吴恩达谈 AI 影响：不雇佣不懂 AI 的工程师

安德鲁，非常感谢你今天在达沃斯加入我们。

Andrew, thanks so much for joining us today here in Davos.

谢谢。我几小时前刚下飞机，所以在这里的第一个会议就是与经济时报的会面。

Thank you. I just got off a plane a few hours ago, so my first meeting here is with ET.

我想先问你，AI 对印度意味着什么，特别是对印度最大的产业之一——IT 服务业。一个主要的担忧是，AI 的快速进步意味着很多软件编码和开发，甚至中级开发，现在都由机器完成。像 Claude Code 这样的工具被认为非常先进。这对像印度 IT 业这样的大型服务业意味着什么？

I'm going to start by asking you about what AI means for India and specifically for one of India's biggest industries, which is the IT services industry. A big concern has been that rapid advancements in AI mean that a lot of software coding and development, even mid-level, is now done by machines. Tools like Claude Code are cited as getting really advanced. What does this mean for a large services industry like India's IT industry?

我认为对长期颠覆的担忧是有道理的。风险与机遇在于，借助 AI 工具，不擅长 AI 的人的生产力会越来越明显低于擅长的人。所以坦率地说，今天我不会雇佣一个不精通 AI 工具的软件工程师。但人们不太理解但正在蔓延的是，对于许多其他职位——从会计师到营销人员、HR 到 IT 后端管理——公司都要求熟练使用 AI 技能，甚至可能到编程的程度。所以坦率地说，今天当我招聘营销人员或 HR 时，我强烈倾向于雇佣那些能用 AI 编写软件以成为更好的营销或 HR 专业人士的人。对于印度这样一个拥有庞大专业服务和外包产业的国家，风险与机遇在于：如果这群世界长期信任的聪明勤奋的人能够快速提升技能，那么印度将继续提供巨大价值。但如果技能提升不够快，那么我认为工作颠覆可能会很严重。所以我感到印度和其他几个国家面临很大压力，需要找到正确的培训方法。

I think the worries about secular disruption are well placed. The risk and opportunity is that with AI tools, individuals not skilled with AI are increasingly dramatically less productive than those who are. So frankly, today I would not hire a software engineer that does not know how to use AI tools in a very sophisticated way. What is not well understood but is spreading is that for many other job roles—from accountants to marketers, HR professionals to IT backend management—companies are demanding sophisticated use of AI skills, maybe even to the point of coding. So frankly, today when I hire a marketer or HR professional, I would strongly prefer to hire someone who knows how to use AI to the level of being able to write software to help be a better marketer or HR professional. For India, with a very large professional services and outsourcing industry, the risk and opportunity is that if this group of very smart, hardworking people that the world has trusted for a long time can upskill rapidly, then India will continue to provide tremendous value. But if the upskilling doesn't take place rapidly enough, then I think the job disruption could be significant. So I feel a lot of pressure for India and a handful of other nations to figure out the training to get this right.

我想谈谈基于 LLM 的 AI 模型和聊天机器人的近期演变。我认为世界在 2022 年的 ChatGPT 时刻才真正意识到这一代生成式 AI 工具的可能性。从那以后，当你看到这些模型的演变和它们的能力时，你是印象深刻还是感到失望？

I want to talk about the recent evolution of LLM-based AI models and chatbots. I think the world woke up in a big way during the ChatGPT moment of 2022 to the possibilities of this generation of GenAI tools. Since then, when you look at the evolution of these models and what they're capable of, have you been really impressed or underwhelmed?

我认为进展很好。自第一个版本的 ChatGPT 以来，模型一直在改进。我觉得炒作速度极快，而商业现实也很快，但不像炒作那么快，事情通常如此。是的，我们还需要做更多工作来构建 AI 应用、AI 智能体和基于这项技术的智能体工作流。那种认为我们将在两个季度或一年内实现 AGI 的炒作只是炒作。我严重怀疑任何合理的 AGI 定义会在短期内实现。但与此同时，即使没有达到人类通用智能水平的计算机，我们在业务流程自动化和工作流重构方面还有很多工作要做。我认为许多不同行业将迎来重大产业转型，十年后我们仍将在这方面努力。

I think it's been great. Since the first version of ChatGPT, the models have continued to improve. I feel like the hype has gone incredibly fast, and the business reality has also gone very fast, but not as fast as the hype, as things often are. Yes, more work lies ahead to build AI applications, AI agents, and agentic workflows on top of this technology. This hype that we'll get AGI in two quarters or a year or whatever is just hype. I seriously doubt that's going to happen for any reasonable definition of AGI. But at the same time, without having computers as intelligent as humans in the general sense, there's so much work to do with business process automation and workflow re-engineering. I think there'll be many big industrial transformations coming in many different sectors, and we'll still be working on this a decade from now.

