Cohere 的起飞：面向安全企业的 AI

这些发达民主经济体的增长有所放缓。当增长开始放缓时，就会出现仇外心理之类的东西，比如移民抢走了我的机会，他们抢走了我的那份蛋糕。很多人谈论 AI 毁灭世界，我真的不认为 AI 会毁灭世界。它可能是我们拯救世界的最佳机会。欢迎收听 Upstarts 播客，这是我们每周一期的节目，与新兴创业领袖谈论他们的创业时刻。Upstarts 是那些超越自身实力、挑战现状以改善世界的挑战者，同时也在建立大企业。我是主持人 Alex Conrad，Upstarts Media 的创始人兼编辑。今天很高兴邀请到 Cohere 的联合创始人兼 CEO Aidan Gomez。感谢你的到来，Aidan。

There's been a sort of slowdown in growth for these developed democratic economies. When that starts to taper off, you see the emergence of things like xenophobia, like that immigrant is taking my opportunity. They're taking my slice of the pie. A lot of people talk about AI destroying the world. I really don't think AI is going to destroy the world. It might be our best shot at saving the world. Welcome to the Upstarts podcast, our weekly show where we talk to emerging startup leaders about their upstart moment. Upstarts are challengers who punch above their weight and take on the status quo to improve the world, all while building a big business, too. I'm your host, Alex Conrad, founder and editor of Upstarts Media. And I'm delighted to be joined today by Aidan Gomez, co-founder and CEO of Cohere. Thanks for joining, Aidan.

谢谢邀请。

Thank you for having me.

本期播客由 Mercury 赞助，这是一家从零开始重新设计的银行。Aidan，你经营 Cohere 已经六年半了，你们筹集了 17 亿美元，与一些大政府和公司合作。当你想到 Cohere 今天的样貌，是早期阶段吗？这是你六年前所期望的一切吗？

This podcast is brought to you by Mercury, banking redesigned from the ground up. Aidan, you have run Cohere now for 6 and a half years. You guys have raised 1.7 billion dollars. You work with some big governments, some big companies. When you think about what Cohere looks like today, is it early innings? Is it everything you hoped for 6 years ago?

我认为我们正处于起飞阶段。我们已经存在了一段时间，但过去一年对我们来说确实是突破性的一年，企业开始大规模采用 AI，Cohere 擅长的所有事情——隐私、主权以及对关键行业的关注——真正引起了全球公司高管的注意。去年，所有这些因素汇聚在一起，我们开始了起飞，今年我们预计随着企业和全球经济真正大规模采用 AI，规模会更大。

I think we're in the takeoff phase. So, we've been around for a while, but this past year was really the breakout year for us in terms of enterprises just adopting AI at scale across the entire company, and the importance of all the things that Cohere is good at really sort of hitting the front of mind of executives at companies all over the world. So, privacy, sovereignty, and then this focus on critical industries. And so, last year, all of that came together, and we sort of started our takeoff, and then this year we expect it to get even bigger as enterprises and the broader global economy really truly adopts AI at scale.

那么，对于这些与 Cohere 合作的大公司或 G7 国家，他们希望你解决什么问题？或者说，你的技术在哪里真正帮助他们？简单概述一下。

So, with some of these big companies or G7 countries that are working with Cohere, what is the problem that they want you to solve or where is your technology actually helping them in a simple overview?

范围非常广，因为我们与来自不同行业的客户合作，从制造业到公共部门、金融服务、能源。简单来说，比如让模型每天早上 6 点检查未读邮件，并发送一条短信，总结最重要的信息。这是一个非常通用的用例，任何公司都想要。再到非常具体的，比如在特定船只上，将 North 接入船上的遥测系统，持续监控不同传感器，并围绕传感器超过阈值时该做什么、通知谁、采取什么补救措施构建自动化。用例非常广泛。

It's super broad because we're working with a really diverse range of customers from different industries from manufacturing to public sector to financial services, energy, and it's as simple as, you know, get the model to every morning at 6:00 a.m. check my unread emails and send me a text with a summary of the most important ones that I need to know. So, that's a super general use case any company would want. To a hyper specific one, within a specific ship, I'm going to plug North into the telemetry within that ship, do continuous monitoring on different sensors on board, and build automations around if sensors cross certain thresholds, what to do, who to notify, what sort of remediations to take. It's a wide range of use cases.

但总的来说，为什么他们需要 Cohere 存在？为什么这对他们来说是一个未解决的问题？

But generally speaking, why often do they need Cohere to exist? Like why is that an unsolved problem for them?

我认为是隐私部分。我认为这项技术、这个平台的安全部署是我们的客户不可或缺的。这是他们环境中的硬性要求。他们不能将数据发送到云端。他们受到高度监管，处于高安全环境中。因此，他们需要部署在本地甚至气隙隔离的基础设施上的东西。这是部署层面的差异化，另一部分在于产品本身。它非常易于使用，高度可定制，拥有大量 North 平台独有的企业管理控制功能。

I think it's the privacy piece. I think the secure deployment of this tech, of this platform, is something that our customers couldn't do without. It's something that is a hard requirement in their settings. They can't send data out into a cloud. They're highly regulated. They're in these high-security settings. And so, they need something that's deployed on prem or even air gapped on their infrastructure. There's a deployment level differentiation and the other piece is within the product itself. It's super easy to use. It's very customizable. There's tons of administrative enterprise controls that are unique to the North platform.

我们应该把 Cohere 看作是基础层、基础设施层，公司在其上构建内部应用，或者运行智能体，利用你提供的算力做各种其他工作？最好的可视化方式是什么？

Should we think of Cohere as the base layer, the infrastructure layer that companies build internal apps on top of or run agents doing all sorts of other work with the info your horsepower provides? What would be the best way to visualize this?

我认为它是这些高安全企业采用智能体并将其集成到安全系统中的基础设施。这就是我们所做的。在 Cohere，显然我们自己是产品的内部客户，许多最好的自动化是由不会编码的人创建或发现的。他们只是作为日常工作的一部分，心想‘哦，我完全可以用 North 来做这个。’他们构建一个自动化，然后与他人分享。所以我们希望实现的是，人们发现系统的杀手级应用，然后在组织内病毒式传播，扩展到所有不同团队，North 真正实现了这一点。

I think it's infrastructure for these high security enterprises to adopt agents and integrate it into their secure systems. That's what we do. At Cohere, obviously we're internal customers of our own product and many of the best automations have been created or discovered by folks who don't code. And they just, as part of their day-to-day work, they're like, 'Oh, I could totally use North for this.' They build an automation and they share it with people. And so that's what we want to be able to enable is like people just discover killer applications of the system and then it goes viral within an organization and spreads out to all these different teams and North really enables that.

