OpenAI 与 Anthropic 的直接对决 + 代理的未来 —— 与亚伦·莱维对话

随着OpenAI和Anthropic几乎都在打造相同的产品，它们之间的竞争现在进展如何？

How is the battle between OpenAI and Anthropic shaping up now that they're both basically building the same product?

AI代理的未来会怎样？

And what is the future of AI agents?

让我们在接下来的时间，与Box公司的首席执行官亚伦·莱维一起探讨这个问题。

Let's talk about it with Box CEO, Aaron Levie, right after this.

欢迎收听《大科技播客》，这是一档致力于对科技世界及其更广泛领域进行冷静而深入对话的节目。

Welcome to Big Technology Podcast, a show for cool headed and nuanced conversation of the tech world and beyond.

今天为大家准备了一场精彩的节目。

We have a great show for you today.

我们现在将深入剖析OpenAI和Anthropic之间的竞争，因为它们的产品路线图已经基本趋同。

We're gonna unpack the battle between OpenAI and Anthropic now that their product road maps have pretty much converged.

我们还将讨论AI代理的现状与未来，以及这项技术的发展方向。

And we'll also talk about the future and the present of AI agents and where that technology is heading.

今天我们邀请到的嘉宾是Box公司的首席执行官亚伦·莱维。

And joining us is Aaron Levie of Box, CEO of Box.

亚伦

Aaron

谢谢。

Thank you.

欢迎。

Welcome.

是的。

Yeah.

很高兴来到这里。

Good to be here.

我确实很喜欢你对这场竞争的描述。

I I certainly like the framing on the battle.

我认为，在某种程度上，这是不可避免的结果，因为如果你仔细想想，如果有一个超级智能的AI模型，它最终必然会涵盖所有相同的应用场景。

You know, I think it's, to some extent, was sort of an inevitable outcome because if you think about it, like, if you have this AI model that is super intelligence packed into a model, it eventually has to converge on all of this all the same use cases will be represented by that.

因此，我认为这些实验室最终必须在所有这些应用场景上正面竞争。

And so then I think the labs eventually need to compete head to head for all those use cases.

是的，我很高兴在第一个问题之前就能展开这场讨论。

Yeah I'm glad to get this discussion going even before the first question

还没到提问环节呢，我觉得你的开场白本质上就是一个问题，那为什么不呢？

comes I like I'll fray your intro is basically a question so why not?

没错，但这就是正在发生的事情。

That's right but it is really what's happening.

所以，为了明确一下，我们看到Anthropic在企业市场领先，而OpenAI似乎对此感到满意。

So just to frame it we saw Anthropic take the lead in enterprise and OpenAI seemed satisfied.

对于编程，是的。

For coding yes.

对于编程，而且他们还通过API向企业销售，这最初让我相信，Anthropic的发展轨迹就是AI的发展轨迹，因为如果这项技术对企业有用，那就意味着它的盈利上限会更高。

For coding but also they were selling into enterprises through the API and that was what were my belief initially about Anthropic came that as Anthropic goes so goes AI because if this technology is useful to businesses that means that the cap on the amount of money that it can make is gonna be higher.

所以Anthropic大胆押注企业市场和编程，并取得了巨大成功。

So Anthropic made this big bet on enterprise and on coding and crushed it.

而OpenAI则大力押注消费者市场。

And OpenAI made this big bet on consumer.

顺便说一下，Apache BT 目前可能已经有十亿用户了，尽管还没有正式公布。

Apache BT by the way is probably at a billion users right now even if it's not announced.

他们在那方面做得很好，但后来发生了一件有趣的事：到了十二月，编码模型变得足够强大，能够无需中断地处理长期任务，并且对非技术人员也变得有用。

And they did very well there but then something interesting happened where the coding models in December became good enough to code for kind of long time horizons without interruption and they became useful to even the non technical folks.

是的。

Yep.

然后我们看到这两家公司都开始想要打造一种超级应用风格的产品，这基本上就是问题所在。

And then we saw this emergence of both these companies wanting to build this super app style thing that basically that's sort of what the question is.

它会成为你的助手吗？

Is it gonna be an assistant for you?

它会代替你完成工作吗？

Is it going to be something that does your work?

他们说，这两家公司都希望为你做一切事情。

They say it once they both wanted to do kind of everything for you.

你觉得这会走向何方？

Where do you see that going?

是的。

Yeah.

你如何看待这场竞争的发展趋势？

And how do you see the battle shaping up?

是的。

Yeah.

让我先针对你最初的框架补充两点想法，然后再直接回答你的问题。

So let me let me just inject two couple quick thoughts in your in your initial framing, and then I'll answer the question more directly.

我认为，要代表Anthropic和OpenAI双方的观点，这个故事可能比你最初设想的还要复杂，因为他们实际上认为Chat2BT已经渗透到企业领域，并在企业部署中取得了相当大的成功，这与API业务是分开的。

I think I think probably the to represent both both sides of Anthropic and and OpenAI is I I I think the probably the the story might be even more kinda complicated than than even that initial framing because they actually think Chat to BT leaked into the enterprise and has had actually a lot of enterprise traction of enterprise deployments, which is separate from the API business.

因此，如果你去很多企业看看，他们实际上会把Chat2BT作为员工使用的公司标准LLM。

And so the if you go to a lot of enterprises, they actually will have Chat2bt as their corporate standard for their corporate LLM for employees to use.

所以很难确定该参考哪些数据，但我通常认为，两者在企业市场都表现得极为出色。

So it's hard to decide what data do you end up but looking at but I would I would generally argue that both have done actually extremely well in the enterprise.

而ChatGPT，显然历史上更专注于消费者市场。

And and and ChatuchPut tea, obviously, even more focused on the consumer historically.

而现在，显然你面临着在编码、API以及最终用户的企业知识工作场景中争夺企业主导地位的激烈竞争。

And now, obviously, you have this increased battle for enterprise dominance both with coding, the APIs and the end user kind of corporate knowledge work use case.

