺保持透明度并提供进一步的解释，我们将密切关注。

The firm promised transparency and a further explanation, which we will be on the lookout for.

这一事件发生在AWS大规模中断仅一个月后。

The incident occurred just one month after a widespread AWS outage.

我也想强调一下我们信息平台独家报道的内容。

I also want to highlight some exclusive reporting that we have at the information.

数据平台Databricks，其工具帮助客户开发和使用人工智能，目前正在洽谈以超过1300亿美元的估值融资，较两个月前的融资估值上涨了约30%。

Databricks, a database provider whose tools help customer develop and use AI is in talks to raise money at a valuation of more than $130,000,000,000 that is up around 30% from its valuation in a financing round just two months ago.

我们将在节目笔记中提供该报道的链接。

We will link that story in the show notes.

好的。

Okay.

以下是今天节目的安排。

Here's what's on deck for today's show.

首先，我邀请了主编杰西卡·莱森和微软记者亚伦·霍姆斯，来讨论一则突发新闻：微软、英伟达和Anthropic达成了一项新合作。

First off, I am joined by Editor in Chief Jessica Lesson and Microsoft Reporter Aaron Holmes to discuss some breaking news, a new partnership between Microsoft, NVIDIA, and Anthropic.

杰西卡还将谈谈杰夫·贝佐斯新成立的AI初创公司。

Jessica will also talk about Jeff Bezos' new AI startup.

接着，我们的谷歌记者将深入解析Gemini 3.0的发布与战略部署。

Then our Google reporter will break down the rollout and strategy behind Gemini three point zero.

随后，我们将与我们的广告记者展开一场精彩的讨论，他今天发表了一篇关于Pinterest如何适应AI时代的文章。

We'll then turn to a great discussion with our advertising reporter who published a story today about how Pinterest is adapting to the AI era.

最后，我们将探讨亚马逊150亿美元的债券发行，这一举动引发了关于科技巨头如何为AI领域的巨额资本支出融资的讨论。

And finally, we will end things with a conversation about Amazon's $15,000,000,000 bond issuance, which is sparking conversations about how big tech companies are funding their big CapEx spend around AI.

这期节目内容丰富，我们马上开始。

It's a big show, so let's get right on into things.

今天早上我们收到了一条重磅消息：英伟达将向Anthropic投资高达100亿美元。

We had big news this morning that NVIDIA is investing up to $10,000,000,000 in Anthropic.

微软将投资高达50亿美元，而Anthropic承诺将向微软采购高达300亿美元的计算资源。

Microsoft is investing up to $5,000,000,000 and Anthropic has committed to buy up to $30,000,000,000 in compute from Microsoft.

这是人工智能生态系统高度互联的最新证据。

It is the latest data point in a very interconnected AI ecosystem.

我想请我们的主编杰西卡·莱森和微软记者亚伦·霍姆斯来帮我们逐条分析。

I want to bring on our editor in chief Jessica Lessen, our Microsoft reporter, Aaron Holmes, to help us break it all down.

杰西卡、亚伦，欢迎来到节目。

Jessica and Aaron, welcome to the show.

很高兴你们再次回来。

It's great to have you back.

谢谢，阿卡什。

Thanks, Akash.

又是充满重大新闻的一天。

Another big news morning.

这件事影响广泛，涉及众多参与者。

It is big all around and so many players involved.

亚伦，既然你是微软记者，我想先快速问你一下。

Aaron, given that you're the Microsoft reporter, I want come to you very quickly.

给我们详细解读一下今天早上的公告，哪些地方让你印象深刻。

Break us down to us the details of the announcement this morning and what stood out for you.

是的。

Yeah.

我认为特别有趣的是，微软通过这次对Anthropic的投资，成功从其竞争对手亚马逊和谷歌手中抢走了这家初创公司的业务。亚马逊和谷歌也是Anthropic的股东，此前一直是Anthropic的云服务提供商。

So I think what's particularly interesting about this is that Microsoft is essentially, you know, with this investment in Anthropic, winning some of the startups business away from its rivals, Amazon and Google, who are also shareholders in Anthropic and up to now have been the the cloud providers, for Anthropic.

同时，这也表明微软正越来越接近OpenAI的劲敌——而OpenAI此前一直是微软在人工智能竞赛中的首选合作伙伴。

And at the same time, this shows how Microsoft is getting closer to the archrival of OpenAI, which had been, you know, up to now Microsoft's chosen horse in the AI race.

微软已向OpenAI投资了130亿美元，并拥有重用其模型的权利。

It has invested $13,000,000,000 into OpenAI and has the rights to reuse that startup's models.

我们正逐渐走向这样一个世界：模型提供商不再紧密绑定于一两家云服务商，而是明确愿意与市场上所有云服务商合作，同时还能获得这些云服务商的投资，这非常引人注目。

And we're basically moving closer to this world where instead of a model provider being closely allied with one or two cloud providers, they're clearly willing to work with all of the cloud providers on the market, while also, you know, scoring investments from those cloud providers at the same time, which is really fascinating.

杰西卡，这种交易感觉就像我们见过的那些循环往复的交易，对吧？

Jessica, this kind of feels like one of those circular deals that we've seen so much of, right?

