Hugging Face 首席执行官谈开源 AI、模型路由与竞争的未来

我觉得蒸馏这种做法非常普遍，人人都在用。

I think distillation is a very common, you know, practice that everyone is using.

每个人都在用，但这并不是成功的主因。

It's something that everyone uses, but that is not, you know, the main reason for success.

如果你本身不行，不管有没有蒸馏，你还是不行。

Like, if you suck, you suck with or without distillation.

让我这么说吧，像 Entropic、Pro OpenAI 这样的公司，你明明是全球增长最快的公司，却说你被不公平竞争了，这真不太好说。

It's hard for me to say, like, oh, or Entropic, Pro OpenAI, you're getting unfairly competed with when you're, like, the fastest growing company in the world.

我觉得，他们反而更需要竞争，而不是更少的竞争。

Even using I think they need more competition than less competition.

对。

Yeah.

因为我们正走向一个少数几家公司完全垄断、掌控所有权力、能力和财富的世界。

Because we're heading towards a world where a few companies are completely dominating, concentrating all power, all capabilities, all wealth.

这比它们可能损失几十亿美元的收入要危险得多。

And that's much more dangerous than them maybe losing a couple billion dollars of revenue.

随着人工智能模型越来越强大，各国政府开始提出关于安全、监管以及谁该掌控前沿技术访问权的新问题。

As AI models become more powerful, governments are beginning to ask new questions about safety, regulation, and who should control access to frontier technology.

在本期节目中，西奥·贾菲和索菲亚·普奇尼邀请了Hugging Face的联合创始人兼首席执行官克莱芒·德朗格，讨论他为何认为开源人工智能本质上更安全，Hugging Face年经常性收入达到一亿美元意味着什么，以及下一代人工智能可能由模型路由主导，而非少数几家领先的实验室。

In this episode, Theo Jaffee and Sofia Puccini sit down with Hugging Face co founder and CEO, Clament Delangue, to discuss why he believes open source AI is inherently safer, what Hugging Face reaching $100,000,000 in annual recurring revenue says about the business of open source, and why the next phase of AI may be defined by model routing rather than a handful of dominant frontier labs.

他们还谈到了GPT-5、人工智能监管、本地模型、欧洲的生态系统，以及人工智能全栈竞争日益增长的重要性。

They also discuss GPT-five, AI regulation, local models, Europe's ecosystem, and the growing importance of competition across the AI stack.

我们再次邀请到Hugging Face的克莱芒·德朗格，这是他第二次做客MTS。

And we are back with Clament Delangue from Hugging Face, his second time on MTS.

Hugging Face基本上就是开源人工智能的平台。

Hugging Face is basically the open source AI platform.

克莱芒，欢迎再次来到MTS。

So, Clam, welcome back to MTS.

当然。

Absolutely.

是的。

Yeah.

那你说吧。

So, well, go on.

我觉得我们要说的是一回事。

I think we're about to say the same thing.

对。

Yeah.

我们刚才在聊一个最近的新闻，说政府打算单方面限制GPT-5.6的发布，这以前从来没发生过。

We were we were talking about this interesting recent piece of news where basically the government is going to restrict GPT-5.6's release sort of unilaterally, basically without precedent.

我记不得以前有哪个政府曾要求前沿实验室别发布某个模型。

I don't think the government has ever asked a frontier lab not to release a model before.

更别说让政府来插手决定谁能用这个模型了。

Certainly, a government has not asked a frontier lab to be able to oversee which customers the model is released to.

这听起来太不寻常了。

This seems very unnatural.

你对这事怎么看？

What are your takes on this?

对。

Yeah.

所以这事儿挺有意思的。

So it was it was funny.

我上周在华盛顿特区，所以算是亲眼目睹了当时发生的一些情况。

I was in DC last week, so I was kind of, like, had some sort of to to what what was happening.

有个有趣的插曲是，我们在那儿碰巧遇到了 Tom Brown，他是 OpenAI 的联合创始人之一。

Interesting anecdote is that we we bumped into Tom Brown there, like the cofounder of of.

我们当时就在想，哦，这看起来像是人事变动，或者类似的情况。在消息公开之前，他们稍微调整了一下与白宫沟通的人员。

And we were like, oh, this seems to be like a change change of staff or or or something like that before before it was made public that they they changed a little bit the people talking to to the White House there.

你听我说。

Listen.

我是说，我看到的、我听到的情况是，现在有一个

I mean, what what I've seen, what I've I'm hearing is that there there's a

很大的

lot

美国政府里有很多人对人工智能的风险非常关注，希望采取稳妥的方式推进前沿模型的部署。

of, you know, interest from from a lot of people in the in the USG to really understand the risks of of AI and take kind of like a a safe approach to the deployment of frontier models.

