Aaron Levie 谈为何 Open AI 胜出

开放权重AI常常被说成是对前沿实验室的威胁。

Open weight AI is often framed as a threat to frontier labs.

莱维认为这种说法把经济逻辑搞反了。

Levie thinks that gets the economics backwards.

Box的联合创始人兼CEO与西奥·贾菲和索菲亚·普奇尼一起在MTS节目中，讨论为什么开放模型可能让AI生态系统更具竞争力、关于中国蒸馏模型的争论，以及为什么美国需要更多开放权重AI。

The Box co founder and CEO joins Theo Jaffee and Sofia Puccini on MTS to discuss why open models could make the AI ecosystem more competitive, the debate around distillation in China, and why America needs more open weight AI.

他们还聊到了最新的前沿模型、AI如何扩大而非缩小Box的工程路线图，以及为什么模型路由可能成为企业AI的默认选择。

They also get into the latest frontier models, how AI is expanding than shrinking Box's engineering roadmap, and why model routing could become the default for enterprise AI.

我们现在请到了Box的联合创始人兼CEO亚伦·莱维，Box做很多事情，云内容管理、企业文档、权限、协作，还有很多不同的AI功能。

We are live with Aaron Levie, the cofounder and CEO of Box, which does all kinds of things, cloud content management, enterprise documents, permissions, collaboration, a lot of different AI functions.

他的推特账号很厉害，叫levie。

He has incredible Twitter account, levie.

他在网上活跃很久了，经常写关于AI和其他很多话题的内容。

He's been around on the website forever, and he writes about AI and many other topics.

今天我们有两个大新闻。

We have two huge stories today.

我想知道，我们应该先聊哪个？

I wonder which which should we start with?

占星术。

Astrology.

占星术。

Astrology.

当然是占星术。

Of course, astrology.

对。

Yeah.

占星图像生成领域的一次重大收购。

Huge acquisition in the astrology image gen space.

明白。

Mhmm.

我们会持续关注这件事。

Something we will be monitoring.

这是开源的吗？还是还没有？

Is it an open source play or not yet?

我觉得还没有。

I think not yet.

好的。

Okay.

我觉得占星术可能暂时还是闭源的。

Think astrology might remain closed source for the time being.

哦，这样啊。

Oh, okay.

那太可惜了。

That's too bad.

我们得想办法应对。

We gotta fight that.

是的。

Yeah.

我们确实得跟这个斗争。

We really gotta fight that.

不过咱们还是回到正题吧。

But going to the real story.

今天，黄仁勋写了一封关于开放权重的公开信，Box也签了名。

So today, there was an open weights letter that Jensen Huang wrote and Box signed.

你能跟我们说说你对这封信的理解吗？

So can you tell us sort of, like, your interpretation of this letter?

它到底针对的是什么？

What exactly is it aimed at?

你们想推动什么？

And, like, what are you guys trying to shape?

你觉得黄仁勋想推动什么？

What do you think Jensen was trying to shape?

对。

Yeah.

这封信是不是有点像罗夏测试，就是看你在里面看到了什么？

Is the letter maybe like a it's like a Rorschach test of, like, what do you see in this letter?

嗯，我们读的时候看到的是，希望这是签署人的默认立场，但基本上主要有两点。

Well, what we saw when we read it was hopefully, it's, the default stance of the folks that signed it, but, basically, almost twofold sort of main points.

第一，开放权重AI之所以超级重要，是因为它确实推动了AI的进步。

One, the reason why open weights AI is super important is because it actually drives AI progress.

我们获得了更多创新。

We get more innovation.

我们有了更多选择。

We get more options.

你可以在这些模型的基础上进行构建。

You can build on top of these models.

你可以针对自己的用例来训练它们。

You can train them for your own use cases.

总的来说，我认为开放权重实际上是AI生态系统中非常重要的一部分。

Open weights in general, I would argue, is actually a very important part of the AI ecosystem.

所以我觉得这其实被错误地框定为与封闭权重之间的零和博弈。

So much so that I think it's actually kind of misframed as zero sum with closed weights.

它实际上只是增加了人们使用AI的用例数量。

It actually just adds to the number of use cases that people then do with AI.

所以这大概是第一部分。

So that's kind of probably part one.

然后第二部分，我觉得信里还隐含了一点对美国行动的呼吁。

And then part two, I think embedded in there was a little bit of a call to arms on The U.

美国。

S.

实际上需要在开放权重模型上投入更多，而且我们可能希望更多公司带着这种创新参与进来。

Actually needs to be probably even more invested in open weights models, and we probably want even more companies kind of showing up to the table with this innovation.

所以这封信似乎并不是直接说我们必须支持中国发生的一切，而是关于美国。

And so it didn't seem like it was directly about we must support all of the things that are happening in China as much as U.

美国。

S.

需要继续支持开放权重，我们可能还需要更多的创新。

Needs to continue to support open weights, we probably need even more innovation here.

顺便说一句，我知道蒸馏其实并不总是坏事，它有很多应用场景。

I know, by the way, distillation is actually not always bad, there's lots of use cases around it.

所以咱们对这方面的恐慌情绪可以稍微冷静一下。

So let's maybe calm down on some of the FUD on that.

我们支持所有这些立场。

We support every one of those stances.

你知道，开放模型对AI生态系统非常重要，而且我认为它甚至能推动封闭模型提供商更快创新，为市场提供更多价值。

You know, open models are, I think, very important to the AI ecosystem, and I actually think it it pushes even the closed model providers to innovate faster and be able to deliver more for the market.

所以我觉得开放权重最终让每个人都受益。

So I think everybody is sort of a winner as a result of open weights.

