Mati Staniszewski 谈 ElevenLabs、语音 AI 及构建 AI 的通信层

那里有很多优秀的公司。

Lots of great companies in there.

我们真正想成为的是，如果你想想未来三年或者更长时间，可能会有三个平台，你所有的互动都发生在这些平台上。

What we would love to be is if you think about the next three years or however many years, there will be probably three platforms that you set all the interactions on those platforms.

我们想成为这些互动和沟通的领先平台。

We want to be the leading platform for those interactions, for that communication.

等等，解释一下。

Wait, explain that.

这整个思路都是围绕那个主题展开的。

The whole way that kind of builds on that theme.

你如何与内容互动，如何与数据互动，这些方式都会改变。

How you engage with content, how you engage with data company will change.

我们希望帮助人们和公司拥有一个统一的地方，一个平台，你可以在那里进行设置。

We want to help people, companies have one place, one platform where you can set it up.

你可以设置你的集成，带来你的知识，带来你的品牌，带来公司的资产，然后在整个客户旅程中部署这些。

You can set up your integrations, you can bring your knowledge, you can bring your brand, you can bring your assets from the company, and then deploy that for entirety of customer journey.

不管是营销场景，你通过内容讲故事，把用户的知识收回来，还是销售场景，你想跟客户互动、把产品带给他们，或者在整个客户旅程中提供支持，又或者更简单一点，当你想了解客户旅程时，旅程中每一次符合品牌的互动，我们都希望能捕捉到，并帮你把它做得更好。

Whether that's in the marketing example, where you tell that story through content, capture that knowledge back about the user, whether that's in sales when you're trying to engage and bring your product to the customers, whether you are supporting that across the customer journey, or maybe in a simpler way, as you are trying to understand the customer journey, every on brand interaction that that journey has, we would love to be able to capture and help you make that better for.

好的。

Okay.

那我们具体一点说吧，因为我还是想把这个搞清楚。

So let's make this concrete because I still, I I want to be able to wrap my head around this.

就拿你整合得最深的客户来说吧。

Let's take your, like, most deeply integrated customer.

对。

Yes.

对吧？

Right?

比如说，高盛。

And say, you know, it's Goldman Sachs.

我随便说的。

I'm making this up.

这就是他们如何使用这个产品，以及他们使用的所有不同产品。

And this is how they use this product, all the different products they use.

比如，哪家公司或者其中一家与你们公司合作最深入，请解释一下他们使用的所有由你们技术驱动的产品套件。

Like who is the company or one of the companies that is most deeply integrated with your company and explain how all the different suite of products that they're using that are powered by your technology.

德国电信就是一个很好的例子。

Deutsche Telekom is a great example.

他们会利用很多音频功能来制作播客，以及他们的应用和Magenta。

They will use a lot of the audio work to create a podcast and their app and the Magenta.

他们会用这些来做广告，这样他们就能创建广告并分发给他们的受众。

They will do that for the ads so they can create ads and distribute that to their audience.

他们会利用我们很多11创意功能来实现这一点。

There will use a lot of our 11 creative work to be able to do that.

然后他们会利用我们的11代理功能，用于呼叫中心，当你打电话进来，你想要帮助时。

Then they will use our 11 agents work for in call center where you call in, you want to help.

你会通过语音代理获得帮助，这个代理整合了德国电信的知识。

You get that help through the voice agent integrated with the knowledge from Deutsche Telekom.

这样就能确保，如果有人打电话来了解产品，他们能很快得到信息。

So you can make sure that if somebody is calling in to learn about the product, they get information very quickly.

如果他们打电话是为了咨询退款或最近的账单问题，他们也能得到帮助。

If they are calling to get support on how to get a refund or the recent billing request, they can get that help.

而且最近，他们甚至在网络内部部署了一个智能助手。

And then more recently, they even deployed an agent inside of the network.

如果你是T-Mobile的用户，打电话时可以让助手加入通话，帮你安排预约，或者实时翻译你和对方的对话。

If you are a T Mobile subscriber, if you call, you can ask agent to join the call and help you out, schedule a booking or real time translate your conversation to the other person.

他们实际上会结合我们的智能助手技术，加上实时翻译功能，来实现跨语言沟通。

They'll use our effectively combination of agents work and bringing that real time dabbing to be able to communicate in other language.

在整个过程中，所有信息都会被记录下来，这样你就能了解客户是如何与品牌互动的，涵盖整个旅程，从营销、支持，到更广泛的运营和销售，我们帮助他们完成所有这些。

Across all of that, all of that information is captured back so you understand how customers engage in with, this case, with the full spectrum of the journey, marketing, the support, and then the wider operations, unit sales, and we help them do all that.

好的。

Okay.

那么在这种情况下，因为你提到了收入大幅增长，而且增长非常快。

So in a situation like that, because I think you mentioned obviously a huge increase in your revenue that's growing really fast.

你们是不是在瞄准更大的公司？

It's like you're targeting bigger companies, right?

德国电信的第一个产品是什么？

What was the first product Deutsche Telekom?

德国电信，抱歉。

Deutsche Telekom, sorry.

对，第一个是营销。

Yeah, the first was marketing.

那是三年前的事了。

So it was three years ago.

然后大概两年前，大多数智能体真的非常非常好。

Then kind of the most of them that two years ago, most of the agents were really very, very good.

它们不太可靠。

They weren't very reliable.

它们不够快。

They weren't very quick.

所以营销是最明显的选择。

So marketing was the obvious one.

他们就从那里开始。

So they started with that.

他们部署了那个方案。

They deployed that.

内容质量很高，人们能通过播客参与进来。

The quality of the content was great so people could engage with the podcast.

实际上，就是你提到的那个用例，把笔记带进来，然后能大声朗读出来。

Effectively, the use case that you mentioned of bringing your notes and then being able to read them out loud.

他们每天为所有客户制作这样的内容，让他们通过ElevenLabs的声音听播客了解世界动态。

They were just creating that daily for all the customers so they could read about what's happening in the world through the podcast with ElevenLabs Voices.

这就是第一个用例。

So that was the first use case.

然后当然，支持是语音和集成的结合。

Then of course support is a combination of voice but also the integrations.

那说说你是怎么把他们从一个产品扩展到下一个产品的。

Well talk about how you expanded them from one product to the next.

你具体是怎么做到的？

How do you actually do that?

在这种情况下，主要是我们怎么和他们合作，帮他们把那个产品做出来。

In that case, main and I guess mostly how do we partner with them to help them bring that product to life?

对，他们从第一个开始，但你们有十种不同的产品可以卖，那你们是怎么从一到二，再从二到五，这样一步步扩展的？对，就是这样。

Yeah, they start with one, but you have 10 different products So you can sell how do you go from one to two and then two to five and so on Yeah, and so

我觉得关键的一点是，首先你当然要部署第一个产品，确保它确实能带来价值。

I think the main thing is one, you of course deploy the first one, you'll make sure that there is value behind that.

所以在和任何客户合作的过程中，我们尽量确保不只是验证概念，而是证明影响、证明价值，只有在那之后，我们才尝试让公司更广泛地和我们合作。

So across any customer engagement, we try to make sure that we don't only prove concepts, prove the impact, prove the value, and only after that we try to get the companies working with us at a broader scale.

我们在营销方面证明了影响。

We prove the impact on the marketing side.

要让支持业务运转起来，你真正需要做的不仅仅是音频，你还需要建立集成。

To get the support going, what you actually need to do is not only the audio, you need to build integrations.

