开源 AI：用平价硬件让机器人技术民主化

所以我认为我们应该害怕的是最终陷入一个只有少数公司能做 AI 的世界。那将是一个非常可怕的 dystopian 世界。开源，即公开分享模型和数据集的能力，是对抗这些自然趋势的一种方式。我们相信可以创造一个世界，不仅少数组织能够构建和主导，而是任何组织——成千上万的小型科技公司、初创企业、非营利组织、政府——都应该能够用 AI 进行构建。今天我有幸与 Clim Dong 对话。他是 Hugging Face 的联合创始人兼 CEO，Hugging Face 基本上就是 AI 界的 GitHub。

So I think something that we should be scared about is to end up in a world where only a few companies are able to do AI. I think that would be a very scary dystopian world. Open source which is basically the ability to share models to share data sets openly is a way to fight these natural tendencies. We believe that you can create a world where not just a few organizations are able to build and dominate but really any organization. Tens of thousands hundreds of thousands of little tech of startups of nonprofits of governments should be able to build with AI. Today I have the pleasure of sitting down with Clim Dong. Uh and he is the co-founder and CEO of Hugging Face, which is basically like AI GitHub pretty much.

我们先从中间那个小机器人是什么开始吧？

Uh let's start off with what is that little robot in the center there?

是的，这是 Richie Mini。这是我们几个月前推出的面向 AI 构建者的开源桌面机器人。已经有超过 5000 人预购，我们开始向全球发货这些小机器人。

Yeah, this is a Richie Mini. This is an open-source desktop robot for AI builders that we introduced a few months ago. Already over 5,000 people pre-ordered it and we are starting to ship these little birdies all over the world.

你们已经开始发货了吗？

Are you already shipping them?

是的，我们开始发货了。我上周刚收到我的。它是完全开源的。大部分部件可以 3D 打印，并且完全可编程，意味着它没有预装应用。作为 AI 构建者，你可以自己构建应用，然后在家使用。例如，很多人和孩子一起用，玩捉迷藏或红绿灯游戏。你还可以与社区分享。所以我们希望全球的 AI 构建者能用最新的 AI 模型和开源模型构建应用，并与世界分享。

Yes, we're starting to ship them. I received mine actually last week. It's fully open source. Most of it is 3D printable and fully programmable, meaning it doesn't come with pre-installed apps. As an AI builder, you can build your own apps and then use them at home. For example, a lot of people are using them with their kids, playing hide and seek or red light, green light. And you can share them with the community. So we hope that AI builders all over the world will build apps with the latest AI models and open-source models and share them with the world.

你们是如何决定将一个小型机器人作为首款发货产品，而不是一个完整的人形机器人或更大的东西？

How did you kind of make the decision to do something a little bit smaller as the first robot that you guys would ship instead of like a full-on humanoid robot or something much bigger?

是的。在 Hugging Face，我们就像是 AI 构建者的平台。我们有 1100 万 AI 构建者使用我们的平台。一年半前，我们开始看到越来越多的 AI 构建者涉足机器人领域。他们遇到的一个挑战是缺乏价格合理的硬件来进行实验。

Yeah. So at Hugging Face, we're kind of like the platform for AI builders, right? We have 11 million AI builders using our platform. And a year and a half ago we started to see more and more AI builders playing with robotics. And one of the challenges they were encountering is the lack of affordable hardware for them to experiment with.

是不是早期入门机器人要 2 万美元甚至更贵？没错。我觉得当 AI 构建者刚开始接触机器人时，他们可能不想花 2 万、3 万、5 万甚至 10 万美元买机器人。所以我们决定制造这种价格合理的机器人。我们售价 400 到 500 美元，非常便宜，这样你很容易决定购买，放在笔记本电脑旁边，开始像修修补补一样用开源 AI 机器人做实验。

Is this the whole situation where like the early entry bots are like $20,000 or much more? Exactly. Uh, you know, um, I feel like when an AI builder is starting with robotics, they might not want to buy a 20, 30, $50,000, $100,000 robots. Um, and so that's that's why we decided to build this affordable robots. We sell it for $400 to $500, right? So it's very cheap so that it's easy for you to take a decision to buy it, put it next to your laptop and start experimenting with open-source AI robotics like tinkering.

没错。

Exactly. Yeah.

你认为修修补补在开发这类东西中扮演什么角色？

What role do you kind of see tinkering playing and how these types of things are developed?

我认为这是最重要的事情。我认为这是目前在 AI 领域你能做的最有影响力的事情之一，因为 AI 有很强的能力集中自然趋势。我认为未来很有可能只有少数公司能做 AI，而我们其他人只能成为 AI 用户，而不是真正的 AI 构建者。我认为这是我们应该担心并努力改变的。所以我们在 Hugging Face 构建的一切，都在思考：如何帮助更多人成为 AI 构建者、修修补补者、学习和实验？这也是我们对 Richie Mini 以及开源感到兴奋的原因之一。

I think it's the most important thing. I think it's one of the most impactful things that you can help with in AI right now because AI has very strong natural tendencies of concentration of capabilities. I think there's a strong probability of a future where only a few companies are able to do AI and the rest of us are kind of doomed just to be AI users, not really AI builders. And I think that's something we should be worried about and that we should try to change. So everything that we're building at Hugging Face, we're always thinking, okay, how can it help make more people AI builders, tinkerers themselves, learn, experiment? And so that's one of the reasons why we are excited by Richie Mini and in general by open source of course.

你认为这类东西会如何发展？你们会先构建第一个版本，然后交到很多人手中吗？你说已经有大约 5000 人购买了。你们期望从人们使用和解决小问题中得到什么反馈？

How do you think something like that is going to evolve? Are you gonna basically build the first version, get it into a bunch of people's hands? You said, you know, 5,000 or so people have already bought it. Um, what are you looking to get back even as far as people just using it and working out the kinks?

首先，我们对人们构建应用感到非常兴奋。我们觉得如果成千上万的 AI 构建者开始用这个机器人构建，将会出现能力的爆发。尤其是在 AI 领域，每天都有新模型、新能力。我们希望最终所有这些能力几乎能瞬间转化到机器人领域。我之前提到，下次 Gemini 新模型、GPT 新模型、Anthropic 新模型发布时，人们应该能立即构建他们的机器人应用。哦，Gemini 3 发布了。看我的机器人有了新能力。现在它能读我的书，能识别这些物体，能教我的孩子以前教不了的新话题。所以我真的很期待看到人们不仅在他们收到机器人时，而且随着领域发展，会构建出什么。

So, first we're super excited about people building apps, right? We feel like if thousands of AI builders are starting to build with this robot, there's going to be an explosion of capabilities. Especially in the AI field where there are new models, new capabilities every day. We hope that ultimately all these capabilities will translate almost instantly to robotics. Right? I was mentioning before next time the next Gemini model is out, the next GPT model is out, the next Anthropic model is out, people should be able to build their robotics apps right away. Oh, Gemini 3 is out. Look at the new capabilities of my robot. Now we can read my book. Now we can recognize these objects. Now we can teach my kids about this new topic that it wasn't able to teach before. So I'm really excited to see what people are going to build not just when they receive their robot but also as the field evolves.

这是软件和平台方面。在硬件方面，因为 Richie Mini 的目标是开源，我们希望人们不只是组装自己的机器人。当他们收到时，机器人是未组装的，所以他们必须自己组装。

So that's on the software side on the platform side. And then on the hardware side, because the goal of Richie Mini is to be open source, we hope that people are not just going to assemble their own robots. So when they're going to receive it, it's not going to be assembled. So they have to assemble it.

有点像宜家乐高套装那种情况。

A little bit like the IKEA Lego set sort of situation.

没错。对我来说，我收到 Richie Mini 后花了五个小时才组装好。我们期望人们不仅自己构建、组装，还能自己改进。所以如果发货后我们看到有人给 Richie Mini 装上轮子，或者添加夹爪或手，或者改进电机，我一点也不会惊讶。所以我们希望看到社区驱动的酷炫发展，不是由我们驱动，而是由社区驱动，无论是软件还是硬件方面。

Exactly. For me, it took me five hours to actually build my Richie Mini when I received it. And what we are expecting is people not only to build it themselves, assemble it themselves, but also to improve it themselves. So I wouldn't be surprised if we start seeing when we're shipping it people who are putting Richie Mini on wheels, or people who are adding grippers or hands to Richie Mini, people who are improving the motors. So yeah, we hope to see really cool developments driven by the community, not so much driven by us, but driven by the community, both on the software side and on the hardware side.

Hugging Face 最初并不是像 AI GitHub 这样的平台。一年半前我肯定没想到你们会制造机器人或试图发货一批机器人。是什么指导你们做出公司发展方向的决定？

Hugging Face did not start off as kind of like this AI GitHub platform. Um, and it's I certainly would not have expected a year and a half ago that you guys were building robots or trying to ship a bunch of robots. um what kind of like guides you when you're making decisions on where to take the company?

我们真的是由社区驱动的。在我们觉得能产生影响、释放社区力量的地方。这实际上就是我们如何成为 AI 界的 GitHub 的原因。

Um we're really kind of like driven by by the community, right? Where we feel like we can have an impact, unleash some community power to to the world. And that's actually how how we became kind of like the GitHub for for AI.

我们刚创办公司时，做的是类似电子宠物 AI 的东西，一种对话式 AI，就像 ChatGPT，但在 ChatGPT 之前。那大概是 2016 到 2017 年。后来有一天，我清楚地记得是个周五，我们的联合创始人 Thomas Wolf 告诉我们，谷歌发布了一个叫 BERT 的模型，这是最早流行的 Transformer 模型之一，但它是用 TensorFlow 写的。而当时社区大部分人都用 PyTorch，所以他认为把 BERT 从 TensorFlow 移植到 PyTorch 会对社区很有用。Julia 和我同意试试，看看社区是否感兴趣。Thomas 整个周末都在工作，周一他发推文发布了 PyTorch 版的预训练 BERT。我们在推特上获得了一千多个赞，这对我们来说简直像炸了互联网。但说实话，这对社区确实有用。接下来几周，研究人员开始把他们自己的模型添加到我们创建的项目里。当时，GPT 团队（第一个开源版本）以及现在 Mistral 的人正在做一个叫 ExcelNet 的模型。还有很多其他研究者把模型加到我们的仓库里，人们开始把它当作模型来源。这就是我们今天——AI 构建者平台——的早期雏形。从一开始，我们就由社区驱动。不只是驱动，是社区造就了我们，推动了我们的成功。回想当时，我们只是三个没什么人脉或特殊资源的法国普通人。真的是社区成就了我们，放大了我们发起的一些倡议。我们非常感激，所以每当我们考虑做任何事，总是想着如何能让社区受益，因为正是社区造就了我们。

When we started the company, we were doing something like a Tamagotchi AI, a conversational AI, like ChatGPT but before ChatGPT. This was around 2016-2017. Then one day, I vividly remember it was a Friday, one of our co-founders, Thomas Wolf, told us that Google released a model called BERT, one of the first popular Transformer models, but it was in TensorFlow. Most of the community was using PyTorch, so he thought it would be useful to port BERT from TensorFlow to PyTorch. Julia and I agreed to do it to see if the community was interested. Thomas worked all weekend, and on Monday he tweeted about the release of PyTorch pre-trained BERT. We got about a thousand likes on Twitter, which felt huge for us. But the truth is, it was useful to the community. Then in the following weeks, researchers started adding their models to what we created. At that time, the team behind GPT (the very first open-source version), and the people now at Mistral were working on a model called ExcelNet. Many others added their models to our repository, and people started using it as a source for models. That was the early stage of what we are today: a platform for AI builders. From the start, we've been driven by the community. Not just driven, but the community made us what we are and propelled our success. If you look at us back then, we were three random French guys without much network or special access. It's really the community that made us who we are and multiplied our initiatives. We're very grateful, and that's why when we think about anything we do, we always consider how it can benefit the community, because that's what made us.

在哪些最重要的时刻，你放弃了赚钱或早期变现的机会，以建立一种帮助构建者的文化？

What have been the biggest points where you forgo making money or early monetization opportunities in order to build a culture of helping builders?