所以可以说是

So kind of co

协同工作场景。

work use cases.

协同工作场景是第三个关键突破，最近几个月才真正出现——那就是，如果一个智能体在编程方面极其出色，但其用途并不是开发软件，会怎样？

The co work use case being kind of third one and the big breakthrough that has happened recently, literally just you know, recently in the past few months is this idea of what if you could give what if an agent was really, really good at coding, but the use case wasn't to build software.

它的用途是利用其编程能力、通用工具调用能力以及运行脚本的能力。

The use case was to use its coding skills and general kind of tool calling skills and the ability to run scripts.

如果这个智能体在这些能力上都非常出色，但应用于其他知识工作领域，会怎样？

What if the agent was really good at at all those capabilities, but was applied to the rest of knowledge work?

这会打开哪些类型的应用场景？

And what what kinds of use cases would that open up?

大致的思维模式是：如果每个人都能真正精通使用电脑，能为任何任务编写代码，会怎样？而这个精通电脑、会写代码的人，可能是一名律师、一名市场人员、一名生命科学领域的研究人员。

And and, know, kind of the mental model is like, what if everybody was, like, truly an expert at using their computer and they could write code for any task they wanted to do, but that same, you know, person that was the expert at using their computer and and, you know, writing code was a lawyer, and they were a marketer, and they were a they were in life sciences, and they did research.

这基本上就是当今智能体的力量，而且这种趋势会越来越明显。

That's that's basically the power of of agents today more and more in in terms of where we're going.

因此，协作工作这个理念很早就很好地体现了这一点。

And so the idea and and co work kind of, you know, best manifested this early on.

我想我们肯定会根据一些传闻看到进展。

I think we'll certainly, you know, you know, see based on the rumors.

OpenAI 在这个领域有布局，其他玩家也在思考：如果你有一个通用的知识工作者智能体，它可以使用你电脑上的每一个工具，那会怎样？

OpenAI have a presence in this space and other players is, you know, what if you had an agent that was your general purpose knowledge worker agent, but again, it could it could use every tool on your computer.

它能即时为从未遇到过的新问题编写代码。

It can write code on the fly for a new problem that it hasn't seen before.

它能使用所谓的技能，来调用现有的持续运行的脚本和代码。

It can use things called skills to be able to leverage existing kind of ongoing scripts and and code that it needs to be able to use.

如果你身边有这样一个勤恳的助手，能拥有这样的超能力，那会是什么样的情景呢？

What kind of now superpower would that be, you know, to be able to have as as, you know, this kind of this workhorse that that you have next to you.

这正是人工智能智能体的下一个前沿。

That's kind of the next frontier of of AI agents.

因此，我们显然正从一个你只是与AI进行对话互动的世界，转向一个新的范式：AI代理被赋予任务，它拥有可访问的一组资源，可以接触你的数据、电脑上的软件、云端工具等，然后它能独立工作数分钟、数小时，甚至数天，生成有效的成果供你审查、使用，并融入你更广泛的工作中。

And so I think we're we're clearly moving from a world where you will use AI as this this thing you chat back and forth with, and that was kind of the first manifestation of the chatbot to now a paradigm where the agent is given a task, it has a set of resources it has access to, it has access to maybe your data, your software, tools on your computer, tools in the cloud, and it can go off and work for minutes or hours or maybe even days and go and generate some effective work output that you can then go and use, review, and then incorporate into your broader work.

因此，这是巨大的机遇，因为目标市场从原本的全体工程师，扩展到了如今的每一位知识工作者。

So this is kind of the big prize because it goes from the TAM, the total addressable market being, you know, all of engineers to now the total addressable market is every knowledge worker.

这大概意味着市场规模扩大了30到50倍，从全球人类的数量及其使用场景来看都是如此。

And that's probably about a 30 to 50 x larger market in terms of, you know, humans on the planet and and and their use cases.

所以你觉得这是先从商业领域开始的吗？

So you see this as business first?

这主要会是商业应用。

This is this is gonna be primarily business.

我认为但

I I think But

因为当我和格雷格·布罗克曼对话时，他说这就像一台笔记本电脑，你可以用它来处理个人事务。

it's because Greg Brockman when I had him on it as like a laptop where you could use your laptop for your personal stuff.

你也可以用它来处理企业事务。

You could use your laptop for your enterprise.

是的

Yeah.

我完全同意这种说法。

And and I I fully agree with that framing.

实际上，我认为这会推动它进入企业领域。

And I actually think that will suck it into the enterprise.

我认为我们会看到，这些令牌的价值和投资回报率——要知道，令牌短期内不会便宜，因此在企业环境中，这些令牌的投资回报率会高得多，因为它们会生成一些对GDP产生影响的东西。

I I think I think what we're gonna see is that the the value and the ROI on those tokens, know, the tokens are not gonna be cheap anytime soon, and so the ROI on those tokens will just be much higher in the enterprise because it'll be, you know, generating something that is sort of, you know, impacts the GDP in some way.

因此，我认为我们可能会优先将这些系统应用于这类活动。

And so I think that we will probably prioritize a lot of these systems toward those types of activities.

但我完全同意他的观点，即你会以通用方式使用它。

But I totally agree with his framing that you'll just use it in a general purpose way.

而且，如果你本来就是那种喜欢在生活中自动化、做大量自动化事情的人，你也会在个人层面使用它。

And probably the more that you're the kind of person that already likes to automate your life and, do a bunch of automation things in your personal life, you'll use this also in a personal capacity.

但我认为，它真正的经济价值主要来自企业。

But I think most of the true economic value of it will come from the enterprise.

这些东西能行得通吗？

Is this stuff going to work?

我的意思是，这其实有两个方面，对吧。

I mean there's two things to it right.

一方面是能力，另一方面是人们使用它的意愿。

There's the capability side and then there's also the interest in using it.

所以，再回到我上周跟格雷格提到的那个例子。

So again just going back to one of these examples that I spoke about with Greg last week.

