AI 炒作与现实：关于幻觉与恐惧叙事的对话

能来到这里真是太好了。很高兴见到大家。

That's really good to be here. It's nice to see all of you.

我准备了发言稿。我让 AI 做了两件事：一是把发言整理成有条理的流程，二是你网站上有一个你训练过的自己的虚拟形象来回答问题。所以我提前问了你的虚拟形象所有这些问题，我很好奇你的虚拟形象会在多大程度上像你一样回答。不过这不是要抓你小辫子。总之，我觉得这挺有意思的。我当时想，你的虚拟形象就在那里，它说“跟我聊聊吧”。所以我说“好，这是个好机会”。

I prepared my remarks. The only thing I had AI do, well, I had AI do two things. I had AI organize the remarks into a flow and then also on your website you have an avatar of yourself trained to answer questions. So I asked your avatar all these questions ahead of time and I'm just curious to see to what extent your avatar is going to answer the way that you're going to answer. So, but it's not a gotcha moment. Anyway, I just thought that was interesting. I was like, your avatar was there and it said, "Talk to me." So, I said, "Okay, well, this is a great opportunity."

这很有意思。你知道，我们花了很长时间做那个虚拟形象，到现在大概一年半了。一开始它真的很差，老是说一些我绝不会说的话。后来随着时间推移，它越来越好。它现在仍会说一些我不满意的怪话。但我学到的一个有趣的事情是，当人们和我的虚拟形象交谈时，我的虚拟形象会犯错、会幻觉。我内部团队的一位成员说：“我和虚拟形象聊了，它跟我讲了你做的那件很棒的登山的事。”我说：“我从来没爬过那些山，但人们和我的虚拟形象聊天似乎还是很满意。”

It's interesting. You know, we worked on that avatar for a long time, like a year and a half now. And it was really bad initially. It kept on saying that I would never say that. And then over time, it got better and better. It still says some weird stuff that I'm not happy with. But one interesting thing I learned is when people talk to my avatar, my avatar makes mistakes and hallucinates. One of my internal team members said, "I talked to Avatar and it told me about this great mountain climbing thing you did." I said, "I've never climbed those mountains, but people still seem satisfied talking to my avatar."

我记不清了。我想是 ChatGPT，当时我……这就像在谷歌上搜自己，但我问 ChatGPT 给我一个简短的自我介绍，它说我也是胰腺癌的幸存者，而我从来没得过癌症，更不用说胰腺癌了。所以我就想，这到底是从哪儿冒出来的？

I can't remember. I think it was ChatGPT when I was... It's the equivalent of Googling yourself, but I was asking ChatGPT what... just to give me a quick bio of myself and it said that I was also a survivor of pancreatic cancer and I've never had cancer much less pancreatic cancer and so I was like where the hell did that come from?

是的，它们确实还会幻觉。既然我们聊到这个，你大概知道当前模型幻觉的平均概率是多少吗？

Yeah, they do still hallucinate. What... since we're talking about this, on average do you know what percentage of the time the current models are hallucinating?

嗯，我不知道。它们确实会幻觉，但可能比大多数人想的要少。两三年前，AI 模型远没有今天先进。当时有一些广为流传的 AI 模型幻觉的例子，比如那个因为编造法庭文件细节而惹上麻烦的律师等等。技术已经进步了很多，它仍然会犯错，但可能比大多数人想的要少。

Yeah, I don't know. They definitely do hallucinate, but probably less than most people think. Two, three years ago, AI models were much less advanced than they are today. And there were some widely publicized examples of AI models hallucinating like, you know, the lawyer that got into trouble for made-up court filing details and so on. The technology has gotten much better and it still does make mistakes, but probably less than most people think.

太好了。好的。我准备了一些问题。我们会进入对话。你可能也有一些问题要问我，然后希望我们能接受一些观众的提问。我想从一个非常宽泛的框架性问题开始，我们大约一周前视频聊过。所以我们算是提前热身了，我现在知道你是站在杠杆的哪一边了，但确实有 AI 末日论者和某种 AI 乌托邦主义者。那么，你会说自己在那个光谱上处于什么位置？

Great. Okay. So, I have a couple of prepared questions. We'll get into a conversation. You may have some questions for me and then hopefully we can take some questions from the audience. I just wanted to start with a really broad framing question that we had a... we zoomed a week or so ago. So, we had a kind of pregame and I now know which side of the fulcrum you stand on, but there are AI doomsayers and sort of AI utopianists. And so, where on that spectrum would you say you fall?

无聊的答案是，我其实两者都不是。我觉得我更倾向于中间。过去几年 AI 叙事的发展方式有点奇怪，关于 AI 极具破坏性的末日论或恐惧叙事比大多数人预期的要多得多。有趣的是，少数几家公司决定，煽动对 AI 的恐惧是绝佳的公关，因为事实证明，如果你到处说“我的技术太强大了，可能导致人类灭绝”，这确实是一种让人们相信你的技术真的很强大、因此一定价值不菲的方式。那个叙事瓦解之后，下一个变成了“现在我的技术可能夺走你所有的工作，那我的技术值多少钱？肯定很多，对吧？”这个叙事也在慢慢瓦解。哦，然后最近的一个是“我的技术太聪明了，可以勒索人”。

The boring answer, I'm actually neither of the above. I think I'm much more in the middle. The way that AI narratives have evolved over the last few years has been kind of weird and there's been a lot more doomsaying or a lot more fear narratives about AI being very destructive than most people would have expected. The funny thing that happened was a handful of companies decided that stoking fear about AI was wonderful PR because it turns out that if you go around and say my technology is so powerful I could cause the extinction of the human race. That turns out to be one way to convince people your technology is really powerful and therefore must be worth a lot. After that narrative fell apart, the next one became, now my technology could take all of your jobs, so how much is my technology worth? Got to be a lot, right? And that narrative is slowly falling apart. Oh, and then a more recent one, my technology is so smart it could blackmail people.

我们看到了很多这样的报道。是的。

We've seen lots of stories about that. Yeah.

