字号·· | 护眼
semafor

2.2万亿估值初创公司联合创始人致力于让聊天机器人“回话”The $22B startup co-founder working to make chatbots talk back

点「原文对照」整页切到原文,或双击某段只看那段的原文。

意大利总理乔治娅·梅洛尼上周申请为其声音注册商标,凸显了人们对AI深度伪造的日益担忧:这些伪造声音越来越像被冒充者本人,从政客到首席执行官都难辨真伪。但她也将目光投向了快速发展的商业领域,即帮助知名人士为其声音肖像确立权利,以及那些声称“声音AI”很快将成为许多人处理客户服务和企业销售电话,或与汽车到机器人等设备交互方式的公司。

The Signal Insight When Italian Prime Minister Giorgia Meloni moved to trademark her voice last week, she highlighted growing fears about AI deepfakes sounding ever more convincingly like the people they impersonate, from politicians to CEOs. But she also shone a spotlight on the fast-developing business of helping prominent individuals secure rights to their voice likenesses — and on the companies that say “voice AI” will soon become how many people handle customer service and corporate sales calls or interact with devices from cars to robots.

ElevenLabs本月通过一项3亿美元的员工股份要约,以220亿美元的估值,确立了自身作为这些初创公司中估值最高者之一的地位。

ElevenLabs established itself as one of the most highly valued of those startups this month with a $300 million employee tender, at a $22 billion price tag.

这家总部位于伦敦的公司,已与马修·麦康纳、迈克尔·凯恩爵士等名人合作,为其声音注册商标,并为雅虎基于其联合创始人杰里·杨打造了定制声音。该公司表示,已开发出全面的防护措施,以检测和防止未经授权的声音克隆或其他滥用行为。

The London-based company has worked with celebrities from Matthew McConaughey to Sir Michael Caine to trademark their voices, and created a custom voice for Yahoo based on its co-founder, Jerry Yang. It says it has developed comprehensive safeguards to detect and prevent unauthorized voice cloning or other abuses.

ElevenLabs的大部分收入来自德意志电信和瑞众银行等企业客户,联合创始人马泰乌斯·“马蒂”·斯塔尼谢夫斯基表示,企业应为一个世界做好准备,在这个世界里,客户越来越期望与AI代理对话,而非与聊天机器人打字交流。

Most of ElevenLabs’ revenues come from corporate customers like Deutsche Telekom and Revolut, and co-founder Mateusz “Mati” Staniszewski says businesses should prepare for a world in which customers increasingly expect to speak to AI agents rather than type with chatbots.

“声音将成为未来的关键界面之一;在许多应用场景中,它将是主要界面,”他预测道。“声音更便捷,它传递的信息远比文字多,并且让技术在场但不喧宾夺主。”安德鲁·埃奇克利夫-约翰逊:请举例说明您看到的声音AI的好处。

“Voice will be one of the key interfaces in the future; for a lot of use cases, the primary interface,” he predicts. “Voice is easier, it conveys a lot more information than text would, and it allows the technology to be present but not overwhelming.”Andrew Edgecliffe-Johnson: Give us an example of the benefits you see in voice AI.

马蒂·斯坦尼舍夫斯基:我们为[销售与开发代表岗位的求职者]设置了一个AI面试官。通常情况下,你会访问网站并填写表格。我们增加了[与代理人对话的选项],很多人开始通过对话参与其中,于是出现了两种情况。一是,说话比打字更快捷,这使得我们吸引了更多的人[申请]。二是,人们开始提供比平时更多的信息。到目前为止,我们通过这一渠道已经收到了超过3,000条线索,其中超过500条是合格的,因此它们是优质线索。

Mati Staniszewski: We have an AI interviewer for [sales and development representative job candidates]. Usually you go to the website and fill in a form. We added [the option] to speak with an agent, and a lot of people started engaging through conversation, and two things happened. One, the ease of making it quicker for you to speak than to type resulted in us having a lot more people [applying]. Two, people started leaving a lot more information than they usually would. We’ve had more than 3,000 leads come through this so far, and more than 500 were qualified, so they were good leads.

首席执行官应该怎么做才能确保他们的员工能够充分利用AI代理?

What should CEOs do to make sure their human employees are getting the most out of their AI agents?

如果你没有做好这一点,其他方面都将无关紧要。归根结底,要让代理有效运作,它需要正确的知识。它需要正确的一套步骤和工作流程。它需要理解自己在做什么。就像人类一样,它在早期阶段需要大量的指导,以确保其正常运作。此外,它还需要恰当的防护措施和保障机制,以免它偏离脚本。

If you don’t get this right, the other parts won’t matter. Ultimately, for an agent to work, it needs the right knowledge. It needs the right set of steps and workflow. It needs to understand what it’s doing. Like a human, it needs a lot of coaching in the early phases to make sure it works. And it needs the right guardrails and safeguards, so it doesn’t go off-script.