说实话，我真不怪他们，毕竟过去几年，那些前沿实验室一直在搞营销。

To be honest, I can't really blame them because of the fact that, you know, the frontier labs were basically doing marketing for the past few years.

对吧？

Right?

你还记得吗？当年说GPT-2太危险，不能公开发布。

Like, if you remember, GPT-two was too dangerous to release.

是的。

Mhmm.

对吧？

Right?

那都是四五年之前的事了。

It was, like, what, like, four four, five years ago.

所以我真不怪他们这么做。

And so, you know, I I I don't really blame them, for for doing things like that.

我只是希望，随着时间推移，我们能对什么是安全、什么是不安全有更多透明度，更注重对模型的公开评估这类事情。

I just hope that, progressively, we'll get a little bit more transparency about, you know, what is safe, not safe, so more focus on, you know, transparent evaluation of of models and things like that.

我也希望这种监管能只集中在少数几个前沿通用模型上，说实话，我觉得它们才是最危险的。

I also hope that it's gonna stay contained to, you know, a few frontier, generalist models because, frankly, I think they're the most dangerous ones.

对吧？

Right?

而且这些公司都是万亿级的企业，背后有庞大的华盛顿游说团队。

And also, you know, these companies are trillion dollar companies with armies of, you know, DC people.

所以它们有能力应对这些挑战。

So they they can they can kind of, like, deal with it.

我希望监管范围就限定在这些大公司，别扩展到其他参与者，比如初创企业、小公司、学术界，或者那些没资金、没规模、没能力处理这些问题的人。

I I hope it's gonna stay contained to that and not, you know, permits to a lot of different players, for example, startups, small companies, you know, academia, or like people who don't necessarily have, you know, the money, the size, the ability to really deal with these things.

这才会是我最担心的事。

That that would be kind of like my, my my main concern.

完全同意。

Totally.

是的。

Yeah.

我们刚才在讨论的是，这种做法能不能用在开源公司和开源模型上？

What we were just talking about before was just like, is this could this be applied to open source companies and models in any way?

美国政府有没有可能从实际或法律层面来限制开源公司？

Like, is there a way that the US government could come in and restrict open source companies, either practically or legally?

我不觉得会，因为开源模型本身比现在那些被限制的模型危险性低得多，毕竟这些模型在技术前沿上没那么靠前。

I I don't think so because, you know, I think I think the open source models are inherently less dangerous than kind of like the the models that are getting restricted now, because, you know, these models are a little bit ahead in terms of, like, being closer to to the frontier.

而且开源模型的通用性也没那么强。

Also, open source models are are less generalized.

它们更偏向于特定用途。

They're more specialized.

没错。

Mhmm.

而且目前几乎没人，或者只有极少数人在研究如何用这些模型开发危险的网络安全能力。

And there's just, like, nobody or very few people focusing on building dangerous, like, cybersecurity capabilities.

是的。

Mhmm.

所以这和那些商业实验室的情况有点不一样。

So it's, like, a little bit different than than kind of, like, the proprietary labs.

对，没错。

So I yeah.

我觉得这说不通。

So I I I don't think I don't think it would make sense.

我觉得开放源代码项目不太可能被波及，就是因为这些差异。

I I don't think it's gonna get to to open source just because of some of these differences.

从更宏观的角度来看，开放源代码AI比商业AI安全得多，这一点我之前在不同场合也提到过不少理由。

A more high high level, you know, I I think in in general, open source AI is is much, much safer than kind of, like, proprietary AI or a bunch of of different reasons that I've talked about in in different outlets.

不过呢。

But yeah.

所以我觉得，针对这两类模型的处理方式应该有所不同：一边是封闭的、前沿的商业模型，另一边是整个行业和生态体系——在我看来，后者带来的风险和威胁小得多，我们更应该支持和推动它，让小型科技公司和每个人都能参与进来，共同推动AI的发展。

So so I I think the the approach there is gonna be hopefully, it's gonna be a little bit different between kind of, like, closed source proprietary frontier models, and then the rest of the industry and the ecosystem that has, in my opinion, pose much less danger, much less stress threats where we wanna keep kind of, like, you know, supporting and and and building up to focus on, you know, competition to to enable kind of, like, little tech, small companies, everyone to be able to basically participate in in AI.

是的。

Mhmm.

没错。

Totally.

我同意，目前开源模型确实更安全。

I agree that open source models are less dangerous now.

但要是六个月后趋势继续，某个地方可能会推出一个能力接近顶级水平的开源模型，让政府都坐不住。

But in, you know, six months, if trends continue, you're gonna have an open source model from somewhere with, like, mythos level capabilities that freaks out the government.

到那时，你完全可以想象他们会想限制开源模型。

And you could imagine them at least wanting to restrict open source models at that time.