公平地说，从另一方的角度来看，先不考虑一些封闭玩家的商业和经济利益，我认为确实有一个类别，在未来的能力水平上，关于开放权重的安全性问题有真实的争论。

To be fair to the other side of the argument, I think there's a category forgetting kind of commercial and economic interest for a second of some of the closed players, I think there's a category that obviously has real arguments around the safety elements of open weights at some point in the future of capability.

我不同意那个观点，但我认为这是一个健康的辩论，值得讨论。

I don't agree with that point of view, but I think it's a healthy debate to have and to have conversations around it.

但我几乎毫无保留地认为，更开放的创新和开放权重创新对整个AI领域以及AI的普及都是好事。

But I almost unequivocally take the stance that more open innovation and open weights innovation is good for AI broadly and the diffusion of AI.

那说到蒸馏这个话题，你怎么区分那种只是正常经济活动的蒸馏，和那种越界到某种民事或刑事违法的蒸馏呢？

So diving into the distillation point, how would you separate distillation that is, like, just normal, basically normal economic activity from distillation that crosses a line into, like, some sort of civil or criminal violation?

嗯，我不确定我会不会这么区分。

Well, I don't know if I would make that argument.

我不知道这条线会画在哪里。

I don't know where that line would exist.

我几乎完全站在这一边——我觉得很难论证说，AI模型可以广泛地基于公共互联网训练，但另一个AI模型却不能基于AI模型的输出进行训练。

I'm almost deeply on the side of I think it's very hard to make the argument that AI models should be trained on broadly the public Internet, but another AI model can't be trained on the outputs of an AI model.

我觉得这个论点站不住脚。

I think it's a very tenuous argument to make.

我看不出怎么能说得通。

I don't see how you can kind of pull it off.

但与此同时，我也理解，如果我是实验室负责人，我也会想尽一切办法把这些东西锁得死死的。

At the exact same time, I understand, and if I were running a lab, I would want to, like, lock these things down as much as humanly possible.

所以我会想尽一切防御机制来防止我的模型被蒸馏，但我不认为你能有说服力地论证说这跨越了某种道德界限，因为那样的话，你基本上就是在反对这些模型的基础训练过程。

So I'd be trying to find every defensive mechanism to prevent the distillation of my models, but I don't know that you'd be able to credibly argue that there's some ethical line that is crossed simply because then you would be basically arguing against the underlying training runs of these models.

大多数人都没有选择权，他们的数据是否被用于训练。

Most people didn't get to opt in or out of their data being trained on.

也许在底层服务商的某个服务条款的细则里有所提及，但这件事基本上就是我们在假设世界的通用知识会被训练进这些模型里，而这些通用知识，不管来自Reddit还是来自Anthropic，我看不出它们之间有什么实质性的区别。

Maybe it's in some kind of terms of service from the underlying provider somewhere in the fine print, but, like, this is a thing that is sort of like we are assuming that the general knowledge of the world is going to be trained into these models, And that general knowledge, whether it comes from Reddit or it comes from Anthropic, is like I don't see the distinction between those in any meaningful way.

对。

Yeah.

我的意思是，我们昨天听到了一个观点，基本上就是这个意思。

I mean, we heard a take yesterday that was kind of exactly that.

就是说我们在责怪蒸馏时，关注错了链条上的环节。

It was just like we're focusing on the wrong part of the chain when we blame distillation.

真正应该关注的是，Anthropic如何考虑限制他们的API使用，或者如何非常有效地防止那种——因为我们之前也在讨论，蒸馏的话，他们还是得付费，对吧。

It should really be like, how is Anthropic thinking about how to restrict their API use or how to very effectively prevent the sort of because we were also talking about how, like, distillation, they still have to pay for Yeah.

为API使用付费。

For the API usage.

Anthropic从蒸馏中赚到的钱，比大多数人从原始训练中赚到的还要多。

Anthropic is getting paid more for distillation than most people have ever gotten paid for the original training run.

所以，如果你做个思想实验，想想看，我通过一万个账户直接使用API进行蒸馏，和我们都用Fable或GPT-5.6生成代码，然后这些代码都进了公开仓库，这两种情况对模型训练来说，到底有什么本质区别？

So and if you kind of, like you almost kinda do a thought experiment of, well, what's the difference between distillation in a very direct, like, I'm using the API across 10,000 accounts versus just quite literally if we all generate code from Fable or GPT-five point six and all of that code ends up in public repos, what's the kind of compelling difference between those two scenarios that a model would get trained against?

我至少不太明白，为什么在这两种情况之间会存在某种道德界限的争论。

I'm not sure I at least understand why there would be arguments, why there would be some ethical line between those two.

然后再加上另一个因素，很多人认为中国在这种情况下，总体上正在做突破性的创新，而不是依赖蒸馏，这推动了模型的进步。

Then you layer in another element, which is that there are a lot of people that argue that China in this case are doing breakthrough innovation in general independent of distillation that's causing models to progress.

但我也确信，还会有其他关于正在发生什么的争议和阴谋论，我了解得不够多，无法可靠地论证其他事情是否在发生，以及为什么这些模型在进步。

But, also, I'm sure that there's gonna be other kind of controversies and conspiracies about what's happening, and I don't know enough to be able to kind of credibly argue that other things are or are not happening and why these models are progressing.

显然，我会让实验室去提出这些论点，让政府去按他们认为合适的方式去处理。

I'll let, obviously, the labs to make those cases and the government to kind of pursue that as they see fit.

但对我来说，所有这些都与我们是否想要开放权重模型，以及我们是否想要生态系统中的这种创新无关。

But all of that to me is independent of do we want open weights models, and do we want this innovation in general from the ecosystem?

我个人认为，最好的情况是，有几个人处于人生的某个阶段，他们可以直接说，是的。

I think the the best case scenario personally would be you get a few people that are at a stage in life where they can just be like, yep.