Hugging Face 平台 99%是免费的。我们试图建立一种促进这种文化的模式。谈到做好事，我倾向于不信任人，而是信任系统和激励机制。对我们来说，我们一直在思考如何建立一个公司或系统，其激励机制能促使我们做好事。我们构建 Hugging Face 的方式是，如果我们让开源作为平台更受欢迎，我们作为公司和创始人就会做得更好。这使我们的激励机制保持一致，让我们继续做好事。很多创始人或公司犯的错误是，他们可能想做好事，但最终建立的系统并不奖励做好事，有时甚至几乎是在对抗人性。例如，有些公司构建 AI 的 API 并以此盈利，然后陷入一种竞赛：API 赚的钱越多，他们就越能投资训练。这种竞赛最终让你很难不屈服于赚更多钱的动机，从而更专注于 API、闭源和不分享。所以对我们来说，始终重要的是创建一个激励机制一致的系统，这样我们就能继续专注于我们的使命：让 AI 更开源、更普及，让尽可能多的人成为 AI 构建者，同时也能成功。

The Hugging Face platform is 99% free. We try to build a model that fosters that. When thinking about doing good in the world, I tend not to trust people too much, but instead trust systems and incentives. For us, we always thought about how to build a company or system with incentives for us to do good. We built Hugging Face so that if we make open source more popular as a platform, we will do well as a company and as founders. This aligns incentives for us to keep doing well. Many founders or companies make the mistake of wanting to do good but building systems where doing good isn't rewarded, or it's almost a fight against human nature. For example, some companies build APIs for AI that they monetize, and they get into a race where the more money they make from APIs, the more they invest in training. That race makes it very difficult not to give in to the motivation to make more money and focus more on closed source and not sharing. So for us, it's always been important to create a system with aligned incentives so we can continue to double down on our mission: making AI more open source, available to everyone, turning as many people as possible into AI builders, while also being successful.

这几乎像是在对抗人性。

It's like betting against human nature almost.

是的。早期你们有没有做过什么不寻常的事，来找到文化契合、使命一致的员工？

Yeah. Did you do anything in the early days that was unusual to find employees who were culturally and mission aligned?

Hugging Face 的结构有点不寻常，因为我们分布在全球各地，很多职能也是分散在全公司的。我们几乎没有专门的人才团队、HR 团队或社区管理团队。我们试图让招聘、社交媒体沟通和社区互动成为每个人的责任。例如，我们的社交账号没有专门的社区经理；团队里的任何人都可以从 Hugging Face 的推特账号发推。这分散了责任，让每个人都明白与社区互动是他们的工作。招聘也一样：我们不想让 HR 或人才团队负责，而是希望每个团队成员都去思考世界上有哪些优秀的人他们愿意一起工作，然后直接联系他们。这对我们来说非常有效。很多大科技公司把人专业化，把他们放进盒子里：'你是软件工程师，你写代码；你是营销人员，你做营销；你是公关人员，你做公关。'这就像只通过一个特定的镜头看他们自己的事，而不是从更广阔的公司视角来看。

We have an unusual structure at Hugging Face because we're distributed all over the world, and many functions are distributed across the company. We have very few dedicated talent teams, HR teams, or community management teams. We try to make it everyone's responsibility to hire, communicate on social media, and interact with the community. For example, our social accounts don't have a dedicated community manager; anyone from the team can tweet from the Hugging Face Twitter account. This distributes responsibility and shows everyone it's their job to interact with the community. It's the same for hiring: we don't want HR or talent teams in charge; we want every team member to think about great people in the world they'd love to work with and reach out to them directly. That's worked really well for us. Many big tech companies specialize people and put them in boxes: 'You're a software engineer, you write code; you're a marketer, you market; you're a PR person, you do PR.' That's like looking through a specific lens on just their thing, not a broader view of the company.

是的。而且他们强迫你只做那件事，在我看来这很无聊，也不能帮助你成长，不仅是作为员工，更是作为一个人。所以我们采取了不同的方法，我们相信每个人都能做技术工作，每个人都能沟通，每个人都能招聘。我们更倾向于招聘通才，并帮助他们完成所有这些工作。我认为这对我们来说非常好，因为它也为每个工作领域带来了新的视角。当一个工程师试图自己招聘而不是通过人才团队时，他们会带来不同的想法，不同的吸引人的方式，不同的筛选方式，不同的招聘方式。反过来也是如此。当一个编码经验较少、工程经验较少的人开始思考如何构建产品时，他们会提出不同的想法和视角。我认为这对每个人作为人的发展也非常好，让他们接触到更多事物，在职能、国家、使命方面拥有多样化的体验。所以我们非常喜欢这种方法，这也是我们未来想要继续做的事情。

Yeah. And they kind of force you to only do that, which in my opinion is quite boring and doesn't really help you grow, not only as a worker but as a human being. So for us, we're taking a different approach where we believe everyone is able to do technical work, everyone is able to communicate, everyone is able to hire. We rather try to hire generalists and help them do all of that. I think it's been really good to us because it also brings new perspectives to each line of work. When you have an engineer who is trying to hire themselves instead of a talent team, they come with different ideas, different ways to attract people, different ways to filter people, different ways to hire people. And the other way around is also true. When you have someone with less coding experience, less engineering experience, starting to think about how to build a product, they come up with different ideas and perspectives. I think it's also been really good to develop everyone at face as human beings, to be exposed to more things, to have a variety of experiences in terms of functions, countries, missions. So we really like this approach and that's something we want to keep doing in the future.

这种将决策权下放给一线战士的模式带来了什么样的意外收获？

What kind of serendipity has come from that model where you're outsourcing decision-making to the frontline warrior?

是的。我提到了 Thomas 的创始故事，他说，好吧，这与我们迄今为止所做的完全不同，我想把它移植到 PyTorch。在一个正常的组织中，人们可能会说，你在说什么？这根本不是我们在做的事情。

Yeah. Well, I mentioned the founding story of Thomas, where he said, okay, there's something completely different from what we've been doing so far with this birth release and I want to port it to PyTorch. In a normal organization, people might have said, what are you talking about? It's not at all what we're doing.

会被太快否决。

Been shut down too quickly.

是的。不是因为人们不好，而是你太专注于某件事，以至于过度限制干扰。但我们的文化允许我们说，如果你觉得它令人兴奋，如果你觉得它对社区有用，如果它会带来影响，那就去做吧，我们之后再来看它是否有效。而它确实有效。所以我认为很多这样的举措都得益于这种文化。Richie Mini 也是一个很好的例子。一个拥有正常文化的正常公司可能不会这么做。

Yeah. And not because people are bad, but you become so focused on something that you try to overindex on limiting distractions. But our culture allowed us to just say, if you think it's exciting, if you think it's going to be useful for the community, if it's going to be impactful, just go do it and we'll see later if it works. And it did work. So I think a lot of these initiatives are made possible thanks to this culture. Richie Mini is a good example too. A normal company with a normal culture probably wouldn't have done it.

嗯，但对我们来说，社区对此很兴奋。我们开始看到 AI 构建者，我们有团队成员对这个话题很兴奋。所以，就实验一下，构建它，看看是否有效。而它确实有效。

Um, but for us, you know, it's just kind of like the community is excited about it. We're starting to see AI builders and we have team members who are excited about the topic. So, just experiment with it, build it and see if it works. And it does work.

有点像宜家的设计，你会对家具有更强的依恋感，因为你去挑选它，然后被迫花时间拧上每个螺丝，把它组装起来。你认为这会在某人第一次打开盒子并考虑购买 Richie Mini 时创造出正确的心态吗？

A little bit like the IKEA design where you feel more attached to the furniture because you go choose it out and then you are forced to actually spend time screwing in each bolt and putting it together. Do you think that that kind of creates the right sort of mentality when someone is first opening the box and thinking about getting the Richie Mini in the first place?

绝对。我认为有一种建造的乐趣。

Absolutely. I think there's a certain joy of building.

是的。人们并不总是意识到这一点，但真正按照自己想要的方式建造东西，积极参与世界并共同创造，会给人们带来很多快乐。作为创始人，我确实如此。建造是我经历的最大乐趣之一。所以当你给他们一个工具、一个平台、一些可以建造的东西时，他们非常享受。所以当我们考虑 Richie Mini 时，我们思考了很多，不仅仅是让人们使用它，使用它很好。但我们如何利用它让人们建造并体验这种建造的乐趣。所以这是第一件事。它给人们带来快乐，然后它创造了一种紧迫感，让他们按照自己想要的方式编程。

Yeah. People don't always realize it, but really building things yourself the way you want them to be built, being active in the world and co-creating something brings a lot of joy to people. It certainly does to me as a founder. Building is one of the greatest joys that I experience. So when you give them a tool, a platform, something to build, they enjoy it very much. So when we think about Richie Mini, we think a lot about how not just enabling people to use it, using it is great. But how we can use that as a way for people to build and experience this joy of building. So that's the first thing. It creates joy for people and then it creates urgency for them to program it the way they want to program it.

我认为我们的一些多元化背景以及我们来自世界各地的事实告诉我们，人们并不希望到处都一样。我们一直觉得，如果只有硅谷的少数人决定建造什么，那将是危险的。

I think some of our diverse backgrounds and the fact that we're coming from all over the world taught us that people don't want the same thing everywhere. And we've always felt that it would be dangerous if only a few people in Silicon Valley would decide what to build.

有点像决定未来的方向。

Dictating the direction of the future sort of thing.

完全正确。所以我认为通过让人们成为更多的建造者，你实际上给了他们自主权，让他们在选择建造什么方面有发言权。我认为这对社会、对人类来说极其重要，当你考虑 AI 的未来时，人们应该决定 AI 将用于什么。也许有些人希望 AI 用于聊天机器人，也许用于社交媒体视频创作。但我也认为很多人会想要创造不同的东西，改善他们生活的东西，对教育、科学有用的东西。他们会想用自己的方式建造它。

Exactly. So I think by making people more builders, you actually give them agency and you give them a choice in deciding what to build. I think that's tremendously important as a society for humanity that when you think about the future of AI, people get to decide what AI is going to be used for. Maybe some people will want AI to be built for chatbots, maybe for social media video creations. But I think also a lot of people will want to create different things, things that improve their life, things that are useful for education, for science. And they'll want to build it in their own way.

就像我家里现在有 Richie Mini 可以拍摄。所以对我来说，例如，我希望 Richie Mini 本地运行。我希望 Richie Mini 收集的数据留在我的笔记本电脑和 Richie Mini 上，因为我不想把这些信息发送给任何其他公司。这是我的房子。这是我的私人空间。所以我认为赋予人们建造的能力将允许他们决定这项技术如何构建，并在构建方式上创造一些粒度 and 多样性。你会有一些选择，有时 AI 会本地运行，留在你的硬件上，留在你的笔记本电脑上。有时你会通过 API。有时你会调用一些 LLM 提供商，如 OpenAI、Anthropic。但至少当你自己参与建造时，你可以选择，最终得到你感觉最舒适的解决方案，而不是被迫接受被锁定在某个生态系统中。

Like we have Richie Mini in my house right now being able to film. So for me, for example, I want Richie Mini to run locally. I want the data that is collected by Richie Mini to stay on my laptop and on Richie Mini because I don't want to send this information to any other company. It's my house. It's my private space. So I think giving the ability for people to build will allow them to decide how this technology is built and create some granularity and diversity on the way it's built. You're going to have some options where sometimes AI is going to run locally, stay on your hardware, stay on your laptop. Sometimes you're going to go through an API. Sometimes you're going to call some LLM providers like OpenAI, Anthropic. But at least when you participate in the building yourself, you can choose and you can end up with a solution that you feel most comfortable about instead of having to accept being locked into some ecosystem.

完全正确。

Exactly.

所以我知道史蒂夫·乔布斯有句话，他说专注就是说不。

So I know Steve Jobs has this line where he's like focus is saying no.

你主动选择不走哪些路？

What paths could you have gone down where you actively chose not to go down them?

嗯，比如，随着专有 API 大语言模型的出现，我们确实有走那条路的可能。我们有一个很棒的科研团队，在训练优秀的小型开源模型，我们本可以说‘好了，现在我们要构建更大的专有模型’，因为很多人都在这么做。但我们决定不做，因为正如我之前提到的，我们认为这不会为我们创造正确的激励机制。一旦你开始将模型商业化，做更多开源的动力就会变得模糊。社区也一直希望我们提供更多算力解决方案，更多自己的云服务，但我们还没怎么做。我们决定与云提供商合作，比如 AWS、Azure、Google Cloud，以及 Hugging Face 平台上的推理提供商，如 Fireworks、Together 和许多初创公司。我们觉得有些人做得很好，所以我们不会为了竞争而竞争。相反，我们专注于社区尚未处理的事情，而不是重复造轮子。在算力方面，我们主要与人合作，而不是构建自己的解决方案。

Well, there's been, for example, as proprietary API LLMs emerged, there was the possibility of going that direction. We have a very good science team training good smaller open-source models, and we could have said, 'Okay, now we're going to build larger proprietary models' because a lot of people were starting to do that. We decided not to do that because, as I mentioned before, we felt it wouldn't create the right incentives for us. If you start to commercialize models, the incentives to do more open source become blurry. There's also been interest from our community for us to do more compute solutions, more cloud ourselves, which we haven't done much yet. We decided to partner with cloud providers like AWS, Azure, Google Cloud, and inference providers on the Hugging Face platform like Fireworks, Together, and many startups. We felt that some people were doing it well, so we don't try to compete just to compete. Instead, we focus on things not handled by the community yet, rather than reinventing the wheel. For compute, we've mostly partnered with people instead of building our own solutions.