基本上就是Codecs，OpenAI那个能替你干活的新编码应用工具。

Basically what Codecs OpenAI's you know new coding app that can do your work for you tool.

我还不太知道该怎么称呼它，但举个例子，如果你需要剪辑视频，它就能进入Premiere，自动给你视频加上章节。

I still don't really know how to refer to it but what it can do is just for one example it can if you need to edit a video it can go into Premiere and like put chapters in your video.

但我还觉得，我们真的需要软件来做这种事吗？或者人们会不会更愿意用老办法？这种工具能深入到什么程度？你认为它真能达到能直接剪辑视频，而不仅仅是加章节的地步吗？

But I also think like do we really need like software to do that or aren't people just gonna be aren't people just gonna prefer to do it the old way and how deep can it get like can it do you think this will actually get to the point where it can edit the video not just put the chapters?

是的，我认为这些正是人们新设立的个人评估标准或基准，比如，你什么时候能开始剪辑视频？

Yeah I think these are the these are these new you know these are like the new kind of personal evals or benchmarks that people have of like of, you know, when would when would you be able to edit a video?

我认为多尔卡什，我觉得达里奥也被问过这个问题。

And I think Dorcache, I think asked, you know, even Dario that question.

对吧？

Right?

他说，我们什么时候才能直接编辑整个内容呢？

And he's, you know, when when can we just edit this whole thing?

我们会迎来大量播客相关的评估标准。

We're just gonna get a lot of podcaster benchmarks.

是的。

Yeah.

没错。

Exactly.

没错。

Exactly.

这主要是我们

This is primarily We

应该让会计师来主持这个节目，是的。

should have accountants host this show Yes.

然后他们可以讨论一些真正实际的内容，

And then they can talk about stuff that actually Actually,

更有趣的问题是，所有的AI模型都在训练这些数据。

the the the more funny problem is like all of the AI models are being trained on all of this.

所以AI模型很可能认为，目前经济中最有效的活动是视频。

And so they probably the AI model probably think like the most useful activity in the economy right now Video.

是编辑播客视频。

Is editing podcast videos.

它们的奖励函数已经被优化得如此极致。

And they just like are the the reward function is like so optimized.

顺便说一句，如果它们真的优先考虑这个，我会非常高兴。

By the way, if that's what they prioritize, I would be thrilled.

是的。

Yeah.

是的

Yeah.

对

Yeah.

赶紧搞定吧，各位。

Get it done, folks.

我不知道。

I don't know.

更多的竞争。

More competition.

我不确定你是否想要那样。

I don't know if you want that.

所以这很好，没问题。

So it's it's good to It's fine.

让这种活动成为一种稀缺资源是好事。

It's good to have that as like a scarce activity.

但我并不太担心人们是否会想要这个，因为我觉得这就像传真机的论点：确实总会有一些人坚持不用，但我认为效率总体上总是会占上风，因为随着新技术的出现，人们最终会更重视自己的时间和时间的价值。

But I'm not worried so much about will people want this in the sense of because I think that's kind of like a fax machine argument, yes, there will always be holdouts, but I think efficiency generally always prevails, simply because you end up prioritizing your time and the value of your time as a new technology emerges.

你会想，嗯，我大概真的不想亲自去用传真机，把一张纸塞进去，然后输入一堆数字，而明明只需要把文件作为附件发到邮箱里就行了。

You're like, well, yeah, I probably don't want to literally go to a fax machine, you know, have to put a piece of paper in this thing, you know, type in a bunch of numbers if it's just as an attachment and I send it to an email address.

这简单了十倍。

Like, it's like 10 times easier.

所以我认为，这种变化将会发生在大量工作领域中，将来我们会回过头来看，觉得以前花两个半小时去读一篇研究论文，只是为了找到一个事实，这简直可笑，因为过去我们不知道那个事实会出现在论文的哪个地方。

So I I think we I think that will happen to a large set of areas of work, we'll look back and we'll we'll just consider it laughable that, we spent two and a half hours going and, like, reading some research paper just to find one fact because previously, we didn't know where that fact might be in the paper.

所以我们，你知道的，每个人都有自己的一些小技巧，比如快速浏览。

And so we, like, you know, we had, like we all have our own little tricks that we do some skimming.

我们会大致从空间上找一下相关区域，但仍然要花上一个小时。

We kinda look roughly spatially for the area, but it still takes like an hour.

而AI代理能替我们直接在三秒钟内完成这件事，而且再也回不去了。

Like, an AI agent just does that literally for us in three seconds, and there's no going back.

我们再也不想做那种事了。

Like, we don't wanna do that anymore.

所以问题来了，这种自动化能深入到工作的哪个程度呢？

So the question is like, you know, how deep can that go into work?

这些代理在工作中能持续运行多久，才需要你去审查它们的输出结果？

How long running can that work those agents be across work before you have to sort of review the output that the agent is doing?

这些模型在处理更主观的任务时表现如何？

How how well do these models work on much more subjective tasks?

比如剪辑视频，在很多情况下实际上比编程更难，因为代码现在有一个很好的特性——在评估和训练过程中，你可以立即判断代码是否运行成功、代码是否整洁。

Like editing a video is like is like, you know, gonna be actually in in many cases a harder task than coding because the because again, the code right now is like it has this great property of in the eval process, in the training process rather, you can instantly evaluate did the code run, how clean was the code.

我们有很多工作领域不具备这种即时验证的能力。

We have a bunch of areas of work that don't have they don't have that ability to instantly sort of verify.

因此，对代理来说，奖励机制要复杂得多。

So the reward function is a lot is a lot trickier for the agent.

于是在实际工作流程中，要自动化这类任务就变得相当困难。

And then thus in real life workflow, it's kind of hard to then go and automate that task.

所以我认为，这实际上需要比我们和硅谷有些人想象的更长时间才能实现。

So I think this is actually going take a lot longer to play out than maybe what we and some think in Silicon Valley.