Yann LeCun，另一位 AI 领域的知名人物，最近表示他不认为当前基于 LLM 的模型是通往 AGI 的正确道路，需要完全不同的浪潮。你在这个辩论中持什么立场？

Yann LeCun, another big name in AI, has recently said he does not believe the current wave of LLM-based models is the right way to approach AGI, and a totally different wave is required. Where do you stand on that debate?

所以我认为他是对的。同时，我所知道的最合理的 AGI 定义是能够完成人类任何智力任务的 AI。目前我不知道有任何技术本身是通往 AGI 的路径。AI 已经被定义和重新定义了很多次，几乎失去了意义。当我开始谷歌大脑项目时，我为团队设定的首要任务是构建非常大的神经网络并投入大量数据。AI 的学习方式与人类不同。实际上，我在谷歌大脑之前就开始研究深度学习了。我阅读了大量神经科学论文，试图理解人脑的工作原理，看是否能启发更好的 AI 算法。但这些努力大多没有用。许多国家正在考虑主权 AI。一个非常高效的方式是投资开源和开放权重模型。在过去一年中，开源开放权重模型领域的一股巨大力量来自中国。但美国似乎越来越不欢迎移民，无论是在技术领域还是其他领域。你怎么看？我认为这很糟糕。我认为这是美国一个巨大的非受迫性失误。问题是，既然 AI 已经颠覆了一切，印度能否足够快地提升技能，使其才华横溢、聪明勤奋的创新者和企业不仅能跟上，还能实现跨越式发展并发明下一个东西？

So I think he's right. At the same time, the most reasonable definition of AGI that I know of is AI that could do any intellectual task that a human can. There's no technology that I know of today that by itself is a path to AGI. AI has been defined and redefined so many times it's almost lost meaning. When I started the Google Brain project, the number one mission I set for the team was to do really large neural networks and throw a lot of data at it. The way AI learns is different than the way humans learn. Actually, I started a little bit before Google Brain when deep learning was coming up. I was reading stacks of papers on neuroscience to try to figure out how the human brain works to see if that could inspire better AI algorithms. Those exercises turned out mostly not to be useful. So many nations are thinking about sovereign AI. One way to do that very efficiently is investing in open source and open-weight models. In the open-source open-weight model space, one huge force over the last year has been the models coming out of China. But increasingly, the United States appears to be less welcoming of immigrants, whether in technology or any other field. How do you look at this? I think it's awful. I think it's a huge unforced error on the part of the United States. The question is, now that AI has shaken everything up, can India upskill rapidly enough so that its talented, smart, hardworking innovators and businesses can not just keep up but potentially leapfrog and invent the next thing?

所以如果我们有 AGI，就意味着 AI 应该能够花五年时间写一篇博士论文，或者学会在丛林里开卡车，而人类可能只需要几分钟或几十分钟就能做到。大语言模型本身并不是通往 AGI 的路径，但据我所知，今天没有任何一项技术本身就能通往 AGI。要实现那种 AGI，需要新的技术突破，而我们谁都不知道那具体是什么。所以你可以把任何技术放进去，说今天存在的技术本身不是通往 AGI 的路径，这句话总是对的。但有趣的是，由于 AGI 的炒作，我看到很多企业试图降低 AGI 的门槛，以便更容易达到。这有问题，因为大多数社会认为 AGI 意味着 AI 和人类一样聪明。对我来说，当我们有 AGI 时，意味着 AI 应该能做远程工作者能做的一切。我们离那还很远。当公司重新定义 AGI 降低门槛，然后用一些狭隘的技术定义说他们可能已经实现了 AGI，这会误导公众和商业领袖，让他们认为 AI 的发展轨迹比实际更强大。我看到这误导了高中生，他们因此做出不同的职业选择，因为他们认为 AGI 会让某些职业过时。我还看到公司根据 AI 在几年内会比我认为的更智能这一假设做出非常大的资本配置决策。所以 AGI 被定义和重新定义了太多次，几乎失去了意义。但即便如此，即使没有达到这个很高的 AGI 门槛，AI 也非常令人兴奋，非常有价值，印度、美国和其他国家朝着这个方向建设将会令人兴奋。