我知道在商业中有一些方面你会决定那不是我们的专长。我们想要不同的激励机制。我们要走不同的路。也有像谷歌和 Facebook 这样的公司试图与你竞争，但他们在自己的组织中有不同的激励结构，无法真正构建你创造的那种环境。围绕这个基础设施建立自己的社区，感觉如何？

I know there are aspects of a business where you decide that's not our area of expertise. We want different incentives. We're going to go down a different path. There have also been companies like Google and Facebook that tried to compete with you, yet they have different incentive structures in their own organizations and couldn't really build the same environment you've created. What has that been like for building your own community around this infrastructure?

这非常有趣。我们看到了 AI 的两个阶段。早期阶段，AI 非常小众——当我开始研究 AI 时，我们甚至不叫它 AI；我们在谈论计算机视觉、聊天机器人，非常小众。第二阶段是 AI 变得主流，每个人都在用 AI 构建。竞争一直非常激烈，有很多参与者，而 Hugging Face 是核心社区、核心平台。一直有很多人试图复制我们正在做的事情。想想模型动物园或模型仓库，大概有 50 家初创公司和大型公司试图在他们的产品中做这件事——从谷歌的 TensorFlow Hub，到 Meta、亚马逊、微软，再到每个小型初创公司，都曾试图复制 Hugging Face 的成就。我们构建的东西的美妙之处在于它非常独立和中立。我们不会推广特定的模型、数据集或提供商。我们总是试图提供你能想到的任何模型，让你选择。现在，Hugging Face 上有超过 600 万个模型、数据集和应用。每 8 秒就有一个新的仓库被创建。贡献量惊人。我们已经达到了一个规模，去其他平台没有意义，因为一天之后，另一个平台在模型数量上就会远远落后——每天都有数千个新模型、新数据集、新应用出现在 Hugging Face 上。我们达到了一个类似 GitHub 的平台规模。我们有很强的网络效应和很深的护城河，其他人很难完全复制我们所做的事情，这是一个很好的位置。

It's been super fun. We've seen two phases for AI. The early phase where AI was really niche—when I started working on AI, we weren't even calling it AI; we were talking about computer vision, chatbots, and it was quite niche. The second phase is when AI became much more mainstream, everyone building with AI. It's always been very competitive with a lot of players, and Hugging Face being the central community, the central platform. We've always had a lot of people trying to replicate what we've been doing. If you think about model zoos or model repositories, there have been maybe 50 startups and big companies trying to do that in their products—from Google with TensorFlow Hub, Meta, Amazon, Microsoft, to every smaller startup at some point trying to replicate what Hugging Face has. The beauty of what we're building is that it's very independent and neutral. We're not going to push a specific model, dataset, or provider. We always try to provide any model you can think of and let you choose. Right now, there are over 6 million models, datasets, and apps on Hugging Face. A new repository is created every 8 seconds. The volume of contributions is insane. We've reached a scale where it doesn't make sense to go to any other platform because after one day, another platform would be so behind in terms of numbers of models—thousands of new models, datasets, and apps appear on Hugging Face daily. We've reached a scale for a platform similar to what we've seen with GitHub. We have quite strong network effects and a strong moat, making it really hard for someone else to do exactly what we do, which is a nice position to be in.

对于这类公司，这让我想起了 Reddit，一开始 Alexis Ohanian 和 Steve Huffman 创建假账户来模拟活动。对于社区驱动的企业，这需要很长时间。在你们公司的头六年，收入并不大，只是缓慢进展。然后在某个时刻，你构建了一些不容易复制的东西。这需要很长时间来构建。在 ChatGPT 诞生之前的头五六年，感觉如何？

For companies like this, it reminds me of Reddit where at the beginning, Alexis Ohanian and Steve Huffman were creating fake accounts to simulate activity. With community-driven businesses, it takes a very long time. For the first six years of your company, there wasn't a huge amount of revenue, just a slow progression. Then at some point, you build something not easily replicable. It just takes a long time to build. What was it like for the first five or six years before the creation of ChatGPT?

是的，你说得完全正确。这是一种不同的业务，一种不同的产品，你需要打好基础。你要让这些基础非常稳固。你要在相当长的时间里培育网络效应。同时，你也要在相当长的时间里与社区建立信任。我觉得如果我们只做了六个月，甚至两年，人们不会像现在这样信任我们。

Yeah, you're totally right. It's a different kind of business, a different kind of product where you want to create the foundations. You want to make these foundations very stable. You want to foster the network effects for quite a long time. And also, you want to build trust with the community for quite a long time. I feel like if we were doing what we're doing for just six months, or even two years, people wouldn't trust us as much as they do now.

我觉得时间能建立与社区更强的信任。如果他们看到你多年来一直这样做，没有试图利用他们，没有搞‘拉地毯’骗局，也没有倒闭，那时他们才开始信任平台，愿意投入更多时间和资源，分享更多模型、数据集和他们在平台上构建的应用。所以我认为这非常重要，是这类产品和公司无法伪造或加速的特点之一。如果你作为创始人正在构建这类公司，你必须接受这一点。我们之前谈到激励对齐：你必须按照自己的价值观和热爱来构建这样的平台，这样你才能在多年的构建中保持热情，不被他人可能更快的成功分心。在 AI 领域，现在很容易被 Cursor 这样的公司分心，你看到他们，心想：‘天哪，他们的收入增长这么快，从零到十亿美元只用了一年半。’你可能会想：‘我要停下一切，去做那个。’但如果你相信自己在做的事，享受它，觉得它有意义，你就可以忽略这些，不去追逐那些东西，它们也有自己的约束和挑战。我见过很多人，有的成功，有的不成功，有的像坐火箭，有的完全失败，有的增长更线性。让我印象深刻的是，他们的幸福和享受程度并不一定与他们所处的境况相关。通常更相关的是他们所做之事与自己的价值观、热情和人生使命的一致程度。对我们来说，因为我们如此热爱开源，真正希望让任何人都能成为 AI 构建者，我觉得如果花一年、十年、五十年，都不那么重要，因为我们真的很享受构建的过程。我们觉得自己在成长，在给世界带来积极影响。所以，无论目的地在哪里，我们都很享受这段旅程。

I feel like time builds stronger trust with the community. If they've seen that you've been doing that for a few years and you didn't try to take advantage of them, you didn't try to do a rug pull, you didn't go out of business, I think that's when they start trusting the platform enough to really invest more of their time and resources on the platform and share more models, more datasets, more applications they're building on the platform. So I think it's really important and it's one of the characteristics of these kinds of products and companies that you can't fake or accelerate. If you're building these kinds of companies as founders, you have to make peace with it. And we were talking about alignment of incentives: you have to build such a platform in alignment with your values and what you're excited about, so that you can stay excited during these years of building and not get distracted by other people's maybe faster success. It's really easy in AI right now to get distracted by another company like Cursor, where you look at them and think, 'Oh my god, they grew revenue so fast from zero to a billion dollars in a year and a half.' You can be like, 'I'm going to stop everything I'm doing and do that.' But if you believe in what you're doing, if you enjoy what you're doing, if you feel like what you're doing matters, you can ignore that and not chase these things, which come with their own sets of constraints and challenges. I've met a lot of people, some successful, some not, some in this rocket ship, some completely failing, some with more linear growth. What's always striking to me is that their level of happiness and enjoyment is not necessarily correlated to the scenario they're in. Usually it's more correlated to how aligned what they're doing is with their values, their excitement, and their life mission. For us, because we're so excited about open source, about really enabling anyone to become AI builders, I feel like if it took a year, if it took 10 years, if it took 50 years, it doesn't really matter as much because we're really enjoying what we're building. We feel like we're growing. We feel like we're having a positive impact in the world. So, really enjoying the journey no matter the destination.

我喜欢查理·芒格的一句话：‘信任是世界上最强大的经济力量。’你是如何得出信任是应该优化的正确目标的结论的？

I love this line from Charlie Munger where he says, 'Trust is the most powerful economic force in the world.' How did you come to the conclusion that trust was the right thing to optimize for?

嗯，首先，我们非常感激和幸运。我认为我们可能是世界上最受 AI 构建者喜爱的公司之一。你可以去和任何 AI 构建者聊聊，很少有人不喜欢 Hugging Face，这对我们来说非常重要，也是一种难以置信的认可。我认为这都归结于我一开始分享的：我们觉得自己不过是普通人，没有特别的权利去赢或成功。正是社区的信任成就了今天的我们。所以，专注于这一点，不断滋养这种信任，努力让社区自豪，让他们开心，让我们对他们产生影响力并有所帮助，这对我们来说非常自然。我们并没有想太多。我觉得这更像我们思考问题的一种自然方式。

Well, first, we're really grateful and lucky. I think we're probably one of the companies in the world that AI builders love the most. You can go talk to any AI builders; there will be very few who don't love Hugging Face, which is something really important to us and an incredible validation. I think it all comes down to what I was sharing in the beginning: our feeling that frankly we're kind of random people with no special rights to win or be successful. And that's really the trust of the community that made us who we are. So it feels very natural for us to focus on that and keep fueling this trust, try to make the community proud, and try to make them happy and for us to be impactful and useful to them. We don't think about it too much. I feel like it's more a natural part of how we think about things.

我觉得基本上每一位伟大的创始人都有一些榜样。我喜欢约翰·D·洛克菲勒和 19、20 世纪的强盗大亨。你在建立公司时，谁是你的榜样？

I think basically every great founder looks up to some role models. I love John D. Rockefeller and the robber barons of the 19th and 20th century. Who's kind of like the role model for you when you're building your company?

好问题。我们受到很多产品的启发。当然，GitHub，我们经常谈论 GitHub。我们认为 GitHub 对世界的影响被低估了。想想现在有 1.5 亿开发者在用 GitHub，以及它释放的开源力量，真的很了不起。所以这类事情对我们来说是很大的激励。至于我个人，我非常喜欢哲学，所以有很多哲学家的启发。

It's a good question. We've been inspired by a lot of products. Of course, GitHub, we talk a lot about GitHub. We feel like the impact in the world of GitHub is really underappreciated. If you think of the number of developers, I think it's 150 million developers now using GitHub, and the power of open source that it unleashed in the world. It's really amazing. So those kinds of things are really an inspiration for us. Then when it comes to me, I'm a big fan of philosophy, so I have a lot of philosopher inspirations.

你最喜欢谁？

Who are your favorites?

我最喜欢的书是阿尔贝·加缪的《西西弗神话》，他是法国哲学家。显然，我是法国人，所以有偏见，但我觉得里面有很多适用于创业的教训。西西弗神话讲的是西西弗被诅咒要把一块巨石推上山。每次石头到达山顶，就会滚下来，他必须重新开始，永无止境。在流行文化中，这被定义为一种诅咒，很糟糕的事。但书的结论和最后一句话是：‘我们必须想象西西弗是幸福的。’尽管任务艰巨、重复、看似毫无意义，西西弗是幸福的。原因在于，他的目标和幸福并不来自于到达山顶或停留在山顶，也不来自于拥有一个更高的神。加缪的荒诞主义哲学是，你并不真正需要神或天堂或某种更高的力量，因为任务本身——推动自己，推动石头——我想象，‘为什么要推石头？哦，有趣，看这座山，真美，也许有日落，也许有日出’——仅仅任务本身就能让他快乐。对创业者来说也是如此。当然，你有成功的动力，有到达那里的动力，有建立伟大公司的动力，但也有构建公司的乐趣：招聘优秀的团队成员，尝试新项目，发布产品。我们之前谈到像 Richie Mini 这样的构建乐趣。

My favorite book is 'The Myth of Sisyphus' by Albert Camus, a French philosopher. Obviously, I'm French, so I'm biased, but I feel like there are a lot of lessons applicable to entrepreneurship. The myth of Sisyphus is about this guy Sisyphus who was cursed to push a rock up a mountain. Each time the rock reaches the top, it falls down, and he has to start all over again for eternity. In popular culture, it's been defined as a curse, something really bad. But the conclusion of the book and the last sentence is: 'You have to picture Sisyphus happy.' Despite the hard, repetitive, seemingly meaningless task, Sisyphus is happy. The reason is that his goal and happiness don't come so much from reaching the top or staying at the top, or from having a higher god. The whole philosophy of absurdism from Camus is that you don't really need a god or heaven or some higher power, because the task itself of pushing yourself, the task itself of pushing the rock—I imagine, 'Why is pushing the rock? Oh, it's funny, look at this mountain, it's beautiful, maybe there's a sunset, maybe there's a sunrise'—just the task itself can be joyful for him. It's the same thing for entrepreneurs. Of course, you have the motivation of success, of getting there, of building a great company, but also the joy of building your company: hiring great team members, experimenting with new projects, releasing things. We were talking about the joy of building like Richie Mini.