我不太确定，因为有个奇怪的现象：最危险的技术通常并不是在开源环境中开发出来的。

I'm not I'm not that sure because, there's this weird thing where, basically, the most dangerous things usually are not so much developed in in open source.

你知道，也许就像人们常说的，阳光是最好的消毒剂。

You know, maybe it's this this thing that people say, like, sunlight is the best disinfectant.

有趣的是，有时候安全专家讨论的反而是核武器。

You know, like, funny funny thing is that sometimes, safety people are are talking about the the nuclear bomb.

核武器从来不是在开源环境下研发的，我觉得也永远不可能在开源环境下造出来。

Nuclear bomb has never been built in open source, and I think it would have never been built in open source.

它是在封闭的、由政府主导的团队里，投入了数十亿美元才搞出来的。

It's been built in a closed source, you know, proprietary team with billions of dollars of, you know Mhmm.

投入了海量的资源。

Of of resources.

所以这跟某些参与者比起来，根本不是一回事。

So it's much less like, you know, some players than some others.

我觉得还有一种可能，就是开源会继续发展出针对不同领域的专用模型，而不是去搞那些风险最高的方向。

So I think I think there's also a path where open source keeps building kind of like more specialized models for different domains and not necessarily for the domains that are presenting the biggest risks.

你懂我的意思吧？

You know?

我不觉得模型越强大，就一定对网络安全构成更大威胁。

I I don't think it's automatic that you build a more powerful model that it's more dangerous for cybersecurity.

比如，你可以造出一个更强大的模型，但它对网络安全并不构成威胁——比如说，如果你根本没用网络安全相关的数据去训练它，而这种事，其实根本没人提。

For example, you could build a more powerful model that is not more dangerous for cybersecurity, for example, if you don't train it on psych cybersecurity, which, really, people are not really talking about.

我们为什么不讨论这个，反而去讨论限制发布能力，或者事后添加那些根本没用的自我保护机制——毕竟谁都能绕过它们？

Why why are we not talking about that rather than talking about, you know, removing the ability to release or putting self, like, safeguards after the fact that we know are not really working because everyone can jailbreak them.

所以我觉得，开源社区因为结构跟大实验室完全不同，反而可能走上另一条路，让它更安全，根本不需要像封闭式AI实验室那样依赖严格监管。

So so I think, I I wouldn't be totally surprised if the open source, you know, community just because it's structurally very differently set up than the big labs are actually, taking different directions that keeps it kind of, like, safer and and never really requires the kind of regulation that you need in kind of, like, a closed source, AI AI labs.

但这不也正是在质疑开源AI研究的能力吗？

Well, was this not also kind of an argument against the capability of open source AI research?

我同意，因为它解决的是不同的问题。

I think so because, you know, it solves different problems.

比如我常举的例子是本地模型。

Like, I sometimes take the example of, you know, local models.

对吧？

Right?

本地智能，比如在没有网络的飞机上也能用，照样能提供智能支持。

Local intelligence, you know, being able to be on the flights in airplane mode without network and still be able to get intelligence.

这种能力只有开源才能做到。

You can only get that with open source.

对吧？

Right?

用API根本做不到这一点。

Like, you can't get that with API.

这根本就没办法实现。

This just got, like, literally no way no way no way to do that.

我觉得这只是栈的不同层级。

So I think it's just different different layers of the stack.

对吧？

Right?

实际上，开源是建立在闭源之上的。

And and actually open source is on top of closed source.

很多闭源系统都在用开源模型和开源基础设施。

A lot of the closed source is using open source models and is using open source infrastructure.

这些都是不同的东西。

And it's all different things.

这个类比就像是，开源可能是引擎，而API是整车。

The the analogy is like, you know, open source maybe is the engine engine and the API is the car.

对吧？

Right?

引擎当然不可能像法拉利那样，但正是它在驱动法拉利跑起来。

And and, obviously, the engine is never never gonna be like a Ferrari, but, you know, that's that's what kind of, like, powers powers powers the Ferrari.

没错。

Right.

所以我觉得，我们看待这个问题的方式就该是这样。

So I think that's that's more like the way, the way we approach it.

而且，不擅长做坏事，并不代表就不能把好事做得更好。

And also, you know, being less good at bad things doesn't mean that you can't be better at good things.

也许开源更擅长解决人们真正的问题。

Maybe, you know, open source is better at, you know, solving people's problems.

也许它更擅长做那些真正重要的事，但在搞网络攻击这方面就不如别人了。

Maybe it's better at, you know, kind of like, helping, you know, do stuff really, really, really important, but, it's worse at, you know, creating cybersecurity attacks.

那才是最理想的情况。

That that would be that would be kind of like the ideal ideal case.

现在大家说的前沿，好像都是个笼统的概念。

People right now are talking about frontier as kind of like this general thing.