因为在硅谷，我们会看到人工智能编程的全部潜力，而编程是硅谷最具经济价值的任务，所以我们往往根据AI编程的表现来推断其他所有领域。

Because what would happen in Silicon Valley is we sort of look at all of the power of AI coding, and because that's the most economically useful task within Silicon Valley, we sort of extrapolate most things from how good AI coding is.

既然AI能很好地编写代码，我们就认为它也能很好地处理法律、医疗、生命科学、建筑和设计等其他任务，因为我们正在将AI在编程中展现出的自动化收益推广到这些领域。

And because that then then we're like, AI can do code really well, then it probably can do legal and medical and and, you know, and life sciences and and architecture and design, all of those other tasks because we're kind of extrapolating the automation gains that we're seeing in AI and in coding.

但问题是，这一点已经被很多人在不同时期讨论过，我只是想分享一下我认为大家普遍认同的几个主要方面。

And the challenge is that that and this has been talked about, you know, by a bunch of folks at at at different times, but just to kind of, you know, sort of share a few of the big big buckets that I think everybody has kind of, you know, come down on.

在编程中，一切都是基于文本的。

In coding, you have you know, it's entirely text based.

你可以访问整个代码库。

You have access to the entire code base.

AI代理通常可以访问整个代码库。

The agent generally has access to the entire code base.

这些模型之所以被训练得非常好，是因为编程具有可验证性。

The models are are really, really trained on coding because, again, it's sort of verifiable.

你可以测试代码，看看它是否能正常运行。

You can test the code and see if it works.

在这种情况下，代理的使用者都是技术非常娴熟的人，他们对这些系统了如指掌。

The users of the agents in these cases are highly technical, so they know their way around these systems.

他们知道当代理失控时该如何把它拉回正轨，也知道如何安装它所需的最新插件。

They know when the agent goes kind of crazy, how to put it back on track, they know how to install the latest plug ins that it needs.

但如果你对比其他知识型工作，比如有人只是在做日常的市场营销工作，代理所需的上下文却分散在二十个不同的系统中。

Now you compare to the rest of knowledge work, where it's just somebody doing their daily marketing job, and the context the agent needs is in 20 different systems.

因此，每个系统都必须单独对接，或者你得把大量数据整合起来。

So each of those systems have to be individually wired up, or you have to consolidate a bunch of data.

用户可能并不特别懂技术，所以他们不得不花大量时间学习这些东西，而对非技术人员来说，学习新工具通常并不有趣，反而像是一种负担。

The user maybe is not insanely technical, and so they've got to go spend a bunch of time learning this stuff, and the learning of a new tool is just generally not that much fun for for people that aren't in tech because it's less like that's just like a pain.

他们无法像编码代理那样获得可验证性的优势，因此即使代理完成了一大堆工作，他们也必须在最后逐一审查，以确保所有内容在事实上准确，且产出符合应有的专业标准。

They don't get the same benefit of the verifiability of the coding agent, and so even when the agent goes and does a bunch of work, they have to have to go review the whole thing at the end of it because they have to make sure everything is sort of factually correct or has the right kind of sensibilities in what they produced.

所有这些因素，我们甚至还没谈到公司内部的治理政策和合规要求。

So all of those things are are and we haven't even gotten into like the governance policies, the compliance policies of that company.

因此，所有这些因素加在一起，意味着这类技术在全社会的普及将需要许多年的时间。

So all of those things add up to actually just meaning that the diffusion of these types of technologies will take many, many years as they go through the rest of the world.

我认为硅谷在这方面得有点耐心。

That's the part that I think Silicon Valley is going to have to be a bit patient on.

恰恰相反，正因如此，我认为现在有巨大的机会，因为如果你能构建产品和平台，作为通向这一最终状态的桥梁，尽可能简化企业走完这段旅程的难度，这将带来巨大的机遇。

And actually, that conversely is why I think there's so much opportunity right now is because if you can build products and platforms that are sort of the bridge to that end state and make it as easy as, you know, possible for enterprises to go down that journey, that's just a tremendous amount of opportunity.

所以实验室会这么做，OpenAI会这么做，Anthropic也会这么做。

So the labs are gonna do that and, you know, OpenAI will do that, Anthropic will do that.

还会有一批初创公司，在垂直领域或我们正在做的这种横向领域中做这件事。

There'll be a bunch of startups that do it in either vertical kind of categories or horizontals like what we're working on.

但这些才是主要的机会。

But that's sort of the big opportunities.

你能连接当今世界的工作方式与那个最终状态吗？

Can you bridge how the world works today to that end state?

但我认为，未来几年，大多数人会在工作中使用智能代理，因为效率提升实在太明显，根本无法回避。

But I think that I would expect most people have agents running in their daily life from a workplace standpoint over the coming years just because the efficiency will just be too strong to kind of avoid.

没错。

That's right.

我会提出一个观点，为了讨论方便，甚至可能进展得更快。

And I will make the argument that it might even go faster just for the sake of discussion.

视频编辑看起来非常主观，但实际上今天的技术已经可以做到这一点。

Video editing feels like pretty subjective but actually you can use technology today.

是的。

Yep.

比如说，如果亚伦在说话，我们就切到你的近景。

To be like all right if Aaron is speaking let's have the tight shot on you.

是的。

Yep.

如果我在说话，我们就切到我的近景。

If I'm speaking let's have the tight shot on me.

是的。

Yep.

还有视频中来回切换的部分。

And parts of the video where there's back and forth.

完全正确。

Totally.

我们用广角镜头吧。

Let's go with the wide shot.

是的。

Yep.

而且现在实际上就能做到这一点，这并不需要AI。

And it actually can do that today without without that's not AI.

所以接下来会发生的是，我用了，你知道的，可能是一种轻量级的AI视频编辑工具。

So and then But here's happen here's gonna happen and I used you know I used maybe like a lightweight AI video editing.

我不确定里面用了多少AI技术。

I don't know how much AI is in there.

