AI吞噬世界：Benedict Evans谈下一个平台变革

也许下周我做计费工时表时会用到它，但这并非我整日的工作内容。

Maybe I'll use it next week when I'm making a table of my billable hours, but that's not what I do all day.

而Excel无法完成律师日常处理的事务。

And an Excel is doesn't use do things that a lawyer can do every day.

我认为还有一类人会感到困惑，不确定该如何运用这项技术。

And I think those there's this other class of person that's like, I'm not sure what to do with this.

部分原因在于习惯使然。

And some of that is habit.

部分原因在于认知转变——意识到不必沿用旧方法。

Some of that is, like, realizing, no.

完全可以用新方式替代原有操作流程。

Instead of doing it that way, I could do it this way.

但这恰恰是产品的本质价值所在。

But that's also what products are.

就像2014到2019年期间我在Y Combinator见过的每个创业者——现在肯定也一样——你看着任何入驻团队都能断言：这本质上就是个数据库。

Like, every entrepreneur who comes into a '16 z, when I was there from 2014 to 2019, and I'm sure now, like, you could look at any company that comes in and say, that's basically a database.

那基本上就是个客户关系管理系统。

That's basically a CRM.

那本质上就是Oracle或Google文档，只不过他们意识到某个行业中存在这个问题或工作流程，并研究出如何运用数据库、客户关系管理系统，或者基本上是五到二十年前的概念，为那个行业的人们解决这个问题，进而推销给他们，并研究如何让他们使用它。

That that's basically Oracle or Google Docs, except that they realize there's this problem or this workflow inside this industry and worked out how to use a database or a CRM or basically concepts from five, ten, twenty years ago and solve that problem for people in that industry and go in and sell it to them and work out how they can get it to use it.

这就是为什么，你看这些数据时，根据统计方式的不同，如今美国典型的大公司拥有400到500个SaaS应用。

And so this is why, you know, you look look look at data on this that, you know, depending on how you count it, the typical big company today has four to 500 SaaS apps in The US.

400到500个SaaS应用程序，它们基本上都在做你可以在Oracle、Excel或电子邮件中完成的事情。

Four to 500 SaaS applications, and they're all basically doing something you could do in Oracle or Excel or email.

是啊。

Yeah.

没错。

Yeah.

这就是另一方面，恐怕我有点在独白了。

And that's the other side I'm monologuing, I'm afraid.

但是，这就是另一个问题——你该如何利用这些东西？

But, like, this is the other side of what is what do you do with these things?

你是直接去找机器人让它帮你做事吗？

Do you just go to the bot and ask it to do a thing for you?

还是企业销售员会来找你老板推销产品，让你现在只需按个按钮，就能分析这个你甚至从未意识到自己在执行的需求流程？

Or does an enterprise salesperson come to your boss and sell you a thing that means now you press a button and it analyzes this process that you needed, that you never realized you were even doing.

是的。

Yes.

我觉得这就是为什么会出现AI软件公司。

And I feel like that's, I mean, that's why there are AI software companies.

对吧。

Right.

真的吗？

Really?

他们不就是在做这个吗？

Isn't that what they're doing?

他们正在拆解ChatGPT，就像十年前的企业软件公司拆解Oracle、Google或Excel那样。

They're unbundling ChatGPT just as the enterprise software company of ten years ago was unbundling Oracle or Google or Excel.

你是否认为，就像Excel为会计师所做的那样，AI现在正在为程序员和开发者做类似的事情，但还没有完全为其他职位找到那种日常关键工作流程的解决方案。

Do you have the view that, you know, what Excel did for for for accountants, you know, we're we're sort of AI is now doing for for coders and developers, but hasn't quite figured out that sort of, you know, daily critical workflow for for other job positions.

因此对于非开发人员来说，他们不太清楚为什么我应该每天花很多时间使用这个。

And so it's unclear for people who aren't developers, you know, why I should be using this for many many hours a day.

或者

Or

我认为有很多人的工作任务并不太适合用这个。

I think there's a lot of people who don't have tasks that work very well with this.

是的。

Yeah.

然后还有很多人需要它被包装在产品和工作流程中，需要有人来告诉他们：'嘿'。

And then there's a lot of people who need it to be wrapped in a product and a workflow and tooling and UX and someone to come and say, hey.

你有没有意识到可以用这个来做？

Have you realized you could do it with this?

今年夏天我和Balaji——另一位前16z的人——有过这样的对话。

I had this conversation with in the summer with with with with Balaji, who's another former a 16 z person.

他当时提出了一个关于验证的观点——因为这些东西仍然会出错，而硅谷的人常常对此轻描淡写。

And he was making this point about validation that can you because these things still get stuff wrong, and people in the valley often kind of hand wave this away.

但要知道，有些问题有明确答案，必须得到正确答案或限定范围内的正确选项之一。

But, you know, there are questions that have specific answers where it needs to be the right answer or one of a limited set of right answers.

你能用机制化方法验证吗？

Can you validate that mechanistically?

如果不能，用人来验证是否高效？

If not, is it efficient to validate it with people?

以营销用例来说，让机器生成200张图片再由人工筛选10张优质图，远比让人直接制作10张或100张高效——即便你要生成500张图筛选100张，也比人工制作100张高效得多。

So, you know, with the marketing use case, it's a lot more efficient to get a machine to make you 200 pictures and then have a person look at them and pick 10 that are good than to have, people make 10 good images or a 100 you know, even if you're gonna make 500 images and pick a 100 that are that's a lot more efficient than having a person make a 100 images.

但另一方面，如果是数据录入这类工作——就像我写过的关于OpenAI推出深度研究部门那样。

But on the other hand, if you're doing something like data entry, and as I wrote something about this, about, about OpenAI launched Deep Research.

OpenAI推出了深度研究部门。

OpenAI launched Deep Research.

他们整个营销案例就是围绕收集移动市场数据展开的。

Their whole marketing case is it go goes off and collects data about the mobile market.

我曾经是一名移动市场分析师。

I used to be a mobile analyst.

这些数字全错了。

The numbers are all wrong.

他们展示用途时声称这多么有用，但他们的数字是错误的。

Their use case of look how useful this is, their numbers are wrong.

在某些情况下，数字错误是因为他们直接从数据源转录时就抄错了。

And in some cases, they're wrong because they've literally transcribed the number incorrectly from the source.

另一些情况下，错误是因为他们使用了本不该采用的数据源。

In other cases, it's wrong because they've used a source that they shouldn't have used.

但就像，如果我让实习生来做这件事，实习生可能也会犯同样的错误。

But, like, if I'd asked an intern to do it for me, then an intern would probably have picked that.

关于验证的问题——如果要进行数据录入，比如让机器从200份PDF中抄录200个数字，然后我还得逐一核对这200个数字，那还不如我自己动手做。

And to my the point about, you know, verification, if you're gonna do data entry, if I'm gonna ask a machine to copy 200 numbers out of 200 PDFs, and then I'm gonna have to check all 200 of those numbers, I might as well just do it myself.

是啊。

Yeah.

所以你面对的是一整套错综复杂的矩阵——如何将其与现有问题对应起来。

So you've got, like, a whole swirling matrix of how do you map this against existing problems.

但另一方面是，如何将其与你以前无法完成的新事物对应起来？

But the other side of it is, how do you map this against new things that you couldn't have done before?

这又回到了我关于平台芯片的观点，因为我看到人们审视ITVT或生成式AI时说‘这没用，因为它会犯错’。

And this comes back to my my point about platform chips because, you know you know, I see people looking at ITVT or looking at generative AI and saying, well, this is this is useless because it makes mistakes.