没错，阿卡什。

Absolutely, Akash.

说实话，我对此的反应并没有花那么长时间。

I don't think my reaction to this was honestly what took so long.

很明显，为了满足这些计算需求，每个人都会成为彼此的客户。

It was so clear that to fulfill these compute demands, everyone was gonna be a customer of everyone.

上次我在这里时，谈到了多边主义。

Last time I was on here, talked about multilateralism.

这就是游戏的名称。

That is the name of the game.

我认为，像Anthropic这样的公司面临的融资需求是巨大的。

And I think what's happening is that the fundraising needs are immense for players like Anthropic.

如今，计算需求也是如此。

Now also the compute needs are.

所以，在需求发生变化之前，这种情况是合理的，对吧？

So this makes sense until demand shifts, if demand shifts, right?

一旦需求发生变化，整个行业就都麻烦了。

And then we're all in trouble as an industry.

但我对这件事的反应才是拖这么久的原因？

But my reaction to this was what took so long?

意思是，亚伦是第一个报道微软在Copilot方面正与Anthropic技术紧密合作并深度整合的人，当这个消息公布时，这是一次巨大的转变，因为这是首个迹象，表明OpenAI不再会是消费端和企业端唯一的合作伙伴。

Mean, Aaron was the first to report that Microsoft in terms of Copilot was working closely and integrating Anthropix technology, which was a huge shift when that story came out because it was the first sign that OpenAI wasn't going to be the only partner on the sort of consumer and business facing side.

我认为这些交易总是多方面的，每一方都希望从对方那里获得很多东西。

And I think these deals are always multifaceted where each side wants many things from each other.

我认为这是第二只靴子落地了，这两家公司之间可能还会有更多类似的合作。

And I think this is the second shoe to drop and there will probably be more between those companies.

所以，亚伦，正如杰西卡所说，你是第一个报道微软和Anthropic关系日益密切的人。

So Aaron, as Jessica said, you reported, you were first to report this relationship between Microsoft and Anthropic getting closer.

你认为这现在对微软和OpenAI之间的关系意味着什么？

What do you think this now means for relationship between Microsoft and OpenAI?

我们最近看到微软CEO萨提亚·纳德拉几乎在公开场合大声宣称，他不希望只有一个巨型云客户主导Azure的大部分收入，而过去几年这个客户一直是OpenAI。

We've seen Satya Nadella, Microsoft CEO, essentially shouting from the rooftops recently that he doesn't want to just have one giant cloud customer that's driving the bulk of Azure's revenue, which in the past couple of years has been OpenAI.

在上一次财报电话会议上，我们确实看到他试图向股东保证，微软在AI云业务上的客户不会过于集中。

And we really saw that on the last earnings call that he was trying to reassure shareholders that Microsoft wouldn't be overly concentrated in its, you know, cloud business for AI customers.

所以某种程度上，这笔交易证明了他对此是认真的，因为对方是少数几家愿意像OpenAI那样在服务器上大笔投入的AI实验室之一。

So in some ways, this is, you know, proof that he's serious about that by inking this deal with essentially one of the only other AI labs that is willing to spend as heavily on servers as OpenAI.

同时，微软仍在使用OpenAI的模型。

At the same time, you know, Microsoft is still using OpenAI's models.

它拥有免费在其软件中重用这些模型的权利。

It has the rights to to reuse those models, for free in its software.

因此，我们不会看到这一点消失。

And so we're we're not gonna see that go away.

我认为，微软仍然表示，其战略是将与OpenAI的合作作为其主要的AI合作伙伴关系。

And I think, Microsoft is still saying that its strategy is to to focus on that OpenAI partnership as its sort of primary AI partnership.

但本质上，微软也在表明，它并不害怕多元化，愿意为其客户提供来自OpenAI竞争对手的更多模型。

But, basically, Microsoft is also showing that it's not afraid to diversify and and offer its customers even more models from OpenAI's rivals.

杰西卡，我想听听你对OpenAI和Anthropic战略的异同有何看法，因为我们之前做过一些报道，提到Anthropic专注于效率，并且更紧迫地追求实现自由现金流。

Jessica, I want to get your take on how the strategies of OpenAI and Anthropic have compared and contrasted, because we've done some reporting about the focus on efficiency that Anthropic has and the tighter timeline it has towards getting towards actually generating free cash flow.

直到最近，我们才看到Anthropic在数据中心和计算能力方面发布了更多公告。

We hadn't seen as many announcements from Anthropic until lately around data centers and around compute.

这真的感觉像是他们开始提高这些公告的频率了。

It really feels like they've started to increase the frequency of those announcements.

你如何看待这两家公司战略上的差异或相似之处？

How do you square the differences or similarities in the strategies of the two companies?

Anthropic 和 OpenAI 是非常不同的企业，我认为它们正朝着截然不同的方向发展。

Anthropic and OpenAI are very different businesses, and I think headed in very different directions.

大约一周前，我们的 Anthropic 和 OpenAI 记者 Sri 发布了独家的 Anthropic 财务数据，显示到 2028 年，OpenAI 计划在基础设施上花费 2350 亿美元，而 Anthropic 只有其三分之一。

A week or so, we had the exclusive Anthropic Financials published by our Anthropic and OpenAI reporter, Sri, that showed that I think through 2028, OpenAI is scheduled to spend $235,000,000,000 on infra with Anthropic, a third of that.