但实际上，前沿在不同任务和领域之间是参差不齐的。

The reality is that the frontier is kind of like jagged between kind of like different tasks, different domains.

对吧？

Right?

这个模型在某些方面强，在其他方面就弱。

This one model is gonna be better at some things, and it's gonna be worse at other things.

我觉得这才是我们应该看待问题的方式。

I think that's that's more kind of like how we should approach it.

是的。

Yeah.

你明白吗？

It make sense?

不对。

No.

对。

Yeah.

这我完全能理解。

This totally makes sense to me.

你们刚刚突破了1亿美元的年经常性收入吧？

So you guys just crossed a $100,000,000 in ARR.

是吧？

Right?

我想知道，这对开源的商业模式意味着什么？

So I'm curious as, like, what do you think this means for, like, like, business model of open source?

当然，现在有很多公司在人工智能领域的收入远超我们。

Obviously, lot of companies are are doing much more revenue than than we do in AI these days.

说实话，我们一直没把盈利和收入作为优先目标，因为我们做的更像一个以使用量为基础的平台，目标是让更多AI开发者受益。

You know, It it hasn't really been our priority to optimize for for monetization and for for revenue given what we're building, which is more kind of like a usage based platform, to reach kind of like how many and empower as many AI builders as as as possible.

但在这种小规模下，我觉得已经证明了开源是有商业模型的。

But at, you know, at this at this small scale, I think it shows that, you know, there's a business model for open source.

开源平台也是有商业模型的。

There's a business model for open source platform.

我们其实早就有这个感觉，毕竟之前已经有不少成功的开源公司了。

We kind of knew it, right, because there's there's been before, and there's been kind of, like, a bunch of open source, successful companies.

但我想，这确实是对这一点的验证。

But I I guess I guess it's a it's a validation of of that.

过去几周，我们看到人们对开源模型和本地模型的兴趣和采用率都有了明显增长。

And we've seen I mean, for the past past past few weeks, we've seen quite quite a lot of growth in terms of interest and adoption of not only open source models, but also local models.

这也从侧面说明了这一点。

And so, you know, that that also speaks speaks a little bit to that.

是的。

Yeah.

人们具体都在用本地模型做哪些应用场景？

What are the some of the specific use cases that, like, people are using local models for?

本地模型其实挺免费的，对吧？

So local models are kind of like kind of free, right?

因为它们直接运行在你的手机或笔记本上。

Because they're running on your on your phone or on your laptop.

所以你根本不需要为它们付费。

And so you don't even you don't really have to to to pay for them.

它们要便宜得多。

So they're they're much cheaper.

而且从设计上就更注重保护隐私，确实如此。

They're much more, privacy kind of like preserving Mhmm.

因为你不需要把数据发到任何 API。

By design because you don't have to send your data to an API.

对吧？

Right?

你的数据都留在自己的手机里。

Your data stays on your on your phone.

所以我们看到人们在最重要的场景下大量使用它。

And so we you we see people using it a lot for the things when it matters the most.

比如，如果你要讨论自己的健康隐私，不想让别人知道；或者想分享一些公司机密数据，又不想传给外部的API服务商；又或者你想运行特别重的负载，比如智能体任务，需要7x24小时持续运行——这时候在自己的笔记本、Mac Mini，或者本地硬件上跑，会更持久、更靠谱。

So like, for example, if you want to talk about your private health and you don't want to share that with someone else, if you want to share some of your private company data and you don't want to share it externally to an API provider, or if you want to run really, really heavy workloads, for example, agentic workloads, and really have something that runs twenty four seven, then doing it on, on your laptop or like on the Mac mini or, or kind of like a local hardware, makes it much, much more sustainable.

这些就是主要的应用场景。

So that that would be kind of like the use cases.

我们有个叫Llama CPP的库，是目前最主流的本地AI运行时，很多人用它配合GPT-OSS。

We have this, library called Lama CPP, which is the most used, runtime for for local AI workloads that people are using a lot, and they use using, GPT-OSS.

他们都在自己的笔记本上本地运行Gemma、Llama这些模型。

They're using, GWEN, Gemma four, like, all all these models locally on their on their laptop.

是的。

Yeah.

当然。

For sure.

当然。

For sure.

回到开源模型，你可以想象，政府可能会因为很多开源模型来自中国而想加以限制。

So going back to open models, you can imagine that the government will want to restrict open models because a lot of them come from China.

也许不是通过严格的法律手段，而是通过出口相关的措施。

Maybe not restrict them in a strict legal way, but maybe in, like, an export related way.

你觉得这种情况会发生吗？

So do you think that might happen?

如果真的发生了，这对 Hugging Face 意味着什么？

And if so, what would that mean for Hugging Face?

是的。

Yeah.

我的意思是，开源权重和 API 从根本上是不一样的。

I mean, like, it's so to open open weights are fundamentally different than an API.

