# (77) 「不毛な競争には加わらない」アンソロピック ダリオ・アモデイCEO 単独インタビュー「安全なAI」と日本市場への期待【WBS】 - YouTube
- Transcript ID: `rldzqhnq2-8`
- Source tier: `youtube_auto`
- Original language: `en_or_source_language_unlabeled`
- Translation status: `agent_translated_reviewed` (20/20 chunks)
- Original source: https://www.youtube.com/watch?v=rldZQHnq2-8
- Raw evidence: `data/transcripts/youtube/dario-amodei-tvtokyo-2026-rldZQHnq2-8.json`
> 中文是派生翻译，英文是留存的源文本。中文未复核时，以英文和原始音视频为准。
## 00:00:00
**中文**
我们看到这些做法后会说：我们不必做任何一项，因为我们并不试图赢得大众消费市场，也不想卷入一场你死我活的竞争。出于我们围绕安全的价值观，我们本来就不想做那些事。因此我们选择了一个与价值观非常一致的商业模式。不过，我真正期待的应用，是与生物医学等相关的应用。如果 AI 能够解决许多长期困扰人类的重大疾病。
**English / Original**
We look at it and we say we don't need to do any of that because [music] we're not we're not trying to win the mass consumer market. We're not trying to get into a We're not trying to get into a rat race and because our values of of our values around safety, we don't want to do that stuff anyway. And so we've chosen a business model that aligns very well with our values. But, you know, the applications that I really [music] look forward to are the ones that relate to things like biomedicine. If, you know, many of the greatest [music] diseases which bedevil humanity, AI was able to was able to solve them.
## 00:00:53
**中文**
我们相信，这是科技史上一家软件公司最快的增长。我认为没有任何可比案例能以如此快的速度增长。我们希望——我们不想提高日本公司的企业价值，而作为副作用，那也会提高我们的企业价值。［笑声］
**English / Original**
Which we believe is the the fastest growth for a software company in in the history of technology. Um I don't think there are any comparable examples of the growth happening so [music] fast. We want to we we we we don't want to enhance the enterprise value of companies in Japan and as a side effect that will also enhance our >> [laughter] >> our enterprise value.
## 00:01:32
**中文**
是的，回到 Anthropic 的创立：我和 Anthropic 的联合创始人是最早记录技术规模扩展及其将变得多么强大的人之一。我们创办公司背后的想法是，看着技术进步如此迅速，我们说：“看，我们确实需要把这件事做好。”
**English / Original**
Yes, so, you know, go going back to the founding of Anthropic, you know, we were among the first myself and my co-founders at Anthropic to to document the scaling of the technology and how powerful it would become. And you know, the idea behind how we started the company is, you know, we looked at how quickly the technology was advancing and said, "Look, we we really need to do this right."
## 00:01:54
**中文**
它会非常强大。对公司、与公司合作的人以及整个社会而言，都存在巨大机会，但我们确实需要做好。安全对我们有几层含义。在模型层面，它意味着确保 AI 模型的行为更可预测、更可靠。我们开发了多种方法，从宪法 AI——一种用一组特定原则训练 AI 的技术——到称为机制可解释性的技术，即像用 MRI 观察大脑一样查看模型内部，理解它为何会如此运作，并逐步追踪其思维。
**English / Original**
You know, it's going to be very powerful. There's an enormous opportunity for the company and for people who work with the company and for society overall, but we really need to do it right. Um, and so safety means a number of things to us. At the level of the model, it means making sure that the behavior of the AI models is more predictable, is more reliable. Um, so we developed a number of methods ranging from constitutional AI, which is a technique we use to train AI from a specific set of principles to something called mechanistic interpretability, which is looking inside the model as you would look inside a brain with an MRI to understand why it does the way it does and to trace its thoughts step by step.
