网络能否在生成式AI时代存活？——与Matthew Prince对话

Capital One的技术团队不仅在讨论多代理AI，他们已经部署了一个。它被称为聊天礼宾服务，正在简化购车流程。通过自我反思和分层推理结合实时API检查，它不仅帮助买家找到心仪的车辆，还能协助安排试驾、获得贷款预批以及估算以旧换新价值。

Capital One's tech team isn't just talking about multiegenthic AI. They already deployed one. It's called chat concierge, and it's simplifying car shopping. Using self reflection and layered reasoning with live API checks, It doesn't just help buyers find a car they love. It helps schedule a test drive, get preapproved for financing, and estimate trade in value.

先进、直观且已投入使用——这就是他们的技术实力。这就是Capital One的科技水平。面对席卷互联网却未支付内容费用的生成式AI浪潮，网络将如何反击？今天我们邀请到了Matthew Prince。

Advanced, intuitive, and deployed. That's how they stack. That's technology at Capital One. How will the web fight back against the wave of generative AI that is ingesting all the content on the Internet but not paying for it? We're joined today by Matthew Prince.

他是Cloudflare的首席执行官兼联合创始人，一直致力于拨乱反正。Matthew，很高兴见到你，欢迎来到节目。

He's the CEO and cofounder of Cloudflare and has been on the warpath attempting to right the ship. Matthew, great to see you. Welcome to the show.

谢谢邀请。

Thanks for having me.

能否先请你概述下生成式AI崛起对网络的影响？我们节目已讨论过一些，但我想听听你的见解。

Can you start by giving us a sense as to what is happening to the web with the rise of gender of AI? We've talked about it a bit on the show already, but I wanna hear it from you.

当然。过去三十年来，互联网的商业模式本质上由搜索驱动。搜索行为产生流量，将用户引向他人创作的内容。

Yeah. Absolutely. So the the business model fundamentally of the Internet over the last thirty years has really been driven by search. You search for something that generates traffic. It takes you to content that someone has created.

内容所有者或创作者主要通过三种方式实现价值：直接售卖内容（如今常见订阅模式）、投放广告，或是获得被人关注阅读的心理满足。这正是当今互联网的构建基础。

And then that content owner, that content creator, can drive value really in one of three ways. They can sell the content itself, sell subscription to it. We see plenty of that these days. They can put ads up against it, or they can just get the ego hit of knowing that somebody cares about and is reading their stuff. And that's really how the web has been built today.

它的构建初衷就是为了追逐流量。然而我们看到的是，历史上首次，各大搜索引擎（尤其是谷歌）的搜索量实际上在下降。取而代之的是越来越多人转向AI。与AI的不同之处在于，它不再提供十个蓝色链接让你点击寻找答案，而是试图直接给出答案本身。这意味着人们不再访问原始信息来源了。

It's been built chasing that traffic. What we're seeing, though, is that for the first time in history, searches across the major search engines, Google in particular, are actually on the decline. And what's replacing it is more and more people turning to AI. And the difference with AI is rather than giving you 10 blue links that you click to and find the answer, now what AI does is it tries to give you the answer itself. And that's meaning that people aren't going to those original sources.

如果他们不去原始来源，就意味着你无法再销售订阅服务，无法投放广告，甚至无法得知人们是否真正从你的内容中获取价值。因此我们Cloudflare真正担忧的是，如果创作内容的动机消失，在AI驱动的新未来里，还有谁会去创作内容？

And if they don't go to the original sources, then that means that you can't sell a subscription anymore. You can't put ads up against it. You don't even know that people are actually getting value from your stuff. And so what we're really worried about at Cloudflare is if the incentives for creating content go away, why is anyone going to create content in a new AI driven future?

那么请谈谈这些AI机器人或搜索引擎过去爬取了多少网页，每次爬取带来多少流量，以及如今的情况如何演变。

So talk a little bit about how many pages these AI bots or search engines have crawled in the past and how much traffic they've delivered for each crawl and where it's gone to today.

是的。要知道，三十年前当拉里和谢尔盖开始这个项目时，谷歌与互联网达成的协议本质上是：让我们复制你的内容，作为交换，我们会给你带来流量，你依然可以通过那三种方式之一从中获取价值。Cloudflare拥有过去十年关于谷歌的可靠数据，其中保持高度一致的指标就是谷歌对网络的爬取量——过去十年他们的爬取速率其实非常稳定。

Yeah. You know, think that the the the deal that Google made with the web starting thirty years ago when Larry and Sergey started working on on the project was basically let us copy your content and in exchange, we'll send you traffic that, again, you can drive value in, you know, one of those three ways from. And we have very reliable data at Cloudflare going back ten years looking just at Google. And the metric that has stayed very consistent over time is how much Google crawls the web. They've actually crawled at a very consistent rate over the last ten years.

在这同样的十年间，互联网用户实际增加了20亿。十年前全球约有40亿网民，如今已达到约60亿。按理说这段时间获取流量应该更容易，但事实并非如此。

Over that same ten years, we've actually added 2,000,000,000 Internet users. So we were at 4,000,000,000 Internet users about ten years ago. Today, we're at about 6,000,000,000 Internet users. So you would imagine it's actually gotten easier to get traffic over that period of time. But that's not what's happened.

实际情况是，若以十年前为基准，如今从谷歌获取一个点击、一个访客到你的网站，难度几乎是过去的10倍。变化何在？答案是谷歌开始在页面上直接提供更多答案。比如搜索'Cloudflare何时成立'，顶部会出现答案框显示'2010年9月27日'——这正是我们成立的日期。

What instead has happened is that back in the day, if you if you take sort of take ten years ago as the litmus test, today, it's almost 10 times as hard to get a click to get a visitor from Google to your site. What's changed? The answer is that Google has started providing more answers directly on the page. So if you search for something like, when was Cloudflare founded? There will be an answer box at the top that will say September 2730, you know, is the day that we that we launched.

你甚至不需要点击任何链接。事实上，如今约75%的谷歌搜索查询都能在谷歌站内获得解答。而过去六个月加速这一趋势的变化是，他们推出了AI概览功能。我们追踪了全球各个地区的数据，发现随着AI直接提供答案而无需用户阅读原始内容，谷歌向这些网站输送的流量正在持续下降。

And you don't have to click to any link. In fact, about 75% of queries to Google now get answered on Google itself. And what's changed in even just the last six months that's accelerated this is they've rolled out AI overviews. And we've tracked this from region to region to region to region. And what we see is as AI is giving you the answer without you having to read with the original content, the the amount of traffic that Google is sending to these these sites has gone down and down and down.

这对出版商来说是个好消息。如果说过去十年从谷歌获取流量难度增加了十倍，那么OpenAI则完全是另一个级别的挑战。以OpenAI为例，现在获取流量比十年前从谷歌获取要困难750倍。而对于Anthropic这样的平台，获取流量的难度更是高达3万倍。为什么会这样？

And that's the good news for publishers. If Google has gotten 10 times harder to get traffic from over the last ten years, OpenAI is a whole different beast. In OpenAI's case, it's seven fifty times harder to get traffic than it was from Google just ten years ago. In the case of something like Anthropic, it's 30,000 times more difficult to get that traffic. So why is that?

我认为答案是人们信任眼前的内容。他们在阅读这些衍生内容时，并不会回溯到原始来源。但问题在于，如果没人阅读原始内容，原创者就无法创造价值。他们无法销售订阅服务，也无法出售广告位。

The answer is I think people are trusting the eyes. They're reading this derivative content, and they're not going back to the original source. But the problem is, if you're not reading that original source, then the original sources have no way of generating value. They can't sell subscriptions. They can't sell ads.

他们得不到成就感。长此以往，这将扼杀内容创作的原始动力。这正是我们约十八个月前开始重点关注的问题。就在今天——7月1日，我们宣布将强硬屏蔽所有AI爬虫，除非它们愿意为内容创作者的作品提供实际补偿。

They can't get the ego hit. And and that over time is strangling the very incentives on why content is being created. And that's the problem that we started to really focus on about eighteen months ago. And and then just today, on July 1, we announced that we are, hard blocking AI crawlers unless they will actually compensate content creators for the content that they're creating.

好的。我们稍后肯定会深入讨论你们的技术解决方案。这部分内容即将展开，但现在让我们再谈谈这个问题。我记得你最近分享的数据是，Anthropic每发送一次点击就会爬取约6万页内容

Okay. And we're definitely gonna get into your technological solutions. Yeah. So that's that's coming, but let's talk a little bit more about this problem. So I think the number that you shared recently was Anthropic will crawl something like 60,000 pages

没错。

That's correct.

而OpenAI的数据大概是...你记得是1万次吗

For one click that's sent. And OpenAI was somewhere in the like 10 do you remember 10,000

是的，目前他们每给你带来一次点击，就要爬取1500页内容。

Yeah, it's 1,500 pages now for every one click that they send you.

不得不说，我很惊讶出版商现在才意识到问题，因为这些AI产品实际上还处于起步阶段。是的，我是说，考虑到互联网的规模，Anthropic Cloud的用户量其实非常少。OpenAI每周有5亿活跃用户，这已经相当不错了。

And I have to say, I'm surprised that publishers are seeing a problem now only because these AI products are really in their infancy. Yeah. I mean, Anthropic Cloud isn't used by very many people at all when you think about the scope of the web. OpenAI has 500,000,000 weekly active users. It's pretty good.

但与每天的网络流量相比，这根本不算什么。我想问题可能出在谷歌身上。所以请解释为什么生成式AI还在初期阶段，就已经对出版商造成影响了。

But really nothing compared to the amount of traffic that you see on the web every day. And I guess Google must be the problem. So just explain why this is already showing up for publishers, because this is the infancy of generative AI.

是的，我认为确实如此。但这是我们正在目睹的一场剧变。历史上首次，谷歌的搜索量出现了环比下降。

Yeah, I think it is. But it's one of these sea changes that we can just see happening. So again, for the first time in history, searches to Google actually dropped in the last period over period

这些都是你的数据？

This is all your data?