我打算砸一百亿美元在美国建一个开放实验室，我们有自己的登月计划。

I'm gonna drop $10,000,000,000 on building an open lab in America, and we have our own version of, you know, moonshot.

我觉得这对美国和对整个创新领域都是巨大的贡献。

I think that would be a great service to America and to innovation in general.

所以我一直在等有人这么干的那一天，但到现在，还在等。

So I've been kinda waiting for the date that, you know, somebody does that, but, you know, for now, still waiting.

嗯，其实还有一个基金会。

Well, there's another foundation.

就在这个帖子下面。

Under this post.

他们可能会这么做。

They might do it.

他们可能会这么做。

They might do it.

这里还有另一个角度，就是大多数开放权重模型都来自中国。

There's another angle here, which is, you know, most open weights models are coming from China.

而且有些人担心，如果我们变得太依赖，如果美国创业生态变得太依赖中国的开源权重模型，然后他们突然把开源权重给关了。

And some people worry that if we become too dependent, if the American startup ecosystem becomes too dependent on Chinese open weight models and then they, like, shut off the open weights.

他们停止发布新的开源权重模型，那对美国科技行业来说可能会非常糟糕。

They stop releasing new open weights models, that this could be very bad for the American tech industry.

或者因为这些都是中国的开源权重模型，它们压低了美国公司的利润空间。

Or that because these are Chinese open weight models, they undercut the margins of American companies.

还有些人说，这涉及到国家安全问题。

And it's a national security concern, say some people.

我觉得这些论点都很有道理，但问题是，那怎么办呢？

And I think those are great arguments, but it's like, okay.

那你能做什么呢？

So what do you do about that?

对吧？

You know?

有人说，iPhone应该卖5000美元而不是1000美元，因为不应该依赖中国现有的制造技术。

Some people say that the iPhone should cost $5,000 instead of a thousand dollars because we shouldn't, you know, rely on manufacturing advancements that exist in China.

我也不太清楚。

Like, I don't know.

如果你把另一种情况推演一下，那你基本上就是在说，美国应该用很贵的人工智能，而世界其他地方应该用很便宜的人工智能。

If you kind of play out the alternative, then all you would basically be arguing is America should have really expensive AI, and the rest of the world should have very cheap AI.

然后我们看看接下来五到十年，这种局面在竞争上会怎么发展。

And then we'll see how kind of, you know, that works competitively over the next kind of five or ten years.

对。

Yeah.

所以这可能就是那种事情，你知道，木已成舟了。

So it just may be one of those things, which is, you know, the the cat's out of the bag on that.

我们没法扭转这个局面。

We can't unwind that dynamic.

开源权重模型已经存在了。

Like, open weights models exist.

中国知道怎么造它们。

China knows how to produce them.

我比较倾向于杰森那边的观点。

I lean more on the Jensen side of the argument.

你知道的，每个人都看过那个杰森·多拉凯什的播客，那就像是一个最初的罗夏墨迹测验。有些人觉得多拉凯什的观点很明显是对的，就是应该采取那种立场，而且你应该去质疑杰森。

You know, every everybody had this moment where they watched the Jensen Dorakesh podcast, and that was, like, the original kind of Rorschach test of, like you know, some people parse Dorakesh as, like, obviously, like, that's the exact right position to be taking, and, you know, you should be interrogating Jensen.

而另一些人看完后觉得，为什么多拉凯什没听懂杰森的意思，还让他继续展开讲呢？

And then other people watched it and was like, why is Dorakesh not understanding Jensen's point and kind of letting him kind of, you know, sort of expand on that?

我当时看着就想，对，确实是这样。

I was, like, watching it and just was like, yeah.

黄仁勋显然说对了。

Jensen's obviously right.

你可以不同意他的观点，但他显然说得对，接下来会发生的事情是……

You can disagree with it, but he's obviously right that the following will happen.

如果你把中国封锁在外，中国肯定不会放弃人工智能。

If you block off China, China is, like, not gonna give up on AI.

他们不会就这样决定这个技术领域对他们来说不重要而退出。

They're not going to just decide that this is not that important of technology category for them to play in.

所以，如果假设中国认为AI具有极其重要的战略意义，而且，顺便说一句，你知道，那些搞笑的梗图推文，比如“我们在美国的中国AI研究员比中国实验室还多”之类的。

So if just assume that China considers AI to be very strategically important, and oh, by the way, you know, you know, all these funny meme, you know, tweets of, like, we have more Chinese AI researchers than the Chinese labs do in The U.

美国。

S.

显然，他们能培养出非常优秀的人才来研究AI。

Like, obviously, they can produce incredible talent working on AI.

所以这并不是我们独有的、无法被超越的人才优势。

So this is not something that we own, some impenetrable talent base to be able to compete against.

所以这对中国来说极其重要，具有战略意义。

So it's incredibly important strategically for China.

他们实际上拥有很多核心原材料，包括人才和工业实力，可以逐步建造芯片和晶圆厂等等。

They actually have a lot of the core raw materials needed between raw human talent and industrial might for building chips and fabs and whatnot over time.

至于训练数据，你可以在中国雇十万人去生成海量数据。

Getting the training data, you could just go hire 100,000 people in China to go generate an insane amount of data.

获取数据的方法总是有的。

There's ways of getting the data.

所以不难预见，未来中国肯定会成为人工智能领域的一支顶尖强队，这点完全想得到。

So at some point, you can it's not that complex to think through China becoming a very, very strong, you know, player in AI.

而如果中国真的做到了这一点，全球都开始用他们的模型和框架，那显然会对美国非常不利。

And if they're a very strong player in AI and the world adopts their models and adopts their architecture, that obviously is is going to be bad for the The U.

美国

S.

从长远的经济发展角度来看就是如此。

Economically over the long run.