因此，这是一个数据点，表明它们在计算资源预测方面也有着非常不同的策略，而这当然是它们未来计划的基石。

So that was one data point that actually shows they have very different strategies, even on the compute forecasting side, which of course is the bedrock for what they intend to do.

我的意思是，如今它们的业务领域略有不同，Anthropic 正在大力推动企业市场。

I mean, they're in somewhat different businesses today with Anthropic really pushing the enterprise side.

OpenAI 目前在收入方面正以每小时一千英里的速度全方位推进。

OpenAI is kind of going a thousand miles an hour in all directions right now on the revenue side.

我们看到它们在企业领域进行了巨大的投入。

We see a huge push with businesses.

首席财务官莎拉·弗莱尔参加任何会议时，都会谈论他们拉到了多少新企业客户。

Sarah Fryer, the CFO, cannot sit at a conference without talking about how many new business customers they've nabbed.

现在，应用部门的首席执行官斐济·西莫也出来表示，我们的消费者业务将通过普及高端建议、高端购物建议和高端财务建议来实现盈利。

And now we have the CEO of applications, Fiji Simo, out there saying our consumer business is gonna make money from basically democratizing high end advice, high end shopping advice, high end financial advice.

我认为，鉴于市场上存在大量免费产品，消费者是否愿意为这些服务付费还是一个很大的疑问。

I think that's much more of an open question whether consumers will pay for that or not, given there are a lot of free products out there.

所以，回答你的问题，我认为它们的战略正在分化。

So to answer your question, I think their strategies are diverging.

我认为观察这一点非常有帮助，因为我越来越觉得，明年的一个关键问题将是：如果你减少在计算资源上的投入，你将如何调整？

And I think it's so useful to see because I increasingly think a question for next year is going to be if you spend less on compute, how are you going to adjust?

我们不知道哪些项目最终能够实现。

We don't know which projects are actually going to come to fruition.

我们也不知道华尔街会怎么说，以及会给予这些公司多少资本支出的缓冲时间。

We also don't know what Wall Street is going to say and how much runway they're gonna give these companies on CapEx.

因此，我认为将非常有趣的是，看看这些曾经是我们预测的内容。

And so I think it's gonna be really interesting to look at, okay, these were our forecasts.

是的，需求很高。

Yes, demand is high.

但如果我们更关注利润底线，投资者就会更加严格地审视这一点。

But if we're paying a little more attention to the bottom line, investors are more scrutinizing this.

你在哪里提高效率？

Where do you get more efficient?

你减少研发支出吗？

Do you spend less on research?

你减少推理成本吗？

Do you spend less on inference?

所以观察这一切非常有趣。

So it is fascinating to watch.

显然，两者之间仍然存在激烈的竞争，但它们为我们提供了两种截然不同的模式，来思考未来如何平衡收入与利润。

Obviously still a fierce rivalry between the two, but they give us very two different models on how you sort of top and bottom line going forward.

还有战略合作伙伴关系。

Also on strategic partnerships.

我的意思是，过去几周我们一直坐在这里，试图弄清楚Anthropic究竟如何筹集更多资金。

I mean, we have been sitting here at the information the last couple of weeks trying to figure out exactly how Anthropic is gonna raise more money.

他们有亚马逊。

They've got Amazon.

他们有谷歌。

They've got Google.

我说，哦，天哪，现在他们有了微软。

You know, I'm sort of saying, Oh, duh, now they have Microsoft.

我们怎么就没想到呢？

Why didn't we think of that?

但这与OpenAI筹集资本的方式截然不同。

But that is very different from how OpenAI has gone in terms of how it's raising its capital.

所以比较和对比起来非常有趣。

So it's fascinating to compare and contrast.

没错。

Right.

艾伦，我们让你走之前，简短问一下，杰西卡提到了Anthropic现在与三大超大规模云服务商的关系。

Very quickly, Aaron, before we let you go, Jessica mentioned the relationships that Anthropic now has with all three of the big hyperscalers.

你认为这在某种程度上会影响Anthropic与亚马逊或谷歌的关系吗？

Do you think this has any meaning of sorts with Anthropic's relationship with Amazon or with Google at all?

是的。

Yeah.

这里真正重要的一点是，Anthropic正承诺与英伟达更深入合作，优化其在英伟达芯片上训练和运行模型的方式。

I mean, one thing that's really significant here is that Anthropic is committing to work with NVIDIA to optimize how it trains and runs its models on NVIDIA chips even more so.

到目前为止，谷歌和亚马逊一直非常积极地展示Anthropic正在使用它们的自研AI芯片——谷歌的TPU和亚马逊的Trainium，作为其芯片应用的典范。

And so far, you know, Google and Amazon have been very eager to show that Anthropic is using their in house AI chips, TPUs at Google and, you know, Trainium at Amazon, as kind of like the the poster child for how their chips can be used.

因此，某种程度上，这是英伟达试图夺回部分业务、证明自己依然是前沿AI实验室首选平台的方式。

So in some ways, this is, you know, NVIDIA's way to claw back some of that business and show that it still is, you know, the the number one destination for Frontier AI Labs.