对吧？

Right?

限制它们的方式也完全不同。

The way the way you can restrict it is very different.

比如，如果你从 Hugging Face 上移除了一个开源模型，它还是会出现在 Model Scope 这类中国类似的平台，或者出现在 BT 下载站点上。

So for example, you know, if you remove an open weight from Hugging Face, then it's still gonna be on model scope, for example, which is like the Chinese equivalent, or it's gonna be like on Torrent platforms.

所以我觉得，对开源模型来说，限制的方式跟 API 完全不一样。

So, like, restriction, like, looks very, very different, I think, for for open source than than for APIs.

你根本堵不住，因为它是开源的。

You can't really block because it's open.

按定义来说，总会有办法能接触到这些模型。

So most by by definition, there's gonna be some ways some ways to to access access them.

而且开源模型的来源其实没那么重要，因为既然开源了，分享的人就等于放弃了控制权，也放弃了影响你的能力。

And also, the provenance for open open ways doesn't really matter as much because because because it's open, the the people who are sharing it kind of like give up the control on it and give up kind of like their ability to influence you.

所以对我来说，开源权重从哪儿来根本没那么关键。

So to me, it doesn't matter so much where open weights are coming from.

但 API 就不一样了，比如你用的是中国公司的云服务或 API 提供商，你的数据会传过去，他们随时可以切断你的访问，或者给你结果加偏见。

It's a different game, for example, for APIs to run the inference, because if you're using an API provider or cloud from China, obviously you're sending your data and also they could cut your access or bias your access.

这才是真正大得多的问题。

That's a much, much bigger problem.

但对于开源权重和开源项目来说，来源其实没那么重要，因为你掌握了全部控制权，也拥有完全的透明度。

But for open weights and open source, it doesn't really matter where it comes from because it's it's kind of like, you you get all the control, you get all the transparency.

没人能骗你、误导你、操纵你，或者剥夺你的访问权限。

There's no way to kind of like trick you, bias you, manipulate you, remove your access.

所以我觉得，对于开源项目来说，来源就没那么关键了。

So I think for open source, like the provenance doesn't doesn't matter as much.

是的。

Yeah.

我想知道，你怎么看美国政府限制发布像GPT-4.6这样的模型？

I'm curious, your thoughts on just like the US government sort of, like, restricting the release of, like, GPT 4.6.

你觉得这是因为他们清楚其中的风险，还是根本没搞明白？

Do you think that comes from, like, them knowing the stakes or not knowing the stakes?

你是觉得他们对这事了解得很透彻，还是只是知道自己不懂，所以才采取这些措施？

Like, do you think they're well informed on this matter, or they just, like, know that they don't know and that's why they're taking these measures?

这是个好问题。

It's a good question.

我没法替他们说话。

I can't talk for them.

我觉得无论在哪里，包括不只是美国政府，但在各个地方，对模型和风险的评估都有很多值得学习和进步的空间。

I do think there's a lot of interesting learning and progress to be made everywhere, not just at the USG, but really everywhere on evaluating models, evaluating risks or for models.

我觉得我们目前还没有真正靠谱的评估标准。

I don't think we have really good benchmark for this.

这总的来说是个大问题。

And that's a big problem in general.

我最终还是希望能看到更多的透明度。

Ultimately, I hope we'll have more transparency.

你说对吧？

Right?

有个机构叫Casey。

There's this agency called Casey.

是的。

Yeah.

这太棒了，真是关键所在。

That is is a main it's amazing.

我觉得他们做得非常出色，正在逐步建立起评估模型和制定基准的能力。

I think they're doing an amazing job, and they're building up this capability to really evaluate and work on benchmark and things like that.

我特别期待他们能多承担一些工作，用更科学的方法来评估这些模型。

And and I'm really excited for them to take a little bit more of the workload there and and and kind of, like, take a very scientific approach to, to evaluating these these models.

等他们真正上手了，对Shield领域一定会大有裨益。

And I think when when they when they will, it's it's gonna be really good for the Shields.

是的。

Yeah.

我们确实很快就要跟Casey那边的人聊聊。

We definitely wanna talk to people from Casey soon.

对。

Yeah.

这肯定会非常有挑战性。

It's gonna be really tough.

我得说实话。

I can't lie.

我们会试试。

We'll try.

我们会试试。

We'll try.

是的。

Yeah.

对了，昨天我跟DeepMind的安德鲁·特拉斯克聊过，他有个挺有意思的观点，说在Open Router上有个叫Open Fusion的模型，我觉得是这个名字。

Oh, also, yesterday, I was talking to Andrew Trask from DeepMind, and he had, like, a very interesting viewpoint that he like, there's a model on open router called Open Fusion, I think.