所以黄仁勋的意思是，与其通过施加限制来强行推动并加速这一进程，不如让我们成为现有基础设施建设的一部分。

So Jensen's point is instead of sort of forcing that and catalyzing that to happen even faster by the constraints that you impose, maybe we should actually be a part of the infrastructure build out that's already going on.

我大体上也倾向于这个观点，因为现实情况是，不管我们怎么做，这些模型最终还是会以某种方式被训练出来，只不过到时候它们会在另一套硬件架构上运行。

And I actually generally kind of land on that because what's gonna happen is the alternative is is that these models are still gonna get trained one way or another, but now they're gonna be, you know, trained on a different hardware stack.

久而久之，这套硬件架构很可能会被部署到主权云等场景中。

Ultimately, over time, that hardware stack can be the one that that gets deployed in sovereign clouds and and and whatnot.

所以我认为，根本不存在那种能完全对中国封锁、大幅拖慢他们进度，然后我们就轻松获胜的情况。

And so I think it's I just don't think there's a scenario here where you can kind of close off completely to China and somehow you dramatically slow them down and then kind of hand wavy, we just win.

我就是觉得在这个领域里不会出现那种情况。

I just don't think that plays out in this space.

你知道，很多关于开放权重或者基础设施方面的理论风险，都得假设我们对中国的领先优势大到无法逾越，而且这种优势只会继续扩大，直到对方根本追不上。

Know, all of the, you know, kind of a lot of the sort of theoretical risks of either open weights or, you know, the infrastructure side have to have to assume that we have such an insurmountable lead against China, and that only accelerates past the point where, like, there's just simply no catching up.

而且我觉得，就像我们最近看到的K3和其他模型一样，这个差距可能是在缩小，而不是在扩大。

And I think as we've seen recently with with, you know, K three and and and other models, you know, that that gap is maybe narrowing as opposed to expanding over time.

对。

Yeah.

顺便说一下，那个Dwarkesh对Jensen的采访。

That Dwarkesh Jensen interview, by the way.

真是划时代的。

Generational.

里面有很多金句。

They had such gems in it.

比如Jensen说“我不是失败者”的时候。

Like, when Jensen was like, I'm not a loser.

你知道吗？

You know?

你面对的可不是一个天生就失败的人。

You're not talking to someone who woke up a loser.

对。

Yeah.

我们不是车。

We're not a car.

对。

Yeah.

失败的心态毫无道理。

Losing mindset makes no sense.

没错。

No.

要么那是史上最好的内容之一，我真不明白他们为什么不收费。

Either that was some of the best content ever produced, so I I I don't know why they didn't charge for that one.

对。

Yeah.

感觉就像科技真人秀一样。

It felt like tech reality TV.

他好像真的投入进去了。

Like, he was actually just getting into it.

太精彩了。

It was awesome.

简直是电影级别的。

Absolute cinema.

对。

Yeah.

好吧。

Well, okay.

你觉得如果美国先开源，抢在中国前面开源一个像Kimi K3那样能力的模型，会发生什么？

What do you think would have happened if, like, let's say The US would have open sourced like, beat China to open sourcing, like, a Kimi k three level capability model?

因为我理解那种论点，就是说，我们需要部署美国的开源模型，这些模型要跟中国的开源模型一样好，甚至更好。

Like because I understand the argument of, like, okay, we need to deploy, like, American open source models that are the same, if not better, than the Chinese open source models.

但是，你觉得完全开源到中国实验室也能获取的程度，会有什么样的权衡或代价呢？

But, like, what do you think are the trade offs of completely open sourcing it to the point where obviously Chinese labs would be able to have access to it?

对。

Yeah.

首先，关于开源和闭源，我其实很欣赏围绕这个的所有讨论。

First of all, even the open versus closed, I actually appreciate all the debate that happens on this.

我对这个话题非常热衷，但我也完全理解另一方的所有论点，即使我可能在某些方面更倾向于另一边。

So I'm very passionate about the topic, but I also totally appreciate all of the arguments on the other side, even even, you know, though I I probably lean the other way in some of them.

但其中一些论点确实影响了我的一些看法，我可能会因此调整自己的立场，变得更加细致一些。

But some of them do inform in in in form some of my views, and I and I and I kind of update maybe then to be a little bit more nuanced on my end.

但在美国开源的问题上，

But on the open source U.

嗯，

S.

从开源这边来看，我总体上觉得AI赚钱的大头在推理。

Side, I generally think that the moneymaker in AI is inference.

所以最终，钱不管怎样都会流向基础设施这一层。

And so ultimately, the dollars are going to flow to the infrastructure stack one way or another.

我觉得如果世界上只有一两家实验室，而且训练方法被当作极其保密的秘密，有真正可以保护和申请专利的知识产权，别人永远无法知道，完全被锁死的话。

I think in a world where you only had one or two labs, and somehow there were these insanely kind of closed secrets about training, you had real intellectual property that was protectable and patentable and nobody ever could know about, and it's you know, totally locked down.

那么在那种情况下，我觉得你确实可以论证，技术栈上还有几层可以封闭起来。

Then in that world, I think you could probably argue that, like, you know, there's another couple layers of the stack that you could kind of close off.

但在现实中，美国会有三到五家玩家，它们都把拥有领先模型视为生死攸关的事，那么市场上就有足够的竞争动力，你可以预期token的成本会随时间越来越接近基础设施的成本。

But in a world where we're gonna have, you know, three to five players in The US that consider it sort of existential to them to have leading models, then then you have enough of a competitive dynamic in the market where you have to expect that that the cost of tokens converge closer and closer to the to the cost of the infrastructure over time.

不是降到零。

Not to zero.

我绝对不认为那里没有利润空间，但会越来越接近。

I'm definitely not a believer that there's no margin there, but closer.

所以，你知道，大概在基础设施成本之上加个20%、30%、40%的样子，而不是70%、80%或90%。

So, you know, in the twenty, thirty, 40% range on top of the cost of infrastructure, which is different from 70 or 80 or 90%, let's say.