我觉得这就像在七十年代末看着苹果二代电脑问‘能用这个来运营银行吗？’

And I think that's kind of like looking at, like, an Apple too in the late seventies and saying, could you use these to run banks?

答案当然是不能。

To which the answer is no.

但这本身就是个错误的问题。

But that's kind of the wrong question.

没错。

Right.

就像你能在网景浏览器里构建专业视频编辑功能吗？

Like, could you build video edit professional video editing inside Netscape?

不能。

No.

但这是个错误的问题。

But that's the wrong question.

对。

Right.

后来确实可以。

And later yeah.

二十年后，你就能做到了。

Twenty years later, you can.

但与此同时，它还能做很多其他事情，移动设备也是如此。

But that meanwhile, it does a whole bunch of other stuff, the same with mobile.

比如，你能用手机取代你那五屏的专业编程设备吗？

Like, can you can you use mobile to replace, you know, your, you know, your five screen professional programming rig?

不能。

No.

因此，它无法取代个人电脑。

Therefore, it can't replace PCs.

那么，你猜怎么着？

Well, guess what?

50亿人拥有智能手机，而只有7到8亿人拥有消费级个人电脑。

5,000,000,000 people have got a smartphone, seven or 800,000,000 people have got a consumer PC.

所以我们某种程度上确实做到了。

So we kind of did Yeah.

但它做了不同的事情。

But did a different thing.

关键在于，这种新事物就像你之前提到的颠覆性框架。

And the point of this is, like, the new thing this is, you know, the disruption framing you mentioned earlier.

新事物通常对旧事物重要的方面表现不佳，但它能实现其他功能。

The new thing is generally not very good or terrible at the stuff that was important to the old thing, but it does something else.

没错。

Right.

很多问题其实就在于，好吧。

And a lot of the question is, okay.

它可能并不擅长处理生成式AI擅长的那一类传统任务。

It may not be very good at doing there's a class of old tasks that generative AI is good at.

还有更多传统任务可能是生成式AI不太擅长的。

There's also a lot many more old tasks that generative AI is maybe not very good at.

但还有一大堆你以前根本不会做的事情，生成式AI却非常非常擅长。

But then there's a whole bunch of other things that you would never have done before that generative AI is really, really good at.

然后你该如何发现或想到这些新用途呢？

And then how do you find those or think of those?

这其中有多少是用户面对通用聊天机器人时自己想到的？

And how much of that is the user thinking of it faced with a general purpose chatbot?

又有多少是创业者说'嘿'想到的？

How much of that is the entrepreneur saying, hey.

我刚意识到有件以前做不到的事现在可以做到了，这就是机会。

I've just realized that there's this thing that I can do that you couldn't do before, and here you are.

我给你提供了一个带按钮的产品，它能帮你完成这件事。

I've given you a product with a button that will do it for you.

对。

Right.

这就是为什么会有软件公司存在。

And that's why there are software companies.

没错。

Right.

在移动互联网时代，一些新用例，比如我们开始搭乘陌生人的车。

And and on on mobiles, you know, some of the new use cases, you know, we're, you know, getting in strangers' cars.

我们提到过Lyft和Uber这类服务，或者通过应用约会认识的人，又或是把闲置卧室出租出去等等。

You know, we mentioned Lyft and Uber or sort of, you know, dating people you met via an app or sort of, you know, lending your spare bedroom out, you know, etcetera.

这些都是围绕这些新行为模式建立的全新公司。

And and those were net new companies that that that, you know, were built around those behaviors.

我认为对于AI而言，核心问题仍然是：哪些是全新的行为模式？

And I think, for AI, there's still a question of, you know, what are those net new behaviors?

我们开始看到一些现象，比如人们与聊天机器人互动交谈，而非人类，或者除此之外。

We're we're starting to see some in terms of, you know, people engaging and talking with, you know, chatbots instead of humans or or, or in addition.

然后问题来了，嘿。

And then there's a question of, hey.

这些是由现有的模型提供商完成，还是由全新的公司来完成，无论是在企业端还是消费者端？

Are these done by the model providers that that currently exist, or are these done by, you know, net new companies both on, you know, sort of enterprise and consumer?

嗯，这始终是个问题：一个新事物能在技术栈中走多远。

Well, this is always a question is how far up the stack does a new thing go.

是啊。

Yeah.

而且，你知道，我之前和另一位前a16z成员讨论过这个问题，他指出在九十年代中期，人们曾争论操作系统会包揽一切。

And, you know, I was talking I about this with another former former a 16 z person who pointed out that, like, in the the the mid nineties, people kind of argued that, well, you know, the operating system does all of it.

而Windows应用程序基本上只是薄薄的Win32封装层。

And the Windows apps are basically just kind of thin Win32 wrappers.

没错。

Yeah.

而且，你知道，Office基本上就是一个薄薄的Win 32封装。

And, you know, Office is basically just, you know, a thin Win 32 wrapper.

就像，所有重要功能都由操作系统完成，无论是文档管理、打印、存储还是显示，这些过去都是由应用程序处理的。

Like, all the important stuff is being done by the OS, whether it's, you know, the document management and printing and storage and display, which all stuff that used to be done by apps.

比如在DOS系统下，应用程序必须自己处理打印。

Like, in dot on DOS, the apps had to do printing.

应用程序必须自己管理显示。

The apps had to manage a display.

转到Windows后，应用程序过去90%的工作现在都由Windows完成了。

Move to Windows, like 90% of the stuff that the app used to do is now being done by Windows.

是啊。

Yeah.

所以Office就像一个薄薄的Win 32封装，所有繁重的工作都由操作系统完成了。

And so Office is just like a thin Win 32 wrapper, and all the half stuff has been doing being done by the OS.

但事实证明，这种框架虽然有用，但可能并不是理解当前情况的有效思维方式。

And it turns out, well, that was again, it's like frameworks are useful, but that's not made maybe not a useful way of thinking about what's going on.

同样的情况现在，这个市场需要多少专门的了解，了解它如何运作、它是什么以及你会如何处理它？

And the same thing now, like, how much does this need single dedicated understanding of how that market it works or what that market is and what you would do with that?

我的意思是，我记得我们在16z的时候，投资了一家公司叫Everlaw，它是云端法律发现服务。

I mean, I remember when we were at a 16 z, there was an investment in a company called Everlaw, which is cloud legal discovery in the cloud.

是的。

Yeah.

于是机器学习出现了，现在他们可以做翻译了。

And so machine learning happens, and so now they can do translation.

他们是否担心律师会说，好吧，我们不再需要你们了。

Are they worried that lawyers are gonna say, well, we don't need you guys anymore.

我们直接去AWS找个翻译应用和情感分析应用就行了。

We're just gonna go out and get a translate app and a sentiment analysis app from AWS.

不。

No.

法律事务所不是这样运作的。

That's not how law firms work.

律师事务所想要购买的是解决法律发现、软件管理需求的整套方案。

Law firms wanna buy a thing that solstice wanna buy legal discovery, software management.

你知道，他们并不想自己去编写API调用。

You know, they don't wanna, you know, go out and write their own do API calls.

我是说，极少数大型律所可能会这么做，但普通律所根本不会考虑。

I mean, very, very big law firms might, but, you know, typical law firm isn't gonna do that.

人们购买的是解决方案。

People buy solutions.

他们不购买技术组件。

They don't buy technologies.

同理，这些模型的应用层级能有多高？

And the same thing here, like how far up the stack do these models go?

你能把多少东西封装成即用型工具？

How much can you turn things into a widget?