但总有一个时刻，你会想，实际上，这时候应该切换到另一个人的反应镜头。

But there's always this part where you're like, actually, that's the moment you want to go and look at the reaction of the other person.

即使另一个人在说话，我们也应该确保切到另一位参与者。

Even though somebody else is talking, we should kind of make sure we cut cut to the other participant.

而且你更接近

And you're closer to

这项技术，所以我很好奇，你认为这会是它的发展方向吗？即你构建两个或三个味觉代理，让它们观看视频，然后投票选出更好的版本，如果出现全票通过或二比一的结果，就作为最终输出。

the technology than I am so I'm curious if you think this is the way it develops where you then build like two taste agents or three taste agents and then they watch the video and then they vote on what's better and if you get unanimous or two versus one that's the output.

是的，然后我认为，如果你看一下好莱坞的复杂制作，会发现有层层叠叠的编辑和制片人，甚至我都不清楚所有这些职位的名称，但总有人负责监督编辑，他们审阅最终的剪辑版本，然后还有最终制片人、导演等等。

Yes and then I think what will happen then is if you look at a sophisticated production in Hollywood they have layers and layers of editors and then producers and there's like I don't even know all the names, but like there's somebody who oversees the editors and they look at the final set of edits and then there's the ultimate producer and the director and so on.

我认为，未来视频编辑的角色将会把这些所有职位压缩在一起，由代理以自动化的方式完成剪辑工作。

I think that what will happen is the video editor of the future just compresses all of those roles and the agent is doing the just that that sort of, you know, cutting part in an automated fashion.

但我其实认为，最终还是会有人来把关。

But I actually think that you'll still have that ultimate person.

也许他们要审阅的是五个不同的剪辑版本作为选项，此时他们扮演的角色，就像过去在电视节目中那位最资深的编辑，但现在，这种能力可以普及到每一位播客创作者。

Maybe what they'll review is five different cuts as options and they are now playing the role of the, you know, the the most senior editor in a in a, you know, TV show that that in that that would have happened in the past, but now you bring that same capability to every podcaster.

这在过去是根本不可能实现的。

Like, that was never possible before.

但你确实可以，是的。

But you could yeah.

不。

No.

抱歉。

Sorry.

请说。

Go ahead.

不。

No.

但所以，这就像是编辑并没有真正消失。

But so so then so it's like it's like the editor didn't really go away.

他们只是在从事与以前完全不同的工作。

The what they are just doing is a completely different activity than what they did before.

他们让五个代理生成一堆示例，然后对这些工作进行某种最终的整合，形成最终输出。

They have five agents producing a bunch of examples, and then they are doing some kind of final kind of synthesis of that work into some final output.

好的。

Okay.

因为你会直接感受到。

And Because you'll just feel it.

比如你听一个播客时会想，哦，他们剪辑这个部分真够粗糙的，他们大概只用了AI。

Like you'll watch a podcast and you'll be like, oh, that was like really janky how they cut that thing and then they'll be like, they probably just used AI only.

好的。

Okay.

但在这里，好吧。

But here alright.

所以我想反驳一下，因为我确实认为事情还能走得更远。

So I want to dispute this because I do think that that things can go even further.

对。

Yeah.

而这意味着，如今我们的互联网和世界都是为人类生产的内容而构建的。

Right and what that means is right now we have an internet and a world set up for human produced output.

是的。

Yes.

在知识型工作领域，对吧。

In knowledge work right.

当内容由智能代理生成时会发生什么？我们不妨假设这个设想是可行的。

What happens when it's agent produced output just assuming going with the thought experiment that this could work.

你最终可能会发现，我们就以视频为例吧，我觉得天啊，我们会不断填充

What you might end up having is you know you got you have let's just go with the video and I think God help me we're gonna keep filling

的

the

优化目录里全是这些东西，但好吧，你制作了视频，然后由AI编辑器剪出了很多不同版本。

optimization catalogs with this stuff but okay you put the video so you have this editor, the AI editor cut a bunch of different videos.

你的品味代理会投票选出最好的五个。

You have your taste agents vote on what the five best are.

然后你可能会看到像YouTube这样的平台——我们已经可以看到，你可以测试很多不同的缩略图、不同的标题，还可以运行大量不同视频，然后只展示给你的前一百或一千名观众。

Then what you might end up seeing is a platform like YouTube we already can see you can test a bunch of different thumbnails, a bunch of different Run total and different you can run a bunch of different videos and then it will show it to your like first 100 or thousand viewers

是的。

Yeah.

然后系统会进行优化，这就是YouTube想要的。

And then it will optimize so you'll end up it'll and that's what YouTube wants.

它最终会把最好的视频推送给观众。

It'll end up getting the best video to the audience.

我用这个作为例子，但你可以想象，这种模式会扩展到整个知识工作或大部分知识工作领域。

And I'm using this as an example but you can kind of think it fanning out across all of knowledge work or much of knowledge work.

是的。

Yes.

这引出了一个问题：我们是否希望生活在一个如此系统化、算法驱动、代理驱动的世界里？

And that sort of gets to like the question of do we want to be in such a systematized algorithm driven agent driven world?

嗯，我只是不同意这一切都会发生，所以我无法辩护我们是否应该生活在那样的世界里，因为我根本不认为这种情况会成真。

Well well I just don't agree that all happened so I can't defend do we want to be in that world because I actually don't think that plays out.

你真的不这么认为吗？

You don't think so though?

不。

No.

因为这确实看起来像是我们已经看到，算法已经在为我们做出大量决策了。

Because it does it does seem like we've already seen that let's say algorithms are already making a lot of decisions for us.

百分之百。

100%.

在你意识到之前，代理就已经被释放到工作中了。

Before you know we've even set agents loose on work.

是的。

Yeah.

所以你不认为这会增加吗？

So you don't think that will increase?

我认为会，但我认为它更多会体现在经济上更容易验证的结果上。

I think it will but I think it's gonna be more for probably economically much more sort of testable outcomes.