有趣的是，在高度人工环境中的孤立实验室测试被炒作得好像真有其事。所以也许有时候想想我们测试飞机的方式。想象一下，如果我们在风洞里测试一架飞机，模拟飓风，如果我在风洞里设法制造出飓风条件让飞机解体。但报纸随后说“哦，飞机可以在空中解体。”好吧，这在现实生活中从未真正发生过。顺便说一句，这就是勒索事件的情况。事实证明，你把 AI 放在实验室里，戳它、探它，让世界上最顶尖的科学家们竭尽全力、拼命把 AI 逼到墙角。所以，它别无选择，在所有这些有意识的指令下，只能看起来像是在勒索某人。这就是发生的事情。媒体上的报道是“哇，AI 太危险了，它可能勒索人。”而这件事的破坏性在于，这些叙事中有一点点真实的成分。你知道，是的，飞机在飓风中会在空中解体，但我不知道最近有多少飞机在空中解体，因为我们不会让它们飞进飓风。所以反驳起来非常困难，因为你不能说飞机永远不会在空中解体。你也不能说在任何情况下，即使是实验室测试，你也不能让 AI 做坏事。但这些叙事的放大对美国造成了很大的伤害。因为它阻碍了我们本应追求的巨大机会，反而，我们看到孩子们对 AI 失去兴趣，人们不想使用这些本可以让他们变得更好的奇妙技术，这是错误的。

And it's been interesting how isolated lab testing in a highly artificial environment has been hyped up as if it was a real thing. So maybe sometimes think about the way we test airplanes. Imagine if we were to test an airplane in a wind tunnel and simulate a hurricane and if in a wind tunnel I managed to create hurricane conditions to make an airplane break apart. But the newspaper then says, "Oh, airplanes can break apart midair." Well, it never actually happened in real life. That, by the way, is what happened with the blackmailing episode. It turns out that you put AI in a lab and poke and probe it and have some of the most skilled scientists in the world bend over backwards to try really hard to push the AI into the corner. So, has no choice but given all the conscious instructions and no choice but to kind of look like blackmail someone. That's what happened. What happened in the media was, "Wow, AI is so dangerous. It could blackmail people." And the damaging thing about this is that a lot of these narratives has a tiny gem of truth in it. You know that yes, airplanes in hurricanes will break apart in midair, but I don't know of many airplanes that broke apart in midair, you know, recently because we just don't fly them through hurricanes. And so it's been very difficult to push back because you can't say airplanes will never break apart in midair. And you can't say there's no circumstance under which even a lab test you can't make AI do something bad. But the amplification of these narratives has been very damaging for America. Because it's holding back what should be great opportunities for us to pursue, but instead, you know, we see kids turned off AI, people wanting to not use these wonderful technologies that could make them so much better, wrong.

是的，我可以谈谈这个。我讲两个小故事。我不知道你们有多少人读了上周末《纽约时报》上关于 AI 公司末日论的文章。从外面看，这对很多人来说有点反直觉。为什么 AI 公司会说他们可能导致人类灭绝？我想大概是八个月前。我是一群被邀请到硅谷参加为期两天研讨会的电影制作人之一，很多来自 DeepMind、Anthropic 和 OpenAI 的大人物，还有所有这些不同的工程师，都做了关于 AI 的希望与危险的演讲，但很多内容非常吓人，他们说哦，四到六年内，四十亿人将失业。嗯，他们做了一个举手统计，我不能说出在场一些人的名字，但有一些人，你知道，在 AI 万神殿里，有多少人认为 AI 有可能导致人类灭绝，你知道，这些人中有三分之一举了手。你知道，这是有可能的。总之，我不是在反驳你的话。我的意思是，我亲身经历了这个。

Yeah, I can speak to this. I'll tell two quick anecdotes. I don't know how many of you read the article in the New York Times about the AI companies doom trolling last weekend. And it seems on the outside of it kind of counterintuitive to a lot of people. Why would the AI companies be saying they could cause the extinction of humanity? I think it was about eight months ago. I was part of a group of filmmakers that were invited up to Silicon Valley and for a two-day symposium and it was a lot of the bigwigs from DeepMind and Anthropic and OpenAI and all these various engineers gave talks on the promise and peril of AI but a lot of it was very scary and they were saying oh within four to six years four billion people are going to be out of a job. Well, they took a hand count and I can't name some of the people that were there, but there were people that, you know, in the AI pantheon, you know, how many people think there's a chance that AI will cause the extinction of the human race and you know, a third of these people raised their hands. You know, that there's that a possibility. Anyway, I'm not pushing back on what you're saying. What I'm saying is I experienced firsthand this.

我一直在想，他们到底为什么要吓唬我们？这看起来不合常理。总之，你有很多话要说。顺便说一句，你的通讯非常有用，我记得叫 The Batch，我建议大家订阅。信息量很大。所以我很好奇，你对 AI 导致人类灭绝的概率等问题，已经有了我认为是明智的看法。你愿意分享一下你的观点吗？

I was trying to figure out why the hell are they trying to scare us? It seems counterintuitive. And so anyway, you have a lot to say. By the way, you have a very informative newsletter. I think it's called The Batch, which I encourage everyone to join. It's super informative. And so I'm curious, you've come away with, I think, a sensible view of the odds of AI human extinction and all that. Do you want to share your views as well?