你能把多少需求转化为大语言模型的请求？

How much can you turn things into an LLM request?

那么现在，你究竟需要多少专属用户界面呢？

And how much now does it turn out that you need that dedicated UI?

有趣的是，你可以在谷歌身上看到这一点，因为他们曾认为一切只需一个谷歌搜索就能解决，谷歌会解析查询意图。

Funny thing is you can see this around Google because Google had this whole idea that everything would just be a Google query, and Google would work out what the query was.

猜猜结果如何？

And guess what?

现在你看，谷歌航班就不是一个简单的搜索查询。

You know, now you want me to this Google Flights is not a Google query.

他们在某个节点上做出了改变。

You know, they use a certain point.

关于这点有个有趣的现象——我认为思考图形界面的作用很有意思，它让Office能承载500项功能且全部可被用户发现。

And and one of the one of the interesting things about this, and and I think it's interesting to think about what a GUI is doing, that some of what a GUI is doing, and the obvious thing that a GUI is doing, is that it enables Office to have 500 application 500 features, and you can find them all.

至少你不需要记忆键盘命令了。

Or at least it's you don't have to memorize keyboard commands.

现在你实际上可以拥有无限功能，只需不断添加菜单和对话框即可。

You can now have effectively infinite features, and you can just keep adding menus and dialogue boxes.

最终，对话框会占满整个屏幕空间。

And, eventually, you you run out of screen space for dialogue boxes.

但是，你可以拥有数百项功能，而无需用户记忆键盘命令。

But, like, you can have hundreds of features without people needing to memorize keyboard commands.

但另一方面，当你身处那个对话框或屏幕流程中——无论是在Workday、Salesforce这类企业软件，还是在航空公司网站、Airbnb或其他任何系统中。

But the other side of it is you're in that dialogue box or you're in that screen in that workflow in Workday or Salesforce or whatever the enterprise software is, whatever any software or or or the airline website or or Airbnb or whatever it is.

屏幕上并没有600个按钮。

There aren't 600 buttons on the screen.

屏幕上只有七个按钮，因为公司的一群人已经坐下来思考过：在这里应该向用户提出什么？

There's seven buttons on the screen because a bunch of people at that company have sat down and thought, what is it that the users should be asked here?

我们应该给他们提供哪些问题？

What questions should we give them?

在这个流程节点上应该设置哪些选项？

What choices should there be at this point in the flow?

这体现了大量的制度性知识、学习积累、测试验证，以及对系统运作方式的缜密思考。

And that reflects a lot of institutional knowledge and a lot of learning and a lot of testing, a lot of really careful thought about how this should work.

然后你给某人一个原始提示，你只是说，好吧。

And then you give somebody a raw prompt, and you just say, okay.

你只需要告诉它怎么做这件事。

You just tell the thing how to do the thing.

你会觉得，但你必须闭上眼睛，皱起眉头，从第一性原理思考这一切是如何运作的？

And you're like, but you've kind of gotta shut your eyes, screw your eyes up, and think from first principles, how does this all of this work?

这有点像我一直把机器学习比作给你无限多的实习生。

It's kinda like I always used to talk about machine learning as giving you infinite interns.

你想啊，假设你有个任务要交给实习生，而这个实习生连风险投资是什么都不知道。

You know, imagine you've got a task and you've got an intern, and the intern doesn't know what venture capital is.

他们能帮上什么忙呢？

How helpful are they gonna be?

他们不知道公司会发布季度报告，不知道我们有彭博账户可以查询倍数，更不知道这类数据应该用PitchBook而不是谷歌。

And they don't know that companies publish quarterly reports and that we've got a Bloomberg account that lets us look up multiples and that then you should probably use PitchBook for this data and rather than using Google.

这就是我说的深度研究的意义所在。

This is my point about deep research.

比如，不。

Like, no.

你应该用这个来源而不是那个来源。

You should use this source and not that source.

你是想从头开始摸索，还是希望一群精通此道的人花五年时间为你筛选出屏幕上该有的选项供你点击？

Do you want to have to work that out from scratch, or do you want a bunch of people who know a lot about this stuff to have spent five years working out what the choices should be on the screen for you to click on it?

就像那句老话说的：计算机永远不该问你一个需要你自己去琢磨的问题——它本该自己就知道答案。

I mean, it's the old user interface saying the computer should never ask you a question that you should have to work out, that it should know by itself.

当你面对一个空白原始的聊天机器人界面时，它简直是在事无巨细地追问你。

You go to a blank, raw chatbot screen, it's asking you literally everything.

它不止问一个问题。

It's not just asking you one question.

它是在全方位拷问：你到底想要什么？以及你打算如何实现？

It's asking you absolutely everything about what is it is that you want and how you're gonna work out what how to do it.

所以你看，你提到Chet...就像ChetJPT本质上不算产品，更像是伪装成产品的聊天机器人。

The and so, you know, you're mentioning Chet you know, wrote about ChetJPT isn't sort of a product as much as this chatbot dis is disguised as a as a product.

我很好奇，当我们回顾这种平台转变时，你认为是否会出现另一款类似iPhone或Excel那样定义平台转变的产品，而ChatGPT无法做到？还是说世界需要时间去适应如何使用ChatGPT这类工具？

I'm curious, you know, when we sort of look back at this sort of the, you know, platform shift, do you think that there will be another sort of iPhone sort of esque or Excel esque product that kind of defines the the the feature the sort of platform shift in a way that ChatGPT won't, or or or is it sort of that the world has to catch up to how to use ChatGPT or or something like ChatGPT?

这两种情况都可能成立，因为就像当初人们需要时间才能理解如何使用谷歌地图、谷歌搜索和Instagram一样。

So both of these both of these can be true because there was a lot of like, it took time to realize how you would use Google Maps and what you could do with Google and how you could use Instagram.

所有这些产品都经历了巨大的演变过程。

And all of these products have evolved a huge amount over time.

某种程度上，这是一个逐渐认识到能用它做什么的过程。

So some of it is, like, you grow towards realizing what you could do with this.

比如，你现在意识到那只是个谷歌查询。

Like, you realize that's just a Google query now.

你意识到其实可以那样做。

You realize that you could just do it like that.

然后你发现我花了几个小时做这个，才突然明白，哦，其实可以直接做个数据透视表。

And you realize I spent, you know, hours doing this, and I just realized, oh, I could actually just make a pivot table.

是啊。

Yeah.

另一方面，你仍然期望人们从基本原理中自己摸索出来。

The other side of it is then but you're still then expecting people to work it out themselves from first principles.

而且，你知道，让一群真正聪明的人——比如100个、甚至10000个聪明人坐下来研究这些事物，然后将其作为产品展示给你，这是非常有用的。

And, you know, it's kind of useful to have somebody really a 100 a 10,000 really clever people sitting and trying to work out what those things are and then showing it to you as a as a product.

我认为另一个方面是，你知道，总是存在这些前身。

I think another side of this is, like, you know, there were always these precursors.

就像，在Instagram之前就有很多其他东西。

So, like, there were lots of other things before Instagram.

是啊。

Yeah.

要知道，YouTube最初并不是现在的YouTube。

You know, YouTube didn't start as YouTube.

我想最初是从视频约会开始的。

It started as video dating, I think.

在Tinder彻底颠覆整个模式之前，曾有过许多在线约会的尝试，它们都取得了一定程度的成功。

There were lots of of attempts to do online dating that all kind of worked until Tinder kind of pulled the whole thing inside out.

所以总是有很多东西，那句话怎么说来着？

And so there were always lots of things what's the phrase?

局部最优。

Local maxima.