所以如果是这样的话，实际上如果你有一个开放模型，并且你为那个开放模型提供了首选的推理基础设施，或者你为那个开放模型创建了后训练环境，或者你被认为是企业部署那个开放模型时更安全的品牌，我其实会认为，通过再次驱动那个模型的推理，你几乎能赚到同样多的收入。

So if that's the case, then actually if you had an open model and you powered, let's say, the preferred infrastructure for that open model, or you created the post training environment for that open model, or you were the kind of considered the safer brand for deploying that open model by enterprises, I actually would argue make almost the same amount of revenue just by, again, driving the inference of that model.

所以如果你假设，比如说，我我我其实会认为，即使对于像OpenAI这样的公司，如果你以更稳定的节奏，用开源版本快速跟进你的前沿模型，比如上一代的版本，我实际上认为你会更有竞争力，因为这样会让越来越多的用例留在你的生态系统和你的家族里，而且你可能也能为那些开放模型提供大量的推理服务。

And so if you assume that with things like you know, I I I would actually make the case that even for something like an open AI, if you fast followed your frontier models with open source versions of, let's say, the prior generation at a more consistent pace, I actually think you would be even more competitive with your frontier models because it would keep more and more of the use cases within your ecosystem and your family, And you probably could just power a lot of the inference of those open models as well.

所以我会说，这些只是用AI支持不同用例的不同方式，而不是完全不同的AI经济结构。

So I would I would just argue that these are these are kind of flavors of of ways of enabling different kinds of use cases with AI as opposed to entirely different economic structures of AI.

因为归根结底，不管怎样，你还是要为大型GPU集群付费。

Because at the end of the day, you still are like, you're going to be paying for large GPU clusters no matter what.

开源AI并不意味着我就能在我的笔记本电脑上运行Fable，然后就能绕过对云基础设施的需求。

Open source AI does not mean that somehow I'm going to be running Fable on my laptop, and I'm able to kind of circumvent the need to have cloud infrastructure.

就像，钱会流向……我们刚才已经讨论过这个了。

Like the dollars are going to We eventually just talked about that.

是的，我们讨论过了。

We did.

对。

Yeah.

哦，好的。

Oh, okay.

对。

Yeah.

所以这些钱最终都会流向某个基础设施提供商，而OpenAI或者Anthropic完全有可能成为那个提供商。

So the dollars are going to flow into some infrastructure provider, and there's no reason that open AI can't be that infrastructure provider, or Anthropic can't be that infrastructure provider.

显然，Meta和SpaceX现在也要进入这个领域了。

Obviously, Meta and SpaceX now are going be getting into that space.

所以我不太相信开放权重AI会彻底改变AI的经济结构，它只是提供了更多创新的途径和更多新的应用场景。

So I'm not convinced that open-weight AI dramatically changes the economic structure of AI other than to just provide even more avenues to innovation and more avenues of use cases that begin to emerge.

而且我认为这对整个生态系统来说其实是件好事。

And that I think would actually be a good thing for the ecosystem broadly.

所以我支持美国。

So I'm pro U.

对。

S.

开放权重。

Open weights.

我其实觉得很多闭源实验室应该考虑推出一些跟进更快的开放权重模型，来处理那些后训练用例和一些主权用例。

I actually think a lot of the closed labs should consider having kind of faster follow open weights models to be able to handle the post training use cases and some of the sovereign use cases.

而且我觉得从市场角度来看，这未必会像大家想的那样糟糕。

And I don't think it would actually be as bad as probably is sort of perceived from a market standpoint.

对。

Yeah.

这个话题真的很有意思。

This is a really interesting thread.

我很好奇，你知道，如果这对实验室的商业模式有好处，那他们为什么不做呢？

I'm curious, you know, if this would be good for the business model, the labs, like, why don't the labs do it?

对。

Yeah.

你知道，Anthropic 从来没有开源过任何东西。

You know, Anthropic has never open sourced anything.

谷歌开源了Gemma，这算是有点东西，但并不是最新一代的Gemini。

Google open sources Gemma, which is like something, but it's not like the last generation of Gemini.

我觉得人们大致分成两类。

I think there's two categories that people fall in.

一类是，你得有更长的时间视野，才能相信可以通过开源权重来赚钱。

One is you you sort of you have to have a you know, your your time horizon has to be longer to believe that you can monetize open weights.

因为，比如说，明天如果我发布一个闭源模型，只要我保持闭源，我就能直接赚更多钱，我强迫所有流量都走我的API。

Because, like, tomorrow, if I release a closed model, I just will literally make more money if I just keep it closed, I I force everything to route through my API.

从纸面上看，我会赚更多钱。

That that on paper, I'm gonna make more money.

这毫无疑问。

No no question.

但从长远来看，不管怎样，你都得假设每个任务的token成本最终会被压低，然后你往前推两三个阶段，你会发现闭源和开源其实差不多，因为说到底，真正重要的还是推理成本。

But over the long arc of time where where no matter what, you have to assume that the cost of token per per task, you know, gets driven down regardless, then, you know, you kind of play things out two or three stages, and it it sort of it's almost, you know, kind of a wash whether whether you were closed or open because, again, it's it's really just the inference cost that that ends up mattering.

这是第一部分。

That's that's the first part.

然后，对于Anthropic来说，第二部分是，我确实认为他们把这看作一个重大的安全风险。

And then the second part for anthropic specifically, let's say, is I actually just believe that they they consider this to be a major safety risk.

所以对他们来说，他们并不是在考虑这个。

So so to them, they're not like thinking about this.

我猜他们并不是从经济角度考虑，比如为了市场份额而开源。

I'm I'm guessing they're not thinking about this as like an economic, you know, argument that they should be open source for market share reasons.