我深受启发。我认为这不仅是幸福的配方，也是播客名字“Relentless”的体现。如果你真正将幸福与奋斗、拼搏、建设本身而非结果对齐，你会变得更有韧性，因为让你放弃会变得更难。

I'm really inspired by that. I think that's one potential recipe not only for happiness but also for the name of the podcast is obviously Relentless, right? And I think to be relentless, you can challenge, you can channel some of that. Like if you really align your happiness with the grinds, with the hustle, with the building, more than the outcome, I think you'll become much more relentless because it's going to be harder to make you give up.

我对西西弗斯的理解是：他把石头推上山，石头滚下来，每次再推上去时，如果做对了，石头会稍微更属于你。你可以决定推什么样的石头上山。人生就是不断推石头上山的旅程。

The way I think about Sisyphus is, he pushes the rock up the hill and then it rolls back down. Every time he pushes it back up, it's just a little bit if you do it right. It's basically a little bit more your rock. You get to decide what kind of rock you're pushing up the hill. Life is just the endless journey of pushing the rock up the hill.

当然。

Of course.

如果做对了，你会越来越享受每一天。

And if you do it right, you enjoy every day more and more.

是的。因为你获得了更多自主权、更多自由、更多实现。你越来越把你正在做的事、正在构建的东西与你的本质或你对自己的认知对齐。于是你滋生出这种喜悦和幸福。

Yeah. Because you get more agency, more freedom, more realization. You align more and more what you're doing, what you're building with your essence or who you feel you are. So you grow this joy, this happiness.

有一位哲学家对此做了变体，他叫克莱蒙·罗塞，一位法国哲学家。他的理论是，快乐来自于世界的无意义与你能在其中找到的享受之间的差距。比如你看美丽的日落、你造的机器人或你创立的公司，也许它们是虚荣或无意义的，没有更高的理由或目的，但不知为何你看着它们感到快乐。他的理论是，驱动你的主要力量就是这种无意义或虚荣与你能在其中找到的快乐之间的差距。而如果你做的是极其有意义的事，你也会快乐，但你的理性头脑理解原因，所以它不会在力量上那么驱动你。这是一个非常有趣的理论和哲学家。

There's one philosopher that did a variation of that. His name is Clément Rosset, a French philosopher. His theory that I find interesting is that joy comes from the gap between the meaninglessness of the world and the enjoyment you can find in it. For example, you look at a beautiful sunset, a robot you built, or a company you built. Maybe it's vain or meaningless. There's no higher reason or purpose, but for some reason you look at it and feel joy. His theory is that the major force that drives you is this gap between the meaninglessness or vanity of some things and the joy you can find in them. Whereas if you work on something extremely meaningful, you find joy but your rational mind understands why, so it doesn't drive you as much in terms of force. It's a very interesting theory and philosopher.

是的。我认为任何伟大的公司都是创始人灵魂的活化身。比如 Uber，特拉维斯·卡兰尼克说过‘我就是 Uber’。很多人觉得这傲慢或奇怪，但我觉得这个人花了将近十年把灵魂和心血倾注进去，这赋予它意义。你创造有价值事物的能力就是一系列随时间积累的行动。

Yeah. I think of any great company as a living embodiment of the soul of the founder. Like Uber, there was this moment where Travis Kalanick said 'I am Uber.' A lot of people took that as arrogant or strange. But I think this person basically spent almost 10 years pouring their soul and lifeblood into the thing. It gave it meaning to him. Your ability to create something valuable is just a set of actions over time.

确实如此。我看到很多初创公司和创始人，尤其是首次创业者，他们创办公司或项目，看到一些初步进展和成功。通常开始时他们不考虑什么会让他们成功，而是考虑自己喜欢做什么。当获得成功后，他们的思维开始作祟：开始思考应该做什么才能更成功，才能继续获得更大的数字。他们开始思考应该做什么，而不是想做什么。许多创始人最终创建了一家自己不喜欢为之工作的公司。某天早上醒来，它几乎像个怪物。

And it should be because I see a lot of startups and founders, especially first-time founders, who start a company or a project. They see some initial traction and success. Usually when they start, they don't think about what will make them successful; they think about what they enjoy doing. Then when it gets success, their mind plays a game: they start thinking about what they should do to be more successful, to keep getting a bigger number. They start thinking about what they should do instead of what they would like to do. Many founders end up creating a company they don't like working for. They wake up one morning and it's almost like a monster.

是的。这在初创公司中很常见。我有很多朋友、创始人、CEO，他们几年里开始思考应该做什么，别人告诉他们要增长、扩张、建立成功公司。然后某天醒来，意识到自己建立的公司不再满足他们的需求、幸福或兴奋。通常那时他们就退出了。他们聘请一位年长的 CEO，这对公司来说通常是坏消息，因为之后更难增长。当我与创始人交谈时，我最常见的建议是专注于构建你享受构建的东西，让你兴奋的东西。与联合创始人讨论：我们对这个方向兴奋吗？我们能每天早上醒来，连续五年专注于它吗？还是我们觉得必须这样做，因为投资者或别人告诉我们，或者因为我们有一个大公司的虚假形象？相反，专注于你兴奋且觉得对世界有用的东西。这样公司才能成为并保持创始人的反映。

Yeah. It happens a lot in startups. I have many friends, founders, CEOs who for a few years started thinking about what they should do, what people told them to do to grow, scale, build a successful company. Then one day they wake up and realize the company they built doesn't fulfill their needs, happiness, or excitement. Usually that's when they quit. They hire an older CEO, and it's usually bad news for the company because it's harder to grow after that. When I talk to founders, my most common advice is to focus on building what you enjoy building, what excites you. Talk to your co-founders: are we excited about this direction? Can we wake up every morning for five years focusing on that? Or is it something we feel we have to do because investors or others tell us, or because we have a fake image of what a big company is? Instead, focus on what you're excited about and what you feel is useful for the world. That's how companies become and stay a reflection of the founders building them.

杰夫·贝佐斯说过，生活中大多数决定都是双向门：你可以穿过去，采取行动，看看感觉如何，大多数时候你可以走回来。有些是单向门，但很少。在我的生活中，我的指导原则就是采取行动，沿着路径走几步，看看感觉如何，如果不喜欢就退回来，换个方向。重复几次直到找到感觉对的事。对你来说，有没有某个时刻你沿着一条路走下去，然后决定掉头去做别的事？

There's this line from Jeff Bezos where he says most decisions in life are two-way doors: you can go through, take action, see how you feel, and most of the time you can walk back. Some are one-way doors, but that's rare. In my life, my guiding principle is to just take action, take a few steps down the path, see how I feel, and if I don't like it, walk back and go in a different direction. Do that a couple times until I find something that feels right. For you, has there been any moment where you walked down a path and decided to turn around and do something different?

一直都有。我尝试做的一件事是每周至少做一个实验，如果我想太多就不会去做。

All the time. One of the things I try to do is every week to do at least one experiment that if I think too much about it, I wouldn't do.

但有点像，少一些随机性，让我的兴奋感驱使我做一些本来不会做的事。而且大多数事情我后来都会反悔。你指出了一件我认为非常重要的事，那就是更多地追随你的兴奋感和直觉，而不是理性思维。这是我一直在努力的事情。比如，我一直抵制做待办事项清单的冲动。我没有助理。

But kind of like, less randomness, let my excitement for something drive me to do something I wouldn't do otherwise. And most of them I walk back from. You pointed something I think really important, which is to follow your excitement and your guts more than your rational mind. It's something I've been working on a lot. For example, I always resisted the urge of doing to-do lists. I don't have an assistant.

你为什么决定不做这个？

Why did you decide to hold off on that?

因为我怀疑很多人，包括我自己，当你开始过度理性化，把东西理性地放进待办清单，试图做所有你应该做的事，你最终就不会花足够的时间在你真正兴奋、直觉告诉你要做的事情上。我认为人的思维中有某种魔力和美感：如果某件事足够重要和令人兴奋，你会记住去做。你不需要待办清单。如果你需要把某件事放进待办清单，那可能意味着你不应该做它，或者现在不应该做。这和助理有点类似。我认为很多人请助理来帮他们安排会议，管理他们不想做的事情。

Because I suspect that a lot of people, including me, when you start rationalizing things too much and put things rationally into a to-do list and try to do everything you're supposed to do or should do, you end up not spending as much time on the things you're excited about, that your gut tells you to do. I think there's some magic and beauty in people's minds: if something is important and exciting enough for you to do, you'll remember to do it. You don't need a to-do list. If you need something on a to-do list, it probably means you shouldn't do it or shouldn't do it now. It's a little bit the same with assistants. I think a lot of people are getting assistants to help them schedule meetings, managing stuff they don't want to do.

管理他们不想做的事情。

Managing stuff that they don't want to do.

没错。在我看来，如果你没有时间做某件事，那可能你就不应该做。这意味着它不够令人兴奋或影响不够大。而且自己做事有一种魔力；它让你高效得多。比如你和我，我们是怎么安排这次采访的。我觉得如果我有个助理，他们可能会花好几天：‘那天你能来吗？那天你能来吗？我们该怎么设置？’而实际发生的是：我很兴奋，我说：‘好，这是个好主意。你那天能来吗？好的，就在我家做吧。’事情就变得简单容易了。所以这是我一个有趣的领悟。有时候通过避免复杂化，你实际上能做更多，让你的生活更符合你兴奋的事情，而不是过度推向你应该做的事。

Exactly. In my opinion, if you don't have time to do something, probably you just shouldn't do it. It means it's not exciting or impactful enough for you to do. And there's some magic in doing things yourself; it makes you so much more efficient. For example, you and I, how we scheduled this interview. I feel like if I had an assistant, maybe they would have taken a few days: 'Can you do that day? Can you do that day? What should we do as a setup?' Versus how it happened: I got excited, I was like, 'Okay, that's a good idea. Can you come that day? Yes, let's do it at my house.' And things become easy and simple. So that's one of my interesting learnings. Sometimes by avoiding complexity, you actually can do more and keep your life more aligned with what you're excited about, instead of pushing too far into what you should do.

杰夫·贝佐斯在亚马逊早期和一位高管开过会。那位高管来找他，说自己忙得不可开交，事情太多，没时间做自己喜欢的事。我记得贝佐斯对他说，这基本上是你的工作。你是组织的最高层。你的工作是弄清楚你应该做什么、你真正想做什么，然后找到能支持你做这些事的人。如果你长期没有做到，那就是你的失败。你是怎么弄清楚业务的哪些部分是你该做的？对我来说，我必须坐在这个椅子上。这是我能够做的事，我很高兴去做。播客和制作这个节目的其他所有领域，我并不是真正喜欢做。但我可以去找世界上最擅长那些事的人，填补我的空缺。对你来说，你是怎么做到像这样：‘这是我喜欢的事。其他所有领域我都不喜欢。我要去找那些人。’

Jeff Bezos had a meeting with one of his executives early in Amazon's history. The executive came to him and said he was hugely jammed, there was too much stuff going on, and he wasn't doing the things he loved. I remember Bezos saying to him that basically it's your job. You're the top of the organization. It's your job to figure out what you should be doing and what you genuinely want to do, and then find the people that can support you in doing that. It's your failure if you haven't done that over time. How have you figured out what parts of the business are the thing? For me, I have to be sitting in this chair. It's something I get to do, I'm really happy to do it. There are all these other areas of the podcast and making this production happen that I don't genuinely enjoy doing. But I can go find people that are the best in the world at those things and fill my gaps. For you, what's that been like where you're like, 'This is the thing that I love. All these other areas I don't love. I'm going to go find those people.'

是的。我总是试着思考如何增加价值、增加影响力，并做到比任何人都好十倍，然后专注于这些。当我觉得某件事我无法比别人好十倍时，如果别人对此有热情，我宁愿让他们去做。所以我通常就是这样挑选不同的事情来做。作为 CEO，我会想：‘好吧，哪些是最有影响力、只有我能做的事情？’另外，在 Hugging Face 的文化中，非常重要的一点是去中心化文化和自由文化，这让我们不仅被自己的热情驱动，也被其他人的热情驱动。很多时候，当我们发布新东西时，人们会看着它，试图理解其战略相关性背后的逻辑。你基本上是先看到这个东西，然后反向推导出采取行动或做出举措的逻辑。

Yeah. I always try to think about how can I add value, add impact, and do something 10 times better than anyone else could do, and try to focus on these. When there's something where I feel I can't do 10 times better than someone else, I prefer to let them do it if they're excited. So that's usually how I pick and choose the different things to work on. As a CEO, I'm like, 'Okay, what are some of the most impactful things that I can do that nobody else can do?' And also, something really important in our culture at Hugging Face is the decentralized culture and the freedom culture that empowers us to be driven not only by my own excitement but also by anyone else's excitement about things. A lot of the time, when we release new things, people look at it and try to understand the strategic relevancy for the logic. You basically look at the thing and then work backwards to create the logic for taking an action or making some move.

你基本上是先看到这个东西，然后反向推导，试图为采取行动或做出举措创造逻辑。

You basically look at the thing and then work backwards and try to create the logic for taking an action or making some move.