我们正处于一个被告知可以‘氛围编码’任何东西的时代，有一些非常成功的公司帮助人们从想法中生成应用。区别在于，我们无法在敏感或安全的环境中内部运行 Bolt 或 Level Up 之类的工具，而使用 North 更可行？区别在哪里？

We are in a moment where we're being told you can vibe code anything and there are some really successful companies that are helping people come up with an app out of an idea. Is the difference here that we couldn't really run a Bolt or a Level Up or one of these tools internally in a sensitive or secure environment with all our data? So using North is just much more viable? What would be the distinction there?

我认为‘氛围编码’对技术人士很有帮助，他们可以完成 90%甚至 95%的工作。它可以带你走得很远，但仍有最后一英里需要完成，而且通常你需要技术背景才能完成最后一英里。而对于 North，你不需要技术背景就能做这项工作。你有一个构建者视图，可以直接与模型对话，不需要自己是开发者。所以我认为这是一点。另一点是，对于许多这类应用，你不需要创建产品。你不需要创建一整套新软件来解决它。你真正需要的是让它访问正确的工具和正确的认证，然后构建一系列步骤让它完成，这是一个更简单的问题，有更简单的解决方法，而不是为每一个我想创建的东西都去创建全新的代码库和程序，并让编码智能体去构建。所以我认为这是更简单优雅的解决方案。

I think the vibe coding stuff is very helpful for technical folks who want to get 90% of the way there or even 95% of the way there. It can carry you quite far, but there's still a last mile that you need to close and oftentimes you do need to be technical to close that last mile. And for North you don't need to be technical to do that work. You have a builder view and you can just chat to the model and there's no need for you to be a developer yourself to do the work. So I think that's one thing. The other thing is that for a lot of these applications, you don't need to create a product. Like you don't need to create a whole new piece of software to solve it. All you really need is it to have access to the right tools with the right authentication, and then you need to be able to build a series of steps that you want it to accomplish, and that's a much simpler problem, and there's an easier way to solve that rather than for each one of these that I want to create, I'm going to have to create an entirely new codebase and program and ask a coding agent to go build it for me. So, I think it's a more simple and elegant solution to the problem.

很酷。有一个有趣的问题：初创公司是反映创始人的个性、背景和偏好，还是反过来？我很好奇，这是你一直想象中会经营或创建的那种企业吗？

Very cool. There's an interesting question of are startups reflecting the personalities or backgrounds, preferences of their founders, or is it vice versa? And I'm curious, is this the kind of business that you would have always imagined running or building?

嗯，我不知道。我确实很重视自己的隐私，但我也尽量保持开放。

Yeah, I don't know. I definitely like appreciate my own privacy as an individual, but I try to be open.

我认为要让人们信任你，他们必须了解你，所以我会谈论我在哪里长大、如何长大、我的家庭。我尽量保持一定的开放度，但我肯定不是一个出色的营销或销售人才。我不会到处大喊自己的名字或 Cohere 的名字。我的背景是学术型的。我是通过研究进入这个领域的，在研究中如鱼得水的性格与在公司建设中成功的性格截然不同。我一直在尽力适应成为 CEO，不得不担任公司的发言人和销售员。但你无法摆脱本性，我就是一个害羞的书呆子。

I think for people to trust you, they have to know you, so I talk about where I grew up, how I grew up, my family. I try to be somewhat open, but I'm certainly not a great marketing or sales guy. I'm not out there screaming my own name or Cohere's name constantly. My background is academic. I came into this through research, and there's a very different type of personality that thrives in research than in company building. I've been doing my best to adjust to being a CEO and having to be a spokesperson and a salesperson for the company. But you can't run from your nature, which is that I'm a shy nerd.

这没什么不对。他是个很高的害羞书呆子，没错。明白了。

That doesn't go anywhere. He's a very tall shy nerd, yeah. Got it.

那么，当你在学术界时，你有幸或凭实力处于一些最早的大语言模型工作的中心，比如著名的 Transformer 论文。你能谈谈你是如何最初进入这个领域的，为什么它让你兴奋，以及这段旅程是如何真正开始的吗？

Well, so when you were in academia, you were privileged or you had earned your way to be at the epicenter of some of the earliest work with LLMs, the famous Transformer paper. Could you talk about how you first got into this field, why it was exciting to you, and how the journey really starts there.

我在加拿大森林深处长大，没有互联网。后来我们装了卫星天线，因为之前是拨号上网，但那是在 Starlink 之前。那是地球同步卫星，所以如果有云，你就完蛋了。那种技术的稀缺性让它对我极具吸引力。它总是遥不可及。我的朋友在线玩游戏、做各种酷事，而我回家后体验完全不同。这让我深深珍惜计算机，对它们充满热情。我总是试图让我的电脑更快，或者通过有限的连接访问别人能访问的东西。

I grew up in the middle of the woods in Canada without any internet. At some point we got a satellite dish because before that it was dial-up, but this was pre-Starlink. It was geostationary, so if there was a cloud, you were screwed. That scarcity of technology made it super appealing to me. It was just out of reach. My friends were gaming online and doing cool stuff, and when I went home, the experience wasn't the same. That made me deeply appreciate computers and be very excited about them. I was always trying to make my computer faster or access what others were accessing through my limited connection.

成功了吗？你在破解更快的电脑吗？

Was it working? Were you hacking a faster computer?