我确实认为他们从根本上相信，你需要一个实体来控制token的流动，以防止提示注入，并确保你能根据用户的查询路由到不同的模型。

I I actually just think that they they fundamentally believe, no, you actually need one entity to control the flow of the tokens to be able to do, you know, kind of prevent prompt injection and make sure that you can route to different models based on kind of queries people are doing.

在开放权重环境中，你做不到这一点。

You can't do that in an open weights environment.

所以我几乎认为他们可能永远不会开源他们的模型，原因与Anthropic的创立初衷有关。

So I would just almost consider them to likely never to open source their models for, you know, sort of like, you know, kind of reasons that are tied to just the creation of Anthropic in the first

对。

Yeah.

我觉得，关于前沿实验室，这要引出Cloud Opus 4.5，或者现在Cloud Opus 5已经发布了。

I feel like, well, with regards to Frontier Labs, is gonna be a segue to Cloud Opus 4.5 or from Cloud Opus five being out now.

不过，关于前沿实验室模型发布的时机，现在感觉就是个时间问题。

But, yeah, with regards to the timing of the release of the models in Frontier Labs, it just feels like a timing thing now.

我们知道他们内部在开发很棒的模型。

We know they're cooking good models internally.

那么，说到这个，你对 Opus 5 有什么看法，尤其是从知识工作的角度？

So with that being said, what were your thoughts on OPUS five, especially from a knowledge work perspective?

因为你在 Box 有这方面的经验。

Because you have that perspective at Box.

对。

Yeah.

我们过去大概一周、一周半或几周的时间里，一直在对 Opus 5 进行评测，它是个非常棒的模型。

So we we we've been running evals on Opus Opus five for the past kind of maybe week or week and a half or so, a couple weeks, and it's it's a fantastic model.

相比 Opus 4.8，它有了显著的提升，而 Opus 4.8 在发布时已经是同类中最好的了。

It's meaning meaningful jumps over Opus 4.8, which already was kind of best in class at the time period that it was launched.

这在我们这里体现为，我们处理的是企业数据、文档、财务文件、合同、营销素材，以及各个行业的研究材料。

The way this shows up in our world, we deal with corporate data, documents, financial documents, contracts, marketing assets, research materials across every single industry.

所以这可能包括生命科学公司、律师事务所、大型银行。

So this can be life sciences companies, law firms, large banks.

你可以想象，如果你是一家大型企业，你需要在AI模型中具备两个主要方面。

And so you can imagine that the things that if you're a large enterprise, you kind of need two major aspects within the AI model.

你需要深入的领域理解。

You want a deep domain understanding.

比如你必须了解生命科学，你必须了解法律。

Like you have to understand life sciences, you have to understand law.

这些必须内置于模型中。

Like that has to be packed into the model.

然后你必须非常擅长处理大量数据、处理工具、分析能力，这些是横向的。

And then you have to be extremely good at just being able to kind of work with large amounts of data, process it, deal with tools, analytical capabilities that are kind of horizontal.

所以你有通用智能的提升，这很重要，然后你有非常具体的行业智能。

So you have kind of general intelligence going up that matters, and then you have very specific kind of industry intelligence.

而Opus 5在这两个维度上都代表了进步。

And Opus five kind of just represents an improvement on both those axes.

所以我们拿它跟我们做的各种行业测试比了比。

So we tried it against, you know, a variety of different industry tests that we do.

跟4.8比，确实有很明显的提升。

Again, kind of meaningful jumps over 4.8.

我觉得这会成为知识工作领域非常有竞争力的模型。

I'd expect that this, you know, becomes, you know, again, very compelling model across knowledge work.

就像GPT-5.6一样，我觉得它也有同样的实力。

As, you know, the GPT 5.6, you know, I think equally has.

所以Claude这边确实做得非常好。

So def definitely great kind of work on the on the Claude front.

而且我觉得，真正让人兴奋的是，这也是为什么我不认为这是零和游戏。

And and I I think, you know, again, like, what what's exciting is and this is why I don't think it's zero sum.

封闭实验室继续在前沿保持领先，而且我觉得在几乎所有可能的情况下，大部分利润还是会流向封闭实验室。

Like like, the closed labs continue to stay ahead on the frontier, and and I think the vast majority of the dollars in profit will still go to the closed labs in almost all scenarios that this plays out.

所以我觉得，这个月对OpenAI和Anthropic来说都是很棒的一个月。

And so I I, you know, I think it's been a great month for both OpenAI and and Anthropic on that front.

对。

Mhmm.

你觉得Fable在哪些任务上比Opus更好用吗？

Are you finding Fable better for any tasks than Opus?

你知道，感觉上好像稍微好一点，但价格也是两倍。

You know, like, is it seems on vibes, like, very slightly better, but also twice as expensive.

所以Opus看起来在性价比上更优。

So Opus seems like Pareto superior there.

是的。

Yeah.

所以我觉得我们还没有……我们有一些内部基准测试，主要针对一般的知识工作，我其实同意你的看法，可能因为成本差异，你会说Opus 5现在比Fable更好，因为价格差距在那儿。

So I don't think we've we've like, so we have internal benchmarks that deal with kind of general knowledge work, and I would actually agree with you that that probably due to the cost difference, you you probably argue that that that that Opus five is now kind of an improvement over Fable because of the the the cost delta.

在某些我们内部测试过的编码任务上，Fable完全超越了4.8，那种优势是光靠token成本差异弥补不了的。

On certain coding tasks that that we had tested internally, Fable completely out outmatched four eight, like, in a in a in a way that you couldn't make up for by the by the, you know, token cost difference.

就是能对复杂的技术问题给出更优秀的解决方案。

Just like like being able to have superior solutions to hard technical problems.

抱歉。

Sorry.