所以你需要把它和德国电信使用的所有CRM系统连接起来。

So you need to connect it with all the CRM systems that Deutsche Telekom works with.

你还需要在输出端做好集成工作。

You need to work on integrations with the output.

那么怎么把它连接到电话系统，比如SIP中继或者Twilio呢？

So how do you connect it to the phone system, SIP trunking or Twilio?

你要怎么确保这种连接是存在的？

How you make sure that that connection exists?

所以你会花很多时间在集成上，确保他们的逻辑被尊重，比如当你接起电话时，代理的行为符合你的预期。

So you would spend a lot of time on integrations, making sure that their logic is respected on how, when you do pick up the phone call, the agent behaves the way you want it to behave.

所以在这里，我们的前部署工程师会和团队合作，花时间在德国，和Bonn一起并肩工作，共同完成集成，确保代理按照你想要的流程走，并且拥有它应该具备的知识。

So here our forward deployed engineers would partner with the team, spend the time and Bon in this case in Germany, working through side by side together on the integration, just being able to making sure that the agent follows the flow that you want, has the knowledge that it should have.

然后在这个阶段，任何语音代理工作中最难的部分就是实际部署，也就是你如何真正测试它是否有效。

And then in that step, hardest thing in any voice agent work is actual deployment of how you actually test that it works.

所以你需要先用模拟的方式试运行，验证通话，然后在小范围内进行测试，接着再……

So then you need to trial that with initially simulated that, you verify the calls, then you do it at smaller And you're

你们是用全职员工来做这些的吗？

doing this with FTEs?

是的。

Yes.

我们所有这些都用全职员工来做，然后部署，当然之后会逐步扩展。

We do all that with FTEs, then we deploy, and then of course we scale over time.

你是什么时候意识到，用全职员工来做是大幅增加收入的一个重要途径？

When did you realize that a huge path to greatly increasing your revenue was doing FTEs?

在我回答这个问题之前，当然还有最后一点，就是当你部署之后，工作并没有结束，你还需要持续评估、监控，并随时间调整行为。

Before I answer this, and of course the last part is, as you deploy, the job isn't done, you still want to continue evaluate, monitor, and refine the behavior over time.

这适用于所有的用例。

That applies across all of the use cases.

NFD在这方面也很不错。

NFDs for that are great too.

我以前在Palantir工作，所以我们一直很重视如何与客户合作。

So I used to be at Palantir, so it always felt like a big thing of how we want to work with customers.

我几乎不太喜欢用“客户”这个词，因为很多时候我觉得他们都是合作伙伴，大家是一起在解决同一个问题，共同努力去攻克它。

It's almost like don't like the word customers because in many ways I feel like all of them are partners where you are working together on the same problem and try to solve that.

所以从一开始，FDE的理念就贯穿其中，就是要具体化，和合作伙伴并肩作战，一起解决他们的问题。

So there was a lot of the philosophy of the FDEs was there from the start to make it specific, working with the partners side by side on their problem

这个“他们”真的很有意思。

and This their is really interesting.

那你跟我说说，你在Palantir观察到哪些做法是有效的？

So tell me what you observed that was working at Palantir.

你当时就想，如果我自己创业，我也要这么做。

You were like, Oh, if I start my company, I want to do that too.

为什么？

Why?

我觉得最了不起的地方在于，从一开始你就应该站在客户那边，对他们的难题充满热情，并且深入了解他们。

I think the most incredible thing was that you were meant to be on the same side of customer obsessed of their problem and try to understand them deeply from the beginning.

在去Palantir之前，我曾在BlackRock工作。

So I used to work at BlackRock before Palantir.

我在BlackRock的时候，刚入职那会儿，因为合规和安全的要求，头一两个月，你发邮件之前，团队要先审核，还要培训，然后才能发出去。

BlackRock, when I joined, given the compliance security, the first month or two, when you send an email, it gets verified by your team, it gets trained before you send it externally.

那流程挺复杂的，就是为了确保邮件没问题。

It's like a pretty detailed flow of making sure that it's good.

但在Palantir，我入职培训完第一个月，我就想，好了，你现在得去了解客户，跟他们一起工作。

In Palantir, when I joined after the onboarding in the first month, I'm like, Okay, you now need to work and understand a customer.

你得飞到阿伯丁，去北海，跟他们并肩工作，了解情况，然后把经验带回来。

You're flying to Aberdeen, to North Sea, to work side by side with them, understand what's happening, and then bring it back.

这对我来说完全是震撼，这种方法和文化，我几乎从没在BlackRock那边跟客户打过交道。

And that was a complete shock for me and the approach and culture of like, I have never almost interacted with a customer on the BlackRock side.

现在入职才几周，就要去那儿，跟他们待在一起，简直不可思议。

Now here I am in the first weeks, meant to go there and be next to them, which was crazy.

我觉得这很棒。

I thought it was great.

我觉得就是这种心态，每次你都要去前线，我姑且这么叫，去前线跟他们合作，把知识带回来，搞清楚真正的问题是什么，哪些问题能解决，然后真的去解决它。

I thought it was that mentality and every time you are going to be there on the, I'll call it frontline loosely on the frontline to work with them and bring that knowledge back, understand what is actually the problem, what's fixable problem, then actually fix it.

这种思维方式在ElevenLabs现在也非常适用，我们所有的FDE和市场团队都在琢磨，怎么才能真正和客户在一起，理解问题并一起解决。

That mentality is so true now at ElevenLabs too, where all of our FDEs, our go to market team too, is going to obsess of like, how can I actually be there with the customer, understand the problem and work with that?

我觉得还有其他一些公司也很不错。

There was other companies that I think were great.

那些小团队，通常有小型部署团队来处理很多这类事情。

The small teams, they had usually small deployment teams that worked on a lot of that.

最好的想法胜出。

The best idea wins.

这个概念在小团队里非常真实，你很快就能汇集你认为对那个客户最好的知识，并且你有权力去实施它。

The concept was very true where in that small team, you very quickly assembled the best knowledge of what you think should be done for that customer and you had power to then actually enact on it.

定义一下小。

Define small.

大约五个人在Palantir就算大了，在ElevenLabs也会相对算大。

About five people is considered in Palantir big and in ElevenLabs is going to be also relatively big.

不过关于FDE，我还想多说一点，因为现在你可能看到每家公司都在做FDE，而在我们这里，FDE是产品团队的一部分。

I will tell you something more though on the FDE side because of course now you probably see every company doing FDEs and in our case, FDEs are part of the product team.

他们不属于市场团队。

They are not part of the go to market team.

他们是产品团队的一部分。

They are part of the product team.

他们深入了解产品的路线图。

They are deeply embedded on understanding what's the roadmap of the product.

但还有第二个原因，一是你希望所有FDE真正解决客户问题，并推动产品朝那个方向发展。

But there's a second reason, which is one, you want all the FDEs to actually solve the customer problem and stretch your product in that direction.

第二件你想让他们做的事，就是把那些知识带回产品。

But second thing you want them to do is bring any of that knowledge back to the product.

这样产品对下一代在其上构建的公司会变得更好。

So the product becomes better for the next generation of companies building on top of it.

我觉得第二部分经常被忽视。

And I feel like the second part is frequently missed.

首先，好吧，如果这些人有效地做了很多艰难的集成工作来解决问题。

It's like first, okay, if these are doing effectively a lot of the hard integration work to solve the problem.