这是谷歌实际报告的数据，在苹果的诉讼案中也得到了印证——他们发现更多流量正在流向其他渠道。我同意这只是小幅下降，但我们看到的是趋势，是发展方向。就连谷歌本身也变得更像AI聊天机器人，而非传统搜索引擎。因此我认为，出版商现在就该警惕了。

This is Google has actually reported this, and it actually came out in the Apple, trial as well, where they're seeing more of this traffic actually going to other sources. And so I agree that it's it is a drop, but what what we're seeing is that is the trend. That is the direction that things are heading. And even Google itself is looking more like an AI chatbot and less like a traditional search engine. And so if that's the case, I think that the time for publishers to panic is now.

如果我们坐等流量被不断蚕食，情况只会越来越糟。长此以往，媒体行业将出现更多兼并，内容产出会越来越少，人们为争夺剩余流量会制造更多耸人听闻的标题。我们必须做出改变来支持出版商，因为我相信未来的网络将由AI驱动，而非搜索驱动。

If we wait where more and more traffic gets strangled, less and less is going to it. Again, I think that that's just going to mean that over time, we'll have more consolidation in the media industry. We'll have less and less content. We'll have actually more salacious, headlines as people are chasing the the content that is is left that's out there. And and we need to actually make a change to make sure that we can continue to support publishers because I do believe the future of the web is going to be an AI driven future, not a search driven future.

而这种AI驱动的未来，既不具备相同的激励机制，也无法支撑传统搜索网络那种商业模式。

And that AI driven future just doesn't have the same incentives and doesn't support the same business model that the old search driven web did.

好的。我要再深入探讨一下这个问题。

Okay. I'm going to poke at this a little more.

是啊。

Yeah.

你提到现在可以搜索谷歌查询CloudFare何时成立，就能直接获得答案。嗯。这是谷歌长期以来一直在做的事情。比如你可以问，马丁·路德·金生日是哪天？

You mentioned that you can now search Google when was CloudFare funded founded, and you'll get the answer. Mhmm. That's something that Google's been doing for a long time. You could ask, when was Martin Luther King Jr. Birthday?

甚至在生成式AI出现前，他们就能提供这类答案。那么是规模发生变化了吗？从消费者角度看，这或许是好事？毕竟在谷歌输入'Cloudflare何时成立'还要跳转到官网查看答案，而谷歌本可以直接显示结果，这确实挺烦人的。

Even before generative AI, they were giving you these answers. So is it that the magnitude has changed? And if so, from the standpoint of a consumer, could this be good? I mean, it's pretty annoying to type this question into Google, when was Cloudflare founded? And then have to click to Cloudflare's website to get the answer that Google could just surface for you.

很大程度上，互联网已沦为谷歌查询的服务工具，许多网站其实没有存在的必要了。

And so much of the web has sort of become effectively the service of Google queries, where websites don't really need to exist.

确实，这种现象已持续很久。如果看数据的话，十年前谷歌爬虫与点击量的比例是2:1。六个月前这个比例升至6:1，这都是因为答案框功能。而这段时间推出的AI概览功能，将这个比例拉大到18:1。就像温水煮青蛙。

Well, so absolutely, this has been happening for a while. And if you look at up until six months ago, the ratio ten years ago of crawls from Google to clicks was two crawls, one click. Six months ago, it was up to six crawls, one click, and that's all because of the AnswerBox. What the AI overviews, which they've rolled out over that time, have done is they've taken it now to 18 crawls to one click. So yes, it is a situation of, you know, the frog boiling in water.

但情况确实在持续恶化。整个媒体行业，出版商的生存越来越艰难。我担心的是，出版商在6:1时已举步维艰，18:1更是雪上加霜。当比例达到OpenAI的250:1或1500:1时他们就会消亡，而像Anthropic这种60000:1的情况则意味着彻底终结。

But that's it has gotten progressively worse. And I think across the media industry, it's gotten harder and harder to actually survive as as a publisher. And so what I worry about is, you know, publishers are struggling at they were struggling at six to one. They're struggling at 18 to one. I think they're dead at 250 or or 1,500 to one that we're seeing with OpenAI and completely dead at 60,000 to one that we're seeing with something like Anthropic.

因此，这就是事物发展的方向，也是一个挑战。我认为你在另一点上也完全正确，即这里的挑战在于这实际上提供了更好的用户体验。这就是为什么越来越多的网络将转向人工智能。能够输入内容并得到实际回应，而不是自己费力搜寻，这确实很棒。那是一种更好的用户界面。

And so that is the direction that things are going, and that's a challenge. I think you're exactly right on the other point as well, that the challenge here is that this is actually a better user experience. That's why more of the web is going to turn to AI. It is great that you can type something in and you can get back an actual response as opposed to having to hunt for it yourself. That's a better user interface.

所以事情绝对会朝那个方向发展。我并非在争论，也不认为有人会主张我们应该摒弃人工智能，或是回到谷歌那十条蓝色链接的时代。但我要说的是，驱动所有这些人工智能系统的燃料——谷歌之所以能告诉你Cloudflare何时成立、马丁·路德·金的生日等信息——是因为有人在从事原始内容的创作工作。这些原创内容正是驱动谷歌、驱动所有人工智能公司的动力源泉。

And so that absolutely is going that direction. I'm not arguing, and I don't think anyone is arguing, that we should just get rid of AI or that we should go back to sort of 10 blue links on Google. What I am saying, though, is that the fuel that runs all of these AI systems, the reason that Google can tell you when Cloudflare was was started or what Martin Luther King's birthday was or something like that is because somebody is doing the work of that original content creation. That original content is the fuel that fuels Google. It fuels all of the AI companies.

如果我们扼杀了这些平台的商业模式，如果我们剥夺了内容创作者的创作动力，那么我们最终也将扼杀人工智能系统。因为如果没有可供训练的内容，人工智能系统在这方面就会显得相当愚钝。因此，我认为大家都同意必须建立激励机制，确保内容创作者能持续获得回报。问题在于，这种激励机制应该是什么样子的？这正是我们一直在深入研究、试图解决的问题。

And if we strangle off the business model of those places, if we strangle off the incentives for content creators to create content, then we're actually going to end up strangling the AI systems as well. Because if there's no content to train on, then the AI systems are going to be pretty stupid for that. And so I think everybody agrees that there has to be incentives that allow content creators to continue to be compensated. The question is, what does that incentive structure look like? And that's, again, what we've been really spending a lot of time looking at, trying to figure out.

肯，我想再问你一个关于网页抓取的问题。

Ken, I just want to ask you one more question about crawls.

好的。

Yep.

我认为有时候抓取是为了将网站纳入你的搜索引擎

I think that sometimes crawl to put a website in your search engine

或者

or

搜索引擎中来自某个网站的页面。这些生成式AI机器人爬取数据是为了做类似的事情，仅仅是为了呈现这些页面上的信息吗？还是部分爬取行为是为了训练它们的模型？因为如果是后者，实际上问题没那么严重，因为这只是在为训练提供数据。我认为问题在于将那种直接查询回答的行为引入搜索引擎。

a page in your search engine from a website. Are these generative AI bots crawling to do something similar just to surface the information from these pages? Or are some of the crawling being done in service of training their models? Because if that's the case, it's actually not as big of a deal because it's just being fed into training. I think the problem is taking that direct query to answer behavior and bringing it into the search engine.

所以你知道这是为了训练还是仅仅为了呈现答案？

So do you know, is it training or is it just surfacing answers?

我认为这可以分为两个不同的部分。肯定有训练的部分，然后还有更接近搜索体验的部分。如果你熟悉这个，可能类似于Rag（检索增强生成）技术，通过实时数据来增强基础模型。但在这两种情况下，实际上都在让内容创作者付出代价——他们需要为这些流量买单。

So I think there are two different parts of this. There's definitely training, and then there's what is closer to a search like experience. If you're familiar with this, it'd be something like Rag or something where you're actually getting that real time data in order to augment the foundational model. I think in both cases, though, you're actually costing the content creator something. There is literally they're paying for that traffic.

他们需要承担爬虫程序抓取数据带来的服务器负载成本。这些被使用的数据本身就是知识产权，是内容提供者的核心资产。AI公司正在利用这些数据训练模型并从中获取价值。

They're paying for that load that the crawlers are pulling off of it. They're also it is the intellectual property. It's the data. It's the content of these providers that they're using to train the models. And so there's value that the AI companies are getting.

如果没有价值，他们就不会进行爬取，对吧？但现状是没有任何补偿或回报。谷歌早期至少还有交换条件：允许我们复制内容，作为回报会给你带来流量。现在的情况就像温水煮青蛙——

If if there weren't, they wouldn't be crawling. Right. But there's no return of any compensation or any reward. Again, in the old days of Google, the trade off was, let us copy your content, and in exchange, we'll give you traffic. What has happened is the frog is boiled in water.

所有人都说'让我们复制你的内容'，却完全不提供任何回报。所以我们主张的是：我们需要一个适应AI时代的新协议。这个协议应该规定——如果你从我创作的内容中获取价值，就应该以某种方式补偿我。单笔金额或许微小，但达到一定规模后，就能持续激励内容创作者产出。如果我们不这样做，不给予创作者经济动力，内容生产终将枯竭。

And now everyone is saying, let us copy your content, and we will give you nothing in return. And so what we're saying is simply, we need a better deal, a deal for a new AI driven future. And that should say, if you are getting value from the thing that I created, then you should compensate me in some way for it. And there may be tiny amounts of money, but at some scale, that actually turns into something that can allow a content creator to continue to have an incentive to create content over time. If we don't do that, if we don't give content creators the incentives to create content, they'll stop creating content.

你提出了一个关键点：当人们抱怨不常在LLM（大语言模型）中看到出版商内容时，其实有时看到的内容正是基于出版商作品的产物。即使属于合理使用范畴（因为内容经过转化），比如从大型科技公司或《纽约时报》爬取的数据被用来预测英语单词序列，最终生成关于夏令营的回答——这背后都有出版商的贡献。每次AI爬虫访问出版商网站，都应该付费。你们和维基百科有合作吗？他们一直在抗议服务器成本呈指数级增长的问题。

So I think you're bringing up a key point here, which is if people are like, well, I'm not necessarily seeing publisher content show up every time I'm on an LLM, what you are seeing sometimes is the product of the publisher that's And been used for even if it's under fair use, totally fine because it's being transformed and something crawled from big technology or The New York Times is now being used to help basically because they're just trying to figure out what word comes next in English language, give you an answer about summer camp. The publishers are actually enabling that. And every time an AI crawler hits a publisher website, they have to pay. And do you work with Wikipedia? Because they've been loud about this, that the server costs that they have to pay have increased exponentially.