嗯，我算是中间派。我认为好处是巨大的。我个人，现在我只是个拍吸血鬼电影的人，所以我知道什么？但我持中间立场。我认为超级智能的可能性可能比人们想的更遥远。我认为 AI 在我们有生之年导致人类灭绝的可能性极低。我个人认为，我们几分钟后可以再谈，AI 取代工人的威胁可能比你说的更大，但这没关系。我们可以谈谈这个。我确实在好莱坞一定程度上看到并经历了这种情况。所以我持中间立场，但过去两三年我也一直在使用 AI。我看到它在很多方面加速了我的工作，无论是通过概念艺术迭代，还是通过视觉特效，或者我在写作和研究过程中使用的不同方式。所以它是一个工具，像任何工具一样，锤子可以用来建造，也可以用来伤人。所以这取决于用途。

Well, I'm kind of in the middle. I think there are massive benefits. I personally, now I'm just a guy who does vampire movies and stuff, so what do I know? But I'm in the middle. I think the possibilities of AI superintelligence are probably farther out than people think. I think the possibility of AI causing the extinction of the human race anytime within our lifetime is highly unlikely. I personally think, and we can talk about this in a few minutes, that AI displacing workers, the threat might be bigger than you're saying it is, but that's okay. We can talk about that. And I've certainly seen that, experienced that in Hollywood to a certain extent. So I'm in the middle, but I've also been using AI myself for the last two or three years. And I've seen lots of ways that it has sped up my work, whether it be through concept art iteration on something I'm doing, or through visual effects, or there are different ways that I use it in the writing process and the research process as well. So it's a tool, and like any tool, a hammer can be used to build and a hammer can be used to hurt someone. So it just depends.

你知道，关于你提到的就业问题，一个炒作叙事是 AI 将导致数百万人失业，这确实有助于让人们相信这项技术非常有价值，但我没有看到这种情况发生，原因如下。我和一家大型劳动力管理公司的 CEO 聊过，他们实际上对最近被裁员的人进行了调查。我们看到所有关于 AI 导致裁员的新闻，对吧？公司说 AI 来了，裁员，等等。他做了一项调查，问那些真正被裁员的人，你是否被 AI 取代了？数字是多少，你觉得？是 50%、20%、10%？实际上只有 1.4% 的近期失业者是因为 AI 而失去工作的。需要说明的是，我非常同情这 1.4% 的人。我非常担心。一小部分人的工作被 AI 取代了，这对他们来说很糟糕，我们需要为此做点什么。给他们一个安全网，为他们提供再培训、提升技能、学习其他东西以找到新工作的途径。所以帮助这 1.4% 的人非常重要。再说一次，这是那种有真实成分的事情。确实有少数人的工作因为 AI 而消失，但炒作远大于现实。更广泛的现实是，如果你看基于任务的工作分析，这些是经济学家会做的非常严谨的研究，我们看看美国经济或全球经济所做的工作，把工作分解成离散的任务，然后弄清楚 AI 到底能自动化什么。事实证明，对于许多工作，AI 可能合理地自动化某人工作中 30-40% 的任务。这确实意味着完全不使用 AI 的人会比使用 AI 的人效率低得多。所以那句“你不会被 AI 取代，但你会被使用 AI 的其他人取代”是绝对正确的。但与此同时，如果 AI 自动化了某人 30-40% 的工作，你确实需要一个人来完成剩下的 60-70%。历史趋势表明，当这种情况发生时，当技术自动化了某人部分工作时，它的互补部分，即需要与之配套的其他东西，变得更加有价值。因此，从历史上看，自动化趋势导致了人们工资的上涨。这就是为什么当我们从马过渡到汽车时，出租车司机的就业机会大大增加，因为出租车司机是汽车的补充。但是，减缓人们采用这项技术的意愿，确实阻碍了人们走向这些通常更好、薪水更高的工作的轨迹。

You know, on the job issue that you mentioned, one of the hype narratives has been AI will lead to millions of people unemployed, and it does help convince people maybe that the technology is very valuable, but I'm not seeing it happen for this reason. So, chatting with CEO of a large workforce management company that actually ran surveys of people that were recently laid off, and we see all this news about AI leading to layoffs, right? Company comes saying AI is coming, laying off people, blah blah blah. He ran a survey and asked of the people that were actually laid off, were you replaced by AI or not? And the number, what do you think it was? Was it 50%, 20%, 10%? It was 1.4% of recently laid off people actually had their jobs replaced by AI. To be clear, I'm very sympathetic to this 1.4%. I'm very worried. A tiny sliver, a small fraction of people have had their jobs replaced by AI, and it sucks for them, and we need to do something about that. Give them a safety net, provide them a way to reskill, upskill, learn other things to find new jobs. So that's really important to help this 1.4%. And again, it's one of these things where there's a gem of truth. There's a tiny number of people whose jobs are going away because of AI, but the hype is vastly greater than the reality. And the broader reality is that if you look at the task-based analysis of jobs, these are very rigorous studies that economists will do where we look at the jobs done by say the American economy or the global economy, break down the job into discrete tasks, and then figure out what can AI actually automate. It turns out that for many jobs AI could plausibly automate 30-40% of the tasks in someone's work. And it does mean that someone that doesn't use AI at all will be much less productive than someone that does. And so that phrase saying that you people won't be replaced by AI, but people will be replaced by someone else that uses AI, that is absolutely true. But at the same time, if AI automates 30-40% of someone's job, you really need a human to do that remaining 60-70%. Historical trends have shown that when this happens, when technology automates part of someone's job, the complement to it, the other thing that needs to go with it, becomes even more valuable. And so historically, the trends have been automation has led to higher wages for people. Which is why when we went from horses to cars, there became tons more jobs for taxi drivers, because taxi drivers are a complement for the car. But slowing down people's willingness to adopt this technology is really hampering people's trajectory to get to these often better, higher paying jobs.

好的，我们换个话题。我很好奇，你给很多政策制定者做过 AI 简报。所以我很好奇，你觉得我们的政府，在你能谈的范围内，对 AI 的了解程度如何？在与政府交谈时，你遇到的最大误解或盲点是什么？

Okay, switching gears for a second. I'm curious, you've briefed lots of policymakers on AI. So I'm curious to what degree do you feel our government, to the extent that you can talk about it, is sufficiently up to speed on AI, and what would you say is the biggest mischaracterization or blind spot that you've run into when talking to our government?

你知道，我觉得在美国，联邦政府基本上已经以比几年前人们所希望的更合理的方式理清了思路。

You know, I feel like in the US, the federal government has mostly got its head together in a more reasonable way than one would have hoped a couple years ago.

听到这个我很惊讶。

I was surprised to hear this.

我明白。是的。

I see. Yeah.