So if we have AGI, it means AI should be able to spend five years to write a PhD thesis or learn to drive a truck through a jungle in maybe minutes or tens of minutes because a human can. LLMs by themselves are not a path to AGI, but there's no technology that I know of today that by itself is a path to AGI. To achieve that type of AGI will need new technical breakthroughs that none of us really know what exactly it is. So you can plug in any technology and say that technology that exists today by itself is not a path to AGI. That will be a true statement. But one interesting thing is, because of the hype around AGI, I've seen a lot of businesses attempt to redefine AGI at a lower bar to make it easier to hit. And this is problematic because most of society thinks AGI means AI that is as smart as humans generally. To me, when we have AGI, it means AI should do everything that a remote worker would be able to do. We're very far away from that. And when companies redefine AGI to lower the bar and then say with some narrow technical finicky definition maybe they've achieved AGI, it misleads the public, misleads business leaders into thinking AI is on a trajectory to be more powerful than it actually is. I've seen this mislead high school students that are now making different career decisions because they think AGI will obsolete certain careers. And I'm seeing companies make very large capital allocation decisions on the basis that AI will be more intelligent in a couple years than I think it actually will be. So AGI has been defined and redefined so many times it has almost lost meaning. But having said that, even without reaching this really high bar of AGI, AI is incredibly exciting and is incredibly valuable, and India and the US and other nations building towards that will be exciting.

我相信 CEO 们一有机会就会问你，‘嘿 Andrew，告诉我们发生了什么，我们应该怎么做……’你从 CEO 那里得到的最常见问题是什么？

I'm sure CEOs ask you all the time wherever they can get hold of you about 'Hey Andrew, tell us what is happening, how should we...' What's the most common question that you get from CEOs?

实际上，我和我的商业伙伴 Kirsty Tan 通过 AI Aspire 为大型企业提供咨询。当我们观察许多行业正在发生的转型时，我与 CEO 们讨论最多的是如何帮助企业采用 AI。我认为首要问题是人员变革：我们如何提升员工技能，带他们一起前进，如何真正做好人员变革管理，帮助那些最终要完成 AI 采用工作的人。其中，我看到很多企业采用了‘百花齐放’策略，但大多效果不佳，这有特定原因。自下而上的创新很重要，我们应继续做，但我发现很多自下而上的创新最终成为点解决方案，有人看到业务的一小部分，将其效率提升 10 倍。这很好，但企业的整体收益是渐进的。所以我们看到很多 5% 或 10% 的效率提升，这不错，但这不是 AI 承诺的。相反，它通常需要自上而下的创新，或者有全局观的自下而上创新，理解企业创造价值的所有步骤，然后重新设计更广泛的工作流程。我看到这能带来更多价值。所以自下而上的创新很好，但有时需要一个有全局观的人，无论是自上而下的领导者还是自下而上的人，来思考工作流程的重新设计。我看到企业做对的地方，能带来更多价值。领导力来自高层。自下而上的影响也很重要，我们也要多做，但领导力来自高层。所以我给每位 CEO 或高管的一个挑战是：如果你认真对待 AI，就自己学习一下。确保你个人除了对 AI 的热情外，还要花几个小时深入了解技术，确保你有知识来驱动团队、分配资源、确定优先级，然后推动业务成果。

Actually, my business partner Kirsty Tan and I do a lot of work through AI Aspire advising large businesses. As we look at the industry transformation happening through many sectors, the biggest thing I end up talking with CEOs about is how to help businesses adopt AI. I would say the number one thing is the people change: how do we upskill people, take them with us, how do we really do the people change management that helps the people who are ultimately going to do the work to adopt AI. Of that group, I would say the biggest category is a lot of businesses have done the 'let a thousand flowers bloom' strategy and it's mostly not working, and there's a specific reason for that. Bottom-up innovation is important, let's keep doing it, but I find a lot of bottom-up innovation ends up with point solutions where someone sees a small piece of the business and makes it 10 times more efficient. That's nice, but the overall gains of the business are incremental. So we see a lot of these 5% or 10% efficiency improvements, which are nice, but it's not what AI is promising to do. In contrast, it often needs top-down innovation or bottom-up innovation with a view of the bigger picture to understand what are all the steps the business executes to create value, to then re-engineer this broader workflow. I'm seeing that deliver much more value. So bottom-up innovation is great, but at some point it takes an individual, maybe a top-down leader or a bottom-up person with that broader view, to think through the workflow redesign. I'm seeing that where businesses get it right, it delivers much more value. Leadership comes from the top. There's a lot to be said for bottom-up influence, let's do lots of that too, but leadership comes from the top. So one challenge I have to every CEO or C-suite person is: if you're serious about AI, learn about it yourself. Make sure that you personally, in addition to excitement about AI, learn enough, spend a few hours going a little bit deeper into the technology to make sure you have that knowledge to drive your teams, allocate assets, prioritize, and then drive the business results.