但这些并非人类访客，而是爬取维基百科的AI机器人。

But those aren't human visitors. They're AI bots crawling Wikipedia.

那就谈谈这个。支持这种爬取行为本身就有实际成本。在我们讨论知识产权或其他问题之前，内容创作者和出版商就不得不承担这些成本。这纯粹是个公平性问题——凭什么要他们承担训练成本来养活那些

So talk about that. There's there's a real cost to just supporting this crawl. And before we even talk about intellectual property, before we talk about anything else, like the the content creators, the publishers are having to bear that cost. And so it is just a simple fairness level. Like, why should they be bearing the cost in order to train, You know,

这些本不该

these shouldn't

市值数十亿的AI公司。应该要有某种价值回馈。但我觉得问题还不止于此，甚至不必动用'合理使用'这类法律术语——目前这个概念本身就存在很大争议。

multibillion dollar, you know, AI companies that are out there. There should be some some value which is given back. But I think it's even beyond that. I don't even think we have to get to use use legal terms like fair use. And I think that's very much up in the air right now.

加州刚出现两个完全相反的判例：训练使用内容是否构成合理使用？不同法院会有不同裁决，就像抛硬币一样。我认为更根本的是：如果你创造价值，就应该获得相应补偿；如果有人将成本强加于你，你就该有权收费抵消。

We literally had two different California cases that came out on both sides of that issue. Is training on content fair use or not? And I think it's going be a coin flip where different courts are going to say different things. And I don't think it's a clear answer there, but I think it's a more fundamental thing, which is if you're doing something to create value, you should be getting some sort of of compensation for that. If if somebody else is is imposing a cost on you, you should be able to charge them to offset some of that cost.

如果不愿支付费用，一开始就不该获取你的内容。现在大家都聚焦法律诉讼——比如纽约时报在起诉——但我觉得在讨论法律问题前，首先应该通过技术手段让内容创作者重获控制权，自主决定是否开放访问或收费。

And if something's not if they're if if someone's not willing to pay that, then they shouldn't be taking your content in the first place. Up until now and and everyone's focused, you know, New York Times is suing, and I mean, bunch of people are doing that. Everyone's focused on the legal issue. I actually think that before we even get to the legal issue, this first step is actually take the technical steps to give content creators back control over the content that they're creating and let them have the choice on, do I want to give access or not? Do I want to charge for this or not?

如果操作得当，应该形成内容创作者与AI公司的交易市场：创作者说'我认为这内容极有价值'，AI公司评估后报价，双方可能达成清算价格，也可能无法达成。

And then done correctly, there should be a marketplace where content creators and AI companies come together and say, Hey, I created this piece of content. I think it's super valuable. And the AI company says, Yeah, maybe it is or maybe it's not, but here's what we're willing to pay. And maybe they meet the clearing price. Maybe they don't meet the clearing price.

但这个市场必须存在，否则就无法传递价值。无法从内容创作中驱动价值。我必须再次强调这一点。如果我们不给内容创作者提供创作内容的激励，他们就会停止创作。

But that marketplace needs to exist because otherwise, there's no way to convey value. There's no way to drive value from content creation. Again, I just need to hammer this point home. If we don't give content creators an incentive to create content, they'll stop creating content.

顺便说一句，听起来你并非AI技术的怀疑者。你认为这种非AI生成式AI技术将会——

And it sounds like, by the way, so you're not a skeptic of the AI technology. You believe that this Not AI generative AI thing is going to

不仅如此，显然它将成为未来网络的交互界面。我们将从过去主导互联网的搜索界面，转向未来网络的交互界面，那将很大程度上是AI。我相信AI会变得越来越好。实际上我认为，如果方法得当，内容的创作方式可以提升AI质量，并且你可以为此设立激励机制。我担心的是，为了让AI变得更好，你必须拥有原创内容。

only that, mean, is already clear that it's going to be the interface of the future of the web. So we're going to move from what has been the dominant Internet face of the future of the past of the web, which was search, to what the interface of the future of the web was going to be, which is very much going to be AI. So I believe AI is going to get better and better and better. I actually think that done correctly, content can be created in such a way that will make AI better and that you can create incentives for for doing that. What I worry about is in order for AI to get better, you have to have original content.

人们必须走出去创造这些内容。而目前，我们正在扼杀所有内容创作的激励，这不仅伤害内容创作者，最终也会伤害AI公司。

People have to be going out and creating that. And right now, we're strangling off all of the incentives for that content creation, which not only hurts content creators, it will ultimately hurt the AI companies as well.

Capital One的技术团队不仅在讨论多代理AI，他们已经部署了一个。它叫做聊天礼宾服务，正在简化购车流程。通过自我反思和分层推理结合实时API检查，它不仅帮助买家找到心仪的车辆，还能协助安排试驾、获得融资预批准以及估算置换价值。

Capital One's tech team isn't just talking about multi agentic AI. They already deployed one. It's called chat concierge, and it's simplifying car shopping. Using self reflection and layered reasoning with live API checks, It doesn't just help buyers find a car they love. It helps schedule a test drive, get preapproved for financing, and estimate trade in value.

先进、直观且已部署。这就是他们的技术实力。这就是Capital One的科技。昨晚我与一位从事大型语言模型数据标注或数据创建工作的人交谈。在期待这次对话时，他说，有一天你们的工作可能会看起来几乎和网络出版商完全一样，你们可能会雇佣博士，让他们撰写信息。

Advanced, intuitive, and deployed. That's how they stack. That's technology at Capital One. So I was speaking with someone who works in data labeling or data creation for large language models last night. In anticipation of this conversation, was like, you know, one day what you're doing might look almost exactly like what web publishers are doing, where you might be hiring PhDs and having them write their information.

你们直接将那些信息输入到LLM的训练集中。可能会有历史学家。比如一个世界历史网站，为这个网站撰写网页的历史学家们，也许有一天他们会撰写那些世界历史文章，但不是发布到网上，而是出售或直接输入到ChatGPT中。如果网络消失，只剩下内容创作者向大型语言模型出售内容，我们会失去什么吗？

You feed that right into an LLM's training set. And there might be, let's say, historians. If you take like a world history website, the historians that are writing the web pages for that world history website, they must be just like maybe one day they're going to be writing those world history articles and instead of publishing them to the web, selling them or feeding them right into ChatGPT. Do we lose anything if the web goes away and it's just content creators selling stuff to large language models?

是的。我认为《黑镜》式反乌托邦未来的本质并非内容创作停滞、记者或研究者消失，而是可能回归类似美第奇家族时代的格局——由五家大型AI公司掌控，各自豢养记者团队、研究团队，成为实质上的知识垄断机构。这些公司会为旗下学者支付薪水，可能分化成保守派AI公司和自由派AI公司。这种模式其实长期存在于媒体和信息控制的历史中。

Yeah. You know, I think the black mirror kind of dystopian future is is not that, you know, content will stop being created and journalists will stop existing and researchers will stop existing. I think the black mirror future is that we actually go back to something like the time of the Medicis, where we have maybe five big AI companies, and they each employ a set of journalists and a set of researchers and a set of set of folks that they become effectively the institutions of knowledge and they they have they have salaries for all their their academics that are on staff. They probably each have different, you know, maybe one of them is the conservative AI company and one of them is a liberal AI company. You can again, you can very much see that has actually been the natural state of media and the natural state of controlling information for quite some time.

可以预见所有研究资源都会向各家AI公司集中，学者们终将成为OpenAI、Anthropic、谷歌或微软的雇员。这是糟糕的结局，因为网络本应促进信息民主化。我们需要在学者被AI公司收编之前建立新机制——让AI公司为真正提升其模型价值的内容付费。

And you could imagine that all of that research actually consolidates behind each individual AI company. And every different academic out there is basically an employee of OpenAI or Anthropic or Google or Microsoft. I think that's a pretty bad outcome because, again, I think that the web has been so amazing at distributing and democratizing access to information that I think we want to create that incentive. And so I think what we're trying to do is say, what's the step, a few steps before the sort of all the academics are employed by one of the AI companies? And I think the answer is you allow the AI companies to pay for the content that is actually valuable to them, that fills in their models and makes their models better.

同时要激励独立记者和研究者持续产出能增强AI的内容。虽然难以实现，但我理想中的未来是：人类应免费获取内容（当前付费墙实在太多），而AI必须高价购买——因为AI每消化一次内容，都是在为数十万乃至数百万人类服务。

And then you create incentives for independent journalists, independent researchers to actually be able to create that content, to augment those AIs, while still being valuable. This won't happen, but my sort of optimistic version of the future is humans should get content for free again because we've kind of paywalled way too much, frankly. And robots should pay a ton for it. Because again, every time a robot ingests something, it's in service of hundreds of thousands, if not millions of different humans. So robots should pay for that content.

我们应该重建人类免费获取知识的体系。尽管前路艰难，但这才是最理想的未来图景。

We should get back to a place where then humans get that for free. Again, that's I think it's going be hard for us to get there. But that's, again, the future that I think is actually the kind of optimal future.

持批判态度的听众可能会说：马修，出版商依赖网络流量是方向性错误，靠CPM分成的广告模式早已过时。我们曾有嘉宾直言，若仍迷信流量变现，他早该破产。真正需要的是愿意订阅简报、收听播客、参加活动的忠实用户——这意味着我们已超越流量换美元的旧模式，因此AI构不成生存威胁。你如何回应？

So someone hearing you and looking at this through a critical lens might say, look, Matthew, publishers, depending on web traffic, are barking up the wrong tree, that selling eyeballs for CPM fractions has not been a good business for a long time. In fact, we had a guest on the show that recently said, listen, like I'm he's a journalist, but he's like, if I thought that traffic was the way to go, I'd be out of business a long time ago. And what you really need is an audience that will, let's say, subscribe to your newsletter or listen to your podcast, maybe come to your events. And we've already moved past this business model of trading traffic for dollars, in which case this isn't an existential threat. What would you say to that?