这让我进入了网页开发。我想知道为什么网页加载不快，所以我会检查元素等等。这就是我进入编程的原因。起初只是浏览网站，试图理解屏幕上的所有文本。然后我因为感兴趣而报了一门课。我在安大略省布莱顿开了一家小公司，为当地企业建网站。有一位很好的女士经营一家针织店，卖纱线和图案，所以我帮她建了一个在线目录。我开始了那个小生意，然后我决定这就是我想做的。于是我进入多伦多大学学习计算机科学，那是我当地的大学。多伦多是最近的大城市。我碰巧在深度学习的中心上大学。杰夫·辛顿在那里，他因创建深度学习而获得了诺贝尔奖和图灵奖，深度学习是让我们达到今天水平的突破性系统。

It got me into web dev. I wanted to understand why a webpage wouldn't load faster, so I would inspect element, etc. That's what got me into coding. First it was just going through websites and trying to understand all the text on my screen. Then I took a course because I was interested. I started a small business in Brighton, Ontario, building websites for local businesses. There was a nice lady with a knitting store, yarn and patterns, so I helped her build an online catalog. I started that little business, and then I decided this is what I want to do. So I went into computer science at U of T, my local university. Toronto was the nearest big city. I happened to go to university at the epicenter of deep learning. Geoff Hinton was there, the guy who won the Nobel Prize and the Turing Award for creating deep learning, which was the breakthrough system that let us reach where we are today.

你是立刻就觉得‘这是我想全身心投入的领域’，还是更循序渐进？

Were you immediately like, 'This is an area that I want to throw myself into' or was it more gradual?

我痴迷其中。我记得醒着的每一分钟都在想它。我会拿着一叠论文走来走去。最初，读该领域的一篇论文要花我一个月时间，因为我会遇到不认识的词，比如‘贝叶斯’或‘高斯’，然后我得去研究那是什么。没有大语言模型可以走捷径。只有维基百科页面和研究论文。渐渐地，我对材料越来越熟悉和自如。我去了温哥华的一家初创公司，从事多声部转录工作，即听音乐并将其转录成乐谱。这在当时是一个极其困难的问题，但我们使用了神经网络，并取得了相当不错的结果。那是在 JAX、TensorFlow、PyTorch 等任何框架出现之前。我不得不用 Metal（苹果的 GPU 语言）编写前向和反向传播的内核。我必须以最低级别学习一切。

I was obsessed. I remember every waking minute of my day thinking about it. I would walk around with a stack of papers. Initially, reading one paper in the field would take me a month because I'd run into a word I didn't understand, like 'Bayesian' or 'Gaussian', and I'd have to research what that was. You didn't have LLMs to shortcut. It was just Wikipedia pages and research papers. Gradually, I got more familiar and comfortable with the material. I went to a startup in Vancouver where I worked on polyphonic transcription, which is listening to music and transcribing it into sheet music. It was a super difficult problem at the time, but we used neural networks and got pretty good results. This was before JAX, TensorFlow, PyTorch, any of those frameworks. I had to write kernels in Metal, Apple's GPU language, for forward and backward propagation. I had to learn everything at the lowest level imaginable.

你基本上是在写机器码。

You were basically writing machine code.

其实不是，但非常底层，愚蠢地底层。然后我回到多伦多，给辛顿发消息说：‘嘿，我有个想法。你做的这个激活函数，我觉得有更好的。’他真的回复了，并把我引荐给了多伦多大学机器学习组的罗杰·格罗斯等人。从那里开始，我和他们一起工作，最终到了山景城的谷歌，参与了 Transformer 论文。剩下的就是历史了。然后我去牛津读博士。我从硅谷回到多伦多时遇到了我的联合创始人。碰到了尼克和伊万。尼克在谷歌为杰夫工作，是杰夫的第一位下属。然后从牛津，我打电话给尼克和伊万说：‘伙计们，我们得做点什么。’我看到了语言模型 Scaling（规模扩张）的最初几个结果。这完全是零到一的突破。我们几乎无法让这些模型连成一个句子，然后机器就能输出看似流畅、表面上像人写的文本。现在我们觉得这理所当然。太正常了。但那时，读机器写的东西简直是超现实的。

Not really, but super low-level, stupidly low-level. Then I came back to Toronto and messaged Hinton, saying, 'Hey, I have this idea. This activation function you made, I think there's a better one.' He actually responded and sent me to Roger Grosse and others in the U of T ML group. From there, I started working with them, eventually ended up at Google in Mountain View, and was on the Transformer paper. The rest is history. Then I went to Oxford for my PhD. I met my co-founders when I came back from the Valley to Toronto. Ran into Nick and Ivan. Nick was working for Jeff at Google, Jeff's first report. Then from Oxford, I called Nick and Ivan and said, 'Guys, we have to start something.' I saw the first few results of language model scaling. It was very much this zero to one. We could barely get these models to string a sentence together, and then it was plausible fluent, ostensibly human-written text coming out of a machine. Now we take that for granted. It's so normal. But back then it was surreal reading a machine.

我认为人们有点理所当然地认为我们在和一些石头和金属说话。在这个阶段，我们实际上是在和我们的宠物石头说话，这太疯狂了。是的，它已经正常化了。我们每天都这么做。但天哪，那时它简直令人震惊。

I think people sort of take for granted the fact that we're talking to some rocks and metal. We're literally talking to our pet rock at this stage, which is so insane. Yeah, it's normalized. We do it every day. But good god, back then it was just mind-blowing.

那你还有什么想做的呢？我们聊聊 Transformer 论文，也就是《Attention Is All You Need》。我不知道最好的类比是什么，也许是 AI 研究的大宪章。这是一篇非常重要的论文，你是作者之一。对于不深入了解 AI 的人来说，这篇论文的 TLDR 是什么？

And so like what else could you want to work on? We talk about the Transformer paper, which was called Attention Is All You Need. I don't know what the best comp would be, maybe the Magna Carta of AI research. Very seminal paper that you were a part of. For folks who are not deep in AI world, what would be the TLDR of what this paper was?