所以，我的意思是，显然我们有很多要谈的，但两年前，AI 灭绝的叙事在华盛顿特区流传开来。这些是少数 AI 公司试图进行监管俘获的一部分，他们想说 AI 会导致人类灭绝，所以要监管。特别是，事实证明，对一些领先的美国 AI 公司的主要威胁之一是，当你花费数百亿美元训练大型 AI 模型（称为基础模型）时，一些公司，包括中国和其他地方的海外公司，正在训练模型并在互联网上免费发布供任何人使用。我们称之为开放权重模型。但如果你花数十亿美元训练一个 AI 模型，然后别人在互联网上发布一个等效的模型供任何人免费使用，那真的很烦人。这确实降低了你的投资价值。所以大约两年前，我实际上对游说的强度感到非常惊讶，他们煽动对 AI 的恐惧，试图通过令人窒息的法规，确保任何人都难以在互联网上免费发布他们的 AI 模型供任何人使用。幸运的是，很多 AI 社区在联邦层面基本上已经击退了这些努力，但由于涉及数十亿甚至可能数万亿美元的利益，你可以想象游说的资源和强度是巨大的。还有非常可信的报道称，美国的一些对手正在忙于资助非营利组织，以煽动对 AI 安全的不必要恐惧，试图减缓美国在 AI 方面的创新。

So, I mean, clearly we have a lot of ground to cover, but two years ago, the AI extinction narratives were making the rounds in Washington DC. And these were part of the regulatory capture attempts of a handful of AI companies that want to say AI will lead to human extinction, regulate. And in particular, it turns out that one of the major threats to some leading American AI companies is when you spend tens of billions of dollars training large AI models called foundation models. It turns out that some companies, including overseas ones in China and elsewhere, are training models and releasing them on the internet free for anyone to use. We call these open-weight models. But it's really annoying if you spend billions of dollars to train an AI model and then someone else posts an equivalent one on the internet free for anyone to use. It really degrades the value of your investment. So about two years ago, I was actually kind of really surprised by the intensity of the lobbying to drum up fears on AI to try to pass stifling regulations to make sure that it's difficult for anyone to publish their AI models on the internet free for anyone to use. Fortunately, a lot of the AI community has largely beaten these efforts back at the federal level, but the lobbying, because there are billions or more likely trillions of dollars at stake, you can imagine the resources and the intensity of the lobbying is intense. There are also very credible reports that some of America's adversaries are busy funding nonprofits to stoke up unnecessary fears over AI safety to try to slow down American innovation in AI.

关于我们的对手资助美国和其他欧洲组织来拖慢我们的可信报道，也非常令人担忧。话虽如此，我认为白宫和国会都已经和 AI 进行了足够的对话。你知道，舒默主持了一个两党 AI 洞察论坛。包括我在内的一群人跟参议院和国会进行了交流。我认为本来可能会更糟。我更担心的是州一级的 AI 监管，因为很多游说活动现在已经转移到了州一级，因为你可以通过州一级的扼杀性监管法案。这可能会对整个国家产生有害影响，而且州议会太多了。所以不断压制州一级的提案要困难得多。

The credible reports about our adversaries funding American and other European organizations to slow us down is also very concerning. Having said that, I think both the White House and Congress have had enough talks with AI. You know, Schumer ran a bipartisan AI insight forum. A bunch of people including me spoke with Senate and Congress. And I think it could have been much worse. I worry much more about state-level regulation of AI because a lot of the lobbying efforts have been now at the state level because you can get a state-level stifling regulation passed. It could have deleterious effects across the entire nation and there are so many state houses. So constantly squashing back referee proposals to the state level has been much harder.

我们听到的一个关于不要通过过度监管扼杀创新的说法是，至少在美国，我们正在和中国竞赛。这是我们经常听到的事情之一，如果我们不先到达那里，中国就会崛起，然后西方就完了。我很好奇，因为你是在美国和中国的 AI 项目中都工作过的人。所以我在我们之前的准备中问过你的一个问题就是，你如何定义两国在 AI 方面的不同方法？

So one of the narratives that we hear in terms of not stifling innovation through overregulation is that at least in America, we're in a race with China. That's one of the things that we often hear, and that if we don't get there first, China will become ascendant and then whatever the West will be over. I'm curious because you're someone who's worked in AI initiatives for both the US and for China. So one of the questions that I asked you in kind of our pregame was how would you define the different approaches between the two countries in terms of AI?

几年前我不会猜到的一件事是，中国在 AI 创新方面最终会比美国开放得多。所以再说一次，我非常了解美国生态系统，因为我在 Google Brain 和 Sam 都工作过，我可能是地球上唯一一个 Sam 和 Dario 都为其工作过的人，所以我觉得我很了解所有团队的人。但在 2022 年底 ChatGPT 发布的时候，美国明显领先于中国。但自那以后，中国打得非常好，大力投入开源开放权重生态系统，中国的团队会做研究，并把研究论文免费发布在互联网上供任何人阅读。顺便说一句，我告诉你，很多美国团队会阅读大量中国研究论文，并从中学习和受益。现在领先的美国团队使用的很多技术，坦率地说，都是来自从中国学到的研究创新。所以我认为知识是双向共享的。但中国做得非常好的一件事是，很多公司会以开放权重模型的形式发布这项技术，这意味着它们被免费发布在互联网上供任何人使用，包括美国公司。

So one thing that I would not have guessed a few years ago was that China would end up being much more open in the way it innovates in AI compared to the United States. So again, I know the US ecosystem really well since I've worked at both Google Brain and Sam and I might be the only person on the planet that both Sam and Dario have worked for, so I feel like I know all the teams of people well. But at the moment that ChatGPT was released at the end of 2022, America was decisively ahead of China. But China since then has played its hand really well, leaning into the open-source open-weight ecosystem in which teams in China will do research and publish their research papers free on the internet for anyone to read. And by the way, I'll tell you, a lot of the American teams read tons of the Chinese research papers and learn from them and benefit from them. Tons of the techniques used by leading American teams now frankly were from research innovations learned from China. So I think there's very much two-way sharing of knowledge. But one thing that China did really well was a lot of its companies will release this technology as open-weight models, meaning they're published on the internet free for anyone to use, including American companies.