订阅制本质仍是流量变现，只是把广告收入换成订阅收入。当AI公司能抓取并摘要你的播客内容时，谁还会付费订阅？如果AI助手能直接告知'这期播客/简报的所有要点'，用户何必亲自获取原始内容？

I think, I mean, even then you're still trading traffic for dollars. You're just trading it for subscription dollars, not ad dollars. That will go away as well because what will happen is the AI company will ingest the podcast and then summarize it on their page. And why would they ever buy a subscription to your podcast? Why would they ever sign up for your newsletter if their AI agent can just simply say, tell me everything that was relevant in this particular podcast or newsletter?

因为收听过程本身具有娱乐价值。人们部分是为了享受学习过程中的休闲体验才消费内容。

Because there's an experience of listening that's enjoyable. People do that in some part for for the entertainment and the leisure value. Think that's how they learn.

我认为AI公司在打造这种体验方面会做得非常出色

I think the AI companies will do a very good job at creating that experience

也是。所以你觉得他们会直接形成竞争关系吗？

as well. So think they'll just create like competing?

噢，毫无疑问。你知道，这是肯定的。

Oh, absolutely. You know, for sure.

我过去一直觉得这简直是天方夜谭的疯狂想法，直到我听了Notebook。是啊。而且我YouTube页面上有好几个人留言问，你是不是授权Notebook使用你的声音了？我说没有。但你会这么问本身就挺让人担忧的。

I used to think that this was like such a pie in the sky and lunatic idea until I listened to Notebook. Yeah. And like we've had multiple people on my YouTube page be like, did you license your voice to Notebook? And I'm like, no. But the fact that you're saying that is pretty concerning.

完全同意。我再次强调，这就是不可避免的未来。我们将需要高度个性化的播客，用我们觉得最令人安心的声音来呈现。而AI会为我们创造这些内容。当然，它们仍需依靠原创内容创作者提供创意、话题和每日新闻作为素材来源。

Totally. And again, I think that's the inevitable future. We're going to want to have hyper customized podcasts that are in exactly the voice that we find the most reassuring. And AIs are going to create that for us. And again, they're going to be fed by original content creators that are out there that give them the ideas, give them what what to talk about, give them the news of the day.

我认为我们必须超越订阅这种商业模式。我们需要建立新机制，让内容创作者能因其创作获得报酬。我的理解是：每个大语言模型都像一块瑞士奶酪——虽然内容丰富，但存在巨大空洞。真正对它们有价值的，正是那些能填补这些空洞的内容。

What what I think is we have to move even past the business model of subscriptions. We've got to get to something else where you as a content creator are being compensated for the content. The way I think about it is every one of these LLMs is a little bit like a block of Swiss cheese. They've got a lot of stuff there, but there are big holes that are in it. And content that is value for valuable for them are the ones where they actually fill in those holes in the Swiss cheese.

因此我设想未来会出现这样的场景：AI能主动识别自身知识体系中的'瑞士奶酪空洞'，然后让内容创作者针对性地生产填补内容。我最喜欢的例子是：几周前我在斯德哥尔摩与Daniel Ek会面——在规模化补偿创作者方面，没人比Daniel做得更多。作为Spotify创始人，他们在这方面成就斐然。在我们长谈中他提到：'你知道Spotify的做法之一吗？我们会实际分析用户在平台上的搜索行为...'

And so what I would imagine in the future is that you're able to actually surface what are the places where there are holes in the Swiss cheese as an AI and then allow content creators to create content that fills that in. My favorite example of this is I was I was in Stockholm a couple of weeks ago meeting with Daniel Ek because there really is nobody who has done more to compensate creators at scale than Daniel. Daniel's the founder of Spotify, and they've done just an amazing job at doing this. And he told me a story in our long conversation. He said, you know, one of the things that we do at Spotify is we actually take the searches that people run at Spotify.

你知道吗，有些需求就像——我想要一首雷鬼节奏的歌，唱的是当你妹妹开走你的车时那种糟透的感觉。

You know, there are things like I want a song with a reggae beat about how much it, you know, sucks when your, you know, your sister runs away with your car.

这事确实发生过。

It's happened.

是啊，总之。结果发现平台没有优质内容来填补这类空白。而有些内容创作者年收入数千万美元，专门为当前搜索结果不佳的需求创作内容——因为Spotify会展示那些未被满足的需求清单。我觉得这其实很美，非常了不起，他们正在揭示人类真实的需求所在。

Yeah, whatever. And it turns out that they don't have good things to fill that in. And there are content creators out there that are making tens of millions of dollars a year just creating content for those searches that don't have good results right now because Spotify surfaces that list of things where they don't they don't give those results. I think that that's actually beautiful. I think that's actually really amazing where they are showing where is there something that there is human need for?

然后我们如何创造内容来满足这种需求，并通过他们的平台实现盈利？我认为AI领域也存在同样机遇，这些AI能明确告知：'我能判断这条新内容对我的价值'。你可以据此排序，从而创建交易市场——AI会说：'听着，这条新信息价值连城，我愿意为此付费'。如果运作得当，这将推动更多原创内容的生产。

And then how can we actually then create content to fill that human need and then monetize it through what they're doing? I think the same opportunity exists in the AI space, where these AIs actually are able to say, I can tell you how valuable this new piece of content is for me. And you can rank it. And then that allows you to create a marketplace where they can say, listen, that new piece of information is so valuable that I'm willing to pay you for that. And I think that done correctly, that then gets us to more original content creation.

这会减少跟风模仿式的新闻报道，在研究领域也同样——或许能让我们进入一个重视原创研究并获得回报的境地，而不是靠耸人听闻取胜。

It gets us to less sort of me too copycat style journalism. Same thing in research gets us to maybe a place where we're we are doing original research and getting rewarded for being more original as opposed to being more salacious.

对，YouTube也有类似功能，比如'洞察'或'灵感'标签页。

Yeah, it's YouTube has a similar thing where there's an insights or an inspiration tab.

没错。

Yep.

他们会给你标题、描述和缩略图。然后说，人们正在搜索这个。你去把它做出来。

And they give you like the title, the description and the thumbnail. And they're like, people are searching for this. You go out and make it.

没错。我认为这实际上正是让人类变得更好的宝贵之处，而不是又一篇追逐最耸人听闻标题的故事。

Yeah. That's exactly right. And I think that that's actually is that incredibly valuable thing that that's making humanity better as opposed to, you know, yet another story that's just chasing, you know, the most salacious headline that you you can get.

所以你在谈论出版商可能向AI出售爬取权限的构想。这还假设内容是稀缺的。我想和你探讨另一个观点：如果我们保持现有内容量不变，这确实是个好主意。但我们现在看到的是生成式AI引发的内容创作大爆发，这有点讽刺。

So you're talking about this idea where publishers might sell the ability to crawl to AIs. That is also assuming that content is scarce. And so I want to run this other idea by you, which is that if we had the same amount of content that we have today, that's a great idea. But what we're seeing now is this explosion of content creation that's made through generative AI. It's kind of funny.

每次看到我们讨论的这些推荐——YouTube之所以做这些推荐，显然是因为存在流量红利。我敢肯定现在就有YouTuber把这些输入ChatGPT生成脚本，用VO3和Google处理后发布视频来赚取流量。我们正处在这场内容大爆发的中心。实际上你的数据应该比我推测的更准确——这几乎像是对网络的DDoS攻击。如果内容创作能力受限于人类生产力，你才有筹码与AI公司谈判。

Every time you see these suggestions that we're talking about, YouTube is making these suggestions because clearly there's traffic to be had. I'm sure there are already YouTubers today that are feeding that into ChatGPT, spitting out a script, running that through VO3 and Google, and then posting the videos and cashing in on traffic. So there's just going to be and we're in the middle, I believe, of this explosion of content. Actually, you probably have better data on that than my suppositions. It almost feels like a DDoS of the web where if the ability to create content is constrained by a human's ability to create content, then you have something to bring to these AI companies.

但如果人机协作内容成为常态，数量将极其庞大。届时即便你生产高质量内容，对这些生成式AI公司也无足轻重。你怎么看？

But if human plus bot content starts to become the norm, there's going to be so much. Then even if you're creating high quality stuff, it's not going to matter very much to these generative AI companies. What do you think about that?

我认为首先，纯AI生成内容存在明显问题。大量研究表明用AI数据训练AI就像迈克尔·基顿的老电影《复制娇妻》——每次复制都会更劣质。这种情况短期内不会改变。未来机器人或许能进行实地采访或深度研究？当然可能。

I think it's still mean, so first of all, I think that there's the pure AI generated content. There's lots of research that shows that training AI on AI data is sort of like that old Michael Keaton film Multiplicity, where basically every copy of something gets worse and worse and worse. And and again, that that feels like that's going to still be still be the case for quite some time. May in the future robots be able to go out and do, you know, interesting reporting from the field? May they be able to do, you know, interesting research for sure?

但就目前而言，只有记者和研究者能产出的原创内容、深度研究和独特见解，才是填补AI认知漏洞的关键。那些海量低质内容？如果评分系统得当，它们就该得到最低限度的回报。我爱好滑雪，所以每年有段时间住在犹他州帕克城。

But today, I think that that interesting research, that interesting original content, that interesting insight that that comes from the work that right now only journalists and researchers and others can do is still the most important thing for filling in those gaps in the in the Swiss cheese of AIs. What is just, again, high volume, low value content? My hunch is that that's if we score it correctly, going to be exactly what it what it is, which is low value content. And so it should it should be rewarded very minimally. I like to ski, so I live part of the year in Park City, Utah.

你来对地方了。

You're in the right place.

我在乎。我非常关心雪情预报。犹他州有位叫埃文·塞耶的预报员，他写的天气预报精确到令人难以置信，会直接告诉你这条雪道降雪量多少、那条雪道降雪量多少。我甚至付费订阅他的内容，因为对我而言这价值连城。

I care. I care enormously about the snow forecast. There is a forecaster in Utah named Evan Thayer. He writes these incredibly precise weather forecasts where he will literally tell you it's going to snow this much on this run and this much on this run. And again, I actually pay for his content because that's super value for me.