我认为这使得亚马逊和谷歌更难继续宣传他们的芯片对Anthropic至关重要。

And I think that it makes a little bit harder for Amazon and Google to continue to pitch their chips as being key to Anthropic.

然而，与此同时，Anthropic今天表示，AWS仍然是其最重要的云服务商。

However, at the same time, Anthropic said today that AWS is still its most important cloud provider.

所以我们得看看情况会如何发展。

So we'll have to see how that plays out.

对。

Right.

好了，亚伦，感谢你今天来做客。

Well, Aaron, I want to thank you for coming on.

我们就不多留你了，我知道你还有更多报道要做。

We will let you get back to the newsroom because I know you have more reporting to do.

再次感谢你加入我们。

Thanks again for joining us.

杰西卡，稍等一下，因为我想聊聊本周的另一个新闻——杰夫·贝佐斯正式进军人工智能领域，成立了他的新初创公司。

Jessica, stick around for a minute because I do want to talk about another story this week, which is Jeff Bezos is officially now in the AI game with his new startup.

关于这家初创公司，你听到了哪些消息？他们是怎么看待这件事的？

What are you hearing on the ground about this startup and how they're thinking about it?

大概六个月前，我们就开始听说杰夫正在为一家庞大的人工智能公司融资。

So we started hearing maybe about six months ago that Jeff was raising for a monster AI company.

《纽约时报》本周报道，这家公司的代号是‘普罗米修斯’，专注于人工智能与制造业。

And the Times obviously reported this week that that company is codenamed Prometheus and focused on AI and manufacturing.

我立刻震惊了，因为今年三月，我和亚伦·吴率先披露了拉里·佩奇成立了一家名为Dynatomics的新公司，专注于人工智能与制造业，特别是飞机领域。

And instantly, my head went off because, of course, in March, myself and Aaron Wu with the information broke that Larry Page has a new company called Dynatomics focused on AI and manufacturing, specifically planes.

所以我认为这是一个非常有趣的时刻。

And so I think it is a fascinating moment.

今年年初，你曾问我，拉里和杰夫是否会运营人工智能与制造业公司？

You had asked me at the beginning of this year, would Larry and Jeff be operating AI and manufacturing companies?

我可能当时会说，我觉得不会。

I probably would have said, I think not.

拉里并不是首席执行官，不过我听说他每周会去办公室好几天。

And Larry isn't the CEO, although I hear he's in the office several days a week.

杰夫是联席首席执行官，我们拭目以待这件事会如何发展。

So Jeff is the co CEO and we'll see how this plays out.

关于‘普罗米修斯’，还有很多有趣的信息。

There's a lot of interesting information around Prometheus.

首先，我知道里克·克劳斯纳是一位知名的AI科学家，曾与尤里·米尔纳共同涉足长寿领域，现在也参与了这个项目。

For one, I now know that Rick Klausner, is sort of well known AI scientist, was involved in the longevity space with Yuri Milner, is also involved in this project.

我不明白为什么一家制造公司需要一位科学家，但你知道，这从来不妨碍这些亿万富翁们大胆设想。

I'm not sure why you need a scientist in a manufacturing company, but you know, that's never stopped these billionaires from thinking big.

所以我的预测是，我们还要等很久才能看到Prometheus项目有任何实质进展。

So my prediction is that it will be a long, long time before we see sort of anything from Prometheus.

我认为，杰夫很可能正在召集一些非常优秀的人才来研究这些问题。

I think, know, Jeff is probably starting to assemble really interesting people to study these problems.

他需要加速与太空相关的制造进程。

He needs to accelerate manufacturing with regards to space.

就在这个消息公布时，我看到埃隆昨天在X上称他为模仿者。

I see that Elon called him a copycat on X yesterday as news of this broke.

而且我

And I

我觉得现在还为时过早。

think it's early.

你知道，我听说这一轮还没结束，但我认为杰夫要为它拿到大量投资并不难。

You know, I hear this round isn't even done, but I think obviously it won't be hard for Jeff to get a lot of checks for it.

所以我的体会是，人们对利用人工智能来制造东西确实感到非常兴奋。

So my takeaway is that there is some real excitement over sort of AI for building things.

这个领域有很多事情正在发生。

There are a lot of things in this space.

我对这一点已经说得够多了，但我现在生活在Waymo的世界里，我认为这在湾区是一个非常有趣的例子，展示了物理实体、制造、人工智能和机器人技术如今如何处于前沿。

I'm sort of a broken horse on this, but I now live in Waymo Land, which I think is a really interesting example here in the Bay Area of how kind of physical things and manufacturing and AI and robotics are really front and center on the frontier right now.

对。

Right.

你提到你的孩子们乘坐Waymo去上学，而且他们根本不在乎。

You you said that your your kids were taking Waymo's to to school, and they're they're not even phased

到学校了，阿卡什。

to by school, Akash.

真希望如此。

I wish.

我的孩子还很小。

My kids are so young.

上周末，我们匆匆去了一趟冰淇淋店。

It was very we we took a quick lap to the ice cream store over the weekend.