这个模型其实是把好几个不同的模型融合在一起。

And it's this like fusion model of basically a bunch of different models.

它在很多方面都展现出很强的能力，效率也高出不少。

And that like had a lot of advanced capabilities and like surpassed inefficiency in a lot of ways.

你觉得以后会看到更多这样的模型吗？

So do you think we're going see more of that?

我觉得是的。

I think so.

对。

Yeah.

对。

Yeah.

我觉得我们现在看到的情况是，很多公司都意识到这太危险了。

I think what we're seeing right now is that a lot of people, companies are realizing that it's too dangerous.

风险太高了。

It's too risky.

完全依赖一个模型毫无道理。

It doesn't make any sense to rely exclusively on one model.

是的。

Mhmm.

为什么？

Why?

因为这个模型随时可能被撤掉。

Because this model can be taken away.

这个模型可能存在偏见。

This model can be biased.

这个模型可能会拒绝回答，或者给你错误的信息。

This model can refuse or tell you the wrong things.

对吧？

Right?

这其实我们以前也见过，比如Sable Five被下架之前就是这样。

That's that's also what we've seen before, right, with with Sable five before before it was taken out.

对吧？

Right?

当时在某些领域，它是故意设计成误导你的，让你困惑。

Is that there was some domain where it was intentionally, by design, kind of, like, telling you the wrong things, right, to confuse you.

没错。

Mhmm.

所以我觉得人们逐渐意识到，我们需要依赖多种不同的模型。

And so I think people are realizing that that we need to rely on a multitude of of models.

是的。

Yeah.

对吧？

Right?

因此，我觉得这正是推动我们更多采用路由机制的原因。

And so I think that's that's driving to to to this to this outcome of doing more routing.

但去年年底，斯坦福大学发布了一项有趣的研究，显示有70%的人向ChatGPT提出的问题，其实完全可以在本地笔记本上准确回答。

But there was an interesting study from, from Stanford published last year, end of last year, that was showing that 70% of the queries that people ask to chat GPT could be accurately answered locally on your laptop.

明白了。

Okay.

也就是说，像这类问题，其实完全免费就能解决。

So for free, like, you know, questions there.

如今大多数人使用前沿模型做的工作，其实用更便宜、更快、更灵活、更可控的模型就能完成。

Most of the questions you ask for, or most of the AI workloads that people do today with frontier models could be done by models that are cheaper, faster, more customizable, more controllable.

对吧？

Right?

是的。

Mhmm.

他们不这么做，说实话，挑模型太麻烦了，得想好用哪个便宜的，但反正你有订阅，费用有人补贴，你也不在乎。

And they don't do it because, frankly, it's a pain to take the picker model picker and be like, okay, this one, I'm gonna go for, like, a cheaper one because, you know and so you're subsidized, so you don't have to care because you have your subscription.

所以你就把所有请求都导向同一个模型。

So you would you you direct everything.

就像把所有问题都丢给爱因斯坦一样。

It's like directing everything to Einstein.

对吧？

Right?

嘿，爱因斯坦。

It's like, hey, Einstein.

爱因斯坦是谁？

What's what's Einstein?

今天天气怎么样？

What's the weather today?

你知道的。

You know?

正常情况下，你早就不耐烦了，直接回他：滚蛋。

In normal life, you would be like, fuck you.

但因为这是AI，而且有实验室补贴，所有问题都一股脑儿扔给爱因斯坦了。其实理想状态下，应该有不同专家模型，各自专精不同领域，才能更精准地回答你的问题。

I'm not answering your silly question, but because it's AI and subsidized by by the labs, all the questions are getting routed to Einstein versus in an ideal world, you can have different people, different models that are more specialized and and better at answering your your questions in different different domains.

所以这其实就是我们现在看到的情况。

So that's kind of like, yeah, what we're seeing.

与其让你自己选模型，我觉得直接智能分流才是更聪明、更好的做法。

And the the way to route instead of giving you the model picker, I think is a very, very smart option and a better one.

Lovable现在也开始这么做了。

Lovable is starting to do that too.

对吧？

Right?

就是在后台自动做路由分配。

Like doing the routing under the hood.

我觉得这可能会改变价值分配格局，现在大部分收益都集中在前沿模型上。

And I think it will it's possible that it's gonna redistribute a lot of the value capture from frontier models, which have been the case now.

对吧？

Right?

以前大部分收入确实都集中在前沿模型上。

Like, majority of the revenue capture was on frontier models Mhmm.

未来会转向更多长尾模型，我觉得这样才更合理。

To a more, like, long tail of of models, which, in my opinion, makes much more sense.

这才真正体现了人工智能的价值。

It's like the it's like AI AI mattering.

对吧？

Right?

我们之前处于AI的初期阶段，大家都用同一个庞大复杂的模型，背后全是封闭的API。

Like, we were in the first phase of AI where it's very simple, very simplistic, everyone using just one giant, intricate model behind proprietary APIs.