四AMR升级到五。

Four AMR up is five.

在能用到五之前，如果我们比较一下五在那方面的进步有多大，我手头没有足够的内部编码测试来做对比。

Prior to having access to five, if we kind of compared, like, how big of a leap five was on on that, I don't have enough internal coding tests to to kind of compare against.

所以我得看看四八到底有多强，是不是真的在编码上更胜一筹，而他们刚发布的很多基准测试都是那种通用知识工作、代理工具使用类的。

So so I had to see if, like, if how much is four eight meant to be, you know, kind of a superior coding model versus a lot of the benchmarks that they just came out with were these kind of general knowledge work, kind of agentic tool use type benchmarks.

所以我觉得这还是个开放性问题。

So so I think it it kind of open question.

Fable的一个问题，大家也都在推特上提过了，就是它经常会在某些功能上把你降级到四八。

The one problem that Fable had, and everybody's kind of tweeted about this already, is is it will often, you know, again, kind of push you down to four eight for certain capabilities.

然后这就成了问题，因为你很难提前知道，是不是因为它在查看权限、访问控制代码的时候触发了什么安全警告，然后它就慌了，结果就没给你Fable级别的智能。

And then that becomes a problem because you're not you know, it's kind of hard in advance to know, like, did the did the thing kind of, you know, trigger some security warning because it was looking through permissions, you know, access control code, and so it kind of freaked out and and, you know, isn't then giving you kind of fable level of intelligence.

所以我觉得Anthropic确实得解决这个问题。

So I do think that Anthropic is going have to kind of work through that.

我听说生物领域的一些人实际上没法用Fable，因为它总是把他们的查询降级得太频繁，所以你得不到模型的原始智能。

I've heard of folks in the biospace that effectively can't use Fable because it just, again, pushes their queries down too frequently, so you're not getting the raw intelligence from the model.

所以我很欣赏Anthropic最终提出的方案，关于我们应该如何与政府处理这类动态，建立一个多点框架，我们共同确定哪些能力属于不同风险级别，并针对这些能力测试模型。

And so I appreciated Anthropoc's ultimate proposal on how we should kind of work through these types of dynamics with the government, have a kind of a multipoint framework where we agree on what are these capabilities that are of different risk levels and test the models for them.

但有一个有趣的问题：谁来决定这个风险框架，我们又怎么达成一致，认为这些风险确实是我们感知到的真实风险？

But there is an interesting question, which is like, Who gets to decide that risk framework, and how do we agree that the risks are actually real risks that we perceive?

因为我认为，如果你现在对Fable做一个时间快照，他们降级并阻止你做的事情的水平，对AI的未来来说是不可持续的。

Because I would argue that if you snapshot it in time fable right now, the level of things that they push down and prevent you from doing, that would be untenable for the future of AI.

如果这种环境持续下去，人们根本就不会使用AI。

People will will simply not use AI if if this is the kind of ongoing environment.

对。

Mhmm.

所以，我们现在应该怎么决定呢？

And so so, like, how should we decide now?

也许他们看到这些风险非常微妙，所以这就是为什么我倾向于认为，现在锁定这些模型还为时过早，考虑到我们处于这么早期的阶段。

Like, like, maybe they see these risks are very highly nuanced, and so that's why I generally lean to where like, it's all very, like way too early to be locking these models down, you know, given how early we are.

那这对Box的软件工程有什么影响呢？

So how is it affecting software engineering at Box?

比如，你们是招更多人还是更少人？

Like, are you hiring more or fewer people?

你们对软件工程师的技能要求是怎么变化的？

Like, how do the skills that you're looking for in software engineers change over time?

你预计在不久的将来这些要求会怎么变？

How do you expect them to change in the near future?

对。

Yeah.

我仍然非常看好软件工程。

We're I'm I'm still very bullish on software engineering.

你知道，我们使用这些模型的方式就是让我们比以前做得更多。

We're we're you know, the way that we've used these models is is simply to just to do way more than we were before.

这些变化非常定性，可能只有每天参与产品路线图评审的人才能完全理解。

The the kinds of and it's just, like, so qualitative, and only only people that are, you know, probably, like, in the product road map reviews each day can, you know, maybe kind of fully process.

但是，当你做产品规划的时候，情况就完全不一样了。当你只考虑纯人力限制的时候，和现在AI能解锁的东西相比，你的思维方式会彻底改变，你会去考虑做哪些事情。

But, like, there's something totally different when you're when you're doing a a product planning session, when you are thinking in a world of just sheer human based constraints versus now what AI unlocks, it completely changes your mindset about the things that you'll go and tackle and take on.

我们现在有几十个项目，如果AI不存在，我们绝对不会去做。

We have multiple dozens of projects right now that we absolutely would not be doing if AI didn't exist.

我们根本不会启动这些项目。

We would not have lit up the projects.

我们会说，不行，那太复杂了。

Like we would have said, No, that's too complex.

不值得做。

It's not worth it.

实际上，你会遇到两种有趣的情况。

Like you actually have two interesting scenarios.

要么你因为事情太小不值得做而拒绝，要么你因为事情太难而拒绝。

You either say no to the very small things because it's not worth it, or you say no to the really big things because they're simply too hard.

所以，你的大多数软件项目都停留在中间地带，就是那种，好吧。

And so most of your software projects kind of stay in this middle band, which is like, okay.

感觉像是那种可能一到六个月就能完成的事情。

It feels like something that maybe you could do in one to six months.

这些就是我们以前在AI出现之前会去处理的那类事情。

Those are the kind of things that we would use to kind of tackle pre AI.

现在的情况是，你可以去处理那些需要多年的项目，因为它们不再需要那么长时间了，而那些像是一周就能搞定但又不是什么大问题的事情，我们以前根本不会去做。

Now what happens is you can tackle the multi year projects because they're not multi year anymore, and the things that are kind of like, well, that would take a week, but it's not that big of an issue, so we're just never going to do it.