## 00:02:34
**中文**
使用这些方法，我们能让模型更可预测、更可靠，也更可能正确。我们发现，全球企业非常重视这一点。在消费者领域这没那么重要；但当你服务全球企业时，如果是在讨论药物开发，就必须确保精确、确保正确；若讨论一家公司的财务分析、如何制造一种新材料，也都必须正确；如果讨论
**English / Original**
And so by using these methods, we're able to make the models more predictable, more reliable, and more likely to be correct. And what we found is that the businesses of the world, the enterprises of the world, value that a lot. It's not as important in the consumer world, but when you want to serve the businesses of the world, you know, if if you're having a debate about drug development, you know, you you need to make sure you're precise. You need to make sure you're getting it right. If you're having a debate about financial analysis of a company, you need to get it right. If you're having a debate about, you know, how to how to manufacture a new material, you need to get it right. If you're having a debate
## 00:03:08
**中文**
怎样最优地做电子商务，也必须正确。因此企业真正信任我们能提供更可靠、适合其需要的技术；它不是为娱乐而设计，而是为它们创造严肃的经济价值而设计。是的。其中一些当然体现在：我们有企业聚焦而不是消费者聚焦。因此即使我们的消费者产品，也更聚焦于向人们提供信息、帮助他们完成事情，聚焦于生产力和具体工作。
**English / Original**
about how to best do e-commerce, you have to get it right. And and so enterprises really trust that we can give them a technology that is more reliable, that suits their needs, that isn't designed for entertainment, but it is but is designed for them to create serious economic value. Yes. Um, so some of those, you know, are are covered in the fact that we have an business focus rather than rather than a consumer focus. So even our consumer product, it's more focused on providing people information, helping them to get things done, um, you know, things that are that are that are more oriented towards um,
## 00:03:51
**中文**
但人们也发现，由于我们拥有大量 AI 人才，并非常专注于以良好方式打造它，Claude 往往更善解人意、更有同理心，是更周到的模型。人们说它听起来不像某些竞争对手那样机械。因此有一部分消费者发现，Claude 是真正吸引他们的模型。
**English / Original**
uh, productivity and uh you know, concrete um efforts. But, people have also found because, you know, we have we have a lot of AI talent and we really focus on crafting this in in a good way. Um people have also found that, you know, Claude is often more understanding. It's more empathetic. Um it's it's a more thoughtful model. People say that it sounds less robotic than some of some of some of the competitors. And so, there's a segment of consumers who have found that, you know, Claude is Claude is a a model that that really appeals to us.
## 00:04:24
**中文**
我们看到这种情况会随时间扩展。是的。我会说，它最主要与我们选择和企业而非消费者合作有关。某些多模态能力对企业很重要，例如分析图像或视频，对客户服务、安保或其他应用都非常重要。但生成图像或视频更像娱乐，因此我们不做，因为那更面向消费者。不过，我们的企业聚焦在许多情况下也源自我们对安全的重视，对吧？我们选择企业聚焦是因为
**English / Original**
And we see that expanding over time. Yes. Um so, I would say what it has most to do with is that, you know, we're we're choosing to work with businesses rather than consumers. So, some types of multimodality are important to businesses. So, for example, analyzing images or analyzing videos can be very important, for example, in customer service or security or other applications. Um but, generating images or videos, that's more of an entertainment thing. And so, we don't we don't do that because it's more consumer-focused. But again, our enterprise focus in many cases comes from our value on safety, right? We're choosing the enterprise focus because
## 00:05:07
**中文**
它与我们的安全价值观相容，而不是反过来。因此，当我们看到人们为在消费领域竞争而做的这些事——制作让人上瘾的短视频、让模型极具黏性、让人与模型形成恋爱关系——我们会说：“我们不必做任何一项，因为我们并不试图赢得大众消费市场，也不想卷入一场你死我活的竞争。”
**English / Original**
it's compatible with our values on safety, not the other way around. Uh uh and and so, when we look at these things that people do to compete in the consumer space, right? Making these uh you know, making these short-form videos that people get addicted to, the models becoming super engaging, people having romantic relationships with models, we look at it and we say, "We don't need to do any of that because we're not we're not trying to win the mass consumer market. We're not trying to get into a We're not trying to get into a rat race.