比如，我不认为世界上所有的计算资源都会被用来以十种不同方式编辑播客并运行它们。

Like, just don't think that that of all the compute supply in the world that what we're gonna do is spend our compute on editing podcasts 10 different ways and running those.

我的意思是，我只是用它作为一个例子。

I mean, I'm just using it as an example.

它可能会变成这样，比如说市场营销。

It could end up being like, let's say it's marketing.

你提到了市场营销。

You brought up marketing.

市场营销是个很好的例子。

Marketing is a great example.

这说得通。

That makes sense.

它已经变得越来越数学化了。

It's already becoming mathematical.

我只是有点想说

I was sort of just

我正是在反思你举的这个例子。

specifically reflecting on your one example.

我觉得这在其他很多领域也都会确切地发生。

I think this will exactly happen in a bunch of other areas.

这将在金融领域发生，会在营销领域发生，会在医疗健康领域发生，也会在生命科学领域发生。

It's going happen in finance, it's going happen in marketing, it's going happen in healthcare, it's going happen in life sciences.

我们将用它来进行药物发现。

We're going to use it for drug discovery.

我之前和一位生命科学领域的首席执行官交谈过，我们现在能够做到的是，能够在我们想要检测的所有领域中，进行大约10到100倍更多的实验。

I was talking to a life sciences CEO and what we're going to now be able to do is we will be able to run on the order of 10 to 100 times more experiments across everything that we want to go detect.

然后你会逐步缩小范围，只挑选出那些真正值得进行完整临床试验和全面实验的项目。

And then you'll sort of narrow those experiments down to the ones that you actually want to do the full clinical trial process on and the full level of experimentation on.

但我们能够并行开展实验，让智能体在所有具有经济价值的工作领域中运行，这只会对社会带来巨大益处。

But our ability to experiment and have agents run-in parallel across all areas of economically valuable work is only going to be a boon to society.

我们将发现以前无法发现的药物。

We will discover drugs that we wouldn't have discovered before.

你肯定会看到更多前所未有的创新。

You'll certainly get much more novel.

你或许会争论这到底是好是坏，但你将看到金融服务出现更多新颖的方式，因为你能够更加精准地捕捉市场趋势和市场动态。

Maybe you could debate if this is good or bad, you'll get more novel ways of doing financial services because you'll be able to be even more kind of hyper tuned to market trends and what's happening in the market.

当然，营销领域也是如此，如果营销人员能更好地找到他们的客户，这绝对是好事。

Certainly marketing, just think it's only a good thing if marketers can find their customers better.

因此，在我看来，算法驱动的广告是更精准地找到需要你服务的客户的必然结果，这纯粹是好事。

And so to me, algorithmically driven advertising a is corollary to being able to better find customers that want your services and that is just only a good thing.

如果你是个小企业主，我能精准定位到我咖啡店附近那些喝咖啡的人，有针对性地投放广告，而不是盲目撒钱却毫无效果，这绝对是好事。

If you're a small business and I can only find the the people from my coffee shop that drink coffee in this neighborhood and I can target them and I can now spend money to get those customers instead of just blasting dollars and then not getting any efficacy, that's only a good thing.

对吧？

Right?

所以我认为，代理能够承担更多这类工作，对社会来说完全是净收益。

So I think that the idea of agents being able to do so much more of this is a completely net positive for society.

我认为在其他一些领域，算法可能会带来一些问题，但我并不担心那些代理在后台并行为我们工作的场景。

And I think there's other areas where algorithms can kind of be tricky, but I'm not worried about the ones where it's sort of like agents running in parallel doing work for us in the background.

我认为资金通常会流向那些对社会真正有用的地方。

I think the dollars will generally flow to the areas where that ends up being useful for society.

而且，许多这些代理或聊天机器人都是基于相同的上下文运行的。

And a lot of these agents or even chatbots are working off the same context.

有些人使用ChatGPT时开始变得想法一致，因为AI都是从相同的上下文中提取信息，给出的都是平均化的答案和观点。

There's been some stories about how people using ChatGPT are all starting to think the same because it's sort of pulling from the same context and giving them answers and perspective from the same average of averages.

所以这可能是另一个问题。

So that could be another issue.

我认为，关于我们要把多少生活交给这些系统，这里面有很多问题。

I think there's lots of there's plenty of issues with the idea of, you know, how much of our life do we put into these systems?

我们对它们的依赖程度到底有多深？

How much do we rely on them for every little thing?

安德烈·卡帕西有过一条有趣的推文，他说他让AI帮他审阅一些内容，要求它批评自己，但后来他又让AI做完全相反的事，结果AI居然也能为相反的观点提供同样有力的论证。

Andre Karpathy had this, you know, funny tweet where he sort of said, you know, I I I had AI go and review something and I asked for for, you know, it to critique me, but then I had it do the exactly the opposite and it and it and it sort of found it it it created just as good of a justification on the exact opposite of what it had said, you know, on on the other side.

我们经常看到这种情况，也就是说，我通常会代表自己说话。

We see this a lot, which is, you know, I'll mostly represent myself.

我不知道我妻子是否愿意被牵扯进来，但我和我妻子经常用ChatGPT来帮忙育儿。

I don't know if my wife wants to be pulled into this, but, you know, I slash we use like Chatuche BT for parenting a lot.

有趣的是，你只要换一种提示方式，就能得到完全相反、截然不同的答案。

And it's funny because like you just know how you could prompt it and get a completely one eighty different answer on on the facts of the situation.

所以你真的需要理解这些系统是如何运作的，才能确保自己不会只是得到基于你提问的、那种平均化的标准回答。

And so you actually have to like you you really have to understand how these systems work so you can ensure you're not just getting, again, what what is the what is the sort of, you know, mean response based on your prompt?

你真的需要从中抽离出来。

You really need to pull out of it.

到底在这种情况中，你真正该做什么呢？

What is it what really, you know, should you do in this particular situation?