但如果你不做第二步，也就是真正学到专业知识并把它带回产品，那实际上就只是服务而已。

But if you don't do the second step of how you actually learn the expertise and bring it back to the product, then it's effectively just services just the

等等，关于第二部分再多说一点。

Wait, say more about that second part.

所以你是通过反复试验、和他们紧密合作来获取信息的。

So you're gleaning information through trial and error, working closely with them.

然后你可以把这些知识带回来，传播给你其他的成千上万的客户。

Then you can take that back and spread that knowledge across the other tens of thousands of customers that you have.

是的，就是这样。

Is Yes, that

你说的正是这个意思。

what you're talking exactly.

好。

Okay.

没错。

Exactly.

而且这是一种产品知识，比如你怎么操作，你知道的，有些简单的事情。

And it's a product knowledge of like how you operate with, you know, there are simple things.

比如说你建了一个集成。

Let's say you build an integration.

那我怎么让这个集成对所有人都可用呢？

How do I now have that integration available to everyone?

这是那个的简单版本。

That's a simpler version of that.

但比如说你在医疗领域工作。

But let's say you are working in healthcare space.

在医疗领域，如果你部署了工作，那么你想为合适的用户优化产品体验。如果你不是作为第二个客户来研究，你要确保你有……

In healthcare space, if you deploy the work, then you want to optimize the product experience for the right If you are not studying as a second customer, you want to make sure that you have

等等，你是在把你的客户群当作研发，另一种形式的研发。

So wait a minute, you're using your customer base as almost like R and D, another form of R and D.

你确实通过了解你所在的领域来加速你的研发。

You are definitely accelerating your R and D through understanding the domain that you're working with.

这是因为他们的问题

It's Because their problems

很少有一个公司的问题是独一无二的。

are not very rarely a company's problems is unique to that company.

没错。

That's right.

这其中的好处就是大家都能受益，因为你跟一个客户合作，学到东西，然后把它融入到产品体验中。

And the beauty of that is like kind of everybody benefits because you work with one customer, you learn, you bring that into the product experience.

再跟另一个客户合作，又学到东西，再把它带回产品体验中。

You learn with another one, you learn, you bring it back to the product experience.

这两个客户都因为产品被优化得更好而受益。

Both of those customers benefited from having the product optimized for the better work.

他们拥有的领域专业知识仍然是他们自己的。

The domain expertise they have is still theirs.

他们可以凭借这些专业知识取胜。

They can win based on the domain expertise.

但是基于领域专长来构建产品的体验，是可以抽象化、可以复用的。

But the product experience of how you build based on that domain expertise can be abstracted, can be reusable.

你在Palantir工作还学到了什么？

What else did you learn working at Palantir?

他们也是一个非常不看重头衔的组织，我们在ElevenLabs也延续了这一点，这确实有助于实现“最佳想法胜出”的理念，大家都能感受到，我觉得我在模仿你——

They also were very much of no title organization, we carry over at ElevenLabs, which does help in that best idea wins approach where people do feel, I feel like mimicking you on

那件衬衫。

that shirt.

我也想到了同样的事情。

I thought the same thing.

就像，你不必，如果你不想——

Was like, you don't have to if you don't

不想的话。

want to.

我知道这完全是巧合。

I know it's a complete coincidence.

我们要把这段留下来。

We're leaving that part in.

没有头衔，组织相对扁平，层级很少。

No titles, relatively flat organization, very few layers.

比如，在我和最高管理层之间，我在DPO之前在那里工作，只有四到五层，甚至更少。

It's like, between me and the C suites, I worked there before the DPO, it was four, five steps or less.

所以你总能感觉到距离很近，我记得实际上可能是四层或三层。

So you always felt like a proximity between, I think it was like four or three actually.

所以合作起来很方便，距离很近。

So it was a good proximity of being able to work together.

在ElevenLabs，我们现在最多有五层，这是最大深度。

At ElevenLabs, we have a couple of five today, like maximum depth of how many layers there should be.

希望随着时间的推移，层级会减少，如果AI能像我们希望的那样帮助运营公司，就不会再有更多层级了。

Hopefully over time, actually there will be fewer, no more layers if AI helps you run the organization the way we would like to run.

好，多讲讲这个。

Well, say more about that.

在你们现在的管理方式中，有哪些事情是你希望未来AI能帮你改变的？

What can you not do today in the way you run your organization that you hope AI can change in the future?

我觉得你总是希望从最接近问题的人那里获取信息。

I think there is always you want the information from of the person working as closely to the problem.

比如说，你在开发一个产品，你想直接和开发这个产品的工程师交流，而不是通过他们的经理。

So let's say you're developing a product, want to work, you want to speak with the engineer that actually develops that product rather than their manager.

如果是客户对话，你也想直接从一线人员那里了解客户说了什么，而不是让经理总结后再告诉你。

If it's a client conversation similar, you want to understand what is that client saying from the person on the ground, rather a manager summarizing that information and giving that to you.

我觉得AI会改变这种信息流动的方式，让你能以更精细的粒度获取所有正在发生的事情，比以往任何时候都更全面。

I think that information flow will change with AI where you will have access to everything that's happening in a better granularity than you could ever have before.

它会自动总结这些来回的交流。

It summarizes that back and forth.

其次，我觉得在ElevenLabs，大多数人大概会有接近10个直接下属。

Second, I think that in general, roughly at ElevenLabs, most people will have close to 10 direct reports.

这是一个相当宽的管理跨度，这有助于我们保持那种小团队的工作方式。

It's a pretty wide blight set which helps us build that small team approach that we have.

而且这也只有在你能放大各个团队、所有同事之间发生的事情，总结他们团队的情况、表现如何、有哪些差距时才行得通，这确实有帮助，也能帮你从被动应对转向主动出击。

And that too only works if you can amplify a lot of what's happening across all the teams, all the people that you work with, to summarize information of what's happening in their teams, how this is performing, what are some of the gaps, that definitely helps and helps shift it from reactive to proactive, I think too.

我的意思是，过去很多时候，你得靠特定的人把信息反馈给你。

What I mean by this is frequently in the past, I think you would rely on specific individuals surfacing the information to you.

现在，想想ElevenLabs和我们运营的方式，你可以总结所有正在发生的事情，所有关于什么有效、什么无效的数据。

Now, as we think about ElevenLabs and how we run, you can summarize all what's happening, all the data of what's working, what's not.

然后获取那个信号，我得到那个信号，就能主动介入我认为有问题的地方，这很有帮助。

And get that signal, I get that signal, and I can proactively engage on where I think it's not working, which helps.

这当然对全公司每个人都有帮助，因为无论你在哪里，你都有同样的能力。

It helps, of course, everybody across the company because you have that same ability wherever you are.

我想最后一点，和前面这些紧密相关，也是为什么今天这一切能成为可能。

I think that maybe the last part, which is very related to those, to why it's even possible today.

我们选择对很多数据和文档保持高度透明。

We took a choice to be extremely transparent with a lot of the data that we have, a lot of the docs that we write.

所以几乎每个人都能接触到公司里所有的文档，这最终帮助你建立一个系统，让你真正能利用那些知识。

So everybody has almost access to all the docs that are there in the company, which ultimately helps you create that system where you can actually tap into that knowledge.

好的，我想让你多谈谈这个，因为我觉得这是我想聊的话题之一，因为你相对年轻。

Okay, so I want you to say more about this because I think this is one of things I want to talk to about because you're relatively young.

这是你的第一家公司。

This is your first company.