你能给观众定义一下什么是开放权重吗？你们有多少人知道什么是开放权重？有一些。好的，不错。哦，谢谢，问得好。谢谢。

Can you just define what open weights are for the audience? How many of you know what open weights are? Some. Okay, cool. Not bad. Oh, thank you, good call. Thank you.

所以当你有一个 AI 模型时，它会阅读互联网上的大量文本，查看大量图像，观看大量视频。我知道版权是痛点，但在 AI 从大量数据中学习之后，它会得到一长串数字。我们称这些数字为权重。开放权重模型就是有人在一个 AI 模型上训练了大量数据，然后把所有这些称为权重的数字免费发布在互联网上，供任何人下载和使用。所以当有一个开放权重模型时，我们任何人都可以下载它，你可以下载它，让 AI 在你的笔记本电脑上运行，而不是像 OpenAI、Anthropic、Google Gemini 的领先模型那样的专有封闭权重模型。那些是封闭权重模型，因为我们没有人能看到 AI 学到的数字。我们只能把我们的提示发送给这些公司之一，让他们使用这些权重或数字来计算输出，计算响应，并且只向我们展示响应。

So when you have an AI model, it reads tons of text on the internet, looks at tons of images, watches lots of videos. I know copyright is the sore point, but after AI has learned from lots of data, it results in a long list of numbers. We call those numbers weights. And an open-weight model is when someone has trained an AI model on lots of data and you publish all of those numbers called weights on the internet free for anyone to download and use. And so when there's an open-weight model, any of us could download it, you could download it and have AI run on your own laptop, as opposed to the proprietary closed-weight models such as the leading models from OpenAI, Anthropic, Google Gemini. Those are closed-weight models because none of us can see the numbers that AI had learned. We can only send our prompt to one of these companies and have them use these weights or these numbers to compute the output, compute the response, and show us only the response.

中国公司做的一个聪明举动是发布开放权重模型。事实证明这能帮助整个世界，但你帮助自己最多，因为它能迅速提高你公司内部或生态系统内部的知识传播速度。所以由于这些因素，在过去几年里，中国在 AI 能力上已经领先了。现在美国的封闭权重模型领先于中国模型。但中国在开放权重模型方面领先世界。这意味着，对于很多想要开放替代方案的国家来说，他们正在采用中国模型。所以中国模型在美国的采用率，你知道，有很多，但在非洲等国家，中国的 DeepSeek 和其他几个中国模型被广泛采用，远远超过美国的开放权重模型，后者确实已经落后了。这很重要，有两个原因。但首先，就像我们在好莱坞看到的故事讲述是软实力的重要来源一样，我们是想要讲述关于自由和民主价值观重要性的故事，还是想要讲述反映其他国家价值观的故事？事实证明，当有人问 AI 一个问题，比如 1989 年天安门广场发生了什么，那么他们使用谁的模型，模型就会给出反映那个国家价值观的答案。所以把你的东西传播出去是巨大的软实力影响力来源。第二件事是，AI 是我们构建许多软件产品的供应链中的关键部分。如果越来越多的公司建立在真正由中国主导的供应链上，这对美国有影响，我认为这也会削弱我们改变技术发展方向的能力。

One brilliant move that Chinese companies did was publishing open-weight models. It turns out to help the whole world but you help yourself the most because it rapidly increases the rate of diffusion of knowledge within your companies or within your ecosystem. And so because of factors like that over the last few years, China has raised ahead in AI capabilities. And right now the US closed-weight models are ahead of the Chinese models. But China leads the world in open-weight models. And what this means is that for a lot of nations that want an open alternative, they're adopting Chinese models. So adoption of Chinese models in the US is, you know, there's a lot, but in nations such as Africa, the Chinese DeepSeek and a handful of other Chinese models are adopted very widely, far more than say the American open-weight models which have really fallen behind. And this is important for two reasons. But first, just as storytelling we've seen in Hollywood is an important source of soft power, do we want stories told about the importance of liberty and democratic values or do you want stories told that reflect other nations' values? It turns out that when someone asks a question of AI such as what happened in Tiananmen Square in 1989, well, whose model they are using will cause the model to give an answer that reflects that nation's values. So getting your stuff out there is a tremendous source of soft power influence. And then the second thing is AI is a key part of the supply chain of how we build a lot of software products. And if more and more companies are building on really a Chinese dominant supply chain, that has implications for America that I think will also weaken our ability to change the way the technology goes.

就在过去这两周，美国发生了几件非常奇怪的事情。首先，当 Anthropic 发布其最新模型 Fable 5 时，它设置了限制，阻止 Fable 以它认为可能支持 Anthropic 竞争对手的方式帮助 AI 研究人员。然后大约一周后，几天后，美国政府对 Fable 模型实施了出口管制限制，导致 Anthropic 关闭了全球所有客户的访问。但这向全世界展示的是，像 Anthropic 这样的美国私营公司可以限制其他人如何使用前沿模型，比如不帮助竞争对手。它还向全世界展示，美国可以实施出口管制来强行切断对 AI 技术的访问。所以我在过去一周看到的是，这加速了世界许多国家首都的紧迫感，他们觉得需要确保自己的 AI 供应，坦率地说，美国政府和美国私营公司无法在接到通知的那一刻就关闭。这也导致许多国家关注开放权重模型，因为一旦他们有了所有数字，没有人能把它从他们手中夺走。

And just these past two weeks, something a couple very strange things happened in America. First, when Anthropic released its latest model, the Fable 5 model, it put in place restrictions to stop Fable from helping AI researchers in ways that it thought could support Anthropic's competitors. And then like a week or so later, a few days later, the US government imposed export control restrictions on the Fable model, causing Anthropic to shut down access all around the world to all customers. But what this showed to the whole world is that private American companies such as Anthropic can restrict how others can use frontier AI models, such as not helping competitors. It also showed the whole world that America can impose export controls to yank access to AI technology. And so what I've seen just over the last week is this has accelerated in many capitals around the world the urgency with which they feel like they need to secure their own supply of AI that the US government and private American companies do not have the ability to turn off at a given moment's notice, frankly. And this is also causing many nations to look at open-weight models because once they have all the numbers, no one can take it away from them.

