全栈构建者的崛起：微软 CEO 萨提亚·纳德拉谈超级杠杆型通才

The world is going to be very skeptical of tech and tech companies that say, "Trust us, we've got it. 世界会对那些说「相信我们，我们有把握」的科技公司持极大的怀疑态度， The future is going to be glorious." 未来将会是美好的。」 You kind of have to deliver tangible benefits because it's too important this time around. 你必须拿出切实的成果，因为这次太重要了。 It's too much of the economy for it not to be the case. 它涉及的经济体量太大，不可能不是这样。 True ambition is about making the impossible possible. 真正的雄心壮志，是让不可能成为可能。 I take great inspiration from sort of the people who were managing the Azure network. 我从管理 Azure 网络的那些人身上获得了很大的灵感。 We built in the last 15 months more Azure capacity than we built in the first 15 years. 我们在过去 15 个月里建设的 Azure 容量，超过了前 15 年的总和。 I mean it's crazy. 我是说，这简直疯了。 Our job is not to do Azure networking. 我们的工作不是做 Azure 网络。 Our job is to build the agentic system that does Azure networking, right? 我们的工作是构建能做 Azure 网络的智能体系统，对吧？ The way to get to information, way to educate yourself, way to continuously keep yourself updated has changed so much. 获取信息的方式、自我教育的方式、持续精进的方式 Maybe the next big startup could be someone who builds a new university, a new pedagogy even of how to get someone to go through a curriculum and find economic opportunity. 也许下一个伟大的创业公司，会是有人创建一所新大学、一个新的培训平台。 That's highly valuable. 那将具有极高的价值。 Please welcome Swix Saragoa Allad Gil and chairman and chief executive officer of Microsoft Satinadella. 欢迎 swyx、Sarah Guo、Elad Gil，以及微软董事长兼首席执行官 Hello, son. 你好啊，小伙子。 Uh, I'm so excited to be here. 我太兴奋了，能来这里。 Welcome to a crossover episode of No Priors in Lane Space with Sat Nadella. 欢迎收看 No Priors 与 Latent Space 联合录制的跨界特别集，嘉宾是 Satya Nadella。 Um, congratulations on an amazing build. 嗯，恭喜你们举办了一场精彩的 Build 大会。 No, thank you so much and it's great to be with both of you. 非常感谢，能和你们两位在一起真的很高兴。 I listen to both of you or both the podcast all the time. 我一直在听你们两个的播客。 It's great to be on it. 能上节目太棒了。 Thank you so much. 非常感谢。 So you're just talking about um these amazing uh announcements from across the Microsoft estate all morning for I think three hours. 你刚才在谈论微软各方面那些令人惊叹的发布， What is the uh what's the most important reflection or takeaway you have? 你最重要的感悟或心得是什么？ I I'd say there are uh perhaps the the biggest one for me is let's sort of conceptualize this more as an ecosystem play as opposed to a single model or even a single platform right 我想说，也许最重要的一点，让我们先聚焦在这上面， I mean you know whenever I at least for me having grown up at Microsoft having seen whatever four major platform shifts u I sort of fall into that um camp where a platform is defined by fundamentally its ability to create more value about the platform versus what's captured in the platform. 我的意思是，至少对我来说，在微软成长起来、经历过那些年代， And so if you you view what's happening right now, I think this morning's keynote was how can any company whether it's an AI native company or a traditional enterprise company participate as a firstass participant where they can point to AI they created. 如果你审视现在正在发生的事情，我想今天上午的主题演讲 Right? 对吧？ It's not that they don't use other people's AI. 不是说他们不用别人家的 AI。 Of course they will. 当然他们会用。 But to me, what's the path? 但对我来说，路径是什么？ What's the recipe? 方法论是什么？ How do I do it? 怎么做？ What does a stack look like? 技术栈长什么样？ What does the tooling look like? 工具链长什么样？ What is valuable? 什么是有价值的？ How do you do that? 怎么实现？ That's it. 就是这样。 That's sort of our job to do. 这大概就是我们该做的事。 Yeah. 对。 Ecosystem strategy is uh very complicated, right? 生态系统战略非常复杂，对吧？ Because you end up building certain components, partnering for certain components, supporting them. 因为你最终得自己做某些组件、为某些组件找合作伙伴，同时支持另一些组件。 You just announced this big suite of models. 你们刚发布了一大套模型。 Like tell us a little bit about the uh training strategy for Microsoft. 能说说微软的训练策略吗？ So, so the thing that we wanted to do with the MAI models was to build and as Mustafa talked about first of all a great lineage right starting with pre-training uh with very good data quality uh doing all the abilations making sure because in in some sense it's become even harder to build a clean lineage model just because there's so much stuff out there um that you truly need to ablate out to be able to have a fantastic pre-trained model. 我们在 MAI 模型上想做的，以及 Mustafa 说的，是构建并 In fact, that's one of the challenges of a lot of the openw rate models is they look great on one benchmark or two, but they're not great on practice. 