所以你得学会，有时候要以负面的方式表达，而不是正面的方式。

So you have to, like, do, like you have to, you know, you know, sometimes word things in, like, in a negative fashion versus a positive fashion.

你在提问时，不希望误导这个智能体。

You don't you don't want to like bias the agent as you're writing the question.

你得做很多这类事情，我认为这会是我们社会逐渐学会的东西，就像我们最终学会了如何使用搜索引擎和其他工具一样。

You have to do a bunch of this kind of stuff and and that that'll be I just think that'll be like a a thing we generally learn over time in society just as we eventually learn how to use search engines and and other tools.

是的。

Right.

而且我觉得，当你试图从这些系统中获取关于重大人生问题的答案时

And I think when you try to get a response on a big life question from these things

是的。

Yes.

有一点很重要要注意：它的目标是让你写出另一个提示。

Something that's important to keep in mind is its goal is to get you to write another prompt.

是的。

Yes.

这个奖励机制确实很棘手。

That reward function is definitely tricky.

一般来说，你真正希望的是，尽可能让代理执行类似这样的任务：给我生成一个关于这件事优缺点的表格。

In general, what you really want is the as much as possible, you want the agents to do things like generate me a table of the pros and cons of this thing.

对。

Right.

并且确保你为双方都提出论据。

And make sure that you make arguments for both sides.

然后你需要站在一个位置上，对这些内容进行解读，并根据你认为在你当前情境中相关的信息做出决策。

And then you wanna be really in the position of interpreting that and making a decision based on what you think is relevant in your situation.

我有时候会做一些我必须做的事，比如在处理医疗问题时，我知道在我的提示中，我已经在某种程度上偏向了我预判的代理或对话会走向的方向。

I do things I have to do these things sometimes, like even for like medical questions, where I know that I've in my prompt, I've sort of I've over biased the direction that I know the agent is going to go in or the chat will go in.

所以我会换一个不同的提示，比如：在什么情况下，你会想象这种类型的医疗问题会出现？

So then I do a different prompt which is just like, under what circumstance would you imagine this type of, you know, kind of medical issue would show up?

然后我就看看，嗯，这些情况在这里有没有出现？

And then I and then I kind of just see, okay, are those things showing up here?

首先，如果你只是直接告诉它你的症状，然后问：你觉得这是不是这个病？

First is if you just give it your symptoms and then you'd like, and do you think it's this?

它会回答：是的，这肯定是这个病。

And it'd be like, yes, it's definitely that.

比如你是不是得了埃博拉病毒？

Like do have Ebola.

是的，没错。

Yeah exactly.

不过，要让这些东西真正发挥作用，关键问题是——我认为你之前也提到过，你必须信任它在你生活中的有用性。

The big question though for this stuff to work is and I think you talked a little bit about how useful you want it to be in your life you have to trust it.

是的。

Yes.

而且，要让这些代理真正高效地工作，你还必须放弃很多控制权。

And you also have to give up a lot of control like to make these agents work really well.

对。

Yeah.

想想我们刚才讨论过的例子。

Like think about any example we just we just went through.

你得说：这是我的电脑，这是我的文件，去执行操作吧。

You have to be like here's my computer have my files take actions.

对。

Yeah.

它们在你卸下限制、信任它们替你做事时，表现得更好。

On my behalf and honestly they work better when you take the guardrails off and trust them to do things for you.

你认为，要实现这个产品愿景，这种情况必须发生吗？

Do you think we're like again for this product vision to work that has to happen.

你觉得现在人们是否已经准备好将这种控制权交给这些机器人了？

Do you think we're in a place where it's feasible for people to give up that type of control to these bots?

嗯，这正是扩散过程会比硅谷人士所想的更漫长的地方。

Well, so this is is where the diffusion this general category is where the diffusion will be longer than than where people in Silicon Valley think.

如果你身处硅谷，你我每天看到的那些热门推文，往往都来自一些只有十个人的初创公司。

So if you're in Silicon Valley and you know every tweet that you and I read you know, that goes viral in in the Valley is is often it's coming from like a 10 person start up.

它们基本上是从零开始，完全重新设计了工作方式、工作环境、使用的工具和拥有的数据，能够直接围绕让代理输出结果来构建整个组织。

They have they have basically like they they started from a completely clean slate of of the way that they work, that their environment, the tools they use, the data that they have, and they can just they can build their organization around around getting output from agents.

但你去看看世界其他地方。

And you go to the rest of the world.

想象一家拥有上万名员工、已经存在了几十年的公司。

Take a company that has 10,000 employees, been around for decades.

它们的数据又分散在二十个、三十个、五十个，甚至一百个不同的系统中。

Their data is in, again, twenty, thirty, 50, a 100 different systems.

如果你去问这家公司：这个客户的最新合同在哪里？

If you go and ask that company, where are your latest contracts for this client?

它可能分布在五个不同的地方。

It could be in five different places.

如果你问，最新的营销活动素材在哪里？

If you go and say, where's the latest marketing campaign assets?

它可能存在于十个不同的地方。

It could be in 10 different places.

如果你问，你们正在研发的这项突破性成果的研究资料在哪里？

If you say, where's the research for the new for that new breakthrough that you're working on?

它可能分散在五个不同的存储库中。

It could be in, you know, five different repositories.

所以，如果你现在想在这样的环境中部署一个AI代理，你可以把它想象成一个新员工加入公司——这个新员工聪明得惊人，拥有博士学位，但才刚加入你公司一分钟。

So the challenge is if you're in if you now wanna go deploy an AI agent in that environment, you can almost think about it like like a new employee joining that company and that new employee is like insanely smart, like they have a PhD, but they just joined your company one minute ago.

你给了他访问你们所有工具的权限，并说：三十秒后，我需要你去找到我们正在开发的新产品的相关研究资料。

You've given them access to your tools, and you say in thirty seconds from now, I need you to go and find me the research for this new product we're building.