嗯，我认为这可能会对美国软实力产生不利影响。

Um, and I think this will probably have an unfortunate effect on America's soft power.

你看，我以为那只是因为特朗普对 Anthropic 感到恼火，想报复他们。

See, I just assumed that happened because Trump was annoyed at Anthropic and wanted to get back at them.

是啊，有这种说法。

Yeah, there's that theory.

但明确一下，你是开源权重（open weights）的支持者。

But to be clear, you're a proponent of open weights.

事实证明，如果 AI 最终集中在极少数看门人手中，可能这里的每个人都会变得更糟。极少数看门人会过得更好，因为他们收取很高的过路费，但几乎所有人都会更糟。我有时会拿移动生态系统做类比。如今有两个看门人，Android 和 iOS。如果他们不想让某些创新被尝试，那么没人能尝试。但成为一两个看门人确实非常赚钱。所以，我认为如果像 AI 这样有价值的东西最终落入少数看门人手中，那将是不幸的。明确一下，很多朋友在这些 AI 公司工作，我希望他们真的做得好。我希望他们 IPO 成功，财务上非常成功。我觉得这些都很好。同时，我认为这些企业有一条路可以做得很好，而不必成为阻碍其他人在此技术之上创新的看门人。

It turns out that if AI ends up concentrated in the hands of a very small number of gatekeepers, probably everyone here will be worse off. The very small number of gatekeepers would be better off because they charge a very high toll, but almost everyone would be worse off. I sometimes make an analogy to the mobile phone ecosystem. Today, there are two gatekeepers, Android and iOS. If there are certain innovations they don't want anyone to really try out, then no one's allowed to try them out. But it turns out to be really profitable to be one or two of the gatekeepers. So, I think it would be unfortunate if AI, as valuable as it is, ends up with a handful of gatekeepers. To be clear, a lot of friends work at these AI companies, I hope they do really well. I hope they have successful IPOs, do very well financially. I think all that is great. At the same time, I think there is a path for these businesses to do really well without them becoming gatekeepers in a way that stifles the innovation of everyone else that would like to do things on top of this technology.

那么，除了你之外，美国还有谁在引领开源权重的方法？

So, is there anyone aside from you in America that is leading the charge on the open weight approach?

这很难说。Nvidia 已经成为开源权重模型的主要生产者。他们的模型不是最先进的，稍微落后，但 Nvidia 销售 GPU 有经济动机这样做。Meta 过去发布开源权重模型，但最近转向了闭源权重。还有一些小型非营利组织、学术大学实验室在研究开源权重模型。但由于构建 AI 模型所需的资源，推进起来一直很有挑战。而现任白宫政府出于各种原因（其中一些我认为是合理的，但我们可以辩论）一直不愿意投入足够资源来资助专有模型的开源替代方案。一些中东国家，我认为阿联酋有一些开源权重倡议。印度和一些国家也在尝试。所以我认为这方面其实有很多兴趣，但结果如何，我不知道，还有待观察。

So, it's a tough one. Nvidia has turned out to be a major producer of open-weight models. Their models aren't quite state-of-the-art, a little bit behind, but Nvidia selling GPUs has a financial incentive to do this. Meta used to be publishing open-weight models, but recently switched directions to closed-weight models. And then there are a number of small nonprofits, academic university labs working on open-weight models. But because of the resources needed to build AI models, it has been challenging to get that going. And the current White House administration has been reluctant, for various reasons, some of which I think are sound but we can debate them, to put enough resources to fund an open alternative to proprietary models. Some of the Middle Eastern countries, I think UAE has had some open-weight initiatives. India and a number of nations are trying. So I think there's actually a lot of interest in this, but how well it plays out, I don't know, remains to be seen.

那么中国在这方面可能会拯救我们。

So China might save us in this regard.

嗯，好的，我们换个话题。我还有几个问题。只是让中国控制开源权重供应链会创造一个非常不同的世界，而不是……

Um, okay, switching gears for a second. I have a couple more questions. It's just having China control the open-weight supply chain would create a very different world than the one...

不，我知道我在开玩笑。

No, I know I'm being facetious.

我知道，我知道，我知道。

I know, I know, I know.

我一直在思考的一件事是我所谓的“柏拉图洞穴”类比。就目前模型的训练方式而言，它们实际上是在数字化的数据上训练的，无论是文本、图片、视频还是音频文件。人类的大量作品尚未数字化。显然，西方或可能中国的作品被数字化的比例不成比例地高，但还有大量未数字化的。我们都知道互联网在某种程度上是对我们世界的扭曲模型。所以我一直在思考并担心的一件事，你对此说了一些令人安心的话，就是模型是在对我们现实的不准确描述、扭曲描述上训练的。如果你只看互联网，一切都是脑腐，每个女人都是 36D，每个男人都是银狐，还有我们在 CNN 或任何其他网站上看到的那些 AI 广告。那么，模型基本上是在对我们现实的扭曲视图上训练，不能准确描述我们的现实，这在多大程度上是一个真正的担忧？

One of the things that I've been thinking about is what I call the Plato's cave analogy. In terms of the way that the models are currently trained, they're effectively trained on data that's been digitized, whether that be text or pictures or videos or sound files. There's a massive amount of humanity's works that have yet to be digitized. Obviously a disproportionate amount of Western or possibly Chinese works have been digitized, but there's a ton that hasn't been digitized. We all know that the internet is kind of a distorted model of our world. So one of the things that I've thought about and been worrying about, and you said a couple of reassuring things about this, is that the models are being trained on an inaccurate depiction of our reality, a distorted depiction of our reality. If you just go by the internet, everything's brain rot and every woman is like a 36D and every man is a silver fox, and all the AI ads that were fed when we're on CNN or any other site. So to what extent is that a genuine concern that the models are basically trained on a distorted view of our reality that doesn't accurately depict our reality?