所以它们就一直留在待办清单里，而且排得非常靠后。

And so it just remains on the backlog, just very far down the backlog.

现在你会去做这些，因为只需要两个小时。

You do those because now that takes two hours.

所以最终的结果是，你基本上能够解决更多客户一直以来的问题，而这实际上也提升了你的雄心壮志。

So what happens is you end up basically being able to solve more and more problems that your customers have always had, and that actually just increases your ambition.

所以，你知道，只要有可能，我一般都倾向于增加更多的人才，因为实际上有更多的事情我们想要去尝试和承担。

So, you know, wherever possible, I'm generally in the mode of actually trying to add more human talent to this because there's actually just more things we want to go and take on.

而现在，真正的问题主要是，你知道，你只是受到财务上的限制，因为业务的其他领域你也想发展，也想招人。

And now, actually, the main problem is just we have you know, you just have financial constraints because there's other areas of the business that you wanna grow and that you wanna hire for.

但我总体上觉得，如果你认为你已经不需要软件工程师了，那只能说明你对产品路线图的野心还不够大。

But but I I I I generally think if you're if if you if you think that that you've kind of eliminated the need for software engineers, you're you're just you're there's just no chance you're being ambitious enough with your product roadmap.

所以我们一直在逼自己思考，因为有了AI，我们还能再多做些什么。

And so we're constantly pushing ourselves of now what more can we take on as a result of AI.

对，完全同意。

Yeah, totally.

是的。

Yeah.

即使和一些编程代理交流，他们也会高估完成一个项目所需的时间。

Even talking to some of these coding agents, they will overestimate the amount of time that it takes to do a project.

他们会严重高估，比如说这是一个周末的项目。

Like, severely, they'll be like, this is a weekend long project.

有时候他们会说需要一个月。

Sometimes they say it's a month.

对。

Yeah.

他们说，这得做一个月的项目。

They're like, this is a month long project.

我就想，天哪。

And I'm like, oh my god.

这得花好长时间啊。

Like, this is this is gonna be so long.

结果我三个小时就搞定了。

And then I knock it out in, like, three hours.

对。

Yeah.

那是因为它学了我们AI出现之前所有的Slack消息。

That's because it was trained on all of our Slack messages pre AI.

对。

Yeah.

所以工程师说这得两年的时候。

And so where the where the engineer, you know, said that's gonna be two years.

所以我觉得它肯定是被训练得对项目时间线非常非常保守。

So I think it's it's it's definitely trained to be highly highly conservative on project timelines.

数学方面也一样。

Same with math too.

比如，你问一个人工智能，它要多久才能推翻一个数学上的开放猜想？

Like, if you ask an AI, like, what is your timeline to an AI model disproving an open conjecture in mathematics?

对。

Yeah.

它可能会说，大概五年吧。

It'll be like, maybe five years.

不对。

Nope.

这事儿已经发生了。

It's already happened.

对。

Yeah.

完全同意。

Totally.

我想，最后一个问题，因为我们之前聊过这个，就是，既然模型发布看起来越来越频繁，特别是对企业来说，只忠于一个供应商似乎行不通了。

I guess, like, last question, because we were talking about this earlier, is, like, since it looks like model releases are just happening way more frequently, for enterprises specifically, it seems like loyalty to one provider is not gonna make it, basically.

那你觉得这里的最佳策略是什么？

So what do you think is the optimal strategy here?

我们之前聊过模型路由是未来，以及所有提供这类服务的公司都是未来。

We were talking about how model routing is the future and all of these services that offer that sort of thing are the future.

对。

Yeah.

我的意思是，这基本上就是我的结论，但我对这个结论也特别有偏见。

Mean, kind of that's my conclusion, but I'm also extremely biased in that conclusion.

你越是需要多个模型来完成一个任务或一组任务，那个能理解任务、获取数据并处理工作流的层面就越有价值，这显然对应用AI层面更有利，而我们通常就处在这个层面。

The more that you need multiple models to do a task or a set of tasks, the more value accrues to the layer that can understand the task and get access to the data and handle the workflow, which obviously is a better outcome for the, let's say, applied AI layer, which is where we tend to sit.

这个未来显然很多公司在战略上，甚至在某些情况下在生存层面，都朝着这个方向对齐。

That's a future that obviously a lot of companies are aligned to strategically and kind of existentially in some cases.

所以像 Cognition、Factory、Cursor 和 Replit 这些公司，你们真正希望看到的结果是拥有多种模型。

So if you're Cognition and Factory and Cursor and Replit, etc, obviously the outcome that you want to have happen is that you actually have many models.

它们在各自擅长的维度上表现都很出色。

They're all good on different axes.

有些是专门优化成本的主力模型，有些则是顶级的前沿编排器。你们希望达到的效果是，只有同时使用多种模型，才能高效且经济地完成这项任务。

Some are really the cost tuned kind of workhorse models, and some are the super frontier you know, orchestrators, and you want to have an outcome where you actually need multiple of those models to to to be able to complete the task effectively or at least cost effectively.

从目前来看，这似乎就是我们正在经历的发展阶段。

That's actually the that that appears to be the timeline that we're on.

而且，现在我觉得基本上有五个靠谱的美国玩家。

And And now, again, I think you have basically five credible U.

在美国。

S.

在模型开发领域，这五家分别是 SpaceX、Google、Anthropic、OpenAI，还有 Meta，我可能漏掉了谁。

Players in model development between SpaceX, Google, Anthropic OpenAI, who'd I forget, and Meta.

所以这五家公司都在全力竞争，既要降低智能的成本，又要提升前沿技术的水平。

And so those five players are all on a warpath for both driving down the cost of intelligence and improving the frontier at the same time.