我赶紧把儿童安全座椅安装好又拆下来。

I was, you know, getting that car seat in and out as fast as I could.

但让我惊讶的是，这根本算不上什么事。

But what was amazing to me about that is it was a non event.

这种事已经变得很平常了。

The idea this was normal.

他们很放松。

They were relaxed.

他们想连上自己的Spotify。

They wanted to connect their Spotify.

这就是我们在旧金山街头听到的反馈。

That was our report from the streets of San Francisco.

我能问一下吗

Can I just ask

我想问你一个问题，关于我们在看到的AI与物理世界结合的趋势吗？

you a question about, know, I kind of was thinking about the AI meets physical world trend that we've been seeing?

这显然是拉里·佩奇和杰夫·贝佐斯关注的重点。

This is obviously what Larry Page and what Jeff Bezos are focused on here.

在另一端，我们有像米拉马拉迪这样受过技术训练的人。

On the other end of the spectrum, we have folks like Miramaradi who are sort of technical people by training.

在我看来，这类融资更多是针对基础技术之类的东西。

It strikes me that that type of fundraising is a little bit more for the foundational technology and stuff like that.

两者都在筹集巨额资金。

Both are raising tons of money.

我想知道你怎么看待这种观点：物理世界获得了大量资金，但底层模型仍需改进。

I wonder what you make of sort of this idea that we have physical world getting a lot of funding, but then we have the underlying models that also still need to be improved.

你知道吗，米拉和思维机器实验室真正做的事情仍然非常令人困惑。

You know, what Mira and Thinking Machines Lab are really doing is still very confusing.

而且，这是有意为之。

And by design, I will say.

我的意思是，你可以去问问在那里工作的人，其实并不清楚。

I mean, you can talk to people who are working there and it's not actually clear.

他们完全知道自己在做什么。

They know fully what they're doing.

所以这是有意为之。

So by design.

但没错，你正看到像伊利亚、像米拉马拉迪这样的AI领袖，仍在努力推动模型的边界，或调整现有模型的边界。

But yes, you're seeing these AI leaders like Ilya, like Miramaradi, who are still trying to push the boundaries of the models or tweak the boundaries of existing models.

老实说，阿卡什，直到我们看到他们究竟在做什么，并真正理解这些成果如何被商业界和研究界采纳，否则根本无法评判。

And honestly, Akash, until we see what they're doing and really understand how it's being picked up by businesses in the research community, it's impossible to judge.

我认为投资这些项目也确实非常困难，但他们是在用自己的声誉做赌注，考虑到他们的过往记录，这或许是一个非常稳妥的押注。

I think it's also really hard to invest in, but they're investing off their reputations, which may be a very sound bet given their track records.

但要真正理解他们究竟在做什么，实在是太难了。

But it's just very, very hard to really understand what they're doing.

你知道，你去问上周和他们开过会的两个人，他们会用两种完全不同的方式描述这家公司。

You know, you'll talk to two people who had meetings with them last week and they'll describe the business in two entirely different ways.

我认为，大概在明年年中，我们可能会看到Thinking Machines推出更多产品。

Again, I think probably mid next year, maybe we'll start to see some products from Thinking Machines a little bit more.

但目前，他们主要依靠势头和创始人的影响力来融资。

But right now they're raising funds off, you know, momentum and the founder power.

很好。

Great.

好了，杰西卡，感谢你来参加节目。

Well, Jessica, I want to thank you for coming on.

现在是一个充满活力的时期，我相信还会涌现出更多新闻。

It's a dynamic time, and I trust there's going to be more news to be unearthed.

那我们就不打扰你了，很快再邀请你回来做客。

So we'll let you get back to it, and we'll have you back on the show again soon.

谢谢，阿卡什。

Thanks, Akash.

好的。

All right.

很快见。

See you soon.

好吧。

Okay.

Gemini 3.0 已经上线了。

Gemini three point zero is here.

当然，这是谷歌的新一代旗舰大语言模型。

That is, of course, Google's new flagship large language model.

每次新模型发布后，通常都需要一些时间来评估其表现，届时会有大量评测陆续出炉。

It often takes some time to assess how good these new models are, every time they're launched, there are tons and tons of reviews that continue to come out.

但今天，我的同事、负责报道谷歌的艾琳·吴发布了一些关于它如何融入公司整体战略的优质报道。

But my colleague, Erin Wu, who covers Google, has some good reporting out today on how it fits into the company's overall strategy.

因此，我想请她来帮我们深入分析一下。

And so I want to bring her on to help us break it all down.

埃琳，欢迎再次做客节目。

Erin, welcome back to the show.

很高兴你来参加。

It's great to have you on.

嘿，非常感谢你邀请我。

Hey, thanks so much for having me.

我们来聊聊三零版本吧。

Let's talk about three point zero.

你能告诉我们，公司对三零版本是怎么看待的吗？我们稍后再谈评测，但公司是怎么想的？

What what can you tell us about how well, we'll get to the reviews in a second, but how is the company thinking about three point zero?

也许我们可以从这里开始。

Maybe let's start there.

是的。

Yeah.

我的意思是，三零版本延续了谷歌目前这段非凡的历程。

So I mean, three point zero is continuing this really incredible run that Google's on.