你创办的第一家公司，你马上就进入了人工智能领域，然后你正好处于这次大爆发的完美位置。

Your first company you built, you got into AI right away, and then you were perfectly positioned for this huge explosion.

显然，人们谈论人工智能已经谈了八十年了，但你是在正确的时间、正确的地点，而且我认为，基于公司目前取得的成就，你拥有正确的技能组合。

Obviously, people have been talking about AI for eighty years or whatever, but you're at the right place, right time with, I think, the right set of skills based on what the company built so far.

那么，你认为未来你管理组织的方式会和现在有多大的不同？

So how different do you think in the future the way you run your organization can be than what it is today?

我认为小团队的方法肯定会在未来许多我认为会创建的公司中继续存在。

I think the small teams approach will definitely be there across many of the companies that I believe will be created in the future.

这也有助于一个非常不同的方面，比如，如果你只是考虑采用人工智能技术，在我们的情况下，它不一定是自上而下的指令，比如你必须使用这项技术来变得更好，然后团队去赋能其他人。

It also helps with a very different piece, which is like, if you just think about adopting AI technology, in our case, it doesn't have to be a top down mandate of like, you need to use this technology to get better and then the teams enable others.

由于团队相对较小且独立，人们会自下而上地采用他们认为最好的方法，并且能够立即开始使用。

Given relatively small and independent, people bottoms up adopt what they think is best and are unable to run with it straight away.

所以在这方面帮助很大。

So that helped a lot in that sense.

我会把它交给其他人，我们对此非常坚信。

I'll give it to others which we believe strongly on.

第一，在我们所有的小团队中，甚至包括非技术团队，我们都会引入工程人才。

One, in all the small teams that we have, in all the teams that we have, we bring even non technical teams, we bring engineering talent in those teams.

所以我们的招聘团队、运营团队、法务团队，都会有工程师来帮助自动化一些工作，同时提升其他人使用AI的水平。

So our talent team, our ops team, our legal team, all of that will have engineers that help both automate some of the work, but also elevate everybody else in how they are using AI.

我认为这将来会成为公司间更普遍的模式，越来越多的公司，无论大小，都会尝试给所有团队注入工程资源，让他们变得更聪明。

I think that will be a bigger pattern in the future across the companies too, where I think increasingly some of the biggest and smallest companies will try to infuse all teams with engineering resources so they can get smarter.

在ElevenLabs，有没有什么职能是集中管理的，然后各个小团队可以共享使用？

Is there any function that is centralized at ElevenLabs that then all these small individual teams can then access?

我和Benning Spoons的Luca Ferrari聊过。

Spent some time with Luca Ferrari of Benning Spoons.

他跟我分享了一个大胆的想法，他说，人力资源其实非常重要。

And he had this wild idea to me where he's just like, Well, HR is actually really important.

他觉得技术人员看不起人力资源这件事很荒谬。

He thinks the fact that tech people are dismissive of it is ridiculous.

然后他说，我的HR团队大概有50个人，而且全是工程师。

And he's like, But my HR is like 50 people and they're all engineers.

然后就是一个集中的人力资源部门，其他所有公司，因为他拥有不知道多少家公司，全都用这个部门。

And then it's like one centralized HR that every single other, because he owns I don't know how many companies, they all utilize it.

但就是大概50个，我记得是50个工程师在HR部门。

But it's just like 50, I think, 50 engineers in HR.

然后他旗下的各种结构、团队和公司都从这个部门获取资源。

And then all the different structures, teams, and companies he has like tap into that for that resource.

当然，有些职能是集中管理的。

Of course, there are centralized functions.

法律部门就是一个很好的例子，当然，像我们怎么运作这个部门，需要非常集中的支持，帮助市场和其他团队的人学习怎么销售ElevenLabs。

Legal is a good example of course of like how we are running that needs to be very central enablement of how we help people across go to market and other teams, how they are learning the craft of how to sell ElevenLabs.

就像你说的，产品太多了。

As you said, there are so many different products.

你需要非常注意如何向特定客户传达这一点。

You need to be very particular about how you deliver that to the specific customer.

不要让客户对我们提供的产品或服务感到困惑。

It's not confusing the customer in what we are offering or what we do.

这类职能有不少。

There's a good amount of those functions.

但它们也都和Luca的做法类似，都会包含大量的工程工作，以及如何扩展这种运营。

But they all too, similar to Luca's approach, all of them will have a good amount of engineering in them and figuring out how you scale that operation.

不是要在所有地方都招几百人，而是如何利用现有的少数人，把知识真正传播到各处。

It's not becoming hundreds of people in all locations, but how you can use the few people that you have and really bring that knowledge everywhere.

对我们来说，最大的一个领域可能是营收运营，也就是如何让收入工程有效运作。我们有很多工具可以帮助你获取知识，比如在对话中转录代码，然后把它带回来，让你更容易填写字段，这样你就能捕捉知识，而不必花手动时间。

One of the biggest ones for us is probably around rev ops of how you like the revenue engineering effectively function we have so many tools that help you get the knowledge whether it's code during the conversations transcribed bring it back make it easier for you to fill on a field so you capture that knowledge and then don't have to spend manual time.

当然，我们也越来越多地使用自己的产品，所以我们在尝试如何创建AI代理来替代部分工作，但仍然在团队内运作。

And then of course increasingly we dogfood a lot of our own work so trying to figure out how we can create AI agents to help replace part of that work but still operate within the team.

一个很好的例子就是AI销售开发代表。

A good example is AI SDR.

现在在ElevenApps网站上，人们可以填写下拉表单，留下公司信息，但也可以和语音助手对话，用更快、更不同的方式留下信息。

People going on the website today on ElevenApps, you can fill in the dropdown form and leave information about your company, but you can also speak with voice agent and leave that in a quicker and different way.

我们看到很多人选择这种流程。

We've seen a lot of people go for that flow.

然后有趣的是，会发生两件事。

And then interestingly, two things happen.

第一，人们既让事情变得更简单。

One, people are both making their things easier.

他们更享受这个过程。

They enjoy it more.

第二，他们留下的信息比填下拉表单时多得多。

But two, they leave a lot more information than they ever would for the dropdown form.

他们告诉我们很多关于问题、更广泛的用例，这样我们能更好地连接他们。

They tell us so many about the problems, about the wider set of use cases so we can connect them better.

这真的很有意思。

That's really interesting.

所以因为他们是说话而不是打字，你就能得到更多信息？

So because they're speaking and not typing, you get more information?

百分之百。

100%.

就是说，这当然是一种更快的方式，而且人们也会觉得更自在，而不是去填那个手动下拉表单。

It's, mean, it's of course a quicker way of doing it, but also people just feel more at ease instead of going for this manual dropdown form.

这是留下信息的更好方式。

It's just better way on leaving the information.

但现在，因为我们觉得这将会成为非常重要的一部分，每个公司都会有他们的营收工程职能。

But now, because we think it's going to be such a big part where everybody, every company will have their revenue engineering function.

但在我们这里，我们希望那种AISDR职能成为公司的一部分。

But in our case, we want that kind of AISDR function to be part of the company.

我们希望它能帮助到每个人，无论你在哪里。

We want it to help everybody across wherever you are.

我们是一家全球性的公司。

We are a global company.

所以它是集中式的，然后部署到每个本地团队。

So it's central and then deployed to each local team.

每个本地团队可以根据当地的新情况来调整，但开发当然还是集中进行的。

So each local team can refine it for the local local new ones, but developed, of course, centrally.