实际上，很多开源模型的挑战之一就是它们看起来 So, that's why in fact even in our FDES are are pretty gone really excited about these MAI models because how the heck can a small 5D model hill climb? 所以，这也是我们的 FDES 对这些东西感到非常兴奋的原因， uh and it goes back a little bit to what I think is ultimately the key thing to do which is try to pursue finding that cognitive core. 这在某种程度上回归到了我认为最终最关键的事情， Uh so to me starting with a clean lineage then creating that ability for companies to be able to use this right not just as a generalist but to create their own specialist by building this hill climbing scaffold around it right so it's not just the model but you have a hill climb scaffold around it then you will start building your rle you will start collecting the traces most importantly you'll have private eval because we know all the eval out there are good, interesting, but they're not really that critical at this point because they all can be maxed. 对我来说，从干净的血统出发，然后为公司创造这种能力 And so the point is each company will have its own private eval. 关键在于，每家公司都将拥有自己的私有 eval。 And so that end to end platform story around our models is sort of uh what I think is interesting. 所以，围绕我们模型的端到端平台故事，是我认为最重要的 And then the one other thing Sarah since you brought that up is I do feel there's a new frontier like people talk about the frontier and you're operating at the frontier. 还有一件事，Sarah，既然你提到了，我确实感觉有一个新的框架 Um, interestingly enough, if you add a little temporality to it, you can use, let's say, in in in fact the the Lando Lakes demo we showed was pretty cool. 有趣的是，如果你加一点时间维度，你可以用，比如说， We used whatever GPD 55, right? 我们用了什么 GPD 55，对吧？ Then you collected a bunch of traces and then you took a 5B reasoning model and achieved higher. 然后你收集了一批追踪数据，再用 50 亿参数的推理模型，达到了 Uh, so that is another aspect of what it means to appear, you know, operate at the frontier. 嗯，这是在前沿运营意味着什么的另一个维度。 Yeah. 对。 I I think uh I first of all I have to congratulate you on basically building a frontier neolab inside of Microsoft in two years. 我首先得恭喜你，基本上是在构建一个前沿的 Um I'm wondering you know you have all this AI strategy that you're rolling out. 嗯，我想问，你有这整套 AI 战略正在推进。 I'm wondering what do you know now that you wish you would tell yourself two years ago or two to three years ago. 我想问，你现在知道的事情，有哪些是你希望两年前就告诉自己的？ Three years for the Jensen partnership, two years for uh Mi. Jensen 合作三年，微软那边两年。 Yeah. 对。 I mean I think the the thing when that I reflect quite a bit right which is sort of obviously I got into all this when I got excited by the the scaling laws paper and you know when you know even the open AI partnership came about when those folks said hey we're going to really throw a lot of computer transformers 我想，我回头思考很多的一件事，就是那个很明显的道理， uh and they've helped right the thing that I always look back and say wow these things um do have capability that they're climbing up 它们有帮助，对吧，我总是回头说，哇，这些东西 I mean this you know this crude way of saying it is intelligence is log of compute kind of works. 我是说，有个粗略的说法是，智能大约是计算量的对数， Now what I think we underestimated perhaps is the real world complexity of deploying these so that they actually deliver the value in the real world. 现在，我认为我们也许低估了的，是在现实世界中部署这些技术的复杂性。 Right? 对吧？ So the outcomes as measured by any benchmark is interesting important but the true eval is when people out there are able to do unique things that they only can value and it's very measurable right that I wish we had sort of even like had more in our consciousness right which is as an industry because right now I think when people say wow I don't want a token max 所以，以任何基准测试衡量的结果很重要，但真正的 eval 是 it's an artifact of us not having thought ourselves as an industry that we are using tokens to create value every step of the way. 这是我们作为一个行业没有想清楚的产物，因为我们正在使用 token So I think that's kind of what I wish we had gotten there but I'm glad we are here. 所以我想，这是我希望当时就想到的，但我很庆幸我们现在想到了。 What are some of the use cases that you've seen that have created the most value for your customers because I know that people talk a lot about code and I think it's pretty clear that that's something that's having very large scale impact. 你们见过的哪些用例为你们的客户创造了最多的价值？ Are there other areas that you find in common that your customers are really benefiting? 还有哪些领域是你发现客户真正受益的共同点？ Yeah, I think yeah to your point obviously coding is now got but it's interesting by the way ladish to you even talk about the coding right which is coding is worked so well that we now have to rebuild the IDE right 对，我想是的，从你的角度来看，编码现在显然已经很普遍了，但有趣的是，对吧，即便就编码本身来说，编码效果好到我们现在必须重建 IDE 了，对吧 I mean it's kind of nuts to see what we launched is like oh my god 我是说，看看我们发布的东西，简直说了个天呐， I have these 100 agent sessions 我有这 100 个智能体会话， I the cognitive load it transfers back to me as a human is so excessive that now I need a new UI 它转移给我作为人类的认知负担是如此沉重，以至于我现在需要 oh by the way I like the the chat as the only artifact is also impossible. 顺便说一句，我觉得只把聊天当成唯一产物也是不可能的。 So that's why we need a canvas. 所以这就是我们需要画布的原因。 So it's kind of interesting for all the things about where is software needed or where is UI needed. 所以，这很有趣，对于软件在哪里需要或在哪里 Uh you kind of need that even for code right in a fully agentic world. 即便是在完全智能体化的世界里，代码也需要这个，对吧。 But that said, one of the things that we are starting to see, we started seeing with co-work, but even some of the work we we showed with auto u um autopilot, right, on what you see with claws is a good one because if you sort of think about a lot of human capital is doing the glue work, right? 但话虽如此，我们开始看到的一件事，我们在 Teams 中开始看到的， If you now can augment that with tokens slash agents that are longunning, durable, right? 如果你现在能用持续运行的长时智能体来增强，对吧， then your ability to scale even what is still judgment and glue work gets amplified like coding does. 那么你扩展那些仍属于判断力和粘合工作的能力就会被大幅放大。 Uh so you can like I'm positive that 6 months from now we'll all be saying 嗯，所以我很确定，6 个月后我们都会说， oh wow 哇， like all through night the night there was a bunch of stuff that all these autopilots that I have working on my behalf with my delegated authority so to speak right 整个夜晚，都有一堆东西，那些自动驾驶仪 I can sort of given even my identity did a bunch of work then of course I'll need my new ad to say what did you do like am I did I do this work and so on so I think that that's where compressing of workflows uh complete leading of tasks. 基于我的身份，它做了一堆工作，然后当然我还需要我自己的 Uh that's where I think a lot of the value gets created. 这就是我认为大量价值被创造的地方。 I think you raised a really interesting point which is there's the actual agent is doing the code and then there's a harness around it and that's the environment that's the context that's everything you're setting up as a developer around actually a coding agent. 我觉得你提出了一个很有意思的观点，就是实际做编码的智能体，以及围绕它的执行框架，那是环境、上下文、作为开发者你在编码智能体周围搭建的一切。 What is the harness for the enterprise? 企业的执行框架是什么？ Is there an equivalent concept for broader productivity work or how do you think about that concept sort of 对于更广泛的生产力工作，有没有类似的概念，或者你怎么看待那个概念？ that's right. 正是如此。 So, so in some sense you kind of want the harness to define the models the the data uh and the tools and so that you have a loop across those three and so what we are trying to first of all make sure is each of our products that we build right whether it's GitHub copilot or the security cop the stuff we showed with mdash or even the discovery for science it doesn't matter 所以，在某种意义上，你希望执行框架来定义模型、数据和工具，使你在这三者之间有一个循环，所以我们首先要确保的是，我们构建的每个产品，无论是 GitHub Copilot 还是安全 Copilot，还是我们展示的 mdash 相关的东西，还是科学发现，都一样 all of them are multimodel harnesses um with tools tools access so that you can do this progressive uh disclosure of tools even so that they're token efficient. 它们都是多模型执行框架，带有工具访问，这样你就能做 Uh and then you're feeding it with very rich context because that's sort of the other hard lesson we've learned in the last two years is oh my god 然后你要用非常丰富的上下文来喂它，因为这是另一个关键， the amount of work you need to do to prep the context layer uh such that your plan can execute in the most efficient way is where the magic is. 你需要做大量工作来准备上下文层，使你的计划能够执行。 So we have in our case we have the GitHub harness which essentially we're using across all our products. 在我们这边，我们有 GitHub 执行框架，我们基本上在所有地方都在用， It's available in foundry and we're open like you can use your llama harness whatever or you can use the um uh you know any open harness or any harness of yours and train with your tools and multiple models and your context. 它在 Foundry 上可以用，我们是开放的，你可以用你自己的 Llama 执行框架，随便， And so that's the pitch because right now a lot of dialogue is um hey if I train the harness plus tools and the model together you get eval. 这就是我们的主张，因为现在很多讨论是，嗯，如果我在执行框架上训练的话 And what we are proving out is and the best example of that is what we did with mdash right because when it launched um it found bugs or vulnerabilities that were not found by mythos 我们正在证明的，以及最好的例子就是我们对 mdash 所做的事情，