我想测试一下自己对大语言模型工作原理的理解，关于当前这波生成式技术。从根本上说，大语言模型通过预测下一个词或下一串句子来工作，对吧？通过它们所有的训练，它们看的是……如果它们真正做的只是预测下一个词，那它们怎么能听起来这么聪明呢？还是说这是对它们所做事情的过度简化？

I want to test my own understanding of how LLMs work about the current wave of generative technologies. Fundamentally, the large language models work by predicting the next word or the next string of sentences, right? By all the training they've had, they look at what's... How is it that models are able to sound so intelligent if all they're really doing is predicting the next word? Or is that an oversimplified understanding of what they do?

粗略来说，大语言模型的训练方式：第一步称为预训练，是学习预测互联网上的下一个词元。所以在那之后，因为互联网有很多随机内容，它倾向于复述互联网上的东西。但之后还有另一步，我们用少量数据对模型进行对齐，让它说更有意义的话。例如，它们尽力如实回答问题并给出有帮助的答案，而有时互联网或社交媒体不会给出最有帮助的答案。所以如果你问一个问题，你可能会得到一堆讽刺搞笑的评论。但在第二步，可以使用一种称为强化学习或其他技术来实现，你训练算法变得有帮助、诚实且无害。所以就是这第二步。第一步预训练提供了来自互联网的大量知识，然后第二步让它利用所有预训练的知识变得有帮助。但通过这样基本的步骤，这些模型却能听起来如此聪明，能给你关于复杂问题的策略并解决它们。这似乎有点反直觉，但它就是有效。AI 中的很多事情，AI 历史上的很多事情是：当你构建一个足够大的 AI 模型，一个足够大的深度学习模型，并投入足够多的数据，它就能出色地模仿数据中的任何内容。所以当你给它一堆有帮助的问答数据时，AI 就能很好地模仿听起来应该是什么样子，以便给出有帮助的答案或按照用户的指令以有帮助的方式行事。

Loosely, the way that large language models are trained: step one, called pre-training, is to learn to predict the next token on the internet. So after that, because the internet has a lot of random stuff, it tends to regurgitate internet-like things. But there's another step after that where we take the model and with just a small amount of data align it to say more sensible things. For example, they do their best to factually answer questions and give helpful answers, whereas sometimes the internet or social media doesn't give the most helpful answers. So if you ask a question, you get back a bunch of snarky funny comments or something. But during the second step, which can be implemented using a technique called reinforcement learning or other techniques, you train the algorithm to be helpful, honest, and harmless. So it's that second step. The first step, pre-training, gives a lot of knowledge from the internet, and then that second step causes it to use all that pre-trained knowledge to become helpful. But with such fundamental steps, these models are able to sound so smart, able to give you strategies about complex problems and solve them. It just seems a bit counterintuitive, but it just works. A lot of things in AI, a lot of history of AI has been: when you build a big enough AI model, a big enough deep learning model, and throw enough data at it, it does a remarkably good job mimicking whatever is in the data. So when you give it a bunch of data of helpful answers to questions, then the AI does a decent job mimicking what it has to sound like in order to give a helpful answer to a question or follow instructions in a helpful way in response to a set of instructions that the user gives.

我认为这种 Scaling（规模扩张）AI 模型的配方，我们在大约 15 年前就知道了，甚至更早，在我启动 Google Brain 项目的时候。我给团队设定的首要任务是：做一个非常大的神经网络，用大量数据去训练它。15 或 16 年后的今天，这个配方——在我启动 Google Brain 时非常有争议，人们说‘Andrew，这是个糟糕的职业选择，别这么做’——但我当时有数据让我有信心，即使在那时，这也会成功。我认为我们还没有榨干这个柠檬的全部汁水。在 Scaling（规模扩张）AI 模型方面，还有更多工作要做。

I think this recipe of scaling AI models, we knew it kind of 15 years ago, I guess more than 15 years ago when I started the Google Brain project. The number one mission I set for the team was: let's do a really large neural network and throw a lot of data at it. Now, 15 or 16 years later, this recipe—which was very controversial when I started Google Brain; people said 'Andrew, this is a bad career move, don't do this'—but I had data that gave me confidence even back then that this would work. I think we're still not yet done squeezing the juice out of this particular lemon. There's still more work to be done to scale AI models.