他们的最新模型发布表现相当出色，比如今年早些时候推出的2.5 Pro，广受好评，还有Nano Banana，也就是Gemini。

Like, they've done quite well with all of their recent model releases, 2.5 Pro, which came out earlier this year, very well regarded, Nano Banana, which was Gemini.

谷歌的图像编辑模型也是最早在Twitter上独立走红的产品之一。

Google's image editing model also was one of the first products to go kind of like independently viral on Twitter.

所以他们一直保持着良好的势头。

And so they've been on like a good run.

而Gemini 3，至少他们在发布前提供的各项指标看起来都非常出色。

And Gemini three, at least the metrics that they gave us prerelease, all look really good.

当然，关键问题在于，用户和开发者真正上手使用后会有什么感受。

Obviously, the big question is going to be how do users and developers feel once they actually get their hands on this thing.

但我认为谷歌目前势头正盛，他们希望这种势头能持续下去。

But I think Google is riding high right now and their hope is that this continues that progress.

好的。

Okay.

那么早期的评价如何？人们觉得它怎么样？

And so early reviews, how are people finding it?

是的。

Yeah.

再次，有点难以说清楚。

Again, a little bit hard to say.

我昨天和Figma的首席设计官聊了聊。

I chatted with Figma's chief design officer yesterday.

是谷歌为我们牵线的。

Google connected us.

不出所料，他们非常喜欢它。

Unsurprisingly, they are really liking it.

他们正在用新的Gemini模型进行设计。

They're using the new Gemini model for design.

他们发现，与他们测试过的其他模型相比，它能生成更广泛的设计风格，并且在将设计转化为实际代码时，非常擅长保留设计师的原始构想。

They're finding that it can generate a broader range of styles compared to other models that they've tested and that it's really good at kind of maintaining the fidelity of the designer's original vision when it's translating a design into actual code.

那么，这个模型目前在谷歌的前端层中是如何部署的？

And so in terms of where this model is now implemented in Google's front end layer.

所以有谷歌搜索，它有AI模式，还有Gemini。

So there's Google Search, which has the AI mode, and then there's Gemini.

所以根据你今天早上写的，两者都有。

And so based on on what you wrote this morning, it's it's both of them.

我的意思是，你可以选择其中一个。

I mean, it you can sort of go one or the other.

是的。

Yeah.

你看。

So look.

谷歌的战略是谷歌规模庞大。

Google's strategy is that Google is huge.

谷歌有19万名员工。

Google has a 190,000 employees.

谷歌有七个产品，用户超过20亿。

Google has seven different products with more than 2,000,000,000 users.

因此，他们本质上试图将Gemini整合到所有产品中。

And so essentially, what they're trying to do is they're trying to put Gemini into everything.

因此，这是首次在搜索的AI模式中，该模型从第一天起就立即上线。

So for the first time, this model is launching immediately day one in search for AI mode.

这是他们首次这样做。

That's the first time they've done they've done this.

它也在Gemini中上线了。

It's also launching in Gemini.

如果你是开发者，你可以在AI Studio和Gemini API上获取它。

If you're a developer, you can get it on AI Studio and the Gemini API.

如果你是企业用户，你可以在Vertex和Vertex AI Studio上获取它。

And if you're a business, you can get it on Vertex and the Vertex AI Studio.

他们今天还推出了一款全新的编程工具。

And they're also launching like a brand new coding tool today.

因此，谷歌的战略，就其存在而言，其产品似乎就是把一切东西都整合Gemini，为人们提供大量不同的方式来使用这些功能。

And so the Google strategy, as such as it exists, her product seems to just be making everything, putting Gemini into everything, like making like a ton of different options for people to use this stuff.

好的。

Okay.

出于好奇，咱们就聊聊你和我吧。

And just out curiosity, I mean, let's just talk me and you.

我是说，谷歌这种重叠策略，说你可以随时随地使用3.0版本，就像你听播客的地方一样。

I mean, this this overlapping strategy where Google is saying, you can use three three point o wherever wherever, you know, it's kinda like wherever you get your podcasts.

对吧？

Right?

也就是说，你能在那儿找到它。

Like, that's that's where you can find it.

嗯。

Yeah.

人们觉得这是个好策略吗？

Do people think this is a good strategy?

这有什么风险吗？

Does this have any risks?

我的意思是，谈谈这个。

I mean, talk about that.

是也不是。

Yes and no.

我所做的大量报道都是采访员工、前员工和资深谷歌人，他们都在问：他们在做什么？

A lot of the reporting that I've done has been talking to employees, former employees, longtime Googlers who are like, What are they doing?

整体的产品战略是什么？

What is the overarching product strategy?

为什么你可以把 Gemini 放在这么多不同的地方，但它却不能统一运行，也无法把你在这里的活动和在那里所做的事连接起来，形成一套真正统一的产品体验？

Why can you have Gemini in all of these different places and it doesn't work the same and it doesn't necessarily connect what you're doing here to what you're doing here against what could be a really unified suite of product experiences.

就连最近这次媒体演示中，也是接连展示：这是 Gemini 中全新的旅行规划功能。

Even just in this most recent press demo is like back to back, like, here's how you can do this cool new trip planning feature in Gemini.