## 00:05:36
**中文**
而且出于我们围绕安全的价值观，我们本来就不想做那些事。因此我们选择了一个与价值观非常一致的商业模式。是，是，是。我的意思是，我们总体上认为自己想帮助人们完成事情，帮助人们成为有思想的人。
**English / Original**
And because our values of of our values around safety, we don't want to do that stuff anyway." And so, we've chosen a business model that aligns very well with our values. >> Yes. Yes. Yes. I mean, you know, I I think we generally think of ourselves as wanting to help people get things done. Wanting to help people be thoughtful.
## 00:05:59
**中文**
帮助人们成为最好的自己。我们有一个名为“Keep Thinking（持续思考）”的品牌活动。其理念是，我们不想让 AI 模型钝化你的思考；不想让 AI 模型替你思考，再用愚蠢的东西娱乐你。我们希望模型能帮助你的认知过程，帮助你思考。
**English / Original**
Wanting to help people be the best version of themselves, right? Or we have this brand campaign called keep thinking. And the idea is we don't want the AI models to dull your thought. We don't want the AI models to, you know, do the do do the thinking for you and keep you entertained with stupid things. We want We want the models to um you know, to to kind of to to to help your cognitive processes. To help you think.
## 00:06:25
**中文**
我们相信这是科技史上一家软件公司最快的增长。我认为没有任何可比案例能以如此快的速度增长；在此之前的两年里，我们每年也已实现 10 倍增长。我们认为原因是，全球企业使用 AI 的巨大未满足需求。当我们走访企业——包括过去几天在日本拜访的数家企业——发现它们很高兴看到有一家完全面向服务它们的公司，为它们服务是我们的第一优先级。我们
**English / Original**
We believe is the the fastest growth for a software company in uh in the history of technology. Um I don't think there are any comparable examples of the growth happening so fast. And we've actually seen 10x growth per year for each of the previous 2 years before that as well. Um So, uh the the reason we think that's the case is, you know, there's this great unmet need for the businesses of the world to use AI. I mean, when we go to businesses, including several that we've visited in Japan over the last few days, we find that these businesses are very happy to see that there's a company that is totally oriented around serving them. That it's their it's their it's our first priority to serve them. We
## 00:07:05
**中文**
为它们制作产品、制作模型。它确实像野火一样蔓延开来，这也是我认为我们增长如此之快的原因。由于增长如此之快，公司价值也会随之上升。所以一切归根结底在基本面：公司价值上升，这是赚钱的一种正当、诚实的方式。
**English / Original**
make products for them. We make models for them. Um uh and and you know, it's really it's really caught on by wildfire. And that's why I think we're growing so fast. And you know, because we're growing so fast, that that causes the value of the company to to grow up. So, it all goes down to the fundamentals. The value of the company goes up. It's it it's a good honest good honest way to make money.
## 00:07:47
**中文**
在我们在日本拥有一名员工之前，我们在日本已大约有 1.5 亿美元收入。我们看到日本大小企业对使用我们模型的强烈需求。一些从未与我们交谈过的企业已在使用模型，它们只是用了公共 API。但我们知道，如果想实现真正大型的企业部署，必须与日本企业建立长期关系；我们知道日本就是这样运作的。
**English / Original**
>> You know, before we had a single employee in Japan, we had roughly 150 million dollars of revenue in in in in Japan. You know, we saw a lot of hunger among companies big and small in Japan to use our models. Some were already using our models without having ever even talked to us, right? They just used the used the public API, but we know that if we want to do the really big enterprise deployments, we have to have a long-term relationship with companies in Japan. We know that's how Japan works.