太棒了。但当你思考人工智能的边界是否真正推进时，你可能会想，如果一个模型解决了人类 300 年来未能解决的数学问题，那就在文明进步方面足够突破极限了。你认为我们会从这些模型中看到纯科学领域的这类进步吗？

Incredible. But when you think about whether the boundaries of artificial intelligence have truly advanced, you might think that if a model solves a mathematical problem that humans haven't been able to solve for 300 years, that would be sufficiently pushing the envelope in terms of civilizational progress. Do you think we're going to see those kinds of advancements in pure science from these models?

我认为我们会达到那一步。但不容易。AI 已经在做出贡献——不是独立完成，而是助力科学进步。在生物学中，AI 在提出待测试的蛋白质分子；在材料科学中，类似的事情也在发生；在计算机科学中，我们用 AI 写代码来帮助运行实验。

I think we'll get there. I think it's not easy. AI has already been contributing—not doing it by itself, but contributing to advancing science. In biology, AI is proposing protein molecules to test; in material science, similar things; and in computer science, we use AI to write code to help us run experiments.

但其中很多往往是因为算力的蛮力，能够比人类以前做得更多。但像那种原创思维，那种可能解决非常复杂数学问题的洞察力——我们也会看到 AI 产生那种创造力和原创思维吗？

But a lot of that tends to be because of the brute force of computing that's able to do things more than what humans were previously able to do. But like the original thinking, the insight that might solve a really complex mathematical problem—are we going to see that kind of creativity and original thinking also coming from AI?

它会来的。但我想给出一个重要提醒：你用了‘原创思维’和‘创造力’这些词，而这些是非常深刻的哲学概念。对我来说，什么是创造性的，取决于观察者。据我所知，创造力没有科学定义。所以人类是否有创造力，或者 AI 是否有创造力，是一个重要的哲学问题，而不是一个重要的科学问题，因为我无法测量它。我无法给出明确定义。这类问题，包括 AI 是否有意识的问题——因为意识不可测量——是哲学问题，不是科学问题。但对我来说，如果它的行为在我看来有创造性，我很乐意说 AI 有创造力。如果它的行为在我看来和许多其他人看来是原创的，我很乐意说 AI 具备原创思维能力。但在这里，我可能偏离了我作为商业领袖、科学家和工程师的典型角色，给出了一个哲学而非科学的回答。

It will come. But I want to give an important caveat: you use the words 'original thinking' and 'creativity,' and these are very deep philosophical concepts. To me, what is creative is in the eye of the beholder. There is no scientific definition of creativity, as far as I know. So whether a human is creative or whether AI is creative is an important philosophical question, as opposed to an important scientific question, because I can't measure it. I can't have a clear definition. Questions like this, including questions of whether AI is conscious—because consciousness is not measurable—are philosophical questions, not scientific questions. But to me, if it behaves in a way that looks creative to me, I'm happy to say AI is creative. If it behaves in a way that seems original to me and a bunch of other humans, I'm happy to say AI is capable of original thinking. But I am here maybe stepping away from my typical role as a business leader, scientist, and engineer, and giving a philosophical rather than a scientific answer.

太棒了。我有时会想：AI 之所以工作得这么好，是不是因为这也是人类思维的工作方式？比如孩子通过观察成人并模仿他们来学习；孩子说的早期句子都是模仿。这是人类学习的基础方式吗？这也是 AI 工作得这么好的原因吗？

Incredible. I sometimes wonder: is the reason AI seems to work so well because this is also how the human mind works? Like children learn by watching adults and trying to mimic them; all the early sentences that kids say are imitations. Is that foundationally how humans learn, and is that why AI works so well as well?