未来我会更愿意为那些获得埃文内容授权的AI付费，而非使用没有这些内容的AI。因为这些内容对我而言极其有用、独特且珍贵。我认为随着AI系统增多，人们会愈发追寻原创性内容——这些才是AI最愿意付费获取的东西。这其实很美好：我们不再激励人们制造耸动标题追逐流量，而是激励他们填补知识体系中瑞士奶酪般的空白。

I am going to be more willing in the future to pay for an AI that has actually licensed Evan's content back from him than I would pay for an AI that doesn't have that content. Because again, that content is going to be super useful and unique and valuable to me. And so I think actually what it will do is we as we have more AI systems that are out there, is it will cause you to look for more original creative content. And that's going to be the thing that the AIs are going to be the most willing to pay for. And that, again, I think is actually a beautiful thing where we are instead of creating incentives to create more and more salacious headlines and chase traffic, we're creating incentives to create knowledge that fills in those places in sort of the Swiss cheese where there might be holes.

总体来看，所有AI可能相当准确地反映了人类知识图谱。如果我们能为其评分，指出人类知识的缺口和需要填补之处，这实际上为创作者提供了丰富的内容创作方向，以推动人类认知边界。

Taken in aggregate, all of the AIs are probably a pretty good representation of what human knowledge looks like. And so if we can score them and say, Okay, here are the gaps in human knowledge and and here are the places we need to fill in, that actually gives a really rich place for creators to look to create content which advances human knowledge.

DeepMind正在研发天气预报系统。你提到的犹他州预报员埃文·塞耶这个例子——我们距离直接对AI说'你接入DeepMind气象模型，我想今天滑这条雪道，天气如何？'还有多远？

So, you know, DeepMind is working on weather forecasting right now. This example that you gave of Evan Thayer, the forecaster in Utah, are we that far away from just telling an AI, hey, like you're you're tapping into the DeepMind model and weather forecasts. I want to ski this route today. What's happening?

我认为还有相当距离。但埃文总能通过AI工具结合他的本地知识做得更好。

I think we're probably pretty far away from that. But and but but again, I think Evan is going to be always better using the tools of A. I. Plus his local knowledge to make this better. A.

AI将成为创意工作者的工具，帮助他们更好地讲故事、获取信息、深化研究。至少短期内，我对纯粹通过生成内容训练创造真实价值持怀疑态度。

I. Becomes a tool that creative people use in order to tell stories better, get better information, do more research. And and again, I I am skeptical that in the in the short term, at least, that we're going to have, real value that is created by by training on purely generated content. Okay.

现在到底发生了什么，为什么会这样？在《连线》杂志，我们每天痴迷于深挖这些问题，或许你也是。我是凯蒂·德拉蒙德，《连线》全球编辑总监，正在主持我们的新播客系列《大访谈》。每周，我会与一些最有趣、最具挑衅性和影响力的人物坐下来聊聊，正是他们塑造了我们的当下。《大访谈》的对话充满乐趣。

What the hell is going on right now, and why is it happening like this? At Wired, we're obsessed with getting to the bottom of those questions on a daily basis, and maybe you are too. I'm Katie Drummond, the global editorial director of Wired, and I'm hosting our new podcast series, the big interview. Each week, I'll sit down with some of the most interesting, provocative, and influential people who are shaping our right now. Big interview conversations are fun.

我想要一条鲨鱼

I want a shark that

能吞噬互联网的那种。让它全部瘫痪。毫无过滤，无所畏惧。

That eats the Internet. That turns it all off. Unfiltered and unafraid.

所以在很多方面，我尽力成为你在网上看到的那种难以想象的、源源不断的反动内容的解毒剂。

So in a lot of ways, try to be an antidote to the unimaginable faucet of reactionary content that you see online to the best of my ability.

每周，我们将为你提供这个时代最极致的奢侈品——意义与语境。真或假？你，布莱恩·约翰逊，约翰逊，那个坐在我对面的男人，总有一天，在未来的某个尚未确定的时刻，你会死去。假的。详细说说。

Every week, we're going to offer you the ultimate luxury of our times, meaning and context. True or false? You, Brian Johnson, Johnson, the the man man sitting sitting across across from from me, me, one day, at some point, as of yet undefined in the future, you will die. False. Tell me more.

现在就在你收听《连线》杂志《诡异谷》播客的同一平台收听《大访谈》。订阅或关注你获取播客的任何渠道。

Listen to the big interview right now in the same place you find Wyrde's Uncanny Valley podcast. Subscribe or follow wherever you get your podcasts.

我们已经讨论了你的解决方案。让我们稍微深入探讨一下技术层面。我们是一档科技播客，理应如此。Cloudflare是一家安全公司，帮助网站在面对各种威胁时保持在线。没错。

So we've talked about your solution. Let's dive into the technological side of it a little bit. We are a tech podcast, so we should do that. So Cloudflare, security company, helps websites stay up on the web despite all the threats. Yep.

首先，让我们谈谈您所看到的对网站的威胁，是谁在试图让它们下线。这方面发生了什么？

And let's just, at the very beginning, talk about the threats that you see to websites, who's trying to take them down. What what's happening on that?

是的。保护网站是我们业务的一部分，保护员工在互联网上的活动也是。Cloudflare本质上是一个网络，它具备性能、可靠性、安全性、可用性和隐私保障，坦白说，如果当初我们知道互联网会变成今天这样，这些本应是互联网的基础。但在六七十年代和八十年代制定这些协议时，我们并没有考虑到这些。

Yeah. So we protecting websites is is part of our business. So is protecting employees as they go out across the Internet. So we we flow is fundamentally kind of a network that is built with all the performance, reliability, security, availability and privacy guarantees that frankly, Internet should have been built with had we all known what it was going to become. But but obviously, back in the sixties, seventies and eighties, when we were laying down all these protocols, we didn't think about those things.

因此，Cloudflare基本上是在对互联网进行逆向工程，以在现有基础上提供性能、可用性、安全性、可靠性和隐私保障。如今，Cloudflare的主要用途之一是，当您将网站、网络应用或任何内容上线时，确保它们免受各种威胁。我们看到的威胁有哪些？每天我们都在与中国政府、俄罗斯政府、朝鲜等对抗。

And so Cloudflare is basically reverse engineering the Internet in order to give it those performance availability, security, reliability and privacy guarantees on top of what what is there. And so today, one of the main uses for Cloudflare would be to you have you're putting a website or a web application or anything online. You want to make sure that it's safe from different sorts of threats. And so what are the threats that we see? I mean, every day we go to war with the Chinese government, the Russian government, the North Koreans.

我的意思是，每个人都在试图入侵我们的客户，因为我们的客户是谁？世界上一些最大的银行、一些最大的政府，它们都不断受到这些组织的威胁和攻击。媒体公司实际上只占我们业务的一小部分。我们有一些媒体客户，但比例不大。真正开始于18个月前的是，这些公司说：‘嘿，我知道我们雇你是为了阻止中国黑客，但现在我们面临一个新威胁。’

I mean, everyone is trying to hack into our customers because who are our customers? Some of the largest banks in the world, some of the largest governments in the world, and they are all constantly under threat and constantly under attack from these these organizations. The media companies actually were a pretty small part of our business. We had some media companies that used us, but it wasn't a big piece of it. What what happened starting really eighteen months ago is that those companies said, Hey, I know we hired you in order to stop the Chinese hackers, but we have this new threat that's there.

坦白说，我最初的反应是：出版商们总是在抱怨新技术。比如，到底发生了什么？他们反复说：‘拉数据，拉数据，拉数据。’直到我们真正看到数据，看到AI公司在不提供任何价值回报的情况下获取内容，它们实际上增加了巨大负载，有些情况下甚至因为流量过大导致整个网站崩溃。

And frankly, my initial reaction was publishers. They're always whining about the next new technology. Like, what's what's what's going on? And and and and over and over, they said, just pull the data, pull the data, pull the data. And it was only when we actually saw the data and we saw how AI companies were taking content without giving anything of value in return, that they were actually adding enormous amounts of load, and in some cases, taking whole websites down because of the amount of traffic that they were sending to it.

他们基本上是对网站进行了DDoS攻击。

They basically DDoS the websites.

对网站进行DDoS攻击。你知道，这很疯狂。虽然不是故意的。但正是在那时我们说：‘听着，也许我们可以做点什么。’起初，很多出版商都说：‘这太难了。’

DDoS the websites. You know, it's crazy. Not intentionally. But but that was the point at which we said, listen, maybe there is something that we can do here. And and, you know, at first, I think a lot of the publishers were saying, oh, this is this is so hard.

我们根本没办法阻止它。你知道，帕洛阿尔托住着这些书呆子，他们聪明绝顶。我们究竟能拿他们怎么办？我一直反复强调，我们只能和中国黑客开战，好像我们能用一个C公司阻止几个书呆子似的。我觉得这个观点花了不少时间才真正被理解。

There's no way we can stop it. You know, there are these these nerds and they live in Palo Alto and they're so smart. What are we ever going to possibly do about it? And, and and I just kept saying, guess we we go to war with the, like the Chinese hackers, like we can stop some nerds with the C corporation. And I think it took a while for that message to really get through.

但现在不同了，看到全球绝大多数主要出版商都表态说'我们需要改变模式，我们的内容应该得到补偿'，这真的很有成就感。Cloudflare在技术解决方案上的思路是正确的。

But now that it has, you know, it was it's been really rewarding to see that the vast majority of the world's publishers, major publishers, have said, we need to change the model. We need to be compensated for our content. And Cloudflare has the right idea in terms of the technical solution to do that.

顺便说一下，各位，这是家市值600亿美元的上市公司，算是纽约证券交易所规模较大的网络安全公司之一。但我想问你，既然我们要讨论这个技术方案，你刚才说的很有意思——如果这些AI机器人不仅抓取出版商内容，还开始吞噬银行网站呢？这会不会形成某种天然对立？因为如果所有信息都通过ChatGPT获取，你们保护的其他网站可能就不再需要你们的服务了。

By the way, folks, 60,000,000,000 company listed publicly. So it's one of the bigger cybersecurity companies on the New York Stock Exchange. But I want to ask you, okay, so we're going to get into this technological solution, what you said is interesting because what if these AI bots you ever think there's a war where these AI bots ingest not just the publishers but the banking websites as well? Like, you like a natural enemy to having everything go through that? Because if everything goes through ChatGPT, then these other sites that you secure might not need your services.