你不希望出现[人类与代理之间的竞争]。你希望找到一种方式,让人们来管理代理,并且因做好这项工作而得到公平的回报。而且这不是一次性的过程。你需要不断改进、不断迭代,有时[你]还需要亲自介入。你希望确保存在一个循环:代理 → 人类 → 代理。在这些努力中,关键在于AI代理需要知道何时该引入人类代理。你们是否计划在不久的将来上市?

You don’t want competition between [humans and agents.] You want to find a way to bring people to manage the agents and be rewarded fairly for doing that well. And it’s not a one-and-done process. You keep improving, you keep iterating, and sometimes [you] need to jump in. You want to make sure that there is a loop: agent to human to agent. The key thing across all those efforts is that the AI agent needs to know when to bring in a human agent. Are you looking to go public anytime soon? We would love to go public one day.

我们很希望有一天能够上市。我们希望在未来18到24个月内做好准备 — — 准备就绪,但不一定真的上市 — — 然后根据更广泛的触发因素来做决定。

We would like to be ready — ready, but not necessarily go public — in the next 18-24 months. And then decide based on the wider side of triggers.

您认为,在目前已经有充足资本支持公司增长的情况下,上市有什么优势?

What’s the advantage, do you think, of going public when there’s been plenty of capital to power your growth so far?

我们希望[我们的用户]能够参与其中,你加入得越晚,他们能够参与的能力就越低。我们看到很多人在使用ElevenLabs进行创作,我们也希望他们能够成为其中的一部分。还有其他[优势],但我们正在打造一家能够经受时间考验的公司,而我们合作的公司正在寻找那些计划在50年后、100年后仍然存在的合作伙伴,我们希望成为那样的公司。

We would love for [our users] to have the ability to be part of the journey, and the later you do it, the lower ability they will have. We see a lot of people building with ElevenLabs, and we’d love for them to have an ability to be part of that. There are other [advantages], but we are building a company to stand the test of time, and the companies we work with are looking for partners that are planning to be here in 50 years, in 100 years, and we want to be that company.

迈克尔·凯恩爵士的声音是他作为演员身份的重要组成部分。他为何决定与你合作而不是对抗你?

Sir Michael Caine’s voice is a big part of his identity as an actor. Why did he decide to work with you rather than against you?

我们实际上创建了一个市场,人们可以[录制]自己的声音,我们进行身份验证,他们可以分享它,并因此获得报酬。因此,我们正试图让行业加入进来,[像迈克尔·凯恩这样的人]也希望成为这个未来的一部分,前提是你找到了一种尊重他们工作并让他们能够从中获得报酬[的方式]。

We created, effectively, a marketplace where people can [record] their voice, we authenticate it, they can share it, and they earn compensation as a result. So we’re trying to bring the industry on board, and [people like Michael Caine] want to be part of that future too, assuming you find a way to respect their work and give them the ability to earn compensation [from it].

CEO冒充诈骗似乎非常普遍。你们是否曾因技术而遇到过这种问题?

CEO impersonation scams seem to be very prevalent. Have you ever had that problem with your technology?

凭借我们的技术,你始终能够清楚地了解[AI声音]是如何生成的。你可以采取行动。因此,如果有人[被诈骗],我们可以提供帮助。我目前不熟悉有任何CEO被冒充的案例,但我们有防护措施。所以,如果你试图询问钱款,该请求就会被阻止或被检测出来并标记为人工审核,这样我们就能主动采取行动。此外,如果你想要将声音克隆到高[质量]的程度,你需要经过额外的身份验证机制。因此,你无法轻易克隆一个声音。

With our technology, you always have transparency into how [an AI voice] was generated. You can take action. So, if somebody [was scammed], we could assist. I’m not familiar with any CEO that was impersonated, but we have guardrails. So if you try to ask about money, that gets either blocked or detected and flagged for human review, so we can proactively take action. And then, if you want to clone a voice to a high [quality], you need to go through an additional authentication mechanism. So you cannot just clone a voice that easily.

值得一提的是,斯坦尼舍夫斯基认为公司的欧洲根源 — — 它诞生于波兰,总部设在伦敦 — — 是“巨大的优势”。他说,伦敦拥有大量希望“在全球范围内打造雄心勃勃事业”的人才。此外,ElevenLabs 从一开始就专注于多语言的欧洲市场,因此开发出能够在多种不同语言中同样有效运作的模式“从来不是事后才想到的”。

Notable Staniszewski sees his company’s European roots — it was born in Poland and is based in London — as “a huge advantage.”London has plentiful talent which wants to “build something ambitious on a global scale,” he says. And because ElevenLabs focused from the start on the polyglot European marketplace, creating a model that could work equally well in many different languages “was never an afterthought.”