嗯，这是我在谷歌时完成的。特别是谷歌大脑（Jeff Dean 领导的 AI 研究部门）和谷歌翻译之间的合作。两个团队走到一起，目标是改进翻译，提升翻译的最优水平。架构本身旨在更好、更可扩展地对语言建模。语言是字符或单词的序列，顺序很重要。你想以最高效的方式建模或表示这些数据。所以 Transformer 就是我们要解决的问题：如何为语言构建更高效的模型。结果 Transformer 在语言之外也变得极其有效，你可以把许多不同的数据模态（比如视频）当作序列来处理。这个模型非常高效，以至于在我们用更大的计算集群构建更大模型时，它比替代方案扩展得更好。人们为它优化芯片和库，随着时间的推移，Transformer 接管了一切——图像、音频，整个领域。但它最初是为了一个狭窄的问题设计的：让翻译提升 3%。现在人们问：“Transformer 之后是什么？”我听到这个问题已经很久了。

Yeah, well, it was done when I was at Google. In particular it was a collaboration between Google Brain, the AI research department under Jeff Dean, and Google Translate. The two teams came together and the objective was to improve translation, improve the state of the art of translation. The architecture itself was designed to better and more scalably model language. Language is a sequence of characters or words in order, and the order matters. You want to model or represent that data in the most efficient way possible. So the Transformer was the problem we were trying to solve: how to make a more efficient model for language. Turns out the Transformer became extremely effective beyond language, and you can phrase many different data modalities, like video, as a sequence. The model was so efficient that it scaled better than alternatives as we built larger models on larger computing clusters. People optimized chips and libraries for it, and over time the Transformer took over everything—images, audio, the whole field. But it was designed for a narrow problem: making translation 3% better. Now people ask, 'What comes after the Transformer?' I've been hearing that for a long time.

当你们发表 Transformer 论文时，你有没有想过“这会很轰动”？

When you published the Transformer paper, did you think 'This is going to be huge'?

没有。我的意思是，我们是世界上最小众的名人。大概只有 13 个人认识我们。但我们不知道它会有什么影响，至少我不知道。我不想代表所有作者发言，但我们不能把 Transformer 的成就归功于自己。我们把它发布出来。我们的努力是围绕改进翻译性能，提出一个更适合语言和序列建模的架构。我认为我们做到了。但随后社区接手了它，把它应用到一切领域，为它优化硬件。我们只是启动了这件事，但把它变成今天这个样子的是成千上万其他人的工作。如果不是我们，也会有其他类似的东西出现。当时已经有 WaveNet 和 ByteNet 这样的想法。我很感激社区选择了 Transformer，但别人也会在三个月后做出来。

No. I mean, we're like the world's most niche celebrities. There's like 13 people that know us. But no, we didn't know the impact it was going to have. At least I didn't. I don't want to speak for all authors, but we can't take credit for what happened with the Transformer. We put it out there. Our effort was around improving translation performance, coming up with an architecture better suited to language and sequence modeling. I think we did that. But then the community took it and applied it to everything, optimized hardware for it. We set the ball in motion, but the work has been done by thousands of others who turned it into what it is today. If it wasn't us, something else would have come out that looked and felt the same. There were ideas like WaveNet and ByteNet emerging at the time. I'm grateful the community leaned in on the Transformer, but someone else would have done it three months later.

关于你的创业历程，在牛津时你想“我们必须在这里做点什么。”你是一直在考虑创办公司，还是需要合适的时机和想法才能让你兴奋地去创业？

On your entrepreneurial journey, at Oxford you thought 'We've got to build something here.' Did you always have in the back of your mind that you might start a company, or did it need the right moment and idea?

我的联合创始人 Ivan 一直给我发消息，说他不想在现在的地方工作了，想创业。我当时在读博士，所以我说“祝你好运”，但我不想让他把时间浪费在愚蠢的想法上。所以我们开始互相发想法。最后有一个想法促成了 Cohere：训练一个网络模型。我们不知道这怎么变成生意，但我们看到谷歌内部用语言模型对维基百科建模产生了令人信服的文本。GPT-2 出现了，我们看到在更广泛的文本上训练的结果。所以我们就说：“如果我们把整个互联网——所有图片、视频等——当作数据集来训练呢？一定会产生有趣的东西。”这最终导致了 Cohere。我们看到像 GPT-2 这样的结果，让我们确信这个方向很重要。我们从无法用自然语言与计算机对话的世界，变成了一个似乎可能的世界。

So Ivan, my co-founder, had been messaging me saying he was done where he was working and wanted to do a startup. I was in the middle of my PhD, so I said 'Good luck,' but I didn't want him wasting years on a stupid idea. So we started sending ideas back and forth. Eventually there was one that precipitated Cohere: train a model of the web. We didn't know how it would be a business, but we had seen internal Google results on modeling Wikipedia with language models producing compelling text. GPT-2 came out, and we saw results from training on a broader source of text. So we said, 'What if we treat the entire internet—all pictures, videos, etc.—as a dataset to train on? Something interesting will come out of that.' That led to Cohere. We saw results like GPT-2 and got conviction that this direction was important. We went from a world where you couldn't speak to computers in natural language to one where that seemed possible.

Cohere 的想法或使命有什么让你如此兴奋，以至于你会说‘让我们在这里投入大量时间和机会’？因为你早于并见证了，例如，团队离开 OpenAI 创立 Anthropic，然后 Anthropic 突然飞速增长，而 Cohere 继续沿着自己的道路前进。为什么不干脆说‘我们可以自己做一个更好的 OpenAI’，或者去争取最明显的那块蛋糕呢？

What was so exciting about the idea or the mission of Cohere that you were like, 'Let's commit a lot of time and opportunity here'? Because you predate and saw, for example, the team leave OpenAI and start Anthropic, and then all of a sudden Anthropic is growing super fast, and Cohere is continuing along the journey. Why not just be like, 'We can do a better OpenAI ourselves' or go for the most obvious piece of the pie?

嗯，对我们来说有吸引力的是技术项目。那才是最令人兴奋的。我们想训练这个互联网模型。那才是真正的目标。我们最初不知道如何将其变现。很快，我们确实决定要面向企业。我们希望产生的影响是将这项技术推向经济。这是我们头 12 个月内做出的决定。我仍然认为这是这项技术最引人注目的应用。我们能产生的最佳影响是推动世界和全球经济的增长。如果我们能让人们更高效，这本身就是一件好事。

Well, the attractive thing for us was the technical project. That was what was most exciting. We wanted to train this model of the internet. That was really the objective. We didn't initially know how we were going to monetize that. Quickly, we did decide we want to be B2B. The impact we want to have is driving this technology into the economy. That was a decision we made within the first 12 months. And I still think that is the most compelling application of this technology. The best impact we can have is by driving growth for the world and for the global economy. If we can make people more efficient, that in itself is a good.