我认为，总归需要有个守门人来管控智能体和其他程序如何访问各类在线服务。不同场景面临的挑战也不同：银行可能需要设置防护栏，确保是真实客户在访问账户，交易必须经过真人授权才能执行。

I think I mean, still there there's going to be some gatekeeper for how agents and other things access various services online. And I think the the challenges in each of those cases are different. In the case of a bank, you might want to say, I want to have guardrails that are in place. I want to make sure that this is actually a customer that's accessing account. I want to make sure that they're, you know, they they can they can only conduct transactions that have been authorized by an actual human being or or something like that.

Kleffler实际上提供了这些防护机制，让银行可以安全地向AI开放基础设施。出版商面临的是另一种挑战。以我们为例，官网上有大量开发文档——我们希望这些能被AI收录，当有人在编程平台询问'我想用Cloudflare构建X/Y/Z功能'时，AI就能直接输出解决方案。

Kleffler actually provides those guardrails and and and makes it so that a bank can say, I want to expose my infrastructure to AI, but do it in a way which is safe and secure. I think publishers have a different different challenge. And so, you know, in our case, a way of thinking is is like we have a whole bunch of developer documents which are on our website. We want those to be an AI. We want coding platforms when someone says, Oh, I want to use Cloudflare to build X, Y or Z for it to be able to spit that out.

我们实际做的是精准识别网页上那些具有变现特征的页面——通常看两点：是否设置了付费墙？是否有横幅广告等广告单元？检测到这些特征就会默认拦截。但同时我们也承认AI的价值，要确保AI获取的是人们真正希望它掌握的数据。

What we've done is we've actually tried to identify with real narrow precision what are those pages that are on the web that have some indication that they are going to be monetized. And generally, is look at is it behind a paywall or is it does it have some sort of an ad unit on it, like a banner ad or or some sort of ad that's there? If we detect that, then we're blocking it by default. But we're not doing this. Again, there's value for AI, and we want to make sure that AI is actually getting the data that people want to have have in it.

比如《纽约时报》的'关于我们'页面应该被AI收录，但突发新闻类的新文章就应该受限——除非AI公司为这些内容付费。

So like the about us page on The New York Times probably should go into the AI system. But in brand new article, you know, with breaking news probably should be restricted. And again, unless the AI company is actually paying for that content.

我想问的是，如果所有东西都输入ChatGPT，那么在媒体世界之外，还剩下多少空间需要你们来保护创新思维？

I guess the way I want to ask it is if everything goes into chat GPT, what's less what's left for you to protect thinking outside of the media world?

嗯，再次强调，我认为80%的AI公司都是我们的客户。所以我们也在保护他们。是的。

Well, again, I think that 80% of the AI companies are customers of ours. So we protect them as well. Yeah.

好的，听起来不错。我只是想问问这个问题。我对此感到好奇。不过我们现在继续讨论吧。

Okay, sounds good. Just wanted to ask that. I was curious about it. But let's talk. Okay, now.

所以你们要开发一个技术解决方案来阻止爬取？是的。那么robots.txt文件——就是那种你放在网站头部的代码，如果你不想被爬取的话——它之前不起作用吗？

So you're going to build a technological solution that will block crawling. Yes. And so robots. Txt, which is this code that you put in, like, the header of your site, if you don't want to be crawled, that wasn't working?

是的。我认为robots.txt存在两个问题。首先是有些人直接无视它。如果你无视它，那你还是可以随意爬取。

Yeah. I mean, I think robots. Txt has two problems. The first is some people just ignore it. And so if you ignore it, then you can still crawl all you want.

甚至有些大型正规公司也完全无视robots.txt。而我们真正擅长的是能够明确指出：看，这是robots.txt的规定。你们实际遵守这些规则了吗？

And there's some just there's some even some big legitimate companies that completely ignore robots. Txt. And we're really good at basically being able to say, okay, here's what robots. Txt says. How are you are you actually following what those those, what what sort of the rules of the road are?

如果答案是肯定的，那么robots.txt就是个很好的解决方案。但如果有人无视它，我们就需要设置额外的技术屏障来限制他们的访问。这正是我们在做的。robots.txt的第二个问题是...

And if the answer is yes, then robots. Txt is a great solution. But in the cases where somebody is ignoring it, then we need to actually put in place additional technical barriers to restrict their their their access. And so that's exactly what we're doing. The second problem with robots.

Txt的问题在于粒度不够。以谷歌爬虫为例，它至少执行五项不同功能。首先是检查页面广告，确保若投放宝洁公司产品广告时，不会出现在色情网站等不当位置，即进行品牌安全审查。

Txt is it's not granular enough. So take the Google bot, for example. Google's crawler does five different things, at least. One is it checks if you have an ad on a page and make sure that if you're putting an ad for Procter and Gamble, Procter and Gamble product up, it's not against a pornographic site or something like that. So it does brand safety checks.

第二项功能是为传统搜索（即那些蓝色链接）建立索引而爬取。第三是为答案框生成答案内容。第四是为新推出的AI概览功能采集数据。第五则是为向Gemini模型输入内容而抓取信息。

The second thing that it does is crawls to index for traditional search, the 10 blue links that are out there. The third is that it crawls to create answers that are in the answer box. The fourth is that it crawls to create answers that are in the AI overview, the newer thing that they've rolled out. And the fifth is that it crawls in order to ingest content in order to put it into Gemini.

这爬取量可真不小。

It's a lot of crawling.

所有这些爬取都通过单一爬虫完成。出于多种原因，他们不愿将其拆分为多个爬虫。但现状是网站只能二选一：要么完全屏蔽谷歌（这意味着无法投放广告、不显示在搜索结果中，也不会出现在AI概览或Gemini里），要么使用他们最近新增的小标记——该标记仅声明不将数据用于Gemini，但仍会出现在AI概览和答案框中。

A lot of crawling all through one crawler. And for lots of different reasons, they don't want to split that out into into various crawlers. But right now, they basically make you have a choice. They say you can either block Google entirely, in which case you can't run ads, you don't appear in search, but you don't appear in the AI overviews or Gemini or other things. Or they've recently added a tiny flag, which which basically just says, I'm not going to use this data just for the Gemini piece, but you still appear in AI overviews, you still appear in Answerbox.

我们认为需要更细粒度区分——获取内容与转化内容应有不同。许可协议应规定：未经许可不得进行内容转化，而品牌安全审查和传统搜索的内容获取则另当别论。为此我们正与IEDF及监管机构合作，提议扩展robots.txt协议以实现这种粒度控制。这将使我们能进一步验证机器人行为是否合规。

We think there needs to be more granularity where there is a difference between taking content and transforming it. And a license should say, you can't do that without my permission, versus just taking that content in order to do brand safety checks, taking that content in order to do traditional search. And so what we've proposed, and we're working with the IEDF as well as regulators, is extensions to robots. Txt to give it that granularity. And that actually then allows us to further test to watch, you know, does this robot behave in an appropriate way?

若验证通过，该机器人可能获得更多网络操作权限；若未通过，我们将设置更多限制与屏障来阻止这些行为失当的机器人。

If the answer is yes, then maybe it gets more permissions to do things online. If the answer is no, then we will put more restrictions and blockades in place to stop what are, again, badly behaving robots.

所以你们现在要做的，是在此基础上再筑起一道技术防火墙。

So what you're going to do now is, in addition to that, put a wall up technological wall.

没错。禁止爬取。抱歉。你们中太多人不遵守Robots TXT协议。禁止进入。

That's right. No crawling. Sorry. Enough of you haven't respected Robots TXT. No entrance.

正是如此。就像我们熟知的404错误表示内容未找到，网络成功的标志是返回200状态码。实际上协议最初还规定了402状态码，表示需要付费。我们现在正是利用这个原始规范，当机器人试图访问一个有变现意图的页面时进行响应。

That's right. And so the the original like, we're we're all familiar with, four zero four errors when something is not found. Success on the Internet is a 200 response that comes back to you. There's actually an original one of the protocols set out a four zero two response, and that response says payment required. And so we're actually tapping into that exact original specification to say when a robot tries to access a page where there's an intent to monetize it.

所以要么是订阅内容，要么是含广告页面，我们就能返回402付费要求，然后进行协商。初期主要会是与大型出版商合作，像Reddit、《纽约时报》等已达成内容授权协议，允许特定机器人访问。但长远来看，这将是个动态过程——小型AI公司或出版商可以自主定价，Cloudflare会评估内容对特定AI的价值，再由AI公司决定是否值得购买。

So it's either behind a subscription or it's got ads on it, that there is an ability for us to say four zero two payments required and then there's a negotiation. And in some cases, and at first, that's going to be largely large publishers with large companies doing deals like what Reddit has done or what The New York Times has done or what others have done where they have licensed the content and then certain robots get access to that. But in other cases, and I think over time, that will be a dynamic process where maybe a smaller AI company or a smaller publisher will say, hey, here's what I would charge for this content. Cloudflare will surface like how valuable that content would be for that particular AI. And then the AI companies can decide, is that worth it or not?

可能是微小交易，比如零点几美分，也可能是几美元。对于极具价值的内容，价格可能达数百、数千甚至数百万美元。想象泰勒·斯威夫特发布新歌时歌词的独家获取权——对于面向孤独少女的聊天应用而言，这价值连城，特别是能获得限时独占权的情况下。这正是未来可能形成的市场机制：原创优质内容获得补偿，通过这堵'墙'创造稀缺性后，市场会自然形成清算价格。

And it might be a very small transaction, maybe a fraction of a penny or maybe a few cents, or in some cases, content that is really valuable might be worth hundreds or thousands or millions of dollars. You could imagine Taylor Swift is about to release a brand new song and the lyrics, get published. How valuable is that for an app for teen girls who are lonely and want to talk about things? Probably pretty valuable and especially valuable if you could have exclusive access to it for some window of time. And so that's the sort of thing where I think a marketplace over time can develop, where original valuable content will get compensated and there will be a clearing price in the market once we have that scarcity that's created by that wall.