我在读《从零到一》这本书时，发现了一句我一直很喜欢的话。

I found one of my all time favorite quotes when I was reading the book Zero to One.

那句话是：我注意到的最强大的模式是，成功的人会在意想不到的地方发现价值，他们通过从第一性原理而不是公式来思考商业问题。

The quote says, the single most powerful pattern I have noticed is that successful people find value in unexpected places, and they do this by thinking about business from first principles instead of formulas.

这正是AppLovin在广告平台上所做的。

That is exactly what AppLovin has done with their advertising platform.

AppLovin能让你在手机游戏里接触到超过十亿的潜在新客户。

AppLovin connects you with over a billion potential new customers inside mobile games.

AppLovin能让你抓住用户全部的注意力。

AppLovin allows you to capture undivided attention.

AppLovin的广告是全屏视频广告，平均观看时长是三十五秒。

AppLovin ads are full screen video ads that are watched for an average of thirty five seconds.

这种留存率让其他广告平台望尘莫及。

That is retention that blows other ad platforms out of the water.

而且你几分钟内就能在AppLovin上启动。

And you can launch on AppLovin in minutes.

你设定目标，AppLovin就能帮你实现。

You set the goal and AppLovin achieves it.

不需要复杂的设置，也不需要专业知识，AppLovin能快速扩展。

There's no complex setup, no expertise needed, and AppLovin scales quickly.

他们能把你的广告展示给超过十亿的潜在客户。

They can put your ads in front of over a billion potential customers.

其他企业已经看到了立竿见影的效果，广告支出每天达到数十万美元，收入增加了数百万。

Other businesses have seen immediate results, have scaled to hundreds of thousands of dollars of spend per day, and increased their revenue by millions.

所以，趁你的竞争对手还没都上AppLovin，赶紧开始吧。

So you want to get started quickly before all of your competitors are on AppLovin.

你可以访问applovin.com来开始。

You can do that by going to applovin.com.

就是applovin.com。

That's applovin.com.

你还把ElevenLabs当成一家单一的公司吗？

Do you still think of ElevenLabs as a single company?

你们现在有多少种不同的产品？

How many different products do you guys have right now?

我们有三个核心产品线，另外还有三个正在开发中的项目。

We have three kind of core product lines, and then we have additional three that are developing Developing bets.

但具体有多少个不同的应用呢？

But how many different applications?

嗯，每个产品线里大概有六个核心功能，然后在这些功能里，你当然可以做二三十种不同的事情。

Well, within each of those, we'll have, let's say, six are core and then within that, you can do, of course, maybe twenty, thirty different things.

对。

Yeah.

看起来你们的产品真的很多。

It seems like you have a ton.

对。

Yes.

比如ElevenReader。

Like ElevenReader, for example.

是的。

Yes.

但你知道，就像ElevenReader或ElevenProductions，它们帮助你在人工环节中修正内容，并将其本地化为另一种语言。

But it's, you know, it's like ElevenReader or ElevenProductions which help you with the human loop aspect to correct the content and localize that to another language.

收入主要来自哪里？

Where is all the revenue coming from?

最大的两块是智能体方面和创意方面。

The biggest nines are on the agent side and the creative side.

所以，部署对话式智能体的用例数量，对我们来说现在正在飞速增长。

So the number of use cases that deploy conversational agents is just skyrocketing today for us.

你有没有看到某个特定行业采用得更快，还是没有？

And are you seeing a specific industry that's adopting them faster or no?

我觉得目前对我们来说，金融科技是最快的，非常快。

I think the quickest today for us is FinTech, Super quick.

比如Revolut？

Like Revolut?

对，就是Revolut。

Revolut, exactly.

Klarna、PagBank，这些公司都在以另一种速度发展，或者在美国这边的客户银行。

Klarna, PagBank, like all of those companies are just moving at another speed or customers bank in The US here.

然后是医疗、零售、电商和电信，这四个是最大的领域。

Then the healthcare, retail, e commerce, and telcos are the four of the biggest ones.

金融科技是发展最快的。

FinTech is moving the quickest.

当然，监管方面会有些不同。

Of course, it's slightly different regulatory aspects.

但我觉得医疗和电信是第二快的，而零售和电商今年才开始迎来这股浪潮。

But then I think the healthcare telcos is the second, and retail e commerce has just this year started going through the wave.

我再问你一次。

I'm going ask you one more time.

只是想回到这个话题，看看我们有没有漏掉什么。

Just want to go back to this in case we missed anything.

你在Palantir工作期间，还有没有学到什么我们没聊到、你觉得很有价值的东西？

Is there anything else that you learned working at Palantir that we didn't talk about that you think is valuable?

嗯，这个有点牵强，而且我有偏见。

Well, this is a stretch and I'm biased.

但在Palantir，经常有这样一个概念：要做好工作，你需要对艺术有一点理解，或者人们需要努力成为艺术家，即使我们不是。

But in Palantir, frequently there was this concept of how you need a little bit of the understanding of the art to do your job well, or people need to effectively try to be artists, even if we aren't.

比如他们用过的一个说法是“艺术家聚居地”，尽管这从未公开说过。

Like one of the phrases that was used was artist colony, even if that was ever publicly said.

我以前从未完全体会到这句话的力量，但现在当我们以多种方式将AI和创意领域结合起来时，我对此有了更多的感受。

I never fully appreciated the strength of that, but I do feel like a lot more about this now as we intersect that AI and creative space in many ways.

无论是构建音频研究模型，还是交付方式上，都有很多细微之处。

Whether it's building the audio research models, there's a lot of nuance in how that's delivered.

比如，每一个声音的呈现，像乔治和11。

Like, even every voice that is delivered, like George and 11.

每个人都有自己的偏好，怎么确保捕捉到这些声音，怎么呈现这些偏好，是个很棘手的挑战。

Like, everybody will have their own preference and how you make sure you capture those voices, how you deliver those preferences is a tricky challenge.

然后第二点，当你和一些试图定义如何与世界沟通的品牌合作时，那也需要很多艺术性。

Then two, as you work with some of the brands that are trying to define how they communicate with the world, that too requires some a lot of the art in that.

我们当然也在努力把很多人结合起来。

And we're trying to bring, of course, a lot of the people to combine that.

但把艺术和科学融合在一起，Palantir确实尝试过。

But blending that art and science, Palantir definitely tried to do.

我觉得成功与否让别人来评判吧，但我们也会努力去做，而且希望我们正在朝这个方向迈出好的步伐，无论是我们谈到的声音市场，还是让公司里有很多创意人员和我们一起或在我们周围做项目，几乎像前部署工程师或前部署创意人员那样与客户合作，我觉得这种融合会越来越重要。

I think I'll let others judge to put it successfully, but we also will try to do and hopefully are doing good steps in that direction, whether it's the voice marketplace that we spoke about, whether it's having a lot of creatives in the company building projects with us or around us, the kind of forward deployed engineers or forward deployed creatives almost to work with the customers, I think that blend will be increasingly important.

现在大家都在谈论品味，说品味将成为AI的决定性前沿。

And everybody talks about taste now, how taste will be the defining front of AI.

在这个流行词背后，我觉得有很多道理，确实，越来越多人将能够创造一切。

Behind the buzzword, I think there is a lot of truth to that where, yes, increasingly everybody will be able to create everything.