很多现在非常突出的问题在过去 15 年里一直在酝酿。我来自加拿大，住在英国。日本、韩国和欧洲大部分地区也是如此。这些发达民主经济体的增长放缓了。当这种情况发生时，当人均 GDP 开始下降，蛋糕不再为所有人增长时，你就会看到零和思维开始主导。如果蛋糕不增长，那么我的份额就不会增长，除非我从别人那里夺走。反之亦然，我感到非常受威胁，觉得别人在蚕食我的份额。所以你看到了像仇外心理这样的东西的出现：那个移民在夺走我的机会。他们在夺走我的份额。把他们赶出去。你看到了领土征服之类的事情。战争又回到了欧洲东线，人们在为土地和资源而战。这些特征并非不可避免。我认为它们源于稀缺性，以及人们觉得需要互相争斗才能过得更好。

A lot of the issues that are very prominent right now have been stewing over the past 15 years. I'm from Canada. I live in the UK. The same is true for Japan, Korea, much of Europe. There's been a slowdown in growth for these developed democratic economies. When that happens, when GDP per capita starts to taper off and the pie isn't growing for everyone, you start to see zero-sum thinking dominate. If the pie isn't growing, then my slice isn't growing unless I'm taking it from someone else. And vice versa, I feel very threatened that someone else is eating into my slice. So you see the emergence of things like xenophobia: that immigrant is taking my opportunity. They're taking my slice. Get them out. You see things like territorial conquest. War is back on the eastern front of Europe, and people are fighting for land and resources. These traits are not inevitable. I think they come out of scarcity and people feeling they need to fight each other to get better off.

巨大的希望——很多人谈论 AI 毁灭世界，还有所有这些 X-risk、EA 恐惧。思考这些很重要，但我真的不认为 AI 会毁灭世界。它可能是我们拯救世界的最佳机会。恢复这个 100 年的增长弧线，数十亿人的寿命延长，更健康，活得更久、更好，他们的教育变得更好，获得医疗、技术、财富——一切都在改善，因为蛋糕在增长，我们都随着蛋糕一起增长。但在过去 15 年，随着增长开始停滞，我认为很多令人不安的政治趋势又回来了。所以我的希望是，通过将 AI 融入经济，我们可以恢复增长，并推进这个更广泛的自由民主项目和这些在过去 100 年里提升了这么多人的价值观。

The big promise—a lot of people talk about AI destroying the world and there's all this X-risk, EA fear. It's important that we think about that, but I really don't think AI is going to destroy the world. It might be our best shot at saving the world. Restoring this 100-year arc of growth where billions of people had longer lifespans, were healthier, living longer, better lives, their education got better, access to medicine, technology, wealth—everything was improving because the pie was growing and we were all growing with that pie. But the past 15 years, as that started to stagnate, I think a lot of disturbing political trends have come back in. So my hope is that by integrating AI into the economy, we can restore growth and carry forward this broader project of liberal democracy and these values that have lifted so many over the past 100 years.

这是否有助于解释为什么 Cohere 寻求或优先考虑这些主权客户，以便其他国家受益或拥有自己的 AI 模型、一个他们可以信任且不依赖于一个中心地方的 AI 基础，还是说这更多是工具繁荣之处的偶然结果？

Would that help explain why Cohere has sought out or prioritized these sovereign customers so that other nations benefit or have their own AI models, an AI foundation they can trust that isn't dependent on one central place, or is that more incidental of where the tools flourish?

是的。在过去的四分之一个世纪里，美国和中国以外的国家在技术上的主权能力确实下降了。美国和中国仍然是技术进步的巨大堡垒。但在我来自的加拿大或我父母来自的欧洲，各国在技术雄心和技术能力上逐渐自满或空心化，主权也被掏空。我们如此依赖单一来源。我们如此依赖一两个我们可以购买东西的地方，但我们自己却做不到。现实是我们自己能做到。我们只是变得太安逸了。部分来说，在多伦多创立 Cohere 并保持其加拿大公司身份是一个有意的选择，以支持这种韧性的概念。只有当不止一个参与者能够提供能力时，我们才能拥有有韧性的民主。这并不意味着我们都需要做所有事情——我们不需要都构建语言模型，那效率极低——但我们需要多样化的供应链，多样化的依赖关系，这样就不会全部集中在一个地方。这是我想贡献的，而且我认为这对民主来说是生死攸关的。如果你热爱民主，如果你热爱人民拥有自治权并对自己如何被治理有发言权，你就不能有对任何一个国家的单一依赖线。因为如果那个民主国家出了什么事，突然间一切都会崩溃。整个项目就会瓦解。所以你需要一个去中心化、多样化的贡献者群体来支撑每个核心技术支柱。而由于 Cohere 在这个基础设施层运作，这些其他国家或文化可以在其上构建，并且比从硅谷获得一切作为应用程序感到更安心。

Yeah. Over the past quarter century, sovereign capabilities in technology have really declined outside the US and China. US and China are still incredible strongholds of progress in technology. But where I'm from in Canada or where my parents are from in Europe, there's been a gradual complacency or hollowing out of technological ambition and capability of the various nations and a hollowing out of sovereignty. We depend so much on a single source. We depend so much on one of a couple different places where we can buy things from, but we can't do it ourselves. The reality is we can do it ourselves. We've just grown too comfortable. In part, founding Cohere in Toronto and keeping it a Canadian company has been an intentional choice to support this notion of resilience. We're only going to have resilient democracies if more than one player can provide capabilities. It doesn't mean we all need to do everything—we don't all need to build language models, that's hyper inefficient—but we need a diverse supply chain, a diverse set of dependencies so that it's not all concentrated in one place. That's something I want to contribute to, and I think it's existential to democracy. If you love democracy, if you love people having self-governance and having a say in how they are governed, you can't have these single lines of dependency on any one country. Because if something happens to that democracy, suddenly everything breaks. The whole project falls apart. So you need a decentralized, diversified set of contributors to each core technological pillar. And because Cohere is operating at this infrastructure layer, these other countries or cultures can be building on top of it and feel more comfortable than if they were getting everything as an application from Silicon Valley.