然后又说：这是 AI 模式中的全新旅行规划功能。

And then it's like, here's how you can do this cool new trip planning feature in AI mode.

但它们并不相同，功能也不完全一致。

But they are like not the same and they don't do quite the same things.

所以，我的意思是，确实有一种观点认为，这是一家庞大的公司，官僚体系根深蒂固，根本缺乏像OpenAI那样的统一愿景——OpenAI从零开始，只专注于ChatGPT。

And so, I mean, like, there is the argument that, like, this is a huge company and it's a bureaucracy and it just kind of there's not, like, a central vision in the way that, like, maybe you could have more so with, an OpenAI where they're starting from greenfield and it's just ChatGPT.

就是单纯的ChatGPT API。

Like, it's just the ChatGPT API.

但另一方面，我觉得你也得承认，你知道吗？

But then, like, I think on the other hand, like, there's the argument that, like, you know what?

在某些方面，这其实并不重要。

Like, this doesn't matter in some senses.

他们是一家大公司。

Like, they're a big company.

他们拥有充足的资源。

They have the resources.

他们的效率真的达到极致了吗？

Is this, like, as efficient as they could possibly be with it?

大概没有。

Like, probably not.

但你看，数据一直在向上攀升。

But, like, numbers are up into the right.

人们确实在使用这些东西。

Like, people are using this stuff.

尽管很多人担心，搜索收入并没有下滑。

Like, search revenue despite, like, a lot of people's fear, like, is not going down.

所以我认为还有另一种观点，你知道吗？

And so I think there's also the argument that like, you know what?

这没问题。

It's fine.

比如，

Like,

模型很不错。

the model is good.

说到数据，你能再提醒我们一下 Gemini 目前取得了多少进展吗？让我们有个概念。

And speaking of the numbers, can you just remind us just about how much traction Gemini is getting just so we have a sense there?

是的。

Yeah.

我们最新的数据来自上个月的财报，并在Gemini的新闻稿中再次重申。

So the most recent number we've got is from earnings last month, reiterated again in the press release for Gemini.

但它的月活跃用户达到6.5亿，这已经很多了。

But it's 650 monthly active users, which is a lot.

增长非常快，增长得相当迅速。

It's grown a lot it's grown quite quickly.

这么多人啊。

That's a lot of people.

我的意思是，有这么多人在使用。

I mean, that's that's a lot of people used.

是的。

Yeah.

但它仍然低于ChatGPT。

It's still it's still lower than ChatGPT.

ChatGPT 的每周活跃用户达到八亿，没错。

ChatGPT is at 800,000,000 weekly active users, which Right.

这是一个更有意义的指标，因为它表明人们在日常或每周的生活中更依赖它。

Is a more meaningful metric because it shows that people are relying on this more in their, like, day to day or, like, week to week routine.

但谷歌，但如果你想想 ChatGPT 的品牌认知度，人们在对话中使用 ChatGPT，没错。

But Google But but if you if you think about the the the name recognition of ChatGPT, know, people using ChatGPT in conversation Yeah.

然后你就会自然而然地应用这种背景认知。

And you sort of, you know, apply that that context.

我的意思是，你看。

I mean, look.

用户数量按季度来看并不差，当人们不在对话中使用 Gemini 时也是如此。

Quarters or something of the user numbers is it's not bad when when people are not using Gemini in conversation.

这并不差，但你也得记住，OpenAI 是一家从零起步的公司。

It's not bad, but you also have to remember, I mean, like, OpenAI is a company that came out of nowhere.

谷歌则不同，也许人们不在对话中使用 Gemini，但每个人都在对话中使用谷歌。

Like, Google is this like peep people maybe aren't using Gemini in conversation, but, like, everyone's using Google in conversation.

我认为，人们仍然认为谷歌浪费了本可以成为品牌聊天机器人公司的机会。

I do think there's still this idea that Google squandered what could have been their opportunity to be, like, the name brand chatbot company.

没错。

Right.

好了，亚伦，非常感谢你前来做客。

Well, Aaron, thanks so much for coming on.

随着你听到越来越多关于人们如何反应的信息，我期待你与我们分享这些内容。

As you hear more and more about how people are reacting to it, I look forward to having you share that with us.

这位是我们在《信息》杂志的谷歌记者亚伦·吴。

That is Aaron Wu, our Google reporter here at The Information.

好的。

Okay.

人工智能和社交媒体如今是密不可分的两个术语。

AI and social media are two terms that go hand in hand nowadays.

但我的同事凯瑟琳·佩洛夫的一篇新报道揭示了Pinterest在采用AI功能时采取了更为谨慎的态度。

But a new story from my colleague, Katherine Perloff, reveals how Pinterest is taking a more measured approach to embracing AI features.

这一策略的部分依据是公司声称其客户真正需要的东西。

Part of that strategy is based on what the company says its customers actually want.

为了让我们更深入了解她的发现，我想请凯瑟琳来详细讲述她的故事。

To tell us more about what she found, I want to bring on Katherine to tell us more about her story.

凯瑟琳，欢迎再次做客我们的节目。

Katherine, welcome back to the show.

很高兴你来到这里。

It's great to have you here.

你好。

Hi.