现在，AI领域正进入第二阶段，变得更加成熟，人们开始使用多个模型，采用开源方案，自己动手构建系统。

Now we're moving to the second phase of the AI field more maturing and using several models, using open source, having control building themselves.

我对这一点非常期待。

I'm I'm quite excited about it.

对。

Mhmm.

是的。

Yeah.

昨天另一个大新闻是，Anthropic指控阿里巴巴在进行模型蒸馏攻击，这件事其实有两种看法。

So another big news story of yesterday was, Anthropic accused Alibaba of doing distillation attacks, which is you know, there's two perspectives on this.

一种说法是，这些人就是想窃取我们模型的能力，违反服务条款，用不正当手段获取我们的产品。

One is, like, this is, like these are the these evil people who are, like, stealing the capabilities of our models, violating our terms of service, like, fraudulently accessing our product.

另一种说法是：这怎么说呢？

And then the other is like, what do you mean?

他们注册了账号，也正常付费购买了算力。

They're creating accounts and they're paying for tokens.

你要是说别人偷，那你自己不也是把整个互联网都拿走了吗？

And, you know, you can't accuse people of stealing when you stole the entire Internet.

那你更认同这两种观点中的哪一种？

So which one of these two positions are you closer to?

我觉得蒸馏这种做法其实非常普遍，大家都这么用。

Well, I mean, I think distillation is a very common, you know, practice that everyone is using.

我一点都不意外，Anthropic过去可能也用过蒸馏，拿别人的优秀模型来训练自己的专用模型。

You know, I wouldn't be surprised if Entropic used distillation in the past for some of their models, for some of their specialized models using kind of like someone else who's better.

比如当年OpenAI的代码能力更强时，你就会用他们的模型来辅助训练自己的代码模型。

For example, you know, when when OpenAI was better at coding, like, you use this model to kind of, like, help you a little bit in the training of your your your, coding model.

这事儿人人都在用，但可不是决定成败的关键。

It's something that everyone uses, but that is not, you know, the main reason for success.

如果你本身不行，不管用不用蒸馏，照样不行。

Like, if you suck, you suck with or without distillation.

这顶多算个加速器，真正决定模型好坏的，根本不是这个。

It's just kind of like a little bit, like, accelerating thing, but it's it's not it's not what makes you good or bad at training models.

所以如果你明天停止知识蒸馏，中国的实验室也不会就此垮掉或消失，因为他们的模型本身还是很厉害的。

So if you stop distillation tomorrow, the Chinese labs won't, like, go down and disappear because it's still gonna be good.

这根本不会改变大局。

It's not it's not really gonna change the game.

而且说实话，我确实有点偏向这个观点：如果真存在严重的不公平竞争，比如别人明显占了便宜。

And, you know, I mean, the the the only point that I'm I'm a little bit, you know, biased towards that that if if there was big competition problems, right, where it's like, oh, it's really unfair.

那才算是扭曲了竞争。

It's biasing competition.

但我很难接受这个说法，因为坦白讲，Entropic和OpenAI可是全球增长最快的公司。

But it's it's hard for me to accept this one because, frankly, you know, I mean, Entropic OpenAI, they've been the fastest growing companies in the world.

它们一夜之间就成了市值万亿的巨头，我觉得它们根本谈不上什么竞争压力。

They've become overnight trillion dollar companies, and so I don't think they have competition problems.

你懂的吧？

You know?

你要说Entropic或者OpenAI被不公平地打压了？可它们明明是全世界增长最快的企业啊。

It's hard it's hard for me to be to say, like, oh, or Entropic, Pro OpenAI, you're getting unfairly competed with when when you're, like, the fastest growing company in the world.

对他们来说，竞争还算不错。

Competition has been okay for them.

说真的，我觉得他们更需要竞争，而不是更少。

If anything, I think they need more competition than less competition.

是的。

Yeah.

因为我们正走向一个少数几家公司完全垄断、集中所有权力、能力和财富的世界。

Because we're heading toward a world where, like, a few companies are completely dominating, concentrating all power, all capabilities, all wealth.

这种局面比他们可能损失几十亿美元收入要危险得多。

And that that's much more dangerous than them maybe, you know, losing a couple of billion dollars of revenue.

你知道的。

You know?

对。

Yeah.

我觉得是这样。

I think.

完全同意。

Totally.

这挺有意思的。

That's funny.

所以真的很难，就是很难去感同身受，觉得这是个重要问题。

So it's it's just hard, yeah, it's just hard to, you know, empathize and kind of, like, be think that this is this is an important problem.

我觉得现在人工智能领域里，比这重要得多的问题多的是。

I think there are many, many more much more important problems than that in the world of AI right now.