正是如此。这不仅仅是你不依赖一个地方。你有选择，你可以通过与多个不同方的相互依赖来建立韧性，而不是仅仅一个。这确实是我们看到被掏空的东西——这个广泛的民主联盟内部多方贡献的概念。

Exactly. It's not just you're not just reliant on one place. You have options, and you can build resilience through interdependence with multiple different parties instead of just one. That's really what we've seen kind of hollow out is this notion of multiple parties inside this broad democratic coalition contributing.

嗯。对此有很多表达的不满。不同方贡献不够，有点只是搭一方的努力和费用的便车。我认为这有深刻的道理。我们需要站出来解决这个问题。我认为 Cohere 至少在 AI 方面是其中的一部分，但在许多方面，比如能源独立和制造业以及建设国家的所有这些不同基本部分，我们这些自由民主国家的人需要贡献、站出来并以必要的规模投资，以支持发展这些经济的基本支柱。

Mhm. There's been a lot of voiced frustration about that. Different parties not contributing enough and sort of just riding off the efforts and expenses of one party. I think there's a deep truth to that. And we need to step up to fix that. I think Cohere is part of on the AI file at least, but across many files, like on energy independence and manufacturing and all these different fundamental parts of building a country, those of us in liberal democracies need to contribute and step up and invest at the scale necessary to support developing these fundamental pillars of the economy.

当你来到像我们今天所在的旧金山这样的地方时，你持有这些观点是觉得自己像个新锐或异见者，还是觉得也许更多人已经认同了？现在和 AI 同行分享这些原则是什么感觉？

When you come to a place like San Francisco where we are today, do you feel like an upstart or a contrarian by holding those views or do you feel like, you know, maybe more people have come around to see this or what is it like to kind of share these principles with your peers in AI right now?

我觉得有些人看到了。这绝对不是一个特别普遍的观点。我不知道他们是否认为这真的那么重要。我认为背后有更多的意识形态或政治意识，至少比今天一些极度痴迷增长的 AI 初创公司所担心的要多。似乎 Cohere 在思考这个问题时是有意为之的。我们也处于独特的位置，能够为这种韧性概念做出贡献，对吧？作为一家加拿大公司，我们是少数在外部构建这种能力的群体之一。因此，我们处于独特的位置，能够实现这一使命。我不知道这里的人是否处于实现这一使命的位置。我认为他们没有。

I think some people see it. It's definitely not like I don't think this is a particularly broadly held view. I don't know if they view it as actually that significant. I think there's more ideology behind it or sort of political awareness than I think, at least some very growth-obsessed AI startups today really worry themselves with. It seems like it's very intentionally intertwined with Cohere thinking about this. We're also like just uniquely positioned to be able to contribute to that notion of resilience, right? Like being a Canadian company, we're one of a very small group that is building this capability outside. And so we're uniquely positioned to be able to deliver on that mission. I don't know if folks here are positioned to deliver on that mission. I don't think they are.

你是怎么走到这一步的？最难克服的部分是什么？在节目中，我们谈到新锐时刻，也许那是你最超常发挥的时候，或者你只是觉得必须真正克服一些东西才能把想法变成现实。回顾迄今为止的旅程，Cohere 的那个时刻是什么？

How did you get there? Like what has been the hardest piece to overcome? On the show we talk about the upstart moment where maybe you were punching above your weight the most or you just felt like I have to really overcome something to get that idea to reality. When you think back on the journey so far, what would that moment have been at Cohere?

哦，有很多不同的时刻符合这种特征。有很多障碍需要克服，以及像“做还是不做”这样的战略难题。似乎 B2B 的选择是第一个非常重要的转折点。第二个是是否将公司的一大块出售给某个超大规模云服务商。这是 OpenAI 对微软、Anthropic 对 AWS 所做的，我们看到并考虑过，但最终没有追求。Cohere 一直有这种理念，即使在过去，现在有很多地缘政治事件，但以前没有。我们已经存在了大约六年半。但即使在这些地缘政治事件之前，Cohere 的核心价值主张之一就是独立性。所以不锁定在任何单一云生态系统等。哦，你想在 AWS 上运行？太好了。OCI？太好了。Azure？太好了。有点像与所有人交朋友的方式。我们对此相当固执。我们相当固执地认为我们要服务所有地方。这是我们的价值的一部分。我们一直有这种性质，战略独立性很重要。在公司层面很重要，对吧？如果你是一家大型企业，你不想完全锁定在你的云提供商的一个生态系统中，因为他们会随着时间的推移不断挤压你的利润。你会把自己深深嵌入那个生态系统，永远无法脱身。脱身成本太高。所以公司需要这种战略独立性，以便在不同云之间进行谈判，找到公平且实际有竞争力的价格。在国家层面也是如此，对吧？你不能过度依赖任何第三方。你不仅需要能够自给自足，还需要有多样化的供应商，以防一个出问题，你可以继续与其他合作。然后在个人层面，我认为接触多样性很重要，接触不同的生态系统、不同的产品，不要陷入泡沫或回音室。

Oh, there were many different moments that kind of fit that character. There's been many hurdles to overcome and just like challenging strategic questions of like do we don't we? It seems like the B2B choice was a first really important junction. The second one was do you sell off a huge chunk of your company to one of the hyperscalers? That's the thing that like OpenAI did with Microsoft and Anthropic did with AWS that we saw and considered but ultimately did not pursue. There was always this notion in Cohere even before, like there's a lot of geopolitical stuff happening now that wasn't happening in the past. We've been around like six and a half years. But even before all this geopolitical stuff one of the core value props of Cohere was independence. So not getting locked into any one cloud ecosystem etc. Oh, you want to run on AWS? Great. OCI? Great. Azure? Great. Sort of a friends with everyone approach. And we were pretty dogmatic about that. Like we were quite dogmatic that we're going to serve everywhere. And that is part of our value. We've always had that nature of like strategic independence is important. It's important at the company level, right? Like if you're a large enterprise, you don't want to be completely locked in to your cloud provider one ecosystem because they're just going to keep squeezing you and squeezing you and squeezing you for margin over time. And you'll dig yourself so deeply into that ecosystem that you'll never be able to get out. It'd be too expensive to get out. So the company you want this strategic independence to be able to negotiate different clouds off of each other, find like a fair and actual competitive price. At the company level, at the country level, right? Like you can't over depend on any one third party. You need to be able to not only provide for yourself, but have a diversified set of suppliers in case one goes down, you can continue to work with these others. And then at the individual level as well, I think it's important to expose yourself to diversity, to expose yourself to different ecosystems, different products, and not get into a bubble or into an echo chamber.