我希望我知道人脑是如何工作的。AI 的学习方式与人类的学习方式不同。当今最前沿的 AI 模型吸收的文本和图像比任何人类一生中能消费的都要多。在某些关键方面，它们仍然比任何人类都笨，但也许在其他关键方面，它们能比任何人类做得更多。所以它的学习方式与人类的学习方式非常不同。人类的学习方式对 AI 的学习方式有一些模糊的启发。但我记得当我启动 Google Brain 时，深度学习刚刚兴起，我读了一堆神经科学论文，试图弄清楚人脑是如何工作的，看看是否能启发更好的 AI 算法。那些努力结果大多没什么用，因为当今的神经科学几乎还不了解人脑是如何工作的。所以试图从神经科学中汲取灵感并在计算机中实现，只有一点点希望，但这并不是我们今天推动 AI 进步的主要方式。人们经常类比说，我们不是通过模仿鸟类来建造飞机的。鸟类扇动翅膀，飞机不扇动。但研究鸟类帮助我们理解了空气动力学，而这些空气动力学原理对于弄清楚飞机如何飞行至关重要。所以有一个我感兴趣的大问题——很多人也感兴趣，但我们不知道答案——那就是：智能的本质是什么？智能是如何产生的？它是如何工作的？我希望我知道。也许如果我们发展出智能的理论，那么就像航空理论塑造了我们建造飞机的方式一样，一个更好的智能理论可能会塑造我们建造 AI 的方式。但我们目前就是不知道。

I wish I knew how the human brain works. The way AI learns is different from the way humans learn. Today's cutting-edge AI models have absorbed way more text and images than any human will ever consume in a lifetime. In key ways, they're still dumber than any human is, but maybe in key ways they can also do more than any human can. So the way it learns is very different from the way humans learn. There is vague inspiration from how humans learn that has affected how AI learns. But I remember when I started Google Brain, when deep learning was coming up, I was reading stacks of papers on neuroscience to try to figure out how the human brain works, to see if that could inspire better AI algorithms. Those exercises turned out mostly not to be useful, because neuroscience today barely understands how the human brain works. So trying to take inspiration from neuroscience to implement in the computer has been only a little bit hopeful, but it's not the way we're driving most of AI progress today. People often make the analogy that we didn't build airplanes by mimicking birds. Birds flap wings; airplanes don't. But studying birds helped our understanding of aerodynamics, and those principles of aerodynamics have been completely important for figuring out how airplanes fly. So one grand question that I'm interested in—that a lot of people are interested in, but we don't know the answer to—is: what is the nature of intelligence? How does intelligence arise? How does it work? I wish I knew. Maybe if we ever develop that theory of intelligence, then just as a theory of aerospace has shaped how we build airplanes, a better theory of intelligence might shape how we build AI. But we just don't know right now.

没错，完全同意。Andrew，最近——比如说过去三年里——显然我们今天处于一场 AI 军备竞赛中，许多顶级公司，包括 Anthropic 的 Claude、OpenAI 的 ChatGPT，都在试图超越彼此。每次新模型发布，大家都说‘天哪，这是最好的模型’，结果两个月后又被另一个前沿模型超越。你能谈谈哪些模型真正让你兴奋吗？哪些让你觉得‘哇，这真了不起’？

Right, absolutely. Andrew, what have been the recent—let's say in the last three years—obviously today we are in an AI arms race where many of the top companies, including Anthropic with Claude, ChatGPT from OpenAI, all of them are trying to outdo each other. Every time a new model comes out, everyone is like 'Oh my god, this is the best model,' only to be out-excited two months later by yet another frontier model. Could you talk a little bit about which ones have been truly exciting to you? Which ones have felt like 'Wow, this is quite something'?

哦，所有的模型。我认为它们都非常令人兴奋。这个领域进展很快。我觉得，你知道，你提到了 Claude——他们做得非常出色。我觉得，为了给 Gemini 团队以肯定，Gemini 3 配合 Gemini CLI 已经显著缩小了差距。我认为 OpenAI 也做得很好。Gemini 3 是一个不可思议的模型，Opus 4、Claude Opus 4.5、GPT-5.1、5.2 和 2 也都是非常强大的模型。所以我认为这场竞赛中有很多强劲的选手。这很棒。但具体到印度，我知道许多国家理性地希望确保你的关键 AI 基础设施不被单一外国势力或外国公司控制。所以许多国家在考虑主权 AI——你必须用 AI 掌控自己的命运。我想提供一个视角供思考。

Oh, all of them. I think they're all incredibly exciting. The field is progressing quickly. I feel like, you know, you mentioned Claude—they did a tremendous job. I feel like, to give the Gemini team credit, Gemini 3 with the Gemini CLI has closed the gap significantly. And I think OpenAI is doing a good job as well. Gemini 3 is an incredible model, so is Opus 4, Claude Opus 4.5, GPT-5.1, 5.2, and two are also very strong models. So I think there are many strong horses in the race. This is great. But thinking about India specifically, I know that many nations rationally want to ensure that you don't want your critical AI infrastructure to be solely controlled by a foreign power or a foreign single company. So many nations are thinking about sovereign AI—you have to control your own fate with AI. I want to offer up one perspective for consideration.