问题是，这个人会去翻阅你们所有的系统，但他根本不知道，究竟哪一个才是那份研究计划、营销素材或合同的权威版本。

The problem is that person is gonna go and they're gonna go look through all your all your systems, but they're not gonna know, like, well, which is the one that that that really is the authoritative copy of that research plan or that marketing asset or that contract.

他们不知道这些信息在哪里，因为这些信息是通过非正式的内部经验流传下来的。

They're not going to know where that is because that came through tribal knowledge.

这些信息来自你参加过十多次会议后，才发现自己拿错了东西，或者不得不问同事某个信息的权威来源在哪里。

It came through you knowing over 10 different meetings that you pulled the wrong thing or you had to ask your colleague where is that right source of truth for something.

所以这个新员工没有任何这样的背景知识。

So that new employee has doesn't have any of that context.

它不了解任何那些非正式的内部经验，也不了解公司里已有的工作模式。

It doesn't know any of the any of that tribal knowledge or the work patterns that that have exist have existed at the company.

这个智能代理正处在同样的境地，但情况更糟，因为当它不知道某件事时，它根本意识不到自己不知道。

The agent is in that exact same situation, but they're even worse off because they they are basically they they are they are they really don't know when they don't know something.

于是，当智能代理获得对这十个系统的访问权限后，它会说：‘嘿。’

And so what happens is the agent gets access to those 10 systems and it it says, hey.

你说：‘嘿。’

You you say, hey.

那个新产品的发布时间是什么时候？

When's the, you know, when's the launch of that new product?

它找到的第一个文档或一组似乎涉及该主题的文档，就会直接从中提取信息。

The first document or set of documents it finds that that seemingly talk about that thing, it's just gonna pull from those.

它不会知道，实际上可能还有另外两个系统需要去检查，并将答案与最初找到的进行对比。

It's not gonna know that actually maybe there's two other systems I should go and check and then compare the answers to the first ones that I found.

它只会直接把那个答案提供给你。

It's just gonna go and deliver that answer to you.

因此，作为企业，你面临的挑战在于：你的信息组织得有多好？你对底层流程的文档记录得有多完善？员工或代理能否轻松获取到企业中任何项目或事务的真实信息来源？

And so the challenge though then is that you're at the mercy as an enterprise, you're at the mercy of how well is your information organized, how well did you document your underlying processes, How easy is it for an employee or an agent to get access to the true source of truth to any project or thing going on in your business?

一个人要找到正确信息越困难，代理要找到就越难十倍。

The harder it is for a person to be able to go in and find the right thing, it's going be 10 times harder for the agent.

在现实世界中，不是那些没有任何历史积淀、只有十来个人的初创公司能轻松做到这一点。

And so the real world, not the 10 person startups that that that get this you know, get started without any of that that history.

在现实世界中，大多数企业都面临着这些挑战。

In the real world, most enterprises are dealing with all of those challenges.

因此，当他们尝试部署代理时，代理首先必须连接到所有这些系统。

And so they they go in and they try and deploy an agent and the agent has to, first of all, connect to all of those systems.

然后它还得尝试找出，再次强调，哪里有能给出正确答案的正确信息。

Then it has to try and figure out, again, where is the right information that needs the right answer.

于是你就得依赖于该系统是否已更新为完全准确的信息、正确的数据，以及那份正确的文档副本。

Then you're reliant on that system having been kept up to date with exactly the right information, the right data, that right, you know, the right copy of the the the document.

这才是真正的挑战。

And that's the big challenge.

因此，未来多年里，企业将逐渐意识到，AI的问题本质上是数据问题。

And so we are gonna be in for, again, years and years of enterprises realizing that an AI problem is really a data problem.

为了给AI提供正确的数据，它们必须确保拥有基础设施、软件、工具和系统，所有这些都服务于为代理提供上下文。

And to get the AI the right data, they need to make sure they have infrastructure, software, tools, systems that all are in service of giving the agent context.

一些公司在这方面走在了前列，但许多公司仍在面对这样一个现实：我拥有大量陈旧的基础设施。

And some companies are ahead of the curve on that, but a lot of companies are still kind of reckoning with, I have a lot of infrastructure that's legacy.

代理在这些陈旧工具上表现不佳，因此我无法轻易让代理访问这些数据。

Agents don't work well with that set of legacy tools, and so I can't, you know, easily get agents to access that data.

在我们的业务中，我们每天都能看到这种情况，因为我们正在帮助客户转向现代化的信息管理方式。

We see this every, you know, every day in our business because we're helping customers sort of move to a modern way of managing their information.

但在我们这个行业，企业管理和企业内容方面，公司通常有二三十个不同的系统来存放他们的企业文档。

But where we come from in our industry of, you know, with enterprises managing enterprise content, companies have 20 or 30 different systems where their enterprise documents are.

而这些系统根本无法与智能代理协同工作。

And that just simply won't work with agents.

因此，你最大的挑战可能是：智能代理需要上下文，而这些上下文无处不在。

So that's probably your biggest challenge is the agents need context, that context is everywhere.

你如何确保智能代理获得完成工作所需的准确上下文？

How do you ensure that the agents have exactly the right context they need to do their work?

这将是知识工作自动化面临的主要挑战。

That will be the big challenge for knowledge work automation.

但除了让它们访问这些上下文之外，你还信任它们拥有这些上下文吗？我真的很需要一个智能代理。

And but beyond getting them access to that context it's do you trust them with that context like I need an agent in the worst way.

我的意思是，如果Open Cloud能帮我处理收件箱，读取我所有的邮件，帮我起草当天还没来得及回复的邮件，甚至查看短信，或者从我的播客广告系统中提取信息，提醒我说：‘你还有这些主持人广告没处理。’

I mean I think Open Cloud would be great for me if you go through my inbox, if it could read all my emails draft the responses it thinks that I need to send that I haven't gotten to that day maybe take a look at text messages maybe can pull from my podcast ad system and be like oh you have these host red ads you need to do.

把文本内容输入到聊天机器人中。

Feed the text into a chatbot.