那么我们来谈谈人工智能和Pinterest。

So let's talk about AI and Pinterest.

人工智能对Pinterest这样的公司可能构成多大的威胁？

How much of a threat does AI conceivably pose to a company like Pinterest?

我想这取决于你问的是谁。

I guess it depends on who you ask.

这个故事的重点。

The point of the story.

嗯。

Yeah.

所以你去问公司，他们会说，嗯，我们并不和谷歌竞争。

So you ask the company, like, you know, they're kinda like, well, we don't compete with Google.

我们做的是更专业的搜索类型，因此我们不会和ChatGPT竞争。

We do a more specialized type of search, and so we're not gonna compete with ChatGPT.

但并不是每个人都这么认为。

But not everyone thinks that.

你知道，华尔街有些人确实认为Pinterest有机会通过ChatGPT实现突破，因为人们来Pinterest寻求的问题，比如‘我该怎么重新装修客厅’，或者‘给我看看一套好穿搭’。

You know, there are folks on Wall Street really see the potential to for Pinterest to get their luncheon in by ChatGeeBT because the kind of questions people come to Pinterest with, which is like, how do I redecorate my living room or, you know, show me a good outfit.

你知道，这些问题人们也可以问ChatGPT，甚至可以让ChatGPT为他们生成图片来获取灵感。

You know, those are questions people could ask ChatGPT and could even ask ChatGPT to generate images for them to sort of get ideas.

人们来Pinterest是为了寻找灵感。

People come to Pinterest for inspiration.

与此同时，ChatGPT 目前并不是一个视觉搜索引擎，所以它不像 Google 图片。

Now at the same time, ChatGPT is not a visual search engine at the moment, so it's it's not like Google images.

但没错，它确实能生成图片，这对 Pinterest 构成了威胁。

But, yeah, it obviously produces images, that is a threat to Pinterest.

对。

Right.

另一个值得注意的是，Pinterest 最初是一个网站，它最初是让用户收集网络上他们最喜欢的图片和视频。

The the other kind of notable thing is that Pinterest started as a website, and it started as a way for users to collect, you know, their favorite photos and videos from the web.

而整个网络生态系统正受到 ChatGPT 的威胁。

And the whole ecosystem web ecosystem is being threatened by ChatGPT.

所以这也是为什么 Pinterest 特别处于风口浪尖的另一个原因。

So that's that's another reason why they're sort of kinda uniquely in the line of fire.

好的。

Okay.

那 Pinterest 正在采取什么措施来应对这个问题？

Now what is Pinterest doing to combat this?

它如何利用人工智能来应对这些威胁？

How is it using AI to to hedge against some of these threats?

是的。

Yeah.

你知道，这很有趣。

You know, it's interesting.

所以，Pinterest 一直在使用生成式人工智能。

So Pinterest, you know, they've been using generative AI.

但它们更多地将AI应用于系统内部，而不是想成为一家聊天机器人公司，或赋予用户使用AI创建图片、视频或文本的能力。

They've but they've been using it more in their systems and, like and sort of not to sort of kind of become a chatbot company or give users the ability to, like, create photos or create videos or create text with AI.

像Meta和TikTok已经向用户提供了这些功能。

Like, Meta and TikTok have been giving these abilities to their users.

我认为，Pinterest的想法是：我们只想让生成式AI让Pinterest在自身核心业务上做得更好。

And, you know, I think that, Pinterest is sort of like, we just want generative AI to make Pinterest better at what Pinterest does.

所以，比如在搜索和推荐服务上做得更好，它们推出了一个助手产品，你可以用自然对话的方式与它交流，询问问题，从而获取购物推荐等信息，这背后就用到了AI。

So, you know, servicing better recommendations in search, they release this assistant product, which you can kind of talk to it, or ask it questions in conversational language that pull up, like, you know, maybe shopping recommendations and that uses AI.

但他们并没有把自己变成一个聊天机器人。

But it's not you know, they're not turning themselves into a chatbot.

我跟他们的首席技术官马特·马德里加尔谈过，他说，这并不是用户想要的。

And I talked to their chief technology officer, Matt Madrigal, and, you know, he was like, that's not what our users want.

对。

Right.

我们不想成为Sora。

And we don't wanna be Sora.

我们不想创造Sora。

We don't wanna create Sora.

我只是觉得，这非常有趣，因为我觉得我们在这个节目里经常讨论的是，ChatGPT将如何改变一切？

And I just think, yeah, that's very interesting because I think a lot of the stuff we talk about on this show is sort of like, how is ChatGBT gonna change everything?

我认为这里还有一个开放性问题。

And I think there is an open question.

Pinterest的想法是，它并不会改变一切。

Pinterest is like, it's not it's not gonna change everything.

就像是，你知道的，后端。

It's like but, you know, the back end.

我们从后端开始吧。

Let's start with the back end.

让我们看看它到底能不能运行。

Let's let's see if it even works.

对。

Right.

是啊，每个媒体公司、每个科技公司都必须变成ChatGPT的版本吗？

Yeah, does every media company does every tech company have to become a version of ChatGPT?

是啊。

Yeah.

你知道的。

You know?

看起来好像……我实话实说。

Seems like a I'm I'm gonna be honest.