基本上我们追随人们的热情。所以有时候我们发布一些东西，人们会问：‘从战略上讲，这对 Hugging Face 有什么意义？’我回答：‘哦，你知道，约翰对此很兴奋，就把它做出来了。’这就是背后的逻辑。我一直努力培养这种文化。我认为我们在 Hugging Face 的招聘方式有点不寻常。我们通常不针对特定的职位描述或职位进行招聘。我们试图找到聪明、有热情、并且认同我们愿景和使命的人。通常我们把他们招进来，告诉他们去做他们兴奋的事情。这效果很好，因为他们做对自己有意义的事，他们变得非常有动力，成为自己项目的所有者，最终对 Hugging Face 产生巨大影响。而如果我们反过来，从任务、使命或职位开始，然后试图找人来匹配这个职位，告诉他们该做什么，比如制定日程，我认为那不会那么有效。所以这对我们来说是一个有趣的领悟。给予人们自由、所有权和结构，让他们做自己兴奋的事情，通常好事就会随之而来。

Basically we follow people's excitement. So sometimes we release things and people are like, 'Okay, strategically how does it make sense for Hugging Face?' And I reply, 'Oh, you know, John got excited about it and built it.' That's the logic behind it. And I always try to foster that. I think the way we hire at Hugging Face is a bit unusual. We usually don't hire for a specific job description or job position. We try to find people who are smart, excited, and aligned with our vision and mission. Usually we bring them in and tell them to work on what they're excited about. That's worked really well because they do things that are meaningful to them, they get extremely motivated, they become owners of their projects, and end up having tremendous impact for Hugging Face. Whereas if we were working the other way around, starting from the task or the mission or the job position and then trying to find someone to fit that position and tell them what to do, like creating the schedule, I think that wouldn't work as well. So that's been an interesting learning for us. Giving people the freedom, the ownership, the structure for them to work on what they're excited about, and usually good things come out of that.

查理·麦格雷戈有句话，他说你应该追随你的自然漂移。

Charlie McGregor has this line where he's like, you should follow your natural drift.

嗯，我觉得对于这样的事情，你不会从反向推导开始，说这里有一个巨大的业务，我们要反向推导，创造出这个小东西。嗯，假设今年四月你说你收购了一家公司，打算构建某个版本。你是如何从那里走到真正创造出让你兴奋并推向市场的东西的？

Um I think with something like that, I don't think you start it by working backwards and saying here's a huge business and we're going to, you know, work backwards and create this little guy. um let's say in April this year you said that you acquired a company and you were going to build uh some version of this. How did you kind of go from that to actually find you know creating something that you were excited to get out into the world and actually ship?

是的。所以即使是收购，也是从我们收购的团队的兴奋开始的——波兰团队，他们很兴奋能加入 Hugging Face，并与我们一起构建。然后，在真正跟随他们的思路之后，因为他们是专家，他们在机器人领域工作了差不多十年，当我们收购他们时，他们是少数几个真正在交付人形机器人的团队之一，例如，现在我们正在向 OpenAI、DeepMind 以及类似的研究实验室交付人形机器人。他们基本上在收购后告诉我们，好吧，我认为为 ZA Builders 做这件事会非常令人兴奋。所以我就说了是。作为 Hugging Face 这样公司的 CEO，我的工作很大一部分其实就是说“是”，然后赋能人们，支持他们，并把他们引向我认为能帮助他们增加影响力的方向，消除他们的一些恐惧。

Yeah. So even the acquisition started from the excitement from the team we were acquiring the Poland team uh who were excited about joining Hugging Face right and and building building with us. Um and then after really kind of like uh following their thinking right because they were the experts they've been working on robotics for um yeah almost uh almost 10 years when we acquired them they were one of the few teams that were actually shipping humanoid robots um for example now we're shipping humanoid robots to to open AI to deep mind to to research labs like this Um and they were basically uh you know told us after we we acquired them, okay, uh I think that would be really exciting to to work on for for ZA builders. And so I just said yes. A big big part of of my job as a CEO actually as a in a structure in a company like Hugging Face is really just to to say yes and just kind of like uh empower people and kind of like uh support them uh and and kind of like uh point them towards kind of like some directions that I think can help increase their impact. um remove some of their fears

比如障碍或

like the roadblocks or

是的。比如，我经常做的一件事就是让人们相信他们有能力比他们认为的发布时间更早地发布。对吧？我认为对于这样的事情，自然的倾向是明年再发布，因为总会有这样的想法：好吧，我们需要做这个，这个还没准备好，我们需要在那方面取得进展。所以自然的倾向是推迟发布，而在像 Hugging Face 这样的组织里，我们总是试图说，好吧，你不需要准备好才能发布，你只需要有一个最小可行产品，一个不会完全崩溃的东西，并且设定正确的期望。对吧？不要告诉人们它从第一天起就是完美的，从第一天起就是 AGI。相反，告诉他们这是一个不完美的第一个版本，然后发布它，让人们去玩、去实验。所以我工作的很大一部分就是安抚团队成员，建立他们的信心，让他们能够快速发布，与社区一起实验，一步一步地构建。

Yeah. Like for example uh uh what I do a lot is to reassure people on their ability to release earlier than what they think they should do. Right? I think uh uh the natural tendency for for something like this would have been to release it next year, you know, because there's always this thing like okay, we need to work on this this this is not ready. We need to make progress on that. And so the natural tendency is for people to delay launch versus uh you know in an organization like cooking face we're always trying to say okay you don't need to be ready to to launch right like uh you just uh you just have to have kind of like a minimum valuable product something that is not going to completely break um and you know set the right expectations Right? Don't tell people that it's going to be perfect from day one, that it's going to be agi from from day one, right? But uh instead tell them that it's an imperfect first version and then release it and uh let let the people play with it and experiment with it. So a big part of my job is just to uh reassure and build confidence from the team members that they can release things quickly with the community experiment and and build one step at a time.

我最近在想，创始人的工作在很多时候就是与不确定性共存，而世界上其他人大都不愿与不确定性共存。他们希望尽可能多地——我不会说壁垒，但我会说具体性——在他们所做的事情中。对你来说，你是如何帮助你的团队与不确定性共存，从而更快地交付的？

I was thinking about this recently where I kind of think that the founder's job in a lot of ways is to basically live with uncertainty and everyone else in the world pretty much doesn't want to live with uncertainty. they want to like have as much I wouldn't say walls, but I would say like concreteness, you know, in what they're doing. Um, for you, how are you kind of like helping your teams live with uncertainty and basically ship faster?

嗯，我认为很大程度上是让人们坦然接受错误和失败，对吧？我认为大多数人拖延的原因是害怕犯错的后果，对吧？这让人不舒服。比如发布某样东西或做某件事，你犯了错，有人告诉你，伙计，你完了。就像他们在嘲笑机器人，直接嘲笑你那种情况。是的。当你提前发布时，这种事经常发生。所以就是让他们对此感到舒适。这种事经常发生。比如我之前跟你讲过的 Thomas 早期发布 PyTorch 预训练模型的故事，对吧？他说他周五开始做，周一就发布了，他整个周末都在工作。有趣的是，我记得有两天它根本不能用。周一和周二就是不行。我不记得具体是什么问题了，但就是完全不能用。所以大多数人可能会害怕、紧张，因为我们发布了一个不能用的东西。但你知道，我们对这类事情容忍度很高。当时 Thomas 的反应是最好的，他说，哦，它不能用是因为有些问题，但人们似乎很兴奋，我们验证了这一点，因为人们在转发、点赞那条推文，所以我现在要去修复这些问题，对人们保持透明。两天后它修好了，人们开始使用它，效果很好。同样，对于 LeRobot，我确信当我们开始发货更大批次时，也会有很多问题。所以我们希望在年底前发货大约 4000 台。我确信会有很多问题，提前向用户、测试者、AI 构建者道歉。会有很多问题，我确信会有很多 bug。但我们会与社区一起努力，尽快修复它们，改进它们，与他们一起学习。这通常是我们处理事情的方式。我觉得对我来说，帮助人们不过度拖延做事或发布东西的最大因素，就是让他们对错误、失败和问题更坦然。

Well, I mean, I think a lot of it is is kind of like uh making feel comfortable with uh with mistakes and up, right? like uh the reason I think uh most people delay things is because they're afraid of the repercussions of making a mistake, right? It feels uncomfortable. Kind of like releasing something or doing something and you make a mistake and someone is telling you, man, you you're done. Like there's they're laughing at the robot. They're laughing directly at you kind of situation. Yeah. Um and uh and when you release early, it happens all the time. So it's just uh making them comfortable with that. Uh it happens happens all the time. Like uh I was telling you the early story of uh Thomas releasing, you know, PyTorch pre-trained birds, right? Um he said he would start working on it on on Friday and he released on Monday, right? He he worked on it all all weekend. And the funny story is that I think for for two days it wasn't working at all. It was kind of like a I don't remember on the Monday and Tuesday it just didn't work. Yeah. I I don't remember exactly what was the problem, but it was not working at all. And so most people would have been like, you know, scared and and kind of like uh stressed, you know, because because we released something that wasn't working. But uh you know, I think we have a high tolerance for things like that. Uh and at the time uh Thomas uh had the best reaction which was you know oh it's not working because there are some some problems but people seem excited about it right we validated that because uh people are retweeting liking the the tweet so now I'm going to fix these problems be transparent with people and two days later it was fixed and and people started using it and uh it was great same for you know Richie Mini I'm sure uh when we uh start shipping bigger batches. Uh so we hope to ship around, you know, 4,000 of them before the end of the year. I'm sure there's going to be a lot of problems, right? And I apologize in advance to to users to to beta testers to to AI builders. Um there are going to be a lot of problems. I'm sure there's going to be a lot of bugs. Uh but we'll work hard with the community to to fix them as fast as possible, to improve them, to learn with them. Um, that's usually the way to approach things. And I I feel like that's the biggest thing for me that helps people not delay too much doing things or releasing releasing things. It's just making them more comfortable with mistakes, failures, problems.

我认为很多试图创造 AI 产品的公司，无论是现在还是过去几年，都试图一开始就打造 iPhone 15，一个完全成熟的产品，并声称它完全成熟，但当它达不到时，它就不是 iPhone 15，可能是 iPhone 1，甚至更早的 iPod。而对于像这样的东西，我认为特别是你的公司，因为它有这种开源修补文化，我认为采用你产品的人可能对期望有不同的设定，与其他公司不同。你是如何看待为即将发布的产品设定期望，并让人们知道事情不会完美，也不应该完美，而这正是快速发布的全部目的？

I think a lot of the companies that have tried to create AI products uh today and over the past couple years have kind of tried to create the iPhone 15 to start. this fully fleshed out and say that it's fully fleshed out and then when it doesn't, you know, it's not iPhone 15. It may be iPhone 1 or maybe even the iPod before that. Um, and with something like this, I think with your company in particular, because it's got this open- source tinkering culture, I think people that are adopting the thing that you buy, uh, may have a different sort of expectation set um, than another company. Um, how do you kind of think about setting expectations for a product that you're going to release and kind of letting people know things aren't going to go perfectly and it's not meant to. And that's the entire objective of releasing quickly.

我觉得期望对 AI 来说是一个非常大的问题，特别是因为“AI”这个词本身，以及它是一种具有新能力的新技术这一事实。很容易过度炒作和过度销售你正在构建的东西。

I feel like expectations a very big problem for AI uh specifically because of the term itself like AI um and the fact that it's kind of like a new technology with new capabilities. It's really easy to you know overhype and oversell what you're what you're building.