真正决定好的产品体验和部署体验的，是设计语言听起来怎么样，有没有品味，以及你怎么把它传达出来。

What defines above good product experience, good deployment experience will depend on how the design language sounds, how tasteful it is, how you bring that across.

我觉得Palantir尝试过这么做。

I think Palantir tried to do it.

我们也在尝试这么做。

We are trying to do it too.

对。

Yeah.

我觉得你提到的这个，让我想到一个叫埃德温·兰德的人。

I think what you're getting at is there's this guy named Edwin Land.

他是宝丽来的创始人。

He's the founder of Polaroid.

我之所以知道他，是因为他是史蒂夫·乔布斯的偶像，而乔布斯曾经提到过

I discovered him because he was Steve Jobs' hero, and Steve Jobs was talking

对。

Yeah.

对，关于这个。

About Yeah.

我记得。

I remember it.

是的。

Yeah.

我在Founders节目里反复跟他聊过这个话题，聊了很多次。

I've talked I've talked to him about it on Founders episodes all over and over again.

我应该在工作室里挂一张埃德温·兰德裱好的照片，或者什么的。

I should get a framed picture of Edwin Land and, like, put him in the studio or something.

但正是他提出了史蒂夫·乔布斯后来用的那个理念，就是要在科技和人文艺术的交汇处创建公司。

But he was the one that invented what the idea that Steve Jobs used where he's like, I wanna build a company at the intersection of technology and liberal arts.

我之所以想到他，是因为你提到品味时，他有句名言：品味就像独角兽一样稀有。

And the reason he just came to mind is when you brought up taste, he has this great quote where he says, taste is as rare as a unicorn.

所以虽然人们经常谈论品味，但它确实是一个限制因素。

And so it's like, people talk about all the time, but it is definitely like a limiting factor.

而且我觉得，这永远都会是一个限制因素。

And it will all I think it it, like, will always be like a limiting factor.

但我很喜欢这个想法。

But I love this idea.

我觉得这就是我不在旧金山生活的主要原因之一。

I think this is my issue with a lot of why I don't live in San Francisco.

我偶尔会去那边。

Like, I drop in there.

我们会录很多节目。

We do a bunch of recordings.

当然，我有很多朋友是科技创业者，但我觉得新一代的科技创业者缺乏人文关怀。

Obviously, I have a bunch of friends that are tech founders, but it's just like, I feel this gen like the the new crop of tech founders is like they they they lack the humanity.

就像我喜欢史蒂夫·乔布斯、埃德温·兰德或者本田创始人的一点是，他们总说，我发明技术就是为了提升人类的生活。

Where like one thing I liked about Steve Jobs or Edwin Land or the founder of Honda, they would say over and over again, it's like, I'm just inventing technology to enhance humanity.

对。

Yeah.

我觉得这些人中有很多只是想着，我发明技术是为了取代人类。

Where I think a lot some of these people are just like, I'm inventing technology to replace it.

这让我想到另一件事，想问问你，因为你在Palantir工作过。

So this leads me to another thing that I wanted to ask you about since you worked at Palantir.

你看到Alex Karp是少数几个这样做的人，而且他已经坚持了一年半了，现在他的观点真的被很多人接受了，他就是说，如果你经营这些实验室，还不停地上电视或接受采访，说我在制造核武器级别的技术，而且它会抢走所有人的工作。

You've seen Alex Karp be one of the only people, and he's done it for like a year and a half, now it's like really, really like, I think his perspective has caught on where he's just like, if you're running these labs and you keep going on TV or giving interviews, talking about I'm building like nuclear weapons grade, you know, technology and it's gonna take everybody's job.

参加太多会议了，这个我想问问你。

See too many conferences, which I wanna ask you about.

没有。

No.

我没去过那里。

I haven't I haven't been there.

好的。

Okay.

我刚去过那里。

So I was just there.

我在那儿只待了几个小时，而卡普是第一个发言的。

I was only there for a few hours and and Karp was the one that spoke first.

他就说，你觉得会发生什么？

And he's just like, what do you think is gonna happen?

他说，你的技术会被国有化。

He's like, your technology is gonna get nationalized.

你对他这个观点有什么看法吗？

Do you have any opinion on his perspective on this?

没有直接看法。

Not directly.

当然，我们最近看到了冯·特拉帕克案的情况，因为我们团队大部分在欧洲，花了很多时间跟欧洲团队和政府打交道，思考如何建设主权，以及这会有多重要。

So of course, we've seen what happened in the Von Trappak case recently, given we spent and most of our team in Europe spend a lot of time with the European teams and governments of how we think about building sovereignty and how important this will be.

所以对你的问题没有直接回答。

So no direct answer to your question.

但我非常同意你刚才说的其他部分，就是AI真的需要为人民服务。

I do agree so much of the other part of what you said, which is AI really needs to work for the people.

它需要放大人类的潜力，而不是取代它。

It needs to amplify human potential rather than replace it.

我希望我们会在这方面做很多工作，而且我们确实正在做很多工作。

And I hope we will do a lot of the work and are doing a lot of work in that space.

但我希望在你提到的那些采访和其他公司的对话中，这将成为他们如何将这一点变为现实的越来越重要的主题。

But I hope in some of the interviews that you mentioned and some of conversations from other companies that this will be an increasing theme of how they can bring that to reality too.

我问你这个问题的原因是，我记得我们和Cognition的创始人Scott Wu一起吃过晚饭。

The reason I asked you that question is because think me and you were at a dinner with Scott Wu, founder of Cognition.

我刚和他录了一期节目。

I just recorded an episode with him.

在过去的几年里，我和Scott有过很多接触，甚至在他们推出Devon之前就认识了。

I've spent a bunch of time with Scott over the last few years, even before they launched Devon.

我见过他。

I met him.

我觉得他做得好的地方，也是你做得好的地方，就是你们总是谈论你们的产品为人们带来的积极好处。

And what I think he does well and what you do well is you guys are always talking about the positive benefits that your products are creating for people.

比如，你讲过一位女士在结婚前失去了声音的故事。

Like, think you told the story of a woman had lost her voice before she got married.

然后她和你们合作重新创造声音，这样她就能用自己的声音重新宣读婚礼誓言。

And then she worked with you guys to recreate, and so therefore she could essentially redo her vows in her own voice.

这真是完美的例子。

It's like a perfect example of that.

声音是一种非常不可思议的身份，你提到的那个案例可能是我们最引以为傲的工作之一，我们可以帮助那些因肌萎缩侧索硬化症、癌症而失去声音的人，和他们一起把声音找回来。

Voice is such an incredible identity that probably the case that you mentioned is some of our proudest work, where we can work with people that lost their voices to ALS due to fraud, cancer, and work with them on bringing it back.

那是一个例子。

That one example.

最近，我们和一位失去声音的音乐家合作，他仍然想继续表演。

Recently, we worked with musician that lost his voice, and he wanted to still perform.

所以他努力重新创造那个声音。

So he worked on recreating that voice.

我们举办了一场音乐会，他和他的老乐队一起用AI声音进行了演出。

We set out a concert, he did a concert with his old band together with an AI voice.

现在他正在巡回演出。

And now he's touring.

在英国，他到处巡演。

And in The UK, he's going across places and touring across.

去年我们的国会议员失去了声音，但她仍然想激励其他人继续工作，并且第一次用AI声音在国会发言，试图让这一切成真。

Our congresswoman last year lost her voice and still wanted to inspire others to do the work and was in the congress for the first time with AI voice trying to bring that to life.