很高兴看到印度在人工智能领域的投资增长，以及印度确保其人工智能命运不受印度以外其他实体决策的控制。这完全合理。一个非常高效的方式是投资开源和开放权重模型。构建这些模型成本极高。大多数国家需要的是确保没有人能控制你的基础设施。但有了开源开放权重的基础设施，你不需要完全拥有它；各国可以集中资源，投资开源开放权重模型，这样就能确保其他实体无法控制一个国家在开源开放权重模型中的根基。过去一年的一大趋势是中国推出的开源开放权重模型。美国在封闭专有模型方面领先中国，但中国发布了一些最好的开源开放权重模型，这正成为中国巨大的地缘政治影响力来源。当有人将中国模型融入他们的工作时，当用户提问时，如果答案来自中国模型，它更可能反映生成该模型的国家的观点。这并不意味着印度需要封锁自己的印度中心模型，但如果它投资开放模型……另一个例子：许多编程语言如 Python，这是我用得最多的。很多印度人用 Python 编程。你不需要控制 Python。你只需要确保其他利益可能与印度相悖的国家不控制 Python 规范。因此，战略问题是如何确保对人工智能的访问，以便印度能够进行技能提升。

It's been great to see the rise of investments in AI in India and India making sure its fate in AI isn't controlled by decisions of some other entity outside India. That makes perfect sense. One efficient way is investing in open source and open weight models. Building these models is so expensive. What most nations need is to ensure no one else can control your infrastructure. But with open source open weight infrastructure, you don't need to own it solely; nations can pool resources, invest in open source open weight models, and that ensures someone else can't control a nation's root in the open source open weight model. A huge force over the last year has been the open source open weight models coming out of China. The US is ahead of China in closed proprietary models, but China has released some of the best open source open weight models, and this is becoming a tremendous source of geopolitical influence for China. When someone incorporates a Chinese model into their work, when a user asks a question, if it's answered by a Chinese model, it's more likely to reflect the views of the nation that generated the model. That doesn't mean India needs to lock down its own India-centric models, but if it invests in open models... Another example: many programming languages like Python, which I use most. Lots of people in India code in Python. You don't need to control Python. You just need to ensure that some other country whose interests may be counter to India's isn't controlling the Python specification. So the strategic question is how to secure access to AI so India can do the upskilling.

印度高管和工程师对美国人工智能生态系统的发展也至关重要。除了谷歌的桑达尔·皮查伊或微软的萨提亚·纳德拉等顶尖人物，整个领域还有许多印度裔的重要工程师和先驱，实际上还有许多其他国家的，比如华裔，遍布各地。但美国似乎越来越不欢迎移民，无论是在科技领域还是其他领域。你怎么看？

Indian executives and engineers are also very critical to the development of the AI ecosystem in the United States. Apart from top names like Sundar Pichai of Google or Satya Nadella of Microsoft, across the spectrum there are many important engineers and pioneers of Indian origin and indeed of many nations, Chinese origin, all over. But increasingly the United States appears to be less welcoming of immigrants, whether in technology or any other field. How do you look at this?

我认为这很糟糕。我认为这是美国一个巨大的非受迫性失误。作为一个移民，美国的一大优势是本土出生的人与来自印度和其他国家的许多聪明移民一起工作，为美国和整个世界发明、创造和建设。美国一直非常幸运，这么多来自世界各地的聪明人愿意并且希望去美国工作、创造、发明和建设。美国让事情变得更困难、更不欢迎，无论是针对高中人才，还是坦率地说，针对潜在的未来高中人才——我特别指的是大学生——这似乎是一个重大的非受迫性失误。在更个人的层面上，生活在美国，我有很多来自印度和其他国家的朋友，他们的个人生活陷入混乱或风险。当某人在美国生活了 5 年或 10 年，孩子在那里出生和长大，却仍在等待绿卡，而法规又发生变化时——这太糟糕了。在个人层面上，我一直渴望尽我所能帮助我的朋友或我认识的正在经历这些的移民。人性方面已经非常糟糕。我希望有一天——我知道美国确实需要确保边境安全，确保南部边境的安全一直是一个重要的讨论话题——但过度纠正，巨大的过度纠正，对即使 17 岁的聪明孩子也不友好，他们想来美国上大学，然后长大，有时回印度，有时留在美国，但无论哪种方式都建立联系、发明、创造——我认为这非常可悲。