所以这不只是拦截机制，还是个内容交易市场。出版商可以出售内容，这样既能保证聊天机器人获取有效信息，又能让网络生态繁荣发展。

So it's not just a blocker. It's also this marketplace where you're going to have publishers that will sell their content. So that's a way where you could have useful, effective chatbots and potentially a flourishing web.

完全正确。这正是我们努力实现的目标。我理想中的未来图景是：机器人应为内容支付高价，而人类可以免费获取。

Exactly. And that's and that, I think, is what we're trying to play for. Again, my utopian vision of the future is robots should pay a lot for content and humans should get it for free.

没错。为启动这个计划，在6月3日临近7月1日时，你们在世贸中心一号楼顶举办了发布会，多家知名出版商按下红色启动按钮——包括康泰纳仕、时代公司、美联社、《大西洋月刊》、《广告周刊》和《财富》等重量级机构。

Right. And so to kick this off on June 3, as the day turned July 1, you had a party on the top of the World Trade 1 World Trade Center where a bunch of publishers pressed a red button to get this thing going. And that includes some very big names, Conde Nast, Time, The Associated Press, The Atlantic, Adweek and Fortune are

而且还将成为更多变革的一部分。坦率地说，我们接触的出版商无一不表示这一变革势在必行。你们正走在正确的道路上。整体来看，不仅是已入驻云端、占互联网20%以上的群体，我认为还有20%到30%的大型出版商也都参与其中并付诸实践。令人鼓舞的是，与此同时，我们也在与顶尖的人工智能公司展开对话。

all And going to be part of a lot more. Frankly, there hasn't been a publisher that we've talked to who hasn't said that this is a change that needs to happen. You're on the right path for it. And so across the board, not only the kind of 20% plus of the web that sits behind the cloud already, but I think another 20 to 30% that are these major publishers that are out there are all on board and doing that. And what I think has been encouraging is at the same time, we've been having conversations with the largest AI companies.

它们全都认同内容创作者应当因其作品获得报酬。这一点上它们毫无异议。难点在于具体细则，部分公司正以各种方式推诿。但让我深感振奋的是，在与全球领先的AI巨头、顶尖科技企业交流时，发现它们实际上在积极接纳这个理念。这些公司都清醒认识到内容创作者必须得到补偿。

And all of them agree that content creators need to be compensated for their content. They all agree on that. The devil's in the details, and some of them are pushing back in various ways. But I've been really encouraged that as we have talked to the largest leading AI companies, the largest technology companies in the world, they're actually leaning into this. They all recognize the content creators need to be compensated.

我认为在未来数月里，真正的难点将是如何建立这个市场机制——既要公平对待生态系统中各类供应商，创造公平竞争环境，允许新进入者，而非仅让预算雄厚的大公司获益，还要确保像谷歌这样的传统供应商与新兴供应商获得同等对待。这极具挑战性。但令我备受鼓舞的是，不仅出版商们全力支持，连那些意识到变革必要性的AI公司也在积极对话。

And I think over the months to come, that's when the hard work will go down around how do we actually create this marketplace in a way which is fair for all of the different providers in ecosystem, treats everybody in a way that has a level playing field, allows new entrants, doesn't just reward the largest companies with the biggest budgets that are out there, make sure that, legacy providers like Google are treated the same as newer providers that are there. That's all going to be really tough. But I am incredibly encouraged by the conversations I'm having, not just with the publishers who are all on board, but actually with the AI companies who recognize something needs to change.

他们能认识到这点很有意思，因为外界普遍感觉这些交易公告——比如OpenAI支付数百万美元获取《华尔街日报》或道琼斯的内容授权——更像是避免诉讼的封口费。萨姆·奥特曼显然对《纽约时报》起诉OpenAI很不满，尤其反对时报要求OpenAI保存聊天记录等诉讼策略。不过确实耐人寻味。你认为这种零散交易会最终演变成市场化模式吗？

That's interesting that they're recognizing this because the sense that you get is that you hear these announcements of deals like OpenAI paying x million to The Wall Street Journal to be able to include their articles or Dow Jones. And the sense you get is that they're just kind of payoffs to not get sued. The Sam Altman, very happy, very clearly is not happy with The New York Times pursuing OpenAI and especially actions that The Times are taking in their lawsuit, like forcing open AI to preserve their chat logs, which I think is wrong. But but it is interesting. So what do you think about is there are we going to see an evolution from these one off deals to this marketplace style world?

嗯，这种剧情我们见过多次。Napster时期就是蛮荒时代，音乐行业对其发起大量诉讼，后来iTunes出现，最初定价每首歌0.99美元，最终演变成更接近Spotify的订阅分成模式。我认为历史正在重演，关键是OpenAI等公司愿意为内容付费这个事实。

Well, I think that I mean, we've seen this story many times before. I mean, Napster was along. It was a wild west. There were a bunch of lawsuits from the publishing the music industry targeting Napster and the like, and then along comes iTunes, which starts out as zero nine nine dollars a song, but eventually evolves into what is much more, much closer to a Spotify model of a subscription and a pool of funds that then get distributed out to So all the I think we've seen this story before. And I think that one of the things that's really important is OpenAI and others are willing to pay for content.

他们确实在进行这些交易。但简单说'花钱消灾'并不恰当。与头部AI公司交流时会发现，他们理解内容创作需要付出劳动，创作者理应获得报酬。如果无法通过订阅、广告或自我实现获得收益，就必须建立其他补偿机制。

They do the deals that are there. And I don't think it's right to just say, we'll do a deal to avoid lawsuits. Again, I think that that when you talk to leading AI companies, they understand that people are doing the work to create content. They need to get compensated for that content. And if it's not going to be through subscriptions or ads or ego, it's got to be through something else.

具体实现方式有待探索，但可以确定的是：如果OpenAI为内容付费，而你们却免费提供给其他所有人，这种模式终将崩溃。OpenAI最终会说：'我们想支持你们，但我们不能当冤大头。'

And so exactly how that happens, we'll figure out. But what I know won't work is if OpenAI is paying for your content, but you're giving it away for free to everyone else. It's not going to work. OpenAI eventually is listen, we want we want to support you. We want to help you out, but we can't be the suckers.

我们不能成为唯一付费而你们却免费赠送的群体。因此，任何市场要真正具有价值，稀缺性是必要的。我认为那些对此投入最多的人，正是那些已经与部分而非所有AI公司达成现有协议的人，因为他们意识到，要使这些协议对他们有价值、值得续签甚至扩大合作，就必须存在真正的稀缺性——他们必须获得有价值的东西。你不能向OpenAI收费，却免费提供给Anthropic。必须有所限制，明确所有人都需要付费。

We can't be the only ones paying where you're giving stuff away for free. And so scarcity is needed in order to actually have value in any kind of market. And so I think that the the people who have actually leaned into this that most heavily are the ones that have the existing deals with some, but not all of the AI companies because they realize that for those deals to be valuable for them to renew for them to renew for more, there has to actually be scarcity where they're getting something of value. You can't you can't charge open AI, but give it away for free to anthropic. Something needs to actually restrict it and say everyone needs to pay.

每个人都需要站在公平的竞争环境中，并共同规划未来的发展方向。

Everyone needs to be on a on a level playing field and figure out what that that looks like going forward.

像你们正在实施的解决方案是否可能带来一些附带损害？例如，我看到这些出版物的名字——康泰纳仕、时代杂志、美联社、大西洋月刊——它们目前从搜索引擎获得了大量流量。如果设置这样的拦截机制，是否会影响它们的SEO表现？

Could there be some collateral damage with the solution like you're implementing? For instance, I'm looking at the names of these publications, Conde Nast Time, The AP, The Atlantic. I imagine they get a lot of traffic from search as it is today. So if you put this blocker up, does that impact their SEO, for instance?

是的。因此我们一直非常谨慎地强调：传统搜索引擎目前不会被拦截，甚至AI驱动的搜索也不会被拦截。但我们将为出版商提供工具，帮助他们区分搜索索引和衍生内容。你可以这样理解：就像今天的谷歌体验。

Yeah. So we've been very, very careful to say that the traditional search today is not blocked. And and even AI driven search today isn't blocked. But you're gonna see us give publishers the tools to differentiate between search indexing, and derivative content. So the way I would think about this is the Google experience today.

可能有些出版商仍希望出现在传统搜索结果（十个蓝色链接）中，但不愿进入AI概览或答案框。我们需要这种精细控制能力——比如告诉谷歌：虽然你们使用同一个爬虫，但不同功能应该区别对待。在与谷歌的沟通中，我越来越相信他们理解这种精细控制的重要性。但若他们未能做到，我完全确信全球监管机构会密切关注，并迫使谷歌将其爬虫功能明确分离和公示。

It may be that it publishes, I still want to appear in the 10 blue links, but I don't want to be in the AI overview or the answer box. And the granularity of being able to say, Okay, Google, I understand you use one bot, but we need that to be treated similarly. And again, I am hopeful. And in my conversations with Google, I am increasingly hopeful that they understand the importance of this and giving that granularity. But if for some reason they don't, I am also 100% certain that regulators are paying a ton of attention to this and that around the world, you will see them force Google to split their crawler out into into announcing exactly what it is doing.

重申一次，希望我们能在监管介入前与谷歌达成协议。但不可避免的是，谷歌终将必须声明：如果你不希望内容被用于衍生用途，你可以在保持搜索可见的同时拥有控制权。

Again, I think that that's kind of the hopefully we get to an agreement with Google way before that has to happen. That's inevitably, I think Google is going to have to say, if you don't want us to use your content for derivatives, you have a way of controlling that while still appearing in search.

好的。结束前还有几个宏观问题：互联网的规模扩张速度如何？我们看到的网络规模增长是否在持续加速？

Okay. A couple of big picture questions before we leave. How much bigger is the web getting? And is the web sort of accelerating in the size increases that we see?