事实上，这正是我们当时在iPhone上的处境。

In fact, this is where we were particularly with the iPhone.

因为在此之前，我已经在移动领域工作了十年。

Before because I was working in mobile for the previous decade.

当时我们并没有在等待某个特定事物的感觉。

It didn't feel like we were waiting for a thing.

感觉事情一直在逐步推进。

It felt like it was kind of working.

就像每年网络速度都在提升，手机性能也在进步，每年都能看到一些改进。

Like every year, the networks got faster and the phones got better and you got a little bit better every year.

我们有了应用、应用商店、3G网络、摄像头，一切似乎都在逐年变得更好。

And we had apps and we had app stores and we had three gs and we had cameras and stuff seemed to be you know, every year was a bit better.

然后iPhone问世了，它直接，你知道，直接让整个图表爆炸式增长，原本的曲线是这样走的，突然出现了一条那样的曲线。

And then the iPhone arrives, and it just, you know, just, you know, blow the chart kind of you know, you've got this line doing this, and then there's a line that does that.

不过要记住，iPhone也花了两年的时间才真正成熟起来，价格定位不对、功能组合不对、分销模式也不太奏效。

Although, remember, also the iPhone took, you know, two years before it worked, As you know, the price was wrong and the feature set was wrong and the distribution model didn't quite work.

所以，是的，你可能觉得一切都很顺利，然后突然出现某个事物让你意识到：不。

And so, yeah, you, you know, you can think your you know, you can think everything is going well, and then something comes along and you realize, no.

哦，不。

Oh, no.

不。

No.

不。

No.

谷歌的情况也是如此。

That's which is the same for Google.

懂我意思吗？

You know?

比如，在谷歌之前就有搜索引擎了。

Like, search was a thing before Google.

只是那时做得不太好。

It just wasn't very good.

所以在Facebook之前也有很多社交产品，你知道，正是这些铺垫催生了它。

So so there were lots of so there was lots of social stuff before Facebook, and, you know, that was the thing that that catalyzed it.

所以我认为从确定性来看，这一切都还太早期，感觉未来肯定会出现几十、上百种新事物。

So, you know, I just think deterministically, this whole thing is so early that it feels like, of course, there are going to be, you know, dozens, hundreds of of new things.

否则H、C和Z这些机构就该关门把钱退还给LP了，对吧。

Otherwise, H, C, and Zs would just kind of shut down and give the money back to the LPs because Right.

粉丝经济模式就会接管整个行业。

The the the fan models will just do the whole thing.

而且我觉得你们不会那么做。

And, like, I don't think you're gonna do that.

至少我希望不会。

At least I hope not.

不。

No.

不。

No.

不。

No.

如果要说过去几年有什么遗憾，那就是我们没能更放手一搏。

If we have any regrets from the last few years, it's it's it's not going bigger.

我认为我们当时没有充分意识到各个领域会出现如此高度的专业化——无论是语音生成、图像生成还是任何细分领域，都会涌现出全新的公司，它们会比模型供应商做得更好。甚至每个领域都可能出现多个模型供应商。在Web2时代，我们总是押注品类赢家。

I I think we didn't fully appreciate how much specialization there would be across, whether it's voice or image generation or take any subsector that there would be net new companies created that would be better than the model providers, there would be even multiple model providers that or that in every category, you know, one thing we've always in the Web two era, we've always been on the category winner.

对吧？

Right?

虽然品类赢家通常会占据大部分市场，但这些市场如此庞大，专业化和细分程度如此之高，以至于每个领域都可能产生多个赢家。

And the category winner would take most of the market, but these markets are so big, and the the the there's so much expertise and specialization that in that there one, there can be winners in in every category.

这不仅仅是模型供应商通吃的局面，甚至在每个领域（包括模型供应商本身）都可能存在多个赢家，且专业化程度会持续加深——这些市场足够容纳多个成功者。

It's not just sort of the the model providers take everything, that even in every category, including the model providers, there can be multiple winners and increasing, you know, specialization, and and the the markets are just big enough to to contain multiple winners.

我认为这是对的。

I think that's right.

而且我认为，这些类别本身并不明确。

And I think, you know, the categories themselves aren't clear.

没错。

Right.

而且，你知道，很多人以为这是一个类别，结果发现并不是。

And, you know, many know, things you think this is a category and it turns out, no.

它实际上是完全不同的东西，这些类别会以不同方式被拆分、组合和重新整合。

It was actually that whole other thing, and the categories kinda get unbundled and bundled and recombined in different ways.

我记得1995年我还是学生时，我的电脑上装了大概四五个不同的网页浏览器和网页服务器。

I mean, I remember I was a student in 1995, and though I think I had like four or five different web browsers on my PC web web servers on my PC.

因为蒂姆·伯纳斯-李最初的网页浏览器里就内置了网页编辑器，他以为这更像是网络驱动器，是个共享系统。

Because I mean, Tim Berners Lee's original web browser had a web editor in it because he thought this was kind of like a network drive, and it was a sharing system.

当时并没有意识到这其实不是一个真正的出版系统。

Didn't realize not not really a publishing system.

所以你会把网页放在自己的电脑上，保持电脑开机，这样同事们就能查看你的Word文档或网页了。

So you would have your web pages on your PC, and you'd leave your PC turned on, and that would be how your colleagues would look at your Word documents or your web pages.

因此，我们真的不知道方法——我总是不停地回到这一点上。

And so, again, like, we just don't know how and and I I just kind of keep coming back to this point.

我觉得目前我们提出的问题可能大多是错误的。

I feel like most of the questions we're asking at the moment are probably the wrong question.

不过我注意到你刚才话里有个有趣的线索。

I'm picking up on on a on a strand within what you just said, though.

我最近经常思考的一个有趣话题是OpenAI，因为我对技术断层特别着迷。

The interesting one of the things I'm sort of thinking about a lot is looking at looking at OpenAI because, you know, I'm I'm I'm sort of fascinated by disconnections.

现在我们面临一个有趣的断层现象：如果你看基准测试分数，这些通用基准下各模型表现基本趋同。

And we've got this interesting disconnect now, which is that, you know, if you look at the benchmark scores so you've got these general purpose benchmarks where the models are basically all the same.

但如果你每天花数小时使用，就会形成偏好——比如更喜欢Claude的语调胜过GPT，或者更中意GPT5.1而非4.9版本（管它具体叫什么）。

And if you're, yes, if you're spending hours a day and then then you've got this opinion about, oh, I like Claude's tone of voice more than I like GBT, and I like GBT 5.1 more than GBD 4.9 or whatever the hell it's called.

如果你每周只用一次，根本注意不到这些差异。

If you're using this once a week, you really don't notice this stuff.

而基准测试分数都大致相同。

And the benchmark scores are all roughly the same.

但使用情况却不一样。

And but the usage isn't.

基本上Claude在消费者使用方面几乎为零，尽管它的基准测试分数是一样的。

It's basically the the only the the Claude has basically no consumer usage, even though on the benchmark score, it's the same.

然后是ChatGPT，再往下是Meta和Google。

And then it's ChatGPT, and then halfway down the chart, it's, Meta and Google.

有趣的是，你读所有AI新闻简报时，会发现Meta已经出局了。

And funny thing is, you know, that you read all the AI newsletters, and then, like, Meta's lost.

他们被淘汰了。

They're out of the game.

他们完蛋了。

They're dead.

马克·扎克伯格正花十亿美元请一个研究员来重返赛场。

Mark Zuckerberg is spending a billion dollars a researcher to get back in the game.