I think it's awful. I think it's a huge unforced error on the part of the United States. As an immigrant myself, a lot of the strength of the US is native-born people together with many smart immigrants from India and other nations coming to the US to work together to invent, create, and build for the US and for the whole world. The US has been so fortunate that so many smart people from around the world have wanted and hopefully still want to go to the US to work and create and invent and build. The US making things harder and less welcoming both for high school talent and frankly for potential future high school talent—I'm thinking college students specifically—seems like a significant unforced error. On a more personal level, living in the US, I'm friends with many people from India and other nations whose personal lives are thrown into turmoil or risk. When someone has lived in the US for 5 or 10 years, kids born and raised there, still waiting in line for green cards, and regulations change—it's just awful. On a personal level, I've been eager to do what I can to help my friends or people I know who are immigrants going through these things. The human aspect has been pretty awful. I hope that someday—and I know the US does have to secure its borders, and securing the southern border has been an important topic of discussion—but the overcorrection, vast overcorrection, to being unfriendly to really smart even 17-year-olds who want to come to the US for college, then they grow up and sometimes go home to India, sometimes stay in the US, but either way build connections, invent, create—I think that's been really tragic.

最后一个问题。对于像印度这样的增长型经济体，在人工智能的演进中，你正在目睹一场国际军备竞赛。这需要大量投资。美国和中国显然在引领这场军备竞赛，需要巨大的算力和投资才能取得进展。应该采取什么策略？

One last question. For a country like India, a growing economy, in the evolution of AI, you're watching an international arms race. It takes a lot of investments. The US and China are clearly leading this arms race, and it takes tremendous amounts of compute and investment to get anywhere. What should be the strategy?

好消息是，人工智能具有如此大的颠覆性，以至于许多旧规则不再适用。一些成熟的企业，在美国和中国，非常强大且根深蒂固，但人工智能的地震正在动摇一切。这给了每个国家比以往更好的机会去跨越式发展、发明和做以前不可能的事情。我们都看到印度在这方面做得很好：没有铺设固定电话，印度直接进入了移动电话和智能手机时代，这很漂亮。所以问题是，既然人工智能已经颠覆了一切，印度能否足够快地提升技能，让印度非常有才华、聪明、勤奋的创新者和企业不仅能跟上，甚至可能跨越式发展并发明下一个东西？印度在多次场合都成功实现了跨越式发展。即时商务在印度比大多数国家发展得更快。在美国，当你告诉人们可以在 10-15 分钟内收到杂货时，他们有时……我希望我能在 8 分钟内收到杂货。所以我认为人工智能再次改变了游戏规则。我对印度朋友的问题和挑战是：学习技能，因为创造不可思议事物的机会就在那里。忽略炒作。学习真正的技能，然后抓住机会去创造大胆和新颖的东西。

The good news is that AI is so disruptive that a lot of the old rules of the game no longer hold. Established businesses, some in the US, some in China, are very strong and entrenched, but the earthquake from AI is shaking everything up. This gives every nation a better chance than before to leapfrog and invent and do things that weren't possible before. We all saw India do this beautifully: instead of putting in landlines, India went to mobile, to smartphones, and that was beautiful. So the question is, now that AI has shaken everything up, can India upskill rapidly enough so that very talented, smart, hardworking innovators and businesses in India can not just keep up but potentially even leapfrog and invent the next thing? India has pulled off the leapfrog game well on multiple occasions. Quick commerce took off faster in India than in most nations. In the United States, when you tell people you could get groceries delivered in 10-15 minutes, they sometimes... I wish I could get groceries in 8 minutes. So I think AI has changed the game again. My question and challenge to my friends in India is: learn the skills, because the opportunities to build incredible things are there. Ignore the hype. Learn the real skills, and then grasp the opportunities to build something bold and new.

太好了。在这个愉快的基调上，安德鲁，非常感谢你的时间。

Wonderful. On that cheerful note, Andrew, thanks so much for your time.

谢谢。再见。

Thank you. See you.

今天就到这里。本期节目由 Vine Joshi 制作，Radna 负责音效设计。《晨间简报》新一期每周二、周四和周五在你最喜欢的收听平台上线。敬请关注。

That's it for today. This episode was produced by Vine Joshi. Sound design by Radna. A new episode of The Morning Brief drops every Tuesday, Thursday, and Friday on your favorite listening platform. Stay tuned.