是的。从我们能看到的所有指标来看，它实际上已经趋于平稳，在内容方面甚至有所下降。你会发现新注册的域名变少了，新上线的网站也减少了。我认为很多内容已经转移到了个人平台上。

Yeah. I mean, it's actually been, by all the measures that we can see, it's actually kind of plateaued and has actually flattened in terms in terms of content. You see fewer domains getting registered. You see fewer new websites going online. I think a lot of that has moved to individual platforms.

所以更多内容出现在YouTube、Facebook、TikTok这些平台上。我认为部分原因是这些工具为内容创作者提供了便捷的变现方式，让他们不必操心某些问题。在理想的未来，你会希望内容创作者能摆脱这些平台，自己赚更多钱，同时还能以有趣的方式将内容变现。确实有很多人正在研究这个问题。实际上，我认为谷歌是其中一个组织，它创造了过去三十年网络的商业模式。

So more of that on a YouTube, more of that on a Facebook, more of that on TikTok that that is there. And I think part of that is because those tools have provided content creators easy monetization tools to allow them to to to not have to think about some of those problems. I think that in a in a in an ideal future, you would want content creators to be able to be free from those platforms, to earn more themselves, and but still have abilities to monetize that content in interesting ways. And so, again, I think there are lots of people who are working on on that problem. I actually think Google has been one of the organizations that has, again, created what was the business model of the last thirty years of the web.

但未来三十年网络的商业模式将会不同。我们必须以不同的方式思考。它不会是横幅广告，很可能也不会是订阅制，而是某种不同的形式。

But the business model of the next thirty years of the web is going to be different. And we've got to think about it in a different way. It's not going to be banner ads. It's not really probably going to be subscriptions. It's going to be something different.

因此，这是我们尝试的一种解决方案，但我怀疑它不会成为最终胜出的那个。

And so this is our attempt at one solution, but I doubt it will be the one that emerges.

现在，我很好奇我要做什么，因为这只是一个单人内容运营。总之，你肯定应该

Now, I'm curious what I'm going to do because it's just a one person content operation. So anyway, you should certainly be

向AI收费来授权使用你的声音。

charging AI to to license your voice.

我能注册使用你们的产品吗？

Can I sign up to your product?

当然。毫无疑问。

For sure. Absolutely.

好的，我

Okay, I'm

准备

going to

稍后邮件联系你。关于网络安全问题，显然你提到了如何应对那些企图入侵全球网站的各路政府黑客。他们是否已能利用生成式AI工具或自动化编码技术来提升攻击效率？

email you after this. And then when it comes to cybersecurity, obviously, you talked about how you're dealing with all these governments that would like to hack into sites across the web. Have they been able to use generative AI tools or automated coding to get more to become more effective at what they do?

是的，我认为每当新技术出现时，坏人和好人都会加以利用。我们已经并将继续目睹一些骇人听闻的事件，比如某个家庭被犯罪团伙欺骗，因接到冒充女儿的电话声称'我在墨西哥被捕需要赎金'而汇出毕生积蓄。尤其值得注意的是，我们观察到朝鲜黑客伪装成求职者的情况显著增加。

Yeah, I mean, I think that that any time a new technology comes out, bad guys are going to use it as well as as good guys. And so we have seen and we will continue to see some horror stories around, you know, the family that was tricked by some gang and wiring their life savings because something someone that sounded like their their daughter called and said, I've been arrested in Mexico. You know, I need I need to pay to get out or other things. I think we are seeing a real rise in especially out of North Korea. North Koreans posing as if they were applicants to various various jobs.

这种手段使他们获得系统访问权限，进而实施各种恶意行为。所有这些犯罪活动都得到了AI的助力。但好消息是，像Cloudflare这样的防御方也在运用AI技术，不仅能够识别威胁，还能更早更智能地预判攻击。

And then that is, you know, allowing them access, which they can then use to to do to do any number of nefarious things. All of that, again, assisted by AI. So I think that's been sort of on the bad guy side. The good news, though, is that the good guys, you know, folks like Cloudflare, we have been using AI as well in order to not only detect these things, but get smarter at detecting attacks earlier in the process.

看来这套方案对你们很有效。

That's working for you.

归根结底，谁能赢得AI竞赛，取决于谁拥有最多的数据访问权。而我认为，好人总能比坏人获取更多数据。过去两年半里，我们通过AI让网络变得更安全，始终领先于攻击者。尽管仍会有骇人听闻的事件发生，问题也总会存在。

At the end of the day, who wins in the AI race is whoever has the access to the most data. And I just think that the good guys are always going to have access to a lot more data than the bad guys. So far, I feel like we have made the web more secure with AI over the course of the last two and a half years and stayed way ahead of the attackers. Although, again, there are going to be horrible stories. There are going to be problems that are there.

我认为，要相信你在网上看到的内容是真实的会越来越困难。我们将不得不转向更安全的验证方式，比如身份验证和认证。

I think that it is going to be harder and harder to trust that something that you're seeing online is is actually real. And we'll have to turn to other ways that are more secure about verifying things like identity and and authentication.

好的，最后一个问题。我们还有60秒。你提到你是这项技术的信徒，未来几年的AI发展在你看来会是什么样子？我们短期内会实现通用人工智能（AGI）吗？

Okay, last question for you. We have sixty seconds. You mentioned you're a believer in this technology. What does the next couple of years in AI look like to you? Are we going to hit AGI anytime soon?

你预估的时间节点大概是？

Like what's the time when you're thinking?

我是说...其实我不...我相信现在每投入AI的一美元中，有99美分都在打水漂。但剩下的那1美分会带来真实回报。很难区分什么是纯粹浪费时间，什么不是。我们看到很多数据显示，目前AI系统在商业领域的实际应用并不多，很多商业场景落地非常困难。但很多应用其实出现在缓解孤独感、社交互动等领域。

Mean, don't, I'm not, I am, So I believe today that 99 out of cents out of every dollar spent on AI is just being lit on fire. But that 1¢ that's out there is going to generate real returns. It's very hard to figure out what's kind of just a total waste of time versus what's not. You know, we see a lot of data about how much AI systems are really are being used, not not so much for businesses today. A lot of the business applications have been very tough to take on, but but a lot of times just for for like loneliness and social interactions and things like that.

所以我预计这类应用会大量涌现，成为首批成熟场景。商业应用反而需要更长时间。在输出结果容易验证的领域会发展更快，比如编程——我们的工程师使用AI工具后效率显著提升，但这并没有减少我们的工程师招聘数量。

So I would imagine that a lot more of those things are going to develop and those would be sort of the first uses. I think the business applications are actually going to take longer. And in places where it's easier to verify the output as being legitimate, it's going to be easier. So coding like we see that our engineers are significantly more productive by using with using AI tools than they were before. That's not causing us to hire any less engineers.

只是意味着每位工程师的产出更高。我们有大量待办事项，AI正在帮助我们推进。但另一方面，我对AI客服或AI律师持怀疑态度，这些是更棘手的难题——因为很难判断AI的输出是否真实有效，这些领域没有调试器来验证AI创造内容的真实性。所以你会看到编程等领域出现巨大飞跃。

It's just meaning that every engineer we hire is that much more productive. We have a huge backlog of things to do and and and and AI is helping us do that. On the other hand, you know, I am I'm still quite skeptical that the AI customer support agent, that that is a much harder problem or the AI lawyer, that is a much harder problem because it's it's just harder to tell whether something actually worked or didn't. There's no debugger in those spaces in order to figure out if what the AI is creating was actually true. And so I think you're going see just huge leapfrogs in what are things like coding.

但我认为，我们需要更长时间才能完成那些验证难度稍高的事情。

But I think it's going to take longer for us to do things that are a little bit more difficult to verify.

非常有趣。你对这项技术持深度乐观态度，但仍然认为

Very interesting. You're deeply optimistic about the technology, but still think

99%。

99%.

被点燃了。浪费了。是啊，接下来的发展会很有意思。马修·普林斯，很高兴见到你。

Lit on fire. Wasted. Yeah. It's gonna be very interesting to shake out. Matthew Prince, great to see you.

感谢你参加我们的节目。

Thank you for for coming on the show.

谢谢邀请我。

Thanks for having me on.

好了各位，感谢观看和收听。我们下次在《大科技播客》再见。

Alright, everybody. Thank you for watching and listening. We'll see you next time on Big Technology Podcast.

现在到底发生了什么，为什么会这样？在《连线》杂志，我们每天都痴迷于深入探究这些问题，或许你也是。我是凯蒂·德拉蒙德，《连线》的全球编辑总监，我正在主持我们的新播客系列——大访谈。每周，我都会与一些最有趣、最具挑衅性和影响力的人物坐下来交谈，他们正在塑造我们的当下。大访谈的对话充满乐趣

What the hell is going on right now, and why is it happening like this? At Wired, we're obsessed with getting to the bottom of those questions on a daily basis, and maybe you are too. I'm Katie Drummond, the global editorial director of Wired, and I'm hosting our new podcast series, the big interview. Each week, I'll sit down with some of the most interesting, provocative, and influential people who are shaping our right now. Big interview conversations are fun

我想要一条鲨鱼

I want a shark that

它能吞噬互联网，将其关闭

that eats the Internet that turns it

全部

all

关闭。毫无过滤，无所畏惧。

off. Unfiltered and unafraid.

所以在很多方面，我尽力成为你在网上看到的那种难以想象的反动内容洪流的解药，尽我所能。

So in a lot of ways, I try to be an antidote to the unimaginable faucet of reactionary content that you see online to the best of my ability.

每周，我们都会为你提供我们这个时代的终极奢侈品——意义与背景。真或假？你，布莱恩·约翰逊，坐在我对面的这个男人，有一天，在未来的某个尚未确定的时刻，你会死去。假的。告诉我更多。

Every week, we're going to offer you the ultimate luxury of our times, meaning and context. True or false? You, Brian Johnson, the man sitting across from me, one day, at some point, as of yet undefined in the future, you will die. False. Tell me more.

立即收听重磅访谈节目，就在你发现《连线》杂志诡异谷播客的同一平台。无论你在哪里获取播客，请订阅或关注。

Listen to the big interview right now in the same place you find Wired's uncanny valley podcast. Subscribe or follow wherever you get your podcasts.