但从消费者角度来看，这其实是分发渠道的问题。

But from the consumer side, well, it's it's distribution.

这里有趣的是，我一直在绕圈子讨论的是：对于普通消费者用户而言，模型显然已是商品化产品，目前既没有网络效应，也没有赢家通吃的效应。

And the interesting thing here is that you've got what I'm kind of circling around is is the model for a casual consumer user certainly is a commodity, and there's no network effects or winner takes all effects yet.

这些效应未来可能会出现，但目前还不存在。

They may those may emerge, but we don't have them yet.

像记忆功能这类特性虽能增加用户粘性，但并非网络效应，而且很容易被复制。

And things like memory aren't network effects or stickiness, but they can be copied.

你们是如何竞争的？

How is it that you compete?

你们是否仅仅依靠品牌认知度、增加更多功能和服务来竞争，而用户因此不愿转投其他产品？就像Chrome浏览器的情况那样。

Do you just compete on being the recognized brand and adding more features and services and capabilities and people just don't switch away, which is kind of what happened with Chrome, for example.

Chrome并不具备网络效应，实际上它也没有好太多。

There's not a network effect for Chrome, but it and it's not actually any better much.

或许它比Safari稍好一些，但你知道，人们用Chrome只是因为习惯用Chrome。

Maybe it's a bit better than Safari, but, you know, you use Chrome because you use Chrome.

还是说你会因为分销渠道或别处出现的网络效应而被抛在后面？

Or is it that you get left behind on distribution or network effects that emerge somewhere else?

与此同时，你自己却没有基础设施。

And meanwhile, you don't have your own infrastructure.

所以我想说的是，虽然你们拥有这八九亿的周活跃用户，但这感觉非常脆弱，因为你们真正拥有的只是默认设置的影响力和品牌效应。

So I suppose what I what I'm getting at is, like, you've got these eight or 900,000,000 weekly active users, but you don't have but that feels very fragile because all you've really got is the power of the default and the brand.

你们并没有网络效应。

You don't have a network effect.

你并没有真正的功能锁定优势。

You don't really have feature lock in.

你没有一个更广泛的生态系统。

You don't have a broader ecosystem.

你也没有自己的基础设施，所以无法控制成本基础。

You also don't have your own infrastructure, so you don't control your cost base.

你不具备成本优势。

You don't have a cost advantage.

你每个月都会收到萨提亚寄来的账单。

You get a bill every month from Satya.

所以你必须在两个方向上尽可能快地行动，一方面要开发产品，在模型基础上构建更多东西，这是我们之前讨论过的。

So you've kind of got to scramble as fast as you can in both of those directions to, on the one side, build product and build stuff that on top of the model, which is our earlier conversation.

仅仅只是模型吗？

Is it just the model?

是的。

Yeah.

现在你必须在模型的各个方向上构建东西。

Now you've gotta build stuff on top of the model in every direction.

它是一个浏览器。

It's a browser.

它是一个社交视频应用。

It's a social video app.

它是一个应用平台。

It's an app platform.

就是这个。

It's this.

就是那个。

It's that.

就像那个贴满连线地图的梗图。

It's like the meme of the guy with the map with all the strings on it.

对。

Yeah.

懂吗？

Know?

所有这些事情。

It's all of these things.

我们昨天就要全部搞定。

We're gonna build all of them yesterday.

与此同时，它也是基础设施。

And then in parallel, it's infrastructure.

比如，你知道的，我们和OpenAI达成了协议。

Like and, you know, we we do we've got a deal with OpenAI.

我们很抱歉。

We sorry.

与NVIDIA、Broadcom、AMD、NVIDIA、Oracle都有合作，还有石油美元，因为你们正忙着从这项惊人的技术突破和八九亿的'哇'时刻，转向真正具有粘性、防御性和可持续的商业价值与产品价值。

Deal with with NVIDIA, with with with Broadcom, with AMD, with NVIDIA, with Oracle, and, well, with petrodollars because you're kind of scrambling to get from this amazing technical breakthrough and these eight hundred, nine hundred million wows to something that has really sticky, defensible, sustainable business value and product value.

是的。

Yeah.

那么在你评估超大规模云服务商之间的竞争格局时，你认为哪些问题对决定谁能获得持久的竞争优势，或者这场竞争将如何演变最为关键？

And so as you're evaluating the competitive landscape among the hyperscalers, what are the questions that you're ask that you think are gonna be most important in determining, you know, who who's gonna gain, you know, durable competitive advantages or or how this competitive is going to competition is gonna play out?

嗯，这某种程度上回到了你关于维持优势的观点，我们之前讨论过谷歌的情况。

Well, this kinda comes back to your point about sustaining advantage, and we we talked about Google.

就像，如果我们想想向移动端的转型，尤其是对Meta而言，这最终被证明是具有变革性的。

Like, if we think about the shift to particularly shift to mobile, for Meta, this turned out to be transformative.

它让产品变得实用得多。

Like, it made the products way more useful.

是的。

Yeah.

对谷歌来说，移动搜索本质上还是搜索。

For Google, it turned out mobile search is just search.

地图可能有所改变，YouTube也稍有变化。

And maps changed probably, and YouTube changed a bit.

但基本上，谷歌搜索就是搜索，网页搜索意味着更多人更频繁地进行搜索。

But, basically, for Google search, Google search is search, and the web web search is just mean means more people doing more search more more of the time.

没错。

Yeah.

目前的主流观点似乎是，Gemini下周就能和其他模型一样出色，比如新版本。

And the default view now would seem to be, well, Gemini is as good as anybody else next week, like the new model.

我还没看今天发布的GPT 5.1的基准测试结果。

I haven't looked at the benchmarks for GPT 5.1, which is out today.

它比Gemini更强吗？

Is it better than Gemini?

很可能。

Probably.

下个月它还会更好吗？

Will it still be better next month?

不会。

No.

所以这是理所当然的。

So that's a given.

比如，你有一个前沿模型。

Like, you've got a frontier model.

好吧。

Fine.

那要花多少钱？

What does that cost?

成本由你随便说个数。

It costs you pick a number.

每年2500亿美元，每年1000亿美元。

$250,000,000,000 a year, a $100,000,000,000 a year.

这是什么？我们之前关于资本支出的谈话是什么？

What's this what is our earlier conversation about CapEx?

好的。

Okay.

所以谷歌可以支付这笔费用，因为他们有钱。

So Google can pay that in because they've got the money.

他们从其他所有业务中获得了现金流。

They've got they've got the cash flow from everything else.

这样做了之后，你现有的产品就能获得优化搜索。

And so you do that, and your existing products get you optimized search.

你优化了广告业务。

You optimize your ad business.

你构建，你知道的，你构建新的体验。

You build, you know, you build new experiences.

也许你会发明新一代的人工智能iPhone。

Maybe you invent the new the iPhone of AI.

也许根本不存在人工智能的iPhone。

Maybe there is no iPhone of AI.

也许别人先做出来，你就像安卓那样直接抄袭。

Maybe someone else does it, you do an Android and just copy it.

那也没关系。

So fine.

这是新的移动时代。

It's a new mobile.

我们会继续前进。

We'll just carry on.

搜索就是搜索。

Search is search.

人工智能就是人工智能。

AI is AI.

我们会做新的事情。

We'll do the new thing.

我们会把它做成一个功能。

We'll make it a feature.

我们会继续做下去。

We'll just carry on doing it.

对Meta而言，似乎存在更大的问题：这对搜索意味着什么，对内容、社交、体验和推荐又意味着什么，这使得他们必须拥有自己的模型，就像谷歌一样。

For Meta, it feels like there are bigger questions on what this means to search, on what it means for content and social and experience and recommendation, which makes it all that more imperative that they have their own models just as it is for Google.