但你们本可以更早获得知名度，也许估值更高，至少获得更多典型科技公司的势头。所以这里有一个权衡。你可以说你们可能是在困难模式下玩游戏，因为……

But you would have gained notoriety, maybe a much higher valuation early on, at least like a lot more of some of the typical tech trappings of momentum. So there is a trade-off there. Like you could argue maybe you guys were playing on hard mode a little bit because...

是的，你知道，我们绝对是，在加拿大创业就是困难模式。从那里就开始了。但我的意思是我们不是在寻找最简单的道路。我们在寻找正确的道路，以及对市场、世界、客户有意义的贡献。我们非常真诚地致力于此已经六年半了。在这六年半里，我们没有大幅转型。我们一直在说同样的事情：隐私和安全至关重要，独立性至关重要。市场正在向我们靠拢。我认为我们看到的是，首先，六年前半没有人购买 LLM。没有人知道 LLM 是什么。大约有 100 到 200 人知道 LLM 是什么。然后消费市场起飞了，爆炸了，变得超级流行。我们从未做过消费者服务。我们一直……

Yeah, you know, we're definitely I mean like founding in Canada is playing on hard mode. That's like it started right there. But I mean we're not looking for like the easiest path. We're looking for the right path and something that is a meaningful contribution to the market, to the world, to our customers. And we've been like quite earnestly dedicated to that for six and a half years. We haven't pivoted that strongly in those six and a half years. We've kind of always been saying the same thing that privacy and security is essential, independence is essential. And the market has moved towards us. And I think that's what we've watched is like first off, no one was buying LLMs six and a half years ago. Nobody knew what LLMs were. There was like a group of 100 to 200 people who knew what LLMs were six and a half years ago. And then consumer took off, blew up, was super popular. We never did a consumer service. We were always like...

内部有争论吗？比如我们需要自己的 ChatGPT。

Was there any debate internally? Like we need our own chat GPT.

太多了。一个非常激烈争论的话题。但我们只是在想，我们要花时间追逐别人在做的事情吗？我们要复制这些人的成功吗？我们不想那样做。我们想开辟自己的道路。我们仍然如此。市场向我们靠拢真是太棒了。突然间，主权和独立性成为世界上最重要的事情。企业是利润率最高、最好的业务。我们看到了顺风。比如我们去年收入增长了 6 倍。今年还会有另一个大的倍数。我不知道我是否相信因果报应。但我确实相信，对一个好的、对世界有用的想法的真诚奉献会得到回报。可能不是最快的回报，也不是最容易的回报。可能不是最简单的道路，也不是最快的道路。但它会得到回报。就像黄仁勋，英伟达，他是任职时间最长的上市公司创始人兼 CEO。天哪，30 年来，他就这样……

So much. Like such a hotly debated topic. But we were just like are we going to spend our time like chasing what other guys are doing? Are we just going to like replicate whatever these people are having success with? We didn't want to do that. We sort of wanted to carve our own path. And we still do. It's fantastic that the market has moved towards us. It's suddenly like sovereignty and independence is the most important thing in the world. Enterprise is like the highest margin, best business. And we've seen the tailwinds of that. Like our revenue 6X'ing last year. It'll do another big multiple this year. I don't know if I believe in karma. But I do believe like an earnest dedication to an idea that is like a good one and a useful one for the world will be rewarded. And it might not be the fastest rewarded. It might not be the easiest rewarded. Might not be the easiest path. Might not be the fastest path. But it'll be rewarded. Like for Jensen, Nvidia, he's been like the longest serving public company founder CEO. And man, for 30 years, he just like...

他四年前凭空出现……

He appeared out of nowhere 4 years ago with the...

不，不。他只是知道这很有用。很难描述，但他知道有需求。他本可以只做 CPU。他本可以放弃，融入英特尔所做的一切，或者任何最新的潮流。我敢肯定他被告知过无数次：‘这无关紧要。这永远不会成功。这是小众市场。它不会规模化。人们不需要这个。’但你知道，它现在是世界上最大的公司。当之无愧。没有人像黄仁勋那样有如此长久的信念，并为此付出了更多的努力。

No, no. He was just he knew this was useful. It's hard to describe, but he knew that there was a need. And he could have just done CPUs. He could have just like folded and done rolled into like everything Intel was doing and whoever the you know, whatever the latest fad was. And I'm sure he got told a million times, 'This is irrelevant. This is never actually going to take off. It's niche. It's niche. It's not going to scale. People don't need this.' But you know, it's the biggest company in the world. And well deserved. Like no one has had as long a conviction and has like worked harder for this than Jensen has.

我觉得他配得上他所获得的所有成功。那种对理想的执着，那种几十年来对自己信念的坚持，是我非常钦佩的。你如何平衡你自己的——也许不是‘理想主义’这个词——但你的价值观和你对这个世界的愿景（这似乎对你很重要），与让这些客户或这些主权实体去做你可能不同意的事情？我们正处于这样一个时刻：一家大型 AI 实验室与政府就其工具的使用发生了公开争执。你们被许多政府使用。这是你必须面对的问题，还是你对此感到非常自在？Cohere 的北极星是什么？

And I think he deserves all the success he's gotten. And that sort of conviction to an ideal, that sort of conviction to his beliefs over decades is something I really admire. How do you balance your own sort of maybe idealism isn't the word, but your values here and the vision of the world that seems to matter a lot to you with the sort of idea of letting these customers or this sovereignty do things that maybe you wouldn't agree with? We're in a moment where one of the big AI labs has had a very public spat with the government about use of its tools. You are used by a bunch of governments. Is this something that you have to grapple with or that you feel really comfortable with? What becomes the true north for Cohere?