是的。我觉得你说的每一点我都完全同意，这是对现实的扭曲视图。我想说的是，AI 团队，我给他们很多赞誉，因为他们纠正了部分扭曲。事实证明，互联网上有很多种族主义、性别歧视的文本差异。但因为 AI 团队关注了这一点，并尽可能故意压制了这些偏见来源，事实证明今天的 AI 模型远比普通人更少种族主义和性别歧视。事实上，我个人几乎不知道如何让一个性别歧视的人变得不那么性别歧视。我只是不知道怎么做。但我知道如何让一个性别歧视的 AI 变得不那么性别歧视。我们有技术。所以也许 AI 比人类平均偏见更少是件好事。但这还不够好。然后对于那些 AI 团队没有关注的地方，因为扭曲的形式太多了，AI 模型确实会持续存在这种偏见或扭曲。这对我来说是为什么我们需要人类很长时间的另一个原因。在可预见的未来，我们所有人都会知道大量我们的 AI 模型没有、也不会在我们有生之年拥有的东西或上下文。所以我不知道。顺便说一句，我的团队多年来一直试图让 AI 取代我。我有点希望那那么容易。AI 还不够好。

Yeah. So I feel like everything you said, I absolutely agree, it's a distorted view of reality. And I want to say that the AI teams, I give them a lot of credit for correcting a subset of the distortions. It turns out that there's a lot of racist, sexist discrepancy texts on the internet. But because AI teams have paid attention to this and deliberately squashed those sources of bias as best as they can, it turns out that AI models today are far less racist and sexist than the typical human being. In fact, I personally have really few ideas how to make a sexist human less sexist. I just don't know how to do that. But I know how to make a sexist AI much less sexist. We have technologies for that. So maybe it's a good thing that AI is less biased than humans are on average. But that's still not good enough. And then for the places that the AI teams are not paying attention, because there are just too many forms of distortions, then the AI models do persist with this type of bias or distortion or what have you. And this to me is another reason why we need people for a long time. For the foreseeable future, all of us will know tons of stuff or have tons of context that our AI models do not have and will not have in our lifetimes. And so I don't know. By the way, my team's been trying for years to get AI to replace me. And I kind of wish it was that easy. AI just isn't good enough.

嗯，你的虚拟化身还在产生幻觉。

Well, your avatar is still hallucinating.

是的。是的。是的。没错。也许一个具体的症状是，我不知道你多久用 AI 帮你头脑风暴，对吧？如果我经常说，嘿，帮我头脑风暴一个营销口号或什么好的项目想法。当我这样做时，我发现通常，如果我得到 10 个想法，也许有一两个好主意，一个还行，然后六七个完全糟糕的主意。我会想，AI 怎么会认为那甚至可能可行？我永远不会那样说。这对我来说是 AI 缺少大量上下文的症状，它认为那个荒谬的想法甚至值得提出。所以在我看来，在可预见的未来，AI 将人类，我们都将拥有巨大的上下文优势。意思是会有大量我们知道的东西，关于什么对特定工作、特定任务、你将要处理的特定脚本是重要的。

Yeah. Yeah. Yeah. That's right. And maybe one specific symptom of that is, I don't know how often you use AI to help you brainstorm, right? If I often say, hey, help me brainstorm a marketing slogan for this or what's a good project idea for this. And when I do that, I find that often, if I get 10 ideas, maybe there'll be one or two good ideas, one that's okay, and then six or seven absolutely terrible ideas. And I go, how on earth could AI have thought that that's even possibly doable? Like I would never say that. And this to me is a symptom that AI is missing tons of context that it thought that ridiculous idea was even useful to propose. So what I see is for the foreseeable future, AI will humans, we will all have a massive context advantage. Meaning there'll be a ton of stuff that we know about what's important for a specific job, for a specific task, for a specific script you'll be working on.

AI 并不知道这一点，这也是我不太担心 AI 在很长一段时间内取代我们的原因之一，也许直到某一天，我不知道，我们都随身携带摄像头和麦克风，我们的生活全部数字化，AI 也远超今天的水平。但那似乎非常不可能，而且如果真的发生，也是非常遥远的未来。

The AI does not know which is one of the reasons why I'm not that worried about AI replacing us for a long time, maybe until someday that I don't know, we all carry cameras and microphones and all of our lives are digitized and AI advances far beyond what it is today. But that just seems like a very unlikely and, if it ever comes to pass, a very distant future.

我相信在座的大多数人，至少我本人，都有过这样的经历：给模型一个提示词，无论是生成图像、视频还是文本，AI 明显误解或理解错了，没有给我想要的结果。所以我会想，好吧，我该怎么修改提示词才能让它理解？它并没有真正理解我的要求。幸运的是，对我来说，这种情况可能十次里才发生一次，但也有很多次我会直接放弃，我会说我不知道该怎么让它理解我想问的东西。

I'm sure most of the people here, certainly I have had the experience of giving a model a prompt, whether it's for an image or a video or text, and clearly the AI misinterpreted or misunderstood it, didn't give me the desired result. So I'll think, okay, how do I change my prompt to get it? It's not really understanding what I'm asking. Fortunately for me, maybe that happens a tenth of the time, but there are plenty of times where I'll just give up, where I'll just say I don't know how to get it to understand what I'm trying to ask it.

不过，你愿意分享一下你的工作吗，关于你如何使用 AI 来……

Actually, do you want to share your work though, for how you use AI to...

我的意思是，就拿写作来说，我早期开始做的一件事是，我可能会为某件事写一篇四页的独白，我花了好几个小时精心打磨，那确实是一篇很好的独白，但我知道它太长了。于是我开始不再担心长度，我会直接拿着独白说：“好，我需要它缩短 40%，或者从四页变成两页。”所以我要你生成六个不同版本的两页独白，精简我自己的文字，保留上下文，不管怎样，给我一些变体。我还会说，比如，一个版本往这个方向偏，一个版本往那个方向偏，一两分钟内它就会生成六个不同版本的两页独白。没有一个版本是 100% 完美的，但都很接近。我会把一些短语混搭起来，原本可能需要我一两个小时来删减的独白，现在只花了我五分钟。这就是我经常使用的一个例子。

I mean, just in terms of writing, like one of the things that I started to do early on was I might write a four-page monologue for something and I labored over it for x amount of hours and it's a really good monologue, but I know that it's too long. So I started not worrying about how long it was and I would just take the monologue and say, "Okay, I need this to be 40% shorter or two pages versus four pages." So I want you to kick out six different versions of that two-page monologue, trimming down my own words, keep the context, whatever it is, but give me variations. And I'll also say, like, lean one variation this way, one variation that way, and within a minute or two, it will kick out six different versions of a two-page monologue. And none of them are 100% perfect, but they're close. And I'll mix and match some of the phrases, and what might have taken me an hour or two cutting down that monologue now has taken me five minutes. So that's an example of something that I'll use all the time.