对亚马逊来说，好吧，一方面，它是商品化的基础设施，我们会把它作为商品化基础设施来销售。

For Amazon, okay, well, on the one side, it's commodity infra, and we'll sell it as commodity infra.

另一方面，或许或许退一步说。

And on the other side and maybe maybe stepping back.

如果你不是超大规模云服务商，而是网络出版商、营销人员、品牌方、广告商或媒体公司，你可以列出一系列问题。

If you're not a hyperscaler, if you're a web publisher, a marketer, a brand, an advertiser, a media company, you could make a list of questions.

嗯，就像，你现在甚至不知道问题是什么。

Well, like, you don't even know what the questions are right now.

是的。

Yeah.

如果我向聊天机器人提问而不是问谷歌，会发生什么？

What is this what happens if I ask a chatbot a thing instead of asking Google?

即使是从谷歌的角度来看，我也会问谷歌的聊天机器人。

Even if it's Google from from Google's point of view, what I'll ask Google chatbot.

没关系。

It's fine.

但作为营销人员，这意味着什么？

But as a marketer, what does that mean?

会发生什么？

What happens?

如果我询问食谱而LLM直接给出答案，这对以食谱为业务的公司意味着什么？

If I ask for a recipe and the LLM just gives me the answer, what does that mean if my business is having recipes?

是的。

Yeah.

你是否存在某种分割——这也是一个关于亚马逊的问题。

Do you have a kind of split between and this is also an Amazon question.

购买决策是如何产生的？

How does the purchasing decision happen?

这个购买我原先不知道存在的东西的决定是如何发生的？

How does this decision to buy a thing that I didn't know existed before happen?

如果我用手机对着客厅问'我该买什么？'会发生什么？

What happens if I wave my phone at my living room and say, what should I buy?

这会以哪些过去不可能的方式引导我的消费？

Where does that take me in ways that it wouldn't have taken me in the past?

因此下游存在大量问题，而这些会回溯影响Meta，在某种程度上也影响谷歌。

So there's a lot of questions further downstream, and that goes upstream to Meta and to some extent for Google.

长期来看，这对亚马逊而言是个更重大的问题。

It's a much bigger question in the long term for Amazon.

大型语言模型是否意味着亚马逊终于能实现真正规模化、优质的推荐、发现和建议功能？这是它过去因纯粹的商品零售模式而难以做到的。

Do do LLMs mean that Amazon can finally do really good at scale recommendation and discovery and suggestion in ways that it couldn't really do in the past, because of this kind of pure commodity retailing model that it has?

苹果公司某种程度上处于边缘位置。

Apple Apple sort of off on one side.

有趣的是，两年前他们曾提出一个关于Siri应该是什么样子的极具吸引力的愿景。

You know, interestingly, they produced this incredibly compelling vision of what Siri should be two years ago.

结果证明他们无法实现它。

It just turned out that they couldn't make it.

有趣的是，当时其他公司也做不到。

Interestingly, nobody else could have made it either.

你回头看看他们展示的Siri演示视频，就会想：好吧。

You go back and watch the Siri demo that they gave, and you think, okay.

所以我们需要的是一种多模态即时设备工具，能实时使用跨平台代理电商功能，没有提示注入问题且零错误率。

So we've got multimodal instantaneous on device tool using agentic multi platform e commerce in real time with no prompt injection problems and zero error rates.

嗯，听起来不错。

Well, that sounds good.

我是说，有人实现这个了吗？

I mean, has anyone got that working?

比如，没有。

Like, no.

OpenEye、谷歌和OpenEye，都没能实现那个功能。

OpenEye open Google and OpenEye, didn't have that working.

我是说，我不认为谷歌或OpenEye能复现苹果两年前展示的Siri演示。

I mean, I don't think Google or OpenEye could deliver the Siri demo that Apple gave two years ago.

他们或许能临时做个演示，但无法持续稳定地实现它。

I mean, they could really do the demo, but they couldn't, like, consistently, reliably make it work.

那个演示里的产品至今没出现在安卓系统上。

I mean, that that demo, that product isn't in Android today.

苹果——在我看来苹果面临最耐人寻味的问题，就像克雷格·费德里吉说的：我们连自己的聊天机器人都没有。

And Apple I mean, Apple to me has the most kind of intellectually interesting question, which is so I saw Craig Craig Federighi make this point, which is like, we don't have our own chatbot.

好吧。

Fine.

我们也没有YouTube或Uber。

We also don't have YouTube or Uber.

解释为什么这有所不同，这个问题回答起来比听起来要难得多。

What what explain why that is different, which is a harder question to answer than it sounds like.

当然，答案是如果这从根本上改变了计算的本质，那就是个问题。

And of course, the answer is if this actually fundamentally changed the nature of computing, then it's a problem.

如果它只是像谷歌那样的一项服务，那就不是问题，这也是关于Siri未来走向的关键点。

If it's just a service that you use like Google, then that's not a problem, which is kind of the point about about, you know, where does Siri go?

但这里有个有趣的坦率例子，可以想想微软二月份的经历——整个开发者生态环境脱离了他们的掌控，2001年后就没人再开发Windows应用了。

But the interesting candor example here would be to think about what happened to Microsoft in the February, which is the entire dev event environment gets away from them, and no one builds Windows apps after, like, 2001 or something.

但你需要使用互联网。

But you need to use the Internet.

要使用互联网，你需要一台个人电脑。

To use the Internet, you need a PC.

那你会买什么样的电脑呢？

And what PC are you gonna buy?

那时候苹果基本上还不是主要玩家，或者说刚刚重新进入这个市场。

Well, like Apple's like not really a player at that time and or just getting back into the game.

Linux显然对普通人来说不是一个选择。

Linux is obviously not an option for any normal person.

所以你买了一台Windows电脑。

So you buy a Windows PC.

所以基本上，微软失去了平台，却卖出了数量级更多的个人电脑。

So basically, Microsoft loses the platform more and sells an order of magnitude more PCs.

不是说他们在销售，而是由于微软这次失败，Windows电脑的数量级反而增长了。

Like, not selling them, but in order of they're all in order of magnitude more Windows PCs as a result of this thing that Microsoft lost.

直到移动时代来临，他们才既失去了设备也失去了开发环境。

And then it takes until mobile that, like, then they lose the device as well as the development development environment.

那么这里的问题是：如果所有新事物都建立在AI上，而我通过App Store下载的应用来访问它，这对苹果来说在多大程度上是个问题？

So here's this kind of question is if all the new stuff is built on AI and I'm accessing it in app that I download from the App Store, to what extent is this a problem for Apple?

要让这成为苹果的问题，需要发生更根本性的转变。

And one would have to you you would need a much more fundamental shift in what it was that was happening for that to be a problem for Apple.

即便你设想那种——不是完全的——就像《天降美食》里那样，我们都住进豆荚舱生活的极端场景。

And even if you take, like, the, you know, not the, like, the full, like, the rapture arrives, and we all just kinda go and live sleep in pods like the guys in up.

不是《飞屋环游记》

Not Up.

对

Yes.

那是什么？

What is it?

就是那个机器人收集垃圾的片子

The one with the robot that's capturing the trash.

哪部来着？

Which one is that?

《机器人总动员》

Wally.

《机器人总动员》

Wally.

《机器人总动员》

Wally.

对。

Yeah.

你知道那部电影里住在舱里的人吧。

You know the guys in the pods in that movie.