一个共同的主题是，声音承载着额外的情感冲击力和辨识度。

And common theme, it's like the voice carries such an additional element of emotional impact of recognition.

当你听到某个人的声音时，如果你认识他们，你会立刻认出来。

The moment you hear someone's voice, recognize it if you know someone.

当然，这还有另一面，那就是我们如何保护和安全地应对未来声音可以被创造的情况。

Of course, that has a second part, which is like how we safeguard and how we think about safety across that future where voice can be created.

但，是的，我们在这方面的工作。

But, yeah, the work across there.

今天，我们已经帮助超过一万人恢复了他们的声音，并希望继续这样做。

Today, it's over 10,000 people where we worked on bringing their voices back and hope to continue that.

除了无障碍领域，当然还有教育、更广泛的文化方面。

And outside of the accessibility space, of course, what happens in education, wider culture.

我们目前与800家机构合作，把这项技术带给更多人。

Here we partnered today with 800 organizations to help bring the technology to the people out there.

其中有一个特别疯狂的故事，最近发生的。

One of the crazy ones, actually, that's a crazy story recently.

抱歉打断你。

Sorry to interrupt you.

没事，你继续说。

No, go ahead.

首先，有个传奇人物，叫蒂姆·格林。

There's a like, first of all, a legend, a guy called Tim Green.

他曾经是畅销书作家，也是NFL顶级球员之一，但不幸的是，他得了ALS，很多工作都做不了了。

He used to be a best selling author, one of the top NFL players, and then unfortunately, he got ALS and couldn't do a lot of that work.

你能想象到的所有事情里，他偏偏选择了最极端的——他做了个播客。

From all the things you could imagine him doing, he decided to go the complete extreme and he did the podcast.

所以他开始了播客。

So he started the podcast.

我得找到这个人，追着他采访。

I'm going have to find this guy and chase him down.

不是吧。

No.

你应该采访他。

You should have an interview with him.

我不是那个意思。

That's not what I meant.

那应该是马蒂。

That would have been Mati.

好的。

Okay.

那倒也是。

That's fair.

那我

Then I

就把他干掉。

will destroy him.

不。

No.

好。

Okay.

但是

But

他他他他做得非常好。

he's he's he's he does extremely well.

他他请的嘉宾都很棒。

He he has great guests.

最近有Harv Diltnik上他的播客，还有一些NFL球员。

There's Harv Diltnik recently on on his podcast, some of the NFL players.

而且今年，他因为那个节目拿了艾美奖，所以感觉就是，哇。

And and this year, he won an Emmy for So his it's like, wow.

所以他真的很厉害。

So he's incredible.

等等。

Wait.

不过那个声音是你做的吗？

So you made that voice, though?

对，是我们做的。

We we yes.

我们帮他做了AI声音。

We we do his AI voice.

那太厉害了。

That's incredible.

所以那是我们小砖块对他工作的一大贡献。

So that was one of our proud Small Brick contribution to his work.

这真的太疯狂了，这简直是最极端的事情，而他却做到了。

It's just so crazy how It's like the most extreme thing you can do and he does it.

他激励了很多人，我不知道具体有多少人受到了启发，但有好几百甚至上千人联系我们。

He inspires I don't know how many people got inspired, but we had from hundreds to thousands of people reach out.

他们都感谢他，问怎么才能像他一样找回自己的声音。

That thanks thanks to him on, like, how can we get our voice back too in the same way?

我觉得你说到点子上了，这也是其中一个原因，我偶然发现的，就是播客之所以这么有力量，在于声音，当然我也很喜欢阅读。

Well, I think you hit on something which makes is one of the reasons, and I discovered this accidentally, is, like, what makes podcasting so powerful is the voice, where it's, obviously I love to read.

我的意思是，家里到处都是书，散落在各个角落。

I mean, there's books all around, like scattered around the house everywhere.

我有，怎么说呢，我另一个书房里大概有一千本书，可能有六百本还没读过。

I have like, I don't know, I have like a thousand books in my other library, like probably 600 unread.

有些作者，我真的就是爱上他们的作品，把他们写的东西全都读一遍。

And there's some authors where I literally just fall in love with what they do and read every single thing they've ever written.

但如果你问我对我最喜欢的作家，比如小说界的科马克·麦卡锡，和最喜欢的播客主播之间的情感联系，那简直没法比。

But if you would ask like my emotional attachment to my favorite authors or my favorite writers, like Cormac McCarthy, for example, in fiction, compared to like my favorite podcasters, it's like not at all comparable.

我觉得因为声音，还有那种人性化的元素，我好像就认识他们了。

It's like I feel like I know because of the voice and like the human like element of that.

我就觉得我认识他们的程度，是我永远没法认识我最喜欢的作家的。

I just feel like I know them in a way that I can never know like my favorite, you know, writer.

没错。

For sure.

而且你现在是在对话里，声音承载的维度比文字多太多了。

I mean, you also it's it's it's you're in the conversation now and it's like you you the voice carries so many other dimensions than text.

文字当然可以让你想象、解读，但它没有情感。

Like text, of course, you imagine, interpret, but it doesn't have the emotion.

没有语调。

It doesn't have the intonation.

没有那些不完美的地方。

It doesn't have the imperfections.

没有停顿。

It doesn't have the pauses.

不完美真的很重要，因为你知道，有些嘉宾来我们节目，会说，哎呀，我说‘light’说得太多了。

The imperfection is really important because, you know, we have some guests that come on here and they're like, okay, I said light too much.

其实那就是你说话的方式。

It's like, that's how you speak.

如果你想的话我们可以剪掉，但我说话会含糊不清，会说‘哦’，会发出各种奇怪的声音。

It's like, we can edit it out if you want, but like I slur my words, I say, oh, I make all these kinds of weird things.

但那就是我本来的样子。

Was like, but that is just how I am.

我可不想在播客里出现，然后你见到我本人，发现声音完全不一样。

I don't ever wanna appear on a podcast and you meet me in person, guy didn't even sound the same.

不完美的地方，人们欣赏真实远超过完美。

Like the imperfections, people admire authenticity way more than they do perfection.

确实。

That's true.

其实有趣的是，这甚至也适用于非人类，比如语音助手。

Actually funnily even applies in a non human way, in a voice agent's way too.

一开始，我们想做一个完美的语音助手，一点瑕疵都没有。

Initially, we are trying to create a perfect voice agent that doesn't do any

结果听起来不像人。

It didn't sound human.

确实不像人。

It didn't sound human.

然后当然，最明显的就是那些“嗯”、停顿什么的。

And then of course, the obvious thing that was the clearest is like the ums, the ums, the pauses.

结果跟那个语音助手合作的效果突然就飙升了。

And suddenly the performance of working with that voice agent like skyrocketed.

大家都说，哇，这太像人了。

Everybody was like, oh, this is so human.

这挺好的。

This is good.

我很乐意跟它说话。

I am happy to speak with that.

所以，是的，就是要让它不完美。

So so, yeah, it's making it imperfect.

现在这几乎成了关键要素。

It's almost now the the the element of that.

但声音确实承载着这些信息。

But voice does carry that information.

它以一种完全不同的方式连接你。

It it kind of connects you in a completely different way.

这也是为什么我认为这是一个很难的研究挑战，因为在文本中，没有那么多维度。

Why I think also it's such a hard research challenge because here in text, don't have that many of those dimensions.