我认为有些公司、组织和个人会陷入这个陷阱。这是个问题，因为它引发了人们的恐惧。现在很多人想到 AI，就会联想到科幻场景，比如《机械战警》、《终结者》，所有这些由 AI 造成的疯狂末日场景。而我们在 Hugging Face 看待这项技术的方式，更像是新一代的软件。Andrej Karpathy 曾用“软件 2.0”来描述 AI，我认为这是一种更接地气、更现实的方式来理解 AI，也更容易让人们理解。AI 不是那种只有少数人才能构建并主宰世界的技术黑箱。它更像是一种新的构建范式，一种创建软件的新方式，每个人都可以使用，每个人都可以参与其中，它将解决我们面临的许多不同挑战和问题，并在许多不同领域创造一系列新机遇。有趣的是，在 Hugging Face 上，现在有 1100 万 AI 构建者在使用这个平台。AI 在 Hugging Face 上增长最快的领域不再是文本。很多人都在谈论 ChatGPT、聊天机器人和文本，但 Hugging Face 上增长最快的领域是音频、视频、图像、生物学、化学。所以我认为我们看到的是，AI 将被应用到每一个领域，就像软件已经被应用到每一个领域一样，并成为一种新的范式，为我们人类构建许多解决我们问题、创造新机遇和新能力的东西奠定基础。所以我认为，当你这样呈现 AI 时，它更接近技术的现实。它消除了人们的一些恐惧。它让人们从被动使用技术转变为主动成为行动者、构建者，并尝试自己动手，从长远来看，这对该领域会更好。

And I think some companies, some organizations, some people fall into this trap. Which is a problem because I think it creates fears from people. You know, now a lot of people think about AI and think about sci-fi scenarios, Robocop, Terminator, all these crazy doomsday scenarios created by AI. Versus how we perceive this technology at Hugging Face is more like a new generation of software. Andrej Karpathy talked about Software 2.0 as a way to describe AI, which I think is a much more down-to-earth, much more realistic way to approach AI, and something that people can understand better. AI is not this technology black box that just a few people are going to build and dominate the world. It's more like a new paradigm for building, a new way of creating software that everyone can use, everyone can be a part of, and that is going to solve a lot of different challenges and problems we have, and create a whole set of new opportunities in many different domains. Something interesting on Hugging Face: we have 11 million AI builders now using the platform. The fastest growing domains for AI on Hugging Face is not text anymore. A lot of people are talking about ChatGPT, about chatbots and text, but the domains that are growing the fastest on Hugging Face are audio, video, image, biology, chemistry. So I think what we're seeing is that AI is going to be applied to every single domain, just like software has been applied to every single domain, and become a new paradigm, new foundations for us humans to build a lot of things that solve our problems, create new opportunities and new capabilities. So I think when you present AI like this, it's closer to the reality of the technology. It removes some of the fears from people. It moves them from a default of using the technology to more like becoming actors, becoming builders, and trying to do things themselves, which in the long term will be much better for the field.

你能花点时间谈谈为什么你们的使命是开源，尤其是在当今世界，最大的公司在单个 AI 模型上花费数十亿美元的情况下。为什么开源如此重要？

Can you basically spend some time talking about why your mission is open source, especially in today's world with AI models where the biggest companies are spending billions of dollars on a single model. Why open source matters so much?

是的。在 AI 领域，存在着非常强烈的自然趋势，即权力集中、能力集中和财富集中。你提到算力是一个决定性因素，人才非常重要，资金在 AI 中也确实很重要。所以我认为我们应该担心的是，最终会陷入一个只有少数公司能够做 AI 的世界。可以这样想：如果只有少数公司能够构建软件，世界会变成什么样？我认为那将是一个非常可怕、反乌托邦、危险的世界。所以我们在 Hugging Face 一直认为，我们需要对抗这些非常强烈的激励因素和权力集中的风险。我不知道一些顶级实验室在华盛顿游说花了多少钱。我猜那是一大笔钱。确实是很大一笔钱。而且大型科技公司已经占据主导地位，它们有能力向 AI 投入数千亿美元，这使得任何人都很难与它们竞争。因此，我们在 Hugging Face 相信，开源——即能够向任何人免费开放共享模型和数据集——是对抗这些自然趋势的一种方式。因为当公司、开发者或组织这样做时，他们为每个人提供了构建自己的 AI 产品、AI 功能、AI 组织的基础，并对抗这些权力集中的自然趋势。得益于开源，我们相信可以创造一个世界，在这个世界里，不仅仅是少数组织能够构建和主导 AI，而是任何组织——成千上万的小型科技初创公司、非营利组织、政府、个人——都应该能够用 AI 进行构建。做到这一点的唯一方法就是开源，因为你无法从零开始构建 AI 并保持竞争力。你需要以开源模型为基础，借助开源数据集和开源工具，这样你就不必重建一切。

Yeah. So in AI, there are very strong natural tendencies for concentration of power, concentration of capabilities, and concentration of wealth. You mentioned compute as a defining impact, talent is super important, and money is really important in AI. So I think something we should be scared about is to end up in a world where only a few companies are able to do AI. The way to think about it is: how would the world be if only a few companies were able to build software? I think that would be a very scary, dystopian, dangerous world. So what we always felt at Hugging Face is that we need to fight these very strong incentives and risks of concentration of power. I have no idea how much some of the top labs are spending to lobby in Washington. I imagine it's a lot of money. It's a lot of money. And even the domination of big tech companies already and their ability to pour hundreds of billions of dollars into AI makes it difficult for anyone to compete with them. So we believe at Hugging Face that open source, which is basically the ability to share models and datasets openly for free to anyone, is a way to fight these natural tendencies. Because when companies, developers, or organizations do that, they give everyone the foundations to build their own AI products, AI features, AI organizations, and fight these natural tendencies of concentration of power. Thanks to open source, we believe you can create a world where not just a few organizations can build and dominate AI, but really any organization—tens of thousands, hundreds of thousands of little tech startups, nonprofits, governments, individuals—should be able to build with AI. The only way to do that is thanks to open source, because you can't build AI and be competitive starting from scratch. You need the help of open-source models as a base, open-source datasets, and open-source tools so you don't have to rebuild everything.

但我记得 Anthropic 的 Dario Amodei 说过，如果你看整个公司，它可能亏损，但如果你看某个特定模型，他们投资 20 亿美元，却从中获得 200 亿美元，或者投资 2 亿美元获得 20 亿美元。我认为在那种世界里，如果你看看你提到的那些新模型，它们更小但更专门化，比如针对生物学、化学或机器人学，它们几乎就像小型初创公司。每个模型几乎就像一家初创公司，迭代和发布的速度在这种小东西上可以有所不同。你如何看待这种演变？

But I remember Dario Amodei from Anthropic saying that if you look at the company as a whole, it may be losing money, but then if you look at a model in particular, they invest $2 billion and get $20 billion out of that, or $200 million and get $2 billion. And I think in that type of world, if you look at the new models you're talking about where they're smaller but specific, like to biology or chemistry or robotics, it's almost like little mini startups. Each model is almost like a startup, and the velocity of iteration and shipping can be different on this small thing. How do you kind of see that evolving?

是的，我同意。我们在 Hugging Face 看到的是，当你拥有更小的定制化模型时，有很多优势。迭代和实验更容易，因为模型更小。你更了解它的局限性，知道它能做什么和不能做什么。它通常运行更快，成本也更低。它没有一堆冗余的东西。没错。这个想法是，也许你需要一个非常大的通用模型来做 ChatGPT，试图回答世界上所有的问题。但当你做一个银行客户支持聊天机器人时，你并不需要它告诉你生命的意义。你只需要它告诉你关于你的账户，关于你遇到的一些问题。所以你可以使用一个更小、更便宜、更专门的模型，通常大多数时候它会更擅长回答你的具体问题。

Yeah, I agree. What we see at Hugging Face is that when you have smaller customized models, there are a bunch of advantages. It's easier to iterate on and experiment with because it's a smaller model. You understand better the limitations, what it can do and what it can't do. It's usually faster to run, and usually cheaper to run. It doesn't have a bunch of fluff. Exactly. It's this idea that maybe you need a very large generalist model to do ChatGPT, to try to answer all the questions in the world. But when you're doing a banking customer support chatbot, you don't really need that to tell you the meaning of life. You just need it to tell you about your accounts, about some of the problems you're encountering. So you can use a smaller, cheaper, more specialized model that usually most of the time is going to be better at answering your specific questions.

所以我们最终设想的世界和领域是，不是只有少数几个通用大模型供所有人使用，而是有数百万个更小、更专业、更定制化的模型来解决各种个体问题和用例。这和软件领域发生的情况非常相似。如果你想一想，你可能会认为更大的代码库更好，甚至会出现一个通用的巨型代码库来统治一切。但事实并非如此。实际情况是，每家公司、每个组织都在编写自己的代码。有数百万个不同的代码库。每个人都在构建自己的。所以我们认为 AI 领域也会发生类似的事情。你将看到一个世界，每个组织都基于开源拥有自己定制化的专业模型。开源是基础，每个人都将拥有自己特定的定制化模型。

So ultimately we envision a world and a field where you don't have just a few big generalist models that everyone is using, but instead you have millions of smaller specialized customized models that are solving all kinds of individual problems and use cases. Very similar to what happened for software. If you think of it, you could believe that a bigger code base is better and that there's going to be a general gigantic code base to rule them all. But this is not what happened. What happened is that every company, every organization is writing their own code. There are millions of different code bases. Everyone is building their own. So something that we think is going to happen is similar for AI. You're going to have a world where every organization has their own customized specialized models based on open source. Open source is the foundation, the base, and everyone will have their own specific customized models.

你提到监管的主要卖点基本上就是恐惧。你把这个机器人做得这么友好快乐，是为了对抗恐惧吗？

You mentioned that the main selling point for regulation is basically just fear. Is the reason that you made that robot kind of like this friendly joyful thing? Is that to fight fear?

是的，我认为我们真正看到了 AI 中透明度的美好之处。我认为很多监管者，也包括普通人，都被黑箱吓到了。当你有一个你不理解的系统时，它会产生更多的恐惧，而当你能够看到幕后，看到它是如何工作的并理解它时，它会消除很多恐惧。或者它会将你的恐惧引导到一些实际上合理且有益的恐惧上。我并不是说 AI 是完美的，没有任何风险或挑战。但一旦系统具有透明度，我认为你就可以专注于一些更真实的挑战。例如，当你观察模型是如何构建的以及它们能做什么时，你会发现很多模型都存在偏见。我们讨论了很多关于聊天机器人偏见的问题。如果你还记得，我认为是 Gemini 有很多觉醒偏见。

Yeah, I think we've really seen the beauty of transparency in AI. I think a lot of regulators, but also people in general, are scared by black boxes. When you have a system that you don't understand, it creates more fear versus when you can see behind the scenes, when you can see how it works and understand how it works, it removes a lot of these fears. Or it directs your fears to something that is actually a rational good fear. I don't think we're saying that AI is perfect and doesn't have any risk or challenges. But once you have transparency in the systems, I think you can focus on some of the challenges that are more real. For example, when you look at how the models are built and what they can do, you realize that a lot of them have biases. We talked a lot about chatbot biases. If you remember, I think it was Gemini who had a lot of woke biases.

没法让开国元勋不是黑人。

Couldn't make the founding fathers not black.

没错。这些都是真实的问题。不是科幻片里的终结者问题，而是你现在就需要解决和修复的实际问题。所以我认为透明度对 AI 非常重要。我认为越透明，人们就越能解决一些挑战并创造更多机会。这也是开源对 AI 的一大价值所在。

Exactly. These are real problems. It's not sci-fi Terminator problems, but it's actual problems that you need to solve and fix now. So I think transparency is really important for AI. I think the more transparent you can be, the better people will be at solving some of the challenges and creating more opportunities. So that's also one of the big values of open source for AI.

这有点假设性，但我觉得我最喜欢的三四位创始人之一是约翰尼·洛克菲勒。想象他今天 18 岁或 19 岁。你觉得他会创办什么样的公司？

This is a little bit of a hypothetical, but I think one of my top three or four favorite founders is Johnny Rockefeller. And imagine him as an 18-year-old or 19-year-old today. What do you think what company would he be building?

他肯定会成为一个 AI 构建者。我们经常讨论 AI 用户和 AI 构建者之间的区别。AI 用户可能是通过 API 或黑箱 AI 来构建的人，而 AI 构建者则是能够训练模型、优化模型并自己拥有模型的人。如果我现在要创办一家新公司，我会专注于 AI 在生物学和化学领域的应用。我们看到 Hugging Face 上出现了很多这些领域的新模型，比如细胞预测、分子预测。我认为这些是投资不足的领域，未来几年你会看到很多新的应用，尤其是因为我们在 AI 的时间理论方面取得了越来越多的进展。所以这是 AI 在时间序列和概率方面的应用，而不是文本。我觉得这些应用加上 AI 在文本方面的应用，可能会释放一些新的能力。所以我希望有更多人从事这些课题。聊天机器人很有趣，文本 AI 也很有趣，但太多人在做这个了，媒体的大部分注意力也集中在这上面，可能太多了。最好开始关注其他领域。

He would be an AI builder for sure. We talk a lot about this differentiation between an AI user, which would be someone building maybe with APIs or with blackbox AI, versus an AI builder. So someone who can train models, optimize models, and own models themselves. If I were to start a new company right now, I would focus on AI for biology, AI for chemistry. We're seeing a lot of new models appearing on Hugging Face on these domains, things like cell predictions, molecule predictions. I think these are underinvested topics where you will see a lot of new applications in the next few years, especially because we're starting to see more and more progress on time theories for AI. So it's the application of AI for time series and probabilities rather than text. And I feel like some of that application, plus application of AI for text, could unleash some new capabilities. So I wish more people were working on these topics. Chatbots are fun, text AI are fun, but so many people are working on that, most of the media attention is focused on that, probably too much. It would be good to start focusing on other domains.