是的。

Yeah.

自从我们上次聊过之后，有两篇关于欧洲的重要文章发表了。

So since the last time we talked, there have been two big pieces written about Europe.

有一篇叫《欧洲2031》，其实就是《AI2027》的欧洲版。

There's this essay Europe 2031, which is basically AI 2027, but for Europe.

简单说，如果欧洲现在不抓紧布局人工智能，迟早会被边缘化。

Like, basically, Europe will slide into irrelevance if they don't lock in on AI right now.

另一篇是政策撰稿人安东·莱希写的，题为《登月计划》，主要讲如果欧洲真想做，其实可以建一个前沿实验室。

And then the other was Anton Leisch, who's a policy writer, wrote this piece on subset called The Moonshot, which basically explained how if Europe wanted to do so, they could build a frontier lab.

你觉得现在欧洲还有可能做到这一点吗？

Do you think that it's possible for Europe to do this at this point?

如果他们真下定决心，能建成一个前沿实验室吗？

Could they build a frontier lab if they really tried?

我觉得可以。

I think so.

是的。

Yeah.

对。

Yeah.

没错。

Yeah.

他们其实拥有非常多优秀的资源。

They they have a lot of really great resources.

他们人才济济。

They they have great people.

为什么？

Why?

你们已经有一些很棒的实验室了。

You've already have some great labs.

对吧？

Right?

比如第一个实验室，Mistral。

Like, first lab, Mistral.

这些人都做得非常出色，而且可以说，他们已经处于前沿水平了。

All these people are are doing doing amazingly, and and, arguably, they're they're at the frontier.

他们的干劲十足。

They have they have amazing energy.

你懂的？

You know?

比如法国，核能非常丰富，这很明显。

Obviously, France, for example, you know, nuclear energy, very abundant.

所以他们完全可以为人工智能大量使用清洁能源。

So they they they could really use a lot of clean energy for AI.

所以我觉得他们确实可以做到。

So yeah, I think they could.

关键是要把精力集中起来，打造一个完整的生态体系。

It's just a matter of focusing the energies towards that, building an ecosystem.

我们有时候总爱讲一些故事，说某些公司凭空崛起，全靠自身力量。

Never, sometimes sometimes we, we build the stories of like companies, emerging out of the blue by their sheer power.

但事实上，这更多是整个生态共同作用的结果。

But the reality is if it's it's more an ecosystem.

对。

Right.

所以这就是为什么我持开放态度。

And then you see that's why I'm open.

没错。

Right.

Transformer里的'T'显然来自谷歌开源的Transformer模型。

The T of transformers, obviously, is coming from from Google that that open source transformers.

在我看来，关键在于推动开放研究和开源AI的生态，就像美国那样，逐步让更多公司和机构靠近技术前沿。

And so it's more of a matter of, in my opinion, fostering an ecosystem of open research, open source AI, which is what happened in The US, right, and kind of fostering that progressively to bring more and more companies, organizations closer closer to the frontier.

完全同意。

Yeah, totally.

我很好奇，既然你运营这么大的平台，年轻人平时都怎么用AI？

I'm curious, like, since you run such a large platform, like, what are younger people doing with AI?

年轻人是不是真的变得特别精通，成了AI原住民？

Are people are younger people actually becoming very, very proficient and like AI native?

你观察到这种行为模式是怎样的？

Or like, how do you see this pattern of behavior?

是的。

Yeah.

是的。

Yeah.

我们经常看到。

We see them a lot.

我试着去理解，感觉年轻人很快就度过了作为AI用户的第一个阶段。

I try, it's almost kind of like, I feel like young people went through the first phase of, of being users of AI really quick.

现在，很多人想成为创造者。

And now a lot of them, I think want to be builders.

所以我们看到很多非常年轻的人进入下一个阶段，自己获取模型，开发产品，优化训练过程，甚至自己构建数据集。

So we, we see a lot of, yeah, very, very young people going on a new phase, getting getting models and and, you know, building building products themselves or optimizing training training models themselves, you know, building datasets themselves.

我看到许多年轻人在各个领域都表现出强烈的AI创造热情。

So I I see a lot of, like, yeah, building appetites in AI for for young people in a lot of different domains.

不一定是那些最受关注的领域，而是一些很少被提及的方向，比如气候变化、生物学、化学，甚至社交媒体——这些其实更贴近普通人的生活，我看到很多年轻人正在投身这些领域。

Not necessarily in kind of like the most talked about domains, but also in a lot of like very, you know, topics that are really not talked about, like, climate, climate change, you know, biology, chemistry, you know, really kind of like, even social media kind of like, like a lot of, a lot of topics that we don't really talk, talk about that I feel like are closer to, to everyone's interest, and I I see a lot of young people working on these things.

这是个不错的收尾。

That's a good white pill to end on.