当然，你需要大量数据来构建一个好的语言模型，但你没有每个声音都不同、每个声音对每个人听起来都不同的维度。

You need a lot of data, of course, to create a good language model, but you don't have the dimensions of every voice being different, every voice sounding different to every person.

比如，即使对文本转语音进行基准测试也非常困难，因为通常不同的模型会有不同的声音。

Like even doing benchmarks for text to speech is extremely hard because usually different models will have different voices.

这已经让它们无法比较了。

That already makes them uncomparable.

你年轻的时候就知道自己想当企业家吗？

Did you know when you were younger that you wanted to be an entrepreneur?

我得说，我当时并不知道这是一条路。你呢，知道还是不知道？

I would say I didn't know this was a path for You a did or did not?

不知道。

Didn't.

是因为你是欧洲人吗？

Was Because you're European.

在波兰，确实有一点那种因素。

In Poland, was For sure, a little bit of that though.

我知道你在开玩笑，但我觉得这是真的。

I I know you're joking, I think it's true.

不是。

No.

我是认真的。

I'm being serious.

比如说，承担那种创业的风险，你懂的，就是去开创一些东西。

Taking, like, the the risk of, not, you know, like, starting something.

这在我们那儿从来都不是什么常见的话题。

It it it wasn't like a common conversation ever happening.

我很幸运，有一个很棒的家庭，给了我出国留学的机会，然后这让我眼界大开，觉得，好吧，现在你可以去一些大公司工作了。

I had a lack of having an incredible family that gave me opportunity to study abroad, and then that opened your eyes of like, okay, now you can work with some of the great companies.

然后当你和那些大公司合作时，他们会发现你确实有能力做事。

Then when you work with those great companies, they realize you can actually do things.

Palantir在这方面对我帮助很大，在那里你确实可以去和客户合作，尝试找出问题所在。

Palantir was great for that for sure, where it's like you can actually go and work with the customer, try to figure out the problem.

这在我当时看来风险很大，但你确实能做到。

Can take that seemed very risky to me, but you can do that.

一步一步地，这让我眼界大开，觉得，好吧，也许这就是一条路。

Step by step that opened the eyes of like, okay, maybe this is the path.

也许自己创业是有可能的。

Maybe it is possible to start your own thing.

从你意识到这一点到创办ElevenLabs，中间隔了多久？是一年还是两年？

From the time you had that realization to the time you started ElevenLabs, what was the Was it a year, two years?

具体是多久？

What was it?

我在Palantir工作期间，我的联合创始人Piotr在Google工作，我们会一起做黑客周末项目。

Over my time at Palantir, my co founder is Piotr time at Google, we would do start doing a hack weekend projects together.

这些年来，我们一直尝试探索新技术，一起合作。

So through the years, we tried to explore new technology and build together.

我觉得在几年前，我们就知道我们想一起做点什么，但你想做的事情必须是你认为真正的问题，并且你对此非常着迷。

I think a few years prior, we we knew that we would love to work on something together, But but you want to work something about that you are, like, you think it's a true problem and you're obsessed about.

这个想法在2021年出现了。

And and that came to us in 2021.

所以是几年。

So a few years.

两年。

Two years.

两三年吧。

Two, three years.

你收到过多少次收购要约？

How many acquisition offers have you had?

具体的话，三四次吧。

Like, concrete ones, three or four.

最近一次是什么时候？

When's the last one?

最近一次是去年，大概去年六月。

Last one was last year, like June last year.

你打算卖吗？

Are you going to sell?

不卖。

No.

我问你是因为

Reason I ask you

人工智能正在改变世界。

AI is changing the world.

我们可以站在这个变革的前沿。

We can build the frontier of that change.

我们全力以赴。

We are going all in.

也就是说，要留意你的投资人，

Is mean, pay Their attention to your VCs,

他们的利益和你的不一样。

incentives are different than yours.

我问你这个，是因为有两个原因让我想到这个。

The reason I ask you is there's two reasons this came to mind.

有些东西全靠直觉，甚至说不清楚，但埃文·斯皮格尔就坐在你现在坐的这把椅子上。

Some of this is all intuition, you can't even describe this, but Evan Spiegel sat in that exact same chair that you're in.

你知道吗，当时有人骂我。

You know, people gave me shit.

他们就说，你为什么要采访埃文？

They're like, why do you want to interview Evan?

你看，看看他的市值。

You know, look at his market cap.

我就说，我才不在乎他的市值呢。

Was like, I don't give a shit about his market cap.

我不是那样看公司的。

Like, I don't look at companies that way.

我痴迷的是产品。

I'm obsessed with products.

而且那个人是真心投入的。

And like, that dude has soul in the game.

我真的希望他能成功。

Like, and I hope he wins.

我也不知道。

I have no idea.

其实我什么都不知道，你懂的，我根本不用它。

Like, I I I don't know anything about, you know, just I don't use it.

像Snapchat、眼镜之类的，不管是什么。

Like Snapchat, specs, anything, whatever it is.

他就是与众不同。

It's just like, he is differentiated.

他身上就是有某种东西让我喜欢，我就是希望他能做得好。

I did there's just something about him that I like that I like just want him to work out well.

对吧？

Right?

所以，我现在也有同样的感觉。

And so that and I'm getting the same exact vibe.

我就想，天哪，我希望Mati能赢。

I like, man, I want Mati to, like, win.

我希望他能成功。

I want him to succeed.

就是觉得你特别讨人喜欢。

Just something very likable about you.

然后第二个原因是，显然，Scott Wu，他会来上节目，

And then the second reason is because, obviously, Scott Wu, he's he's coming on the show, like,

每隔几个月就来一次，因为我真的很喜欢Scott。

every few months because I I really like Scott a lot.

就是因为他是个天才。

I'm just because he's genius.

他也非常非常厉害。

He's so so good as well.

不只是这样，他还很会表达，很聪明，也很乐观，但现在所有人都围着他转。

Well, not only that, but he's articulate and brilliant and optimistic, but everybody is running at that dude right now.

所有人。

Everybody.

想收购他公司的人可多了。

The amount of people that want to buy his company.

这是我们之前聊过的事情。

This is something we talked about.

我希望他能坚持下去，因为我想看到更多人直接说，不，我要一直做下去。

I hope he holds out because I would like to see a lot more people just be like, No, I'm in this forever.

不只是为了赚一大笔钱。

It's not just to get a big bag of money.

你去听听优步的特拉维斯怎么说。

You go listen to Travis from Uber.

他从优步赚了几十亿美元。

He made billions and billions of dollars from Uber.

但那并没有让我开心。

That did not make me happy.

我需要有事情做。

I need something to work on.

看，这就是为什么你会问这个问题，因为我也问过斯科特同样的问题。

See, that's why that's your question because I asked Scott that too.

他就说，显然有很多人想挖他，或者想把他整个公司和所有人才都弄走。

He's just like, well, there are obviously a ton of people trying to either get him or his entire company and all his talent.

他到目前为止就说，不，我不卖。

He's just so far said, No, I'm not selling.

喜欢斯科特。

Love Scott.

我觉得他们也应该建立独立公司。

I think they should build independent company too.

我觉得他们有机会成为未来的超级AI之一，或者你叫未来的超级云也行。

I think they have an opportunity to be one of the hyper AIs of the future or however you call the hyper clouds of the future.

所以我觉得他能做到。

So I think he can do it.

团队很棒，我觉得我们也能做到。

Have an incredible team and I think we can do it too.

我觉得，你知道吗，我是认真的。

I think it's, you know, I really mean it.