上周我收到了关于 AI 泡沫的问题。我们处于 AI 泡沫中吗？我的回答是，可能处于 LLM 泡沫中。人们谈论很多 LLM 和聊天机器人，大部分资金都流向那里。而在我看来，我们可以开始关注 AI 的其他领域，比如生物学、化学、时间序列、图像、音频。在这些主题上，我们才刚刚触及皮毛。所以从这个意义上说，我们在 AI 领域还非常早期。

I got the question last week on the topic of AI bubble. Are we on an AI bubble? And I answered that probably in an LLM bubble. People are talking a lot about LLMs, about chatbots. That's where most of the money is going. Whereas in my opinion, we can start focusing on other domains in AI like biology, chemistry, time series, images, audio. On these topics, we're just scratching the surface of what we can do. So in that sense, we're super early in AI.

我记得听过山姆·奥特曼谈论真实趋势和虚假趋势。他被问及 2015 年左右是否处于泡沫中，或者初创公司是否处于泡沫中。我记得他描述自己思考方式的方式是，真实趋势就像 iPhone 1，可能是一个不完美的产品，但早期用户会持续使用它。每天平均使用时间大约 3 小时。而虚假趋势就像 VR 和 AR，大多数早期用户第一天会戴上 5 到 6 小时，第二天 2 到 3 小时，到第三周基本上就放在架子上积灰了。我认为对于这种事情，当人们说我们是否处于 AI 泡沫时，我记得加文·贝克谈到 GPU 的利用率，说没有安静的 GPU，它们都在嗡嗡作响，所以可能不是泡沫。这里正在创造大量价值。

I remember hearing Sam Altman talk about real trends versus fake trends. He got asked if they were in a bubble in like 2015, or if startups were in a bubble. I remember the way he described how he thought about it was basically like a real trend is something like the iPhone 1 where maybe it's an imperfect product, but the early adopters are just constantly using it. The average time per day was like 3 hours on these devices. Whereas a fake trend is like VR and AR where most of the early adopters, they'd put it on and wear it for 5 or 6 hours the first day, then 2 or 3 hours the next day, and by the third week it's basically sitting on the shelf collecting dust. I think with this sort of thing, when people say are we in an AI bubble, I think Gavin Baker was talking about how much the GPUs are utilized and it's like there is no quiet GPU. They're all humming, so probably not. There's lots of value being created here.

对我来说，这很有趣，因为我在 AI 领域已经有一段时间了，我们可以看到演变。通常我以某种方式定义炒作，即感知和使用之间的差距。对我们来说，作为 AI 构建者的平台，从我们创办 Hugging Face 到 GPT 之前，AI 被严重低估了，因为我们已经开始看到一些使用。例如，AI 被用于搜索引擎，你已经开始在谷歌中看到一些 AI。AI 被用于社交媒体，比如推荐系统。AI 被用于自动驾驶汽车的早期阶段，以及你的家庭助手。

For me, it's interesting because being in AI for some time now, we could see the evolution. Usually I define hype between the gap between perception and usage in a way. It felt like for us being the platform for AI builders from when we started Hugging Face to pre-GPT, AI was very much underhyped because we were already starting to see some usage. For example, AI was used for search engines, you were already starting to see some AI in Google. AI was used for social media with recommender systems. AI was used for the early innings of self-driving cars, for your home assistant.

所以即使在 GPT 之前，AI 已经有了一些不错的使用量，但人们并没有怎么谈论它。它处于雷达之下，是小众的。AI 初创公司不多，除了我们的核心用户群，投资者、媒体——当时没人知道 Hugging Face。然后 ChatGPT 出现了，出现了一种追赶。投资者、媒体、非领域内的人开始意识到，多亏了 ChatGPT，AI 将变得巨大且具有变革性。所以我们进入了一个调整期，在我看来，认知与底层使用量趋于一致。这就是我们今天的情况。那些之前不在这个领域或没看到 AI 使用量的人可能会想，‘哦，这肯定有泡沫，因为两年前我没听说过 AI，现在到处都是。’他们被认知变化的速度误导了，但我认为这只是追赶。

So even before GPT, there was already some decent usage of AI, but people weren't really talking about it that much. It was under the radar, niche. There weren't many AI startups, and outside our core user base, investors, media—nobody knew Hugging Face at the time. Then ChatGPT came out, and there was a catch-up. Investors, media, people not in the field started to realize, thanks to ChatGPT, that AI was going to be big and transformative. So we entered an adjustment where, in my opinion, perception aligns with underlying usage. That's where we are today. People who weren't in the field before or didn't see all the AI usage might think, 'Oh, there's definitely a bubble because two years ago I wasn't hearing about AI, and now it's everywhere.' They're misled by the speed of perception change, but I think it was just catch-up.

我不知道过去两三年里创建了多少初创公司。会有很多初创公司最终没有变得非常有价值，这没关系，但这只是人们尝试新事物的创造性破坏过程的一部分。

I don't know how many startups have been created in the past 2 or 3 years. There's going to be a lot of startups that don't end up being super valuable, and that's fine, but that's just part of the creative destruction process of people trying things out.

在 AI 范式下失败的初创公司可能比在软件范式下更多。这完全没问题。

There's probably going to be more startups failing in the AI paradigm than in the software paradigm. And that's completely fine.

那只是意味着更多的兴奋。

That just means there's more excitement.

没错。是的。人们在尝试，也有更多的新奇事物，对吧？我认为 AI 真正有趣的地方在于，因为它是一个新范式，它把很多剧本都扔出了窗外。这几乎就像我们经历了一个周期，大约 20、25 年前我们开始做软件。随着时间的推移，这个领域成熟了，变得更加复杂，所以你开始有了这些剧本——SaaS 剧本、精益创业剧本。你有从学习中得出的最佳实践，因为构建软件的范式在很长一段时间内保持相对相似。现在我们处于一个新范式，AI，我看到很多人犯错误，试图将软件领域的相同剧本应用到 AI。在我看来，更好的方法是完全忘掉，把之前学到的一切、所有剧本都扔进垃圾桶，重新开始。思考，‘好吧，现在在 AI 时代，我如何建立一家初创公司？’这很可能与你建立传统软件初创公司的方式非常不同。但这也是乐趣的一部分——你可以重新发明，尝试新事物，从头开始，而不是使用过去的剧本。

Exactly. Yeah. People trying and also more novelty, right? I think what's really interesting with AI is that because it's a new paradigm, it throws a lot of the playbooks out the window. It's almost like we've had this cycle where we started doing software maybe 20, 25 years ago. Over time, the field matured and became more sophisticated, so you started to have these playbooks—the SaaS playbook, the lean startup playbook. You had best practices from learning that the paradigm of building software stayed relatively similar for a long time. Now we're in a new paradigm, AI, and I see a lot of people making mistakes by trying to apply the same playbook from software to AI. In my opinion, a better approach is to completely unlearn, throw in the trash everything you've learned before, all the playbooks, and start new. Think, 'Okay, now in the AI era, how do I build a startup?' It's probably going to be very different from how you built traditional software startups. But that's part of the fun—you get to reinvent, try new things, and start from scratch instead of using the playbooks of the past.

我记得 Sam Altman 说过一句话，他说第一次创办公司并经历所谓的‘公司杀手’危机时，感觉世界要崩塌了。但当你挺过去之后，到了第七次，你就会想，‘好吧，我挺过了前六次，这次大概也没什么不同。’在 Hugging Face 的旅程中，第一次让你觉得‘天哪，我们要死了吗？’是什么时候？

I remember a line from Sam Altman where he said the first time you're building a company and you experience what can be described as a company-killing crisis, it feels like the world is falling apart. But then you make it through, and by the seventh time, you're kind of like, 'Well, I made it through the first six, so this is probably no different.' What was the first time during Hugging Face's journey where you were like, 'Oh my god, are we going to die?'

是的。嗯，我想大概是在其他人发布了与我们类似的东西的时候。我稍微提到过谷歌发布 TensorFlow Hub，那是我们正在做的事情的直接竞争对手。当然，当你面对像谷歌这样拥有巨大资源和团队的大公司发布与你类似的东西时，那是一个充满压力和恐惧的紧张时刻。

Yeah. Well, I think it's probably when other people released similar things to us. I mentioned a little bit Google releasing TensorFlow Hub, which was a direct competitor to what we were doing. Of course, when you have a massive company like Google with massive resources and teams releasing something similar to what you're doing, it's an intense moment of stress and fear.

你在那一刻做了什么？

What did you do in that moment?

我认为我们非常擅长在当时不急于行动，因为很容易恐慌，从而损害你建立的一些东西，比如文化或声誉。声誉需要 30 年建立，五分钟就能毁掉。所以我们通常先花时间，不急于求成。然后，对我们非常有效的做法是，即使在这些时刻，也将其视为合作而非竞争的机会。例如，对于 TensorFlow Hub，我们没有想着必须竞争并试图打败他们，而是主动联系他们，特别与一位名叫 François 的人交谈，他当时是谷歌 AI 领域最杰出的人物之一。我们开始与他们合作，在 TensorFlow Hub 和 Hugging Face Hub 之间建立集成。这对我们非常有利，因为最终我们找到了让谷歌的一些举措变得有用的方法，特别是对于使用更多谷歌中心化工具的谷歌员工，同时也促使许多用户同时使用他们的产品和我们的产品。所以有点反直觉，但在大多数情况下，我们最终与这些竞争对手合作，而不是正面竞争。这很特定于我们、我们的文化以及我们所做的事情，但效果很好。

I think we've been really good at not rushing too much at the time, because it's easy to panic in a way and damage some of the things you've built, like the culture or reputation. Reputation takes 30 years to build and five minutes to ruin. So usually we try to first take our time, not rush things. Then the approach that has served us quite well is to approach even these moments as an opportunity to collaborate rather than compete. For example, with TensorFlow Hub, instead of thinking we have to compete and try to kill them, we actually reached out to them and talked particularly to a guy named François, one of the most prominent people in AI at Google at the time. We started collaborating with them, building integrations between TensorFlow Hub and the Hugging Face Hub. That served us really well because ultimately we found a way for some of Google's initiatives to be useful, especially for people working at Google using more Google-centric tools, but also driving many of those users to use both their products and what we have to offer. So a bit counterintuitive, but in most of these cases we ended up collaborating with these competitors more than competing frontally. It's specific to us, our culture, and what we do, but it worked quite well.

去年八月，Pavel Durov 飞往法国，下飞机后立即被以虚假借口逮捕，被送进监狱。几天后他被释放，但现在他必须飞回法国，每两周见一次法官，已经持续了将近一年或更久。我认为美国对初创公司和创业精神来说真的很特别。你对法国有什么看法？你希望那里发生什么？

In August last year, Pavel Durov flew to France, got off the plane, and was immediately arrested under false pretenses, taken to jail, and put in prison. Then a couple days later he was released, but now he has to fly back to France and see a judge every two weeks for almost a year or more. I think what we have in America is really special for startups and entrepreneurship. How do you feel about France and what would you like to see happen there?

是的。嗯，我认为对于法国的科技行业来说，过去几年取得了一些进展，特别是因为它已经更接近美国和世界其他地区的一些最佳实践。

Yeah. Well, I think for tech in France, there has been some progress in the past few years, especially because it has gotten closer to some of the best practices in the US and the rest of the world.

所以现在有更多美国风投投资法国的初创公司。更多法国创始人搬到美国，参加 YC，有时又回到法国。法国还有一个疯狂的优势，就是培养出非常优秀的数学家和工程师。我记得你前几天和肖恩聊过这个。我的联合创始人就是很好的例子。他们都上过巴黎综合理工学院，是他们那一代数学方面最优秀的人，因为整个教育体系都基于数学。所以它培养出了非常优秀的 AI 工程师。实际上，如果你看看大多数 AI 公司和大型科技公司，总有一些法国人在 AI 岗位上。例如，Meta 的 Yann LeCun 就是一个著名的例子。所以我认为法国有很多优势。显然，目前的政治局势有点复杂，但如果他们能控制住，我认为他们会对整个领域产生影响。历史上，法国在开源方面有过很好的举措。有一家叫 Mistral 的法国公司，一直在推动开源，发布优秀的开源模型。所以有很多机会，我希望从长远来看，他们能为整个领域做出贡献。

So right now there are more American VCs investing in startups based in France. You have more French founders moving to the US, going to YC, sometimes going back to France. France also has this crazy advantage of producing really good mathematicians and engineers. I think you were talking about it with Sean the other day. My co-founders are a good example of that. They both went to Polytechnique, some of the best people of their generation in math, because the whole education system is really based on math. So it produces really good AI engineers. Actually, if you look at most AI companies and big tech companies, there are always some French people in AI positions. Yann LeCun, for example, is famous from Meta. So I think France has a lot of advantages. Obviously, the political situation is a bit complicated right now, but if they can get that under control, I think they would have a way to have an impact on the field in general. Historically, there have been good initiatives on open source. There's a company called Mistral that is French and has been really pushing the field for open source, releasing good open source models. So there are a lot of opportunities, and I hope that in the long run they can contribute to the field in general.