是的，这非常重要。这项技术将无处不在——每个行业、公共部门、私营部门、国防、民用。我们必须接受这一点。我们在 Cohere 的做法是，我们很幸运，因为我们直接与客户合作，有很多机会教育和接触他们。所以我们能够成为他们的合作伙伴，一起思考在哪里部署是合适的，技术在哪里已经准备好部署，以及需要什么样的保障措施。我们试图将这些构建到模型中，构建到围绕模型的产品 North 中。人类监督是关键组成部分。能够说某些类别的活动或行动我们永远不会完全自动化，我们总会有人类参与其中。我们确保在 North 内部，这可以以极其准确和可靠的方式实施，从而让人类参与决策。关于 AI 如何部署到一些更敏感的领域，有很多伦理问题，当然包括你提到的国防，还有医疗和金融决策。所有这些都对人们的生活产生重大影响。所以我们非常注意这一点，并提供工具来做出有效决策。然后我们也选择与谁合作。实际上，我们可以从自己的立场决定与谁做生意，特别是与哪些国家接触。对于民主国家，美妙之处在于这是一种更好的治理方式。它往往有更好的制衡机制。当有人越界或某个机构违法时，会有问责，然后实施修复，系统通过这些自我修复。所以我们相信民主进程能够产生保护人民的结果，并最终形成最好的政策环境来支持这项技术的正确采用。但当然，我们确实选择与谁合作，与哪些政府接触，不与哪些接触。

Yeah, it's super important. This technology is going to be integrated everywhere—every single industry, public sector, private sector, defense, civil. We just have to accept that it will be. The way we approach it at Cohere is we're fortunate in that because we work directly with our customers, we have a lot of opportunity to educate and engage with them. So we're able to be a partner to them in thinking about where is it right to deploy this, where is the technology ready to be deployed, and what sort of safeguards need to be in place. And we try to build those into the model. We try to build those into North, the product around the model. Human oversight is a key component of it. The ability to say there's a certain category of activities or actions that we're never going to fully automate. We will always have a human in the loop for. We ensure that within North, that can be implemented in an extremely accurate and reliable way so that a human is involved in making that decision. And there are many ethical questions about how AI gets deployed into some of these more sensitive sectors, of course defense like you're referencing, but also healthcare and financial decisions. All of these have meaningful consequences on people's lives. So we try to be very conscious of that and provide the tools to make effective decisions. And then we also choose who we work with. We can actually from our position make a decision of who we do business with and in particular which countries we engage with. With democracies, the fantastic thing is that it's a better way of governance. It tends to have much better checks and balances. When someone does step out of line or when an agency does break the law, there's a reckoning with that and then fixes are implemented and the system self-repairs through those issues. So we trust the democratic process to produce results that protect people and ultimately end up in the best possible policy environment to support the right sort of adoption of this technology. But of course, we do choose who we work with and what governments we engage with versus which ones we don't.

当你思考你的总体使命或胜利或成功的样子时，是用影响力来衡量吗？是用 Cohere 的规模来衡量吗？我不知道万亿美元 IPO 是否是你的目标，就像你的一些同行那样，但你会乐于顺便实现它吗？当你想到最终结果时，愿景板上有什么？

When you think about your overarching mission or what winning or success would look like, is it measured by impact? Is it measured by the scale of how big Cohere grows? I don't know if a trillion-dollar IPO is a goal for you as it seems to be for some of your peers, but is that something that you'd be happy to have incidentally? When you think where this ends up, what's on the vision board?

我不在乎 IPO 的估值。我确实在乎上市，因为我认为人们需要——普通公民需要能够持有公司股份。我认为这很好。如果我们永远保持私有，只有机构投资者才能拥有 Cohere 的所有权，那是一个糟糕的结果。上市的另一件好事是你会受到额外的审查。人们可以看到你的财务状况。你必须报告。你别无选择。你会更——你会得到更好的治理，因为你受到更高的审查。所以我认为 Cohere 应该基于这两点上市：更好的治理、更多的审查，以及更多人能够参与。但我显然希望估值对投资者来说是公平且高的。对你的投资者来说，而不是对你。我不知道。我已经够富有了。没关系。我不知道这是否还那么重要。但对我来说重要的是 Cohere 在世界上的影响力，以及我们能够支持帮助国家更具韧性的更广泛使命。特别是帮助民主国家无论发生什么都能更具韧性。所以如果我们能为这个使命做出贡献，如果我们能帮助国家变得更加主权、拥有更多自主权、感到更有安全感，从而能够抵制不良行为并做出符合他们信仰的价值观的正确决定。那就是成功。我想我们会在过程中得到回报。

I don't care about the valuation of the IPO. I do care about IPOing because I think people need to be able to—private citizens need to be able to take a stake in a company. I just think that's good. If we stay private forever and it's only institutional investors that have access to taking ownership of Cohere, that's a bad outcome. The other good thing about IPOing is just that you get additional scrutiny. People can see your finances. You have to report them. You have no other choice. You're way more—you're better governed because you're under higher scrutiny. So I think Cohere should IPO on the basis of those two things: better governance, more scrutiny, and the fact that more people will get to participate. But I obviously want the valuation to be fair and high for my investors. For your investors, not for you though. I don't know. I'm like rich enough already. It's okay. I don't know if it really matters that much anymore. But it matters to me the impact that Cohere has in the world and that we're able to support this broader mission of helping countries be more resilient. In particular helping democracies be more resilient no matter what comes. So if we can contribute to that mission, if we can help countries become more sovereign, to have more autonomy, to be able to feel like they're on more secure footing so that they can push back against bad behavior and make decisions that are right for the values that they believe in. That's a success. And I think we'll get paid along the way.

太棒了。好吧，我希望你对这些客户产生的影响越来越多地进入公众视野。所以也许即使 Cohere 不是家喻户晓的名字，基于 Cohere 模型的产品或基础设施仍然在产生巨大影响。

Awesome. Well, I hope that the impact you're having with these customers makes its way into the public eye more and more. So maybe if it's not Cohere being the household name, the products or the infrastructure run off of Cohere models is still having that huge impact as well.

是的。嗯，我们赞助了一支 F1 车队。阿斯顿马丁。那是为了让我能去看比赛，是的。但我们的贴纸在那里。所以人们现在会看到它。也许我们的知名度会提高。

Yeah. Well, we sponsored an F1 team. So Aston Martin. That was so I could go to the race, yeah. But our stickers are there. So people are going to be seeing it now. So maybe our profile will raise.

好吧，非常感谢你来到节目，Aidan。

Well, thanks so much for coming on the show, Aidan.

是的，谢谢你邀请我。

Yeah, thanks for having me.