也许我们会成为那样的人，那样也行。

Maybe we'll be the people maybe we'll be like that, in which case, fine.

但存在一种中间情况，就是软件的本质完全改变，不再有应用，你只需去问大语言模型就行。

But, like, there's a sort of a mid case, which is like the whole nature of software changes, and there are no apps anymore, and you just go and ask the LL M a thing.

好吧。

Fine.

你用什么设备去问大语言模型呢？

What is the device on which you ask the LL M a thing?

它可能会配个漂亮的大彩屏，电池大概能用一天。

Well, it's probably gonna have a nice big color screen, and it's probably gonna have like a one day battery life.

可能会用麦克风，应该还有不错的摄像头。

Probably use a microphone, probably a good camera.

是啊。

Yeah.

听起来有点像iPhone。

It's kinda sounds like an iPhone.

对。

Yeah.

我会去买那个价格十分之一但只能用LLM的设备吗？

Am I going to buy the one that's a tenth of the price and just use the LLM on it?

不会。

No.

因为我仍然想要好摄像头、好屏幕和长续航。

Because I'll still want a good camera and this good screen and the good battery life.

所以当你深入思考时，会浮现出一系列有趣的战略问题。

So it's not there's a bunch of kind of interesting strategic questions when you start poking away.

那么这对亚马逊意味着什么？

Well, what does this mean for Amazon?

这些问题与对谷歌意味着什么、对苹果意味着什么、对Facebook意味着什么、对Salesforce意味着什么，或者你知道的，对Uber意味着什么，是完全不同的问题。

Those are completely different questions to what does it mean for Google, or what does it mean for Apple, what does it mean to Facebook, or what does it mean to Salesforce, or what does it mean to, you know, Uber.

然后回到我们对话开始时所说的，你知道，这对Uber意味着什么？

And then right back to what we were saying at the beginning of this conversation, you know, what does this mean for Uber?

嗯，他们的运营效率提高了X个百分点，现在欺诈检测也有效了。

Well, their efficiency get operations get x percent more efficient, and now the fraud detection works.

而且，你知道，好吧，也许他们会有自动驾驶汽车。

And, you know, okay, maybe they're autonomous cars.

那是另一个话题了。

Different conversation.

但假设没有自动驾驶汽车。

But presume no autonomous cars.

那完全是另一回事。

That's a whole other conversation.

否则，作为Uber，这会改变什么？

Otherwise, as Uber, what does this change?

嗯，变化不是很大。

Well, not a huge amount.

我想稍微拉远一点看整个框架。

I wanna sort of zoom out a little bit, the this whole framing.

所以你做这些演示已经有一段时间了。

The so you've been doing these presentations for a while now.

你知道，因为变化太多，你已经把频率提升到每周两次了。

You know, you bumped them up to two times because there's so much is changing.

而且你每次演示的招牌动作就是——你以提出精彩问题而闻名，并记录下哪些是当前最值得追问的关键问题。

And and one of the things you do in each presentation is is you're famous for asking, you know, really great questions and chronicling what what are the important questions to to be asking.

我很好奇，当你回顾时——比如在2022年ChatGPT出现后，或者说GPT-3问世后——你当时提出的问题与现在对比，我们在多大程度上明确了某些问题的方向？这些问题在多大程度上依然相同，或是出现了新的不同问题？

I'm I'm curious as you reflect, you know, maybe post, you know, Chad GBT in 2022 or GBT three rather, The questions you were asking then and you reflect on to now, to what extent, do we have some direction on some of those questions, or to what extent are they the same questions or or new and and and different questions?

或者说，如果我昏迷后醒来，读完你最初的演示稿——比如GPT-3发布后的那一版——再看现在的版本，最令人惊讶的认知更新是什么？哪些发现重塑了那些问题？

Or what what is sort of your you know, if I woke up on a in a coma after reading your, you know, your original presentation, let's say, you know, the one after g p t three launch came out and then seeing this one now, what were the sort of most surprising things or things that we we we learned that updated the those questions?

所以我认为今年我们有很多新问题。

So I think we have a lot of new questions this year.

所以我觉得，你可以列出一份清单，可能包含2023年的六七个问题，比如开源、中国、英伟达、规模扩展是否持续？

So I feel like, you know, you could make a list of as it might be half a dozen questions in '23, like open source, China, NVIDIA, does scaling continue?

图像领域会发生什么变化？

What happens to images?

OpenAI的领先优势还能保持多久？

Does how how long does OpenAI's lead remain?

这些问题在2023到2024年间其实没有太大变化。

And those questions didn't really change in '23 and '24.

其中大部分问题至今仍然存在。

And most of those questions are kind of still there.

比如关于英伟达的问题就基本没变。

Like, the NVIDIA question hasn't really changed.

对吧？

You know?

他们就像，答案不会尝试。

They like, the answer won't try.

答案在于，你知道，是否会有，嗯，会有多少模型出现？

The answer on, you know, will there be well, how many models will there be?

答案是，好吧，那些能投入几百亿、几千亿美元的公司才能拥有前沿模型。

The answer is, okay, there's gonna be who can spend a couple of 100 can can can spend a couple of billion dollars can have a Frontier model.

我认为这在23年就已经相当明显了。

And that was, I think, pretty obvious, know, only 23.

大家花了一段时间才理解这一点。

It took a while for everyone to understand that.

至于大模型和小模型，我们现在已经有在设备上运行的小模型了。

And big models and small models, well, we have small models running on devices.

不，因为小模型的能力发展太快，还来不及将小模型压缩到设备上。

No, because the small models the capabilities keep moving too fast for the small models to get to shrink the small model onto the device.

但这些问题在过去两年半里基本没有变化。

But those questions kinda didn't change for two, two and a half years.

我认为我们现在面临更多产品战略问题，随着看到真实的消费者采用情况，OpenAI和谷歌在不同方向构建产品，亚马逊走不同路线，苹果显然尝试失败后又重新尝试。

I think we now have, I think, a bunch of more product strategy questions as you see real consumer adoption and OpenAI and Google building stuff in different directions, Amazon going in different directions, Apple trying and obviously failing and then then trying again to do stuff.

这个行业里似乎正在发生着比单纯'让我们再建一个模型，投入更多资金'更深层次的事情。

There's some sense of, like, there is something more going on in the industry than just, well, let's just build another model and spend more money.

是的。

Yeah.

现在有更多的问题和决策需要考虑。

There's more questions and more decisions now.

在技术领域之外，特别是在零售媒体方面，也涌现出更多问题：你该如何开始思考如何运用这项技术？

There's also more questions outside of tech in certainly on, like, the retail media side of, how do you start thinking about what you would do with this?

再次强调，我演示文稿中的经典框架是：第一步是将其作为功能吸收，做显而易见的事。

And again, you know, classic framing in my deck is like step one is you make it a feature and you absorb it and you do the obvious stuff.

第二步是尝试新事物。

Step two is you do new stuff.

第三步可能是有人会彻底颠覆整个行业，完全重新定义问题。

Step three is maybe someone will come and pull the whole industry inside out and completely redefine the question.

因此你可以在这里做个思想实验：第一步是假设你是湾区或华盛顿特区某家沃尔玛的经理...

And so you could kind of do like an ImagineIF here of like, step one is, you know, you're you're a manager at a Walmart in the Bay Area or DC or whatever it is.

第一步是找到那个指标。

Step one is find me that metric.

第二步是给我建一个仪表盘。

Step two is build me a dashboard.

第三步是黑色星期五那天，我正在华盛顿特区外经营一家沃尔玛。

Step three is it's Black Friday, and I'm running managing a Walmart outside of DC.