## 00:08:14
**中文**
因此我们在这里开设办公室，人员虽不多但发展很快。这也是我此刻在日本的原因。我计划经常回来。我们确实看到这种潜力能增长许多倍、巨大增长。昨天我参加了 Anthropic 在此组织的面向建设者和初创创业者的峰会，现场的热情显而易见。我们能看到许多不同想法，人们想构建各种事物；许多人想用我们的模型打造新的、创新的东西并让事情发生。因此我们看到这种活力显而易见，也看到
**English / Original**
And so we opened our we opened our office here. We have a few people. very fast. This is why I'm in I'm in Japan right now. I'm planning to come back often. And you know, we we we really see the potential for that to to grow to grow by many times, to grow enormously. I was at a summit for builders, for startup entrepreneurs here yesterday organized by by Anthropic, and the excitement in the room was palpable. We could see you know, so many different ideas for things people wanted to build. So many people who you know, want wanted to you wanted to use our models to build something new and innovative and make something happen. So we can see that the energy is palpable and you know, we see
## 00:08:55
**中文**
这里不仅能带来大量收入，也能让我们与企业协同工作以提高它们的价值。你的问题是，日本会不会提高我们的企业价值？我认为这是错误的问题。我们想提高日本公司自身的企业价值；作为副作用，这也会提高我们的企业价值。
**English / Original**
this as a place not just where we can get a lot of revenue, but where we can we can actually work synergistically with the companies to enhance their value as well, right? Your question was you know, will will Japan enhance our enterprise value? I think that's the wrong question. I think we want to we we we we want to enhance the enterprise value of companies in Japan and as a side effect, that will also enhance our our enterprise value.
## 00:09:23
**中文**
我认为两者都会做到；但再说一次，我们有一种能让两者一致的商业模式，这正是我们的目标。一些例子是：乐天在使用 Claude，包括 Claude Code；松下在设计和软件中使用 Claude；野村综合研究所也在使用 Claude。这些是大公司的例子。此外，我们看到极广泛的个人、个人公司，
**English / Original**
I I think we're going to do both, but but you know, again, we have a business model that aligns those two. That is that is what we're aiming to do. Yes. So some some examples. um are, you know, uh Rakuten is using uh uh Claude including including using Claude code. Um uh uh uh Panasonic uh is is using Claude for design, also for softwares, and the Nomura Research Institute is using Claude as well. So, those are those are some examples of big companies, and then and then beyond that we see um just a very wide variety of again individuals individuals companies,
## 00:10:05
**中文**
个人创业者使用 Claude API 构建新的创新产品，并使用 Claude Code 帮助编写构建这些产品的代码。是，是。我的意思是，我认为这就是我们已在做的，只是做得更多。因此我们将继续在日本扩大存在：在市场进入方面与日本大中小企业合作；在初创方面接触并支持初创企业；以及在技术方面帮助模型在日语和在
**English / Original**
individual entrepreneurs who are using the Claude API to, you know, build some some some new innovative product and are using Claude code to help to help write the code to build it. Yes. Yes. I mean, you know, I think it's it's you know, it's it's what we've done already except more so. So, you know, I I you know, I think we're going to continue to build out a presence in Japan both on the go-to-market side to work with uh you know, to work with Japanese companies large and small, on the kind of startup side to, you know, engage with startups and support startups, and also on the technical side to help make the models better in Japanese and at,
## 00:10:48
**中文**
日本强势的行业中做得更好。我们目前的东京办公室一个月前刚开业。我们正在日本各个领域招募优秀人才。我认为我们会努力深化关系，并与一些日本最大的公司建立长期伙伴关系，同时支持初创生态系统。我认为，对我们而言，这 1.5 亿美元可变成数十亿美元；同时，我们也能为日本公司增加数十亿美元的企业价值。
**English / Original**
you know, industries that are strong in Japan. Um so, you know, our our current um our current Tokyo office was just opened a month ago. You know, we're we're we're recruiting for strong talent across across the board across the board in Japan. Um and I you know, I think we're going to try to deepen our relationships and build long-term partnerships with uh you know, some of the biggest companies in Japan as well as supporting the startup ecosystem. And and and so, you know, I think I think, you know, on our side that 150 million can become billions, and then at the same time I think we can add, you know, many billions in enterprise value to to to the Japanese companies as well.
## 00:11:30
**中文**
是的，我认为 AI 是一种在各个领域都有不同应用的技术。目前我认为代码领域的进展多于其他领域；我能想象一个构建软件将容易得多的世界。因此。
**English / Original**
Yes, um so, you know, I I I I think AI is one of these technologies that has different applications across the board. So, you know, right now I think, you know, code is happening more than more than other areas and I can envision a world where it's just much easier to build software. So,