就像我谈到应该在国内通过 AI 构建者来分散和分布 AI 能力一样，我认为在国际上这也很重要。我觉得不应该只有一个国家能够创造和构建 AI。如果世界上任何国家都能真正为这个领域做出贡献并成为 AI 构建者，我们会过得更好。

The same way I talk about how you should find ways to decentralize and distribute AI capabilities inside countries with AI builders, I think internationally that's really important too. I feel like not just one country should be able to create AI and build AI. I think we'd be better off if any country in the world can really contribute to the field and become AI builders.

有一个有趣、对某些人来说有点可怕的趋势：现在很多最好的 AI 开源模型都来自中国。

There's this interesting, to some people kind of scary trend of a lot of the best open-source models in AI now coming out of China.

是的。这也挺令人惊讶的，因为我记得普遍的看法是，如果美国开源，中国会得到所有好东西，但他们不会开源自己的模型。而实际发生的是，中国在开源他们所有的模型，而美国却在闭源。

Yep. That was kind of surprising too because I remember the common wisdom was if the US open sources, China will get all that great stuff, but then they won't open source their models. What's ended up actually happening is China's open sourcing all their models and the US is keeping closed source.

是的，这非常令人惊讶，因为如果你看 AI 的早期，从 2016 年到 2022 年，美国的 AI 是极其开放的。你看《Attention Is All You Need》、Transformer，对吧？T 就是 ChatGPT 里的 T。所有这些都是开源的，而且在很多方面，正是开源和开放科学让生态系统得以繁荣发展，达到现在的水平。如果 Google 没有开源 Transformer，OpenAI 就不可能基于它构建 ChatGPT。所以它创造了这种更快的相互叠加、公开的模仿，让美国主导了 AI。但大约在 2022 年，发生了一些事情。出于某种原因，人们开始有点害怕开源，人们开始从 AI 中赚更多钱，所以他们把梯子收起来了。

Yeah, and it's super surprising because if you look at the early days of AI, from 2016 to 2022, AI in the US was extremely open. You look at 'Attention is All You Need', Transformers, right? The T is like the T in ChatGPT. All of these were open source, and in many ways, the fact that it was open source, open science allowed the ecosystem to thrive and flourish and get to where we are. If Google didn't open source Transformers, OpenAI couldn't have built ChatGPT based on that. So it created this faster building on top of each other, emulation in the open that really allowed the US to dominate AI. But something happened around 2022. For some reason, people started to be a little bit scared about open source, people started to make more money from AI, and so they pulled up the ladder.

是的。美国这边变得更封闭了，对吧？作为反应，或者也许同时，中国那边变得更加开放，开始在开放科学和开源方面做出更多贡献。

Yeah. Things became more closed in the US, right? And in reaction, or maybe at the same time, in China they became much more open and started to actually contribute more in terms of open science, open source.

你知道为什么会这样吗？

Do you know why that happened?

如果你看中国整体，他们一直非常关注开源。这不太为人所知，但他们一直是开源的重要贡献者，作为从中国外部贡献的一种方式。我认为激励机制与美国有些不同，这使得更多参与者能够分享开源。但我认为随着他们看到开源的潜力和影响，这种情况逐渐加剧。我记得大约两年前，已经有很多中国组织和研究人员在 Hugging Face 上分享模型，但没有得到很多关注和 traction。但渐渐地，人们开始注意到。现在，每当中国在 Hugging Face 上发布一个好模型，影响都是巨大的。例如，你看到了 DeepSeek。DeepSeek 现在在 Hugging Face 上有超过 10 万粉丝。所以到现在，他们真的明白了，当你以初创公司、组织甚至国家的身份发布开源时，你可以产生巨大的影响。这反过来又强化了他们继续这样做的动力和热情。这种模仿：我们从少数组织分享开源，到现在中国可能有 30、40、50 个非常好的组织和公司在分享非常好的模型，不仅限于文本，还有音频、视频、图像，很多不同的领域。

If you look at China in general, they've always been quite focused on open source. It's not so known, but they've always been big contributors to open source, as a way to contribute from outside China. I think the incentives are a little bit different than in the US, which allows more players to share in open source. But I think it intensified progressively as they've been seeing the potential and the impact of it. I remember maybe two years ago, we already had quite a lot of Chinese organizations and researchers sharing models on Hugging Face that weren't getting a lot of visibility and traction. But progressively, people started to notice. Now whenever a good model is released from China on Hugging Face, the impact is massive. You've seen that with DeepSeek, for example. DeepSeek has over 100,000 followers on Hugging Face now. So by now, they've really understood that when you release in open source as a startup, an organization, or even a country, you can have a massive impact. And so it reinforces their motivation and excitement to keep doing so. The emulation: we went from a few organizations sharing in open source to now probably 30, 40, 50 really good organizations and companies in China sharing really good models, not only for text but for audio, video, image, really a lot of different domains.

我希望美国能回到最初的开源和开放科学理念。我们开始看到这方面的一些进展。埃隆·马斯克和 xAI 几周前在 Hugging Face 上开源了上一代 Grok。

I hope that the US will get back to its initial philosophy of open source and open science. We're starting to see some progress in that direction. Elon Musk and xAI open sourced the previous generation of Grok a few weeks ago on Hugging Face.

那是 Grok 3 吗？

Was that Grok 3?

Grok 3。是的，在 Hugging Face 上，这太棒了。OpenAI 去年夏天在 Hugging Face 上发布了他们一段时间以来的第一个开源大语言模型 GPT-OSS，它被 AI 构建者大量使用和采用。像 AI2 这样的组织，不仅发布开源模型，还发布开源数据集、模型训练方式以及训练模型的脚本，所以是完全开源的。Nvidia 也在做越来越多的开源。他们过去一年平均每天发布超过一个新模型或数据集。所以过去一年他们发布了超过 400 个模型和数据集。所以美国有越来越多的组织在做开源。美国政府非常支持开源。这太棒了，因为上一届政府并非如此。

Grok 3. Yeah, on Hugging Face, which is amazing. OpenAI released their first open-source LLM in a while on Hugging Face last summer, GPT-OSS, which is seeing a lot of usage and adoption from AI builders. You have organizations like AI2, which is not only releasing open source models but also open source datasets, how the models are trained, and scripts on how they trained their models, so really fully open source. You have Nvidia that is doing more and more open source. They released, I think on average in the past year, over one new model or dataset per day. So over 400 models and datasets that they released in the past year. So you're starting to have more organizations in the US doing open source. The US administration is extremely supportive of open source. It's something that is amazing because it wasn't really the case with the previous administration.

但他们确实发布了一项行政命令和一项 AI 计划，其中 AI 行动计划的第三点是促进更多的开放权重和开放数据集。所以政府里有很棒的人，比如之前来自安德森·霍洛维茨的人，在推动这件事。这对整个领域非常重要且有益。希望我们能回到美国更多参与开源 AI 的世界，这非常重要，因为如果不这样做，想到很多美国初创公司、企业、非营利学术机构和科学组织可能建立在中国的开源基础上，这有点奇怪和可怕，对吧？而这正是现在正在发生的事情。安德森·霍洛维茨的一位合伙人最近说，他们投资组合中许多正在构建 AI 的公司都基于中国的开源模型，这显然是一个风险。美国的活力建立在中国的模型之上。

But they did an executive order and an AI plan where the third point of the AI action plan was to foster more open weights and open datasets. So you have really great people in the administration, like someone who was at Andreessen before, pushing for that. This is really important and great for the field. Hopefully we can get back to a world where the US is doing more open source AI, which is very important because if it's not happening, it's a bit weird and scary to think that a lot of current American startups, companies, nonprofit academic, and science organizations could be built on Chinese foundations, right? Which is what's happening now. A partner from Andreessen said recently that many of their portfolio companies building AI are building on top of Chinese open source models, which is obviously a risk. American dynamism built on top of Chinese models.

是的，完全同意。这是个问题，因为如果你考虑偏见，很多文化偏见已经嵌入到开源基础中。所以美国公司构建的产品会融入其中一些偏见。如果你考虑控制，这为中国创造了更多控制权。而且如果你考虑整个领域的发展，基础进化得越快，整个领域进化得也越快。现在中国开始在开源 AI 领域占据主导地位，那么认为几个月后他们可能开始在整体 AI 领域占据主导地位也不疯狂，因为这加速了进步的速度。所以对我来说这是一个非常重要的话题。我认为美国应该更多地投资于开源 AI，真正激励开源模型和数据集，以改变当前的趋势。

Yeah, absolutely. Which is a problem because if you think about biases, a lot of cultural biases are embedded into the open-source foundations. So American companies' products will integrate some of these biases. If you think of control, it creates more control for China. And if you think about the development of the field, the faster the foundations evolve, the faster the whole field evolves. Now that China is starting to dominate in open source AI, it wouldn't be crazy to think that in a few months they might start to dominate in AI in general, because it accelerates the velocity of progress. So that's a very important topic for me. I think the US should invest more in open source AI, really incentivize open source models and datasets to change the current trends.

如果你看看机器人技术这样的领域，我想你会对趋势有早期了解。我知道有很多公司筹集了大量资金来建造人形机器人。但我认为很多 AI 领域最聪明的人正在考虑建造不一定采用人类形态的机器人公司。你看到了什么？你认为它会如何发展？

If you look at something like robotics, I imagine you get an early view on the trends. I know there are companies that have raised huge amounts to build humanoid robotics. But I think many of the smartest people in AI are thinking about building robotics companies that don't necessarily use the human form factor. What are you seeing? How do you think it will evolve?

嗯，我认为我们正处于机器人 AI 的早期阶段，有几个非常令人鼓舞的趋势。现在建造好的机器人硬件比以前容易得多，由于多种原因成本也更低。而且为这些硬件配备出色的软件和 AI 能力也变得越来越容易。因此，更好的硬件和更好的软件汇聚在一起，将共同开启机器人技术的新范式。一个挑战是，在这个早期阶段，很多计划都是孤立的。几乎每个机器人实验室都创建自己完全不同的软硬件栈。所以没有像 AI 其他领域那样的协作，这大大拖慢了领域的发展。这就是为什么在 Hugging Face，我们推动更多的开源、共享、标准和最佳实践。我们有一个名为 LeRobot 的库，它已经成为最流行的机器人库之一，用于为任何硬件启用新数据集和模型。我认为这样的举措可以解锁机器人技术的更快进步。如果每个人都开始协作并分享学习成果，这个领域将会加速，从而迎来机器人技术的 ChatGPT 时刻，而 ChatGPT 正是 AI 领域协作的产物。这就是我们对机器人技术的期望。

Well, I think we are in the early stages for robotics AI with a couple of really encouraging trends. It's much easier than before to build really good hardware for robotics, making it more affordable for multiple reasons. And it's becoming easier to have great software and AI capabilities for this hardware. So you have this confluence of better hardware and better software that together will enable a completely new paradigm for robotics. One challenge is that at this early stage, many initiatives are siloed. Almost every robotics lab creates their own software and hardware stack that is completely different from their neighbors. So there isn't the collaboration you see in the rest of AI, which slows the field down. That's why at Hugging Face, we push for more open source, sharing, standards, and best practices. We have a library called LeRobot, which has become one of the most popular robotics libraries for enabling new datasets and models for any hardware. I think such initiatives could unlock faster progress for robotics. If everyone starts to collaborate and share learning, the field will accelerate, leading to a ChatGPT moment for robotics, which was the product of collaboration in AI. That's what we hope for in robotics.

最后一个问题。你克服过的最困难的事情是什么？

Final question. What's the hardest thing you've overcome?

这是个好问题。在创业旅程中，创始人和企业家总会遇到困难。有时人们会美化事物，认为一切都充满魅力和乐趣，但事实并非如此。对我个人影响最大的事情更多是与人员相关的。当我们一起工作了一年、两年、五年甚至更久的团队成员决定去迎接不同的挑战，创办自己的公司，离开 Hugging Face 大家庭或组织时，这对我来说总是很难。我已经学会接受这一点，尤其是因为我看到很多人离开后在外面建立了很酷的东西，有时还会从外部做出贡献，保持联系和连接，因为 AI 世界很小，每个人都在合作。但对我个人作为创始人来说，最困难的事情就是当人们不得不离开或者我们不得不让人离开的时候。那对我来说是最具挑战性的部分。

That's a good question. There are always hard things for founders and entrepreneurs in the startup journey. Sometimes people idolize things and think it's all glamour and fun, which is not true. The things that impact me the most personally are more people-related. When team members we've worked with for a year, two years, five years, sometimes even more, decide to go on to a different challenge, start their own company, and leave the Hugging Face family or organization, that is always tough for me. I've learned to accept it, especially because I've seen many people go on to build really cool stuff outside Hugging Face and sometimes contribute from the outside, staying in touch and connected because the AI world is small and everyone collaborates. But for me personally as a founder, that is the hardest thing—when people have to leave or when we have to let people go. That's the most challenging part for me.