我该担心什么？

What should I be worried about?

这个例子可能不太恰当，但就像亚马逊的第一步是你买了灯泡。

Like and that might be the wrong one, but it's like, you know, step one for Amazon is you bought light bulbs.

你买了气泡膜，所以这里有些打包胶带。

So here's so you bought bubble wrap, so here's some packing tape.

但亚马逊真正应该做的是说：'看起来这个人正在搬家，我们会给他们展示家居保险广告'，这是亚马逊的关联系统无法实现的，因为他们的购买数据中没有这类信息。

But what Amazon should actually be doing is saying, looks like this person's moving home, we'll show them a home insurance ad, which is something that Amazon's correlation system wouldn't get because they wouldn't have that in their purchasing data.

我们目前还处于非常初级的阶段——仍在第一步摸索，但已经开始思考第二步、第三步会是什么样子。

And we're still very much at the, like we're still starting to we're we're we're still on the step one of that, but thinking much more what would the step two, step three be.

除了简单的自动化之外，这项技术还能带来哪些新的收入来源？

What would new revenue be for this other than just, simple dumb automation?

我们能利用这项技术构建哪些新事物？

What would new things that we would build with this be?

这项技术可能会在哪些方面重新定义或改变市场格局？

Where would this actually like, might might actually kind of redefine or change what the market might look like?

这对内容行业的从业者来说显然是个重大问题。

And that's obviously a big question for anyone in the content business.

是啊。

Yeah.

想想看，如果我能直接去问大语言模型这个问题意味着什么？

You know, what does it mean if I can just go and ask an LLM this question?

哪些类型的内容是基于谷歌把这个问题导流给你的前提而存在的？

What kinds of content were predicated on Google routing that question to you?

又有哪些内容其实根本算不上是真正的问题？

And what kind of questions what kind of content isn't really that question?

比如，我是想要一个博洛尼亚肉酱食谱，还是想听斯坦利·图齐讲述在意大利烹饪的故事？

Like, do I want a Bolognese recipe, or do I want to hear Stanley Tucci talking about cooking in Italy?

比如，我是单纯想要这个偏好的内容，还是想弄清楚该买哪个产品？

Like, do I just want the do I want that skew, or do I want to work out which product I should buy?

亚马逊擅长给你提供商品编号，却完全不擅长告诉你该选哪个商品。

Which is Amazon is great at getting you the SKU, terrible at telling you what SKU you want.

我是只想要这份幻灯片，还是愿意花一周时间与贝恩的合作伙伴讨论如何推进这件事？

Do I just want the slide deck, or do I want to spend a week talking to a bunch of partners from Bain about how I could think about doing this?

我是单纯想要钱，还是想与16z的运营团队合作？

Do I just want money, or do I want to work with a 16 z's, you know, operating groups?

是啊。

Yeah.

比如，我到底在这里做什么？

Like, what is it that I'm doing here?

我认为LLM正在以多种方式让这个问题逐渐清晰起来。

And I think the the LLM is starting thing is starting to crystallize that question in lots of different ways.

是啊。

Yeah.

比如，我到底想在这里做什么？

Like, what am I actually trying to do here?

我是只想要一个现在电脑能回答我的东西，还是想要其他电脑无法提供的东西？

Do I just want a thing that a computer can now answer for me, or do I want something else that isn't?

因为LMS能做很多以前电脑做不到的事情。

Because the LMS can do a bunch of stuff that computers couldn't do before.

对。

Right.

电脑以前做不到的事情现在是我的业务了吗？

Is that thing that the computer couldn't do before my business?

是的。

Yeah.

还是我其实在做别的事情？

Or am I actually doing something else?

我们即将以更细致的方式弄清楚，对于许多许多情况来说，真正的待完成工作是什么

We're we're about to figure out what is the in a much more granular way, what what what is the true job to be done for for for many, many of these

是的。

Yeah.

而且，回顾互联网的发展，有一种关于报纸的观察是，报纸审视互联网时，谈论的是专业知识、内容筛选、新闻报道等等，却从未真正承认过：我们本质上是一家轻制造企业兼本地物流运输公司。

And, you know, going back to the Internet, there was, you know, the the sort of observation about newspapers is that newspapers looked to the Internet, and they talked about, you know, expertise and curation and journalism and everything else and didn't really say, well, we're a light manufacturing company and a local distribution and trucking company.

没错。

Yep.

而这正是问题所在。

And that was the bit that was the problem.

在互联网出现之前，这类讨论根本不会进入人们的思考范畴。

And until the Internet arrived, like, that wasn't a conversation you thought about.

随后互联网突然让这一切变得清晰，并实现了前所未有的业务解构。

And then the Internet suddenly makes that clear and suddenly creates an unbundling that didn't exist before.

因此将会出现这样的情况：在LLM出现之前，你从未意识到自己的本质——直到有人带着LLM指出：看，我能用这个实现你从未意识到构成自身防御壁垒或盈利基础的功能。

And so there will be those kinds of, like, you didn't realize you were that before until an LLM comes along and points to someone comes along with an LLM and says, oh, I can use this to do this thing that you didn't really realize was the basis of your defensibility or the basis of your profitability.

就像那个关于美国医疗保险的笑话——美国医疗保险的盈利基础就是把事情搞得极其无聊、困难和耗时。

I mean, it's like the joke about US health insurance that the basis of US health insurance profitability is making it really, really boring and difficult and time consuming.

利润正是来源于此。

That's where the profits come from.

也许并非如此。

Maybe it isn't.

不知道。

Don't know.

不知道。

Don't know.

不同意。

Disagree.

但为了讨论方便，假设这就是你的防御壁垒。

But for the sake of argument, say that's that's your defensibility.

而大语言模型能消除那些无聊、耗时又令人麻木的任务。

Well, an LLM removes boring, time consuming, mind numbing tasks.

是的。

Yeah.

那么哪些行业因此受到保护呢？

So what industries are protected by having that?

而他们当时并未意识到这点。

And they didn't realize that.

你知道，这就像在90年代中期关于互联网，或是十年后关于移动技术提出的那些问题。

And these you know, it's like you could have asked these questions about the Internet in the mid nineties or about mobile a decade later.

通常来说，事后回顾时你会发现，当时提出的问题有一半都是错误的。

And, generally, you'd have half of the questions you'd asked would have been the wrong questions in hindsight.

我记得2000年刚当分析师时，所有人都在问：3G的杀手级应用是什么？

I mean, I remember as a as a baby analyst in 2000, everyone kept saying, what's the killer use case for three g?

3G有什么好的应用场景？

What's a good use case for three g?

结果发现，随时随地口袋里的互联网就是3G的应用场景。

And it turned out that having the Internet in your pocket everywhere was the use case for three g.

是啊。

Yeah.

但当时人们并没有提出这个问题，我确信现在的情况会是：未来会有太多事物被创造出来，当你亲眼目睹时才会恍然大悟——原来应该这样做。

But that wasn't the question that people were asking, and I'm sure that will be the thing now is there's so much that we will that will happen and get built where you go and you realize, oh, that's how you would do this.

你可以把它变成那样。

You can turn it into that.

对。

Yeah.

我相信你在观察创业者时也有过这种体验。

And I'm sure you've had this experience seeing entrepreneurs.

你知道，时不时会有创业者来提案，听完后你会想‘哦，原来可以这样’。

You, you know, you get every now and then, they come in, they pitch the thing, you're like, oh, okay.

你可以把它变成那样。

You can turn it into that.

我原先没意识到它能变成那样。

It didn't I didn't realize it was that.