或者我最近在写一个剧本，里面有个角色被绑架，被扎带绑住，带进了一架高端商用直升机的机舱。我知道我想让这个角色在离海面约 100 英尺的地方逃出机舱，跳海求生。于是我说：“好，市面上最高端的商用直升机有哪些？”选了一款，然后我说：“我想让这家伙像《百战天龙》那样想办法脱身。”那么机舱里有哪些标配的东西可以让他用来逃脱，比如耳机、耳机线等等，还有安全带。就这样，我生成了这段动作戏，结合了我给 AI 的提示、写作和反复迭代，但我在大概一个小时左右就完成了。

Or I was recently writing a script where I had a character get kidnapped and zip tied and taken into the cabin of a high-end commercial helicopter. And I knew that I wanted to have this character about 100 feet over the ocean escape from the cabin of the helicopter and dive to his safety. So I said, "Okay, what are the highest end commercial helicopters out there?" Picked one, and I said, "Oh, I want this guy to kind of MacGyver a situation out of it." So what are all of the things that would likely come standard within this cabin that I can use for him to escape, whether it's the headphones, the wire from the headphones, etc., the safety belts. And so I was able to generate this action sequence, a combination of me giving prompts to the AI, writing, and then reiterating, but I was able to do that in, I don't know, an hour or something like that.

所以我会说，对于长片剧本，根据研究深度，我通常需要 8 到 10 周才能写出初稿。但现在我使用 AI，通常能在四五周左右完成。所以它确实加快了我的写作速度。它并不是替我写场景，而是帮我迭代，用不同的方式使用我的文字。所以我不知道，我觉得它很有帮助。

So I would say that for features, it might take me anywhere from, depending on the level of research, 8 to 10 weeks to generate a draft of a feature. But now I'll use AI and I can usually do it in, I don't know, four to five weeks or something like that. So it's definitely sped up my writing. It's not writing the scenes for me, but it's iterating and using my words in different ways. So I don't know. I've found it helpful.

我知道你还有一家公司，一个基金会，我在这里概括一下，致力于 AI 倡议，伦理 AI 倡议，如果找不到更好的词的话。你说是这样吗？

I know that you also have a company, a foundation, that, I'm paraphrasing here, works on AI initiatives, ethical AI initiatives, for lack of a better word. Would you say that's right?

基金会。

Foundation.

嗯，或者是你资助的项目。

Well, or the projects that you help finance.

当然。

Sure.

你资助的那些。

That you help finance.

我们认真对待负责任的 AI。

We take responsible AI seriously.

是的。是的。负责任的 AI。

Yeah. Yeah. Responsible AI.

事实证明，在我自己的 AI 工作和我领导的 AI 团队中，我们确实偶尔会遇到伦理决策，但可能比大多数人想象的要少得多。当我们遇到这些决策时，坦率地说，我们确实会纠结。我们并不总是知道我们做出了正确的决定。而且有很多案例，我们实际上真的不知道正确的决定是什么。我们做的可能正是你所预期的。我们会进行内部辩论，咨询代表各种不同观点的利益相关者，综合意见，坐下来讨论清楚，然后做出决定，并且，你知道，拼命希望我们做对了。

It turns out that when in my own work in AI and the teams I lead in AI, we do come across ethical decisions occasionally, but probably much less often than most people think. And then when we come across these decisions, frankly, we do wrestle with them. We don't know that we're always making the right decision. And there are a number of cases that we actually honestly don't know what the right decision is. And we do probably what you'd expect. We have internal debate, consult with a range of stakeholders that represent a range of diverse perspectives, synthesize, sit down, hash it out, and then just make a call and, you know, really hope like crazy that we got it right.

其中一些非常清晰。在 AI 基金，我们遇到过一些商业想法，我们会坐下来讨论，说，你知道吗，这个想法会赚很多钱。我们认为这个想法从财务角度来看非常有前景，但我们认为它会让人们变得更糟，这些很容易处理。我们很容易就砍掉这些项目。唯一棘手的，其实也不多，是我们真的无法判断，可能让一些人受益，也可能让一些人受损。那么它最终是否让世界变得更好？这些我们真的很难抉择。

Some of them are really crystal clear. At AI Fund, we have come across business ideas where we sat down and said, you know what, this idea is going to make a ton. We think this idea is very promising from a financial point of view, but we think it's going to make people worse off, and those are easy. We kill those projects really easily. The only tough ones, which is not that many, is we honestly can't tell, makes some people better off, maybe make some people worse off. So does it at net net make the world better off? Those we really struggle with.

我想顺便问一下，你有两个孩子，对吧？他们多大了？

I guess a side question to that is, you have two kids, right? How old are they?

女儿七岁，儿子五岁。

Daughter is seven, son is five.

好的。我有三个孩子，12 岁、15 岁和 19 岁。我 15 岁的孩子强烈反对任何形式的 AI 使用。我的意思是，他就是坚决反对任何 AI 的使用。而我认为自己对 AI 持中立态度。所以我会和他进行这些辩论……我的意思是，我认为很难说 AI 在医疗行业、气候倡议、各种事情上没有积极的好处。但现在有一个很大的趋势。我们都看过毕业典礼演讲中人们嘘 AI 高管的片段之类的。我并不是在淡化这一点，但今天的年轻人，很多年轻人，有一种感觉，他们会被挤出工作岗位，未来不会有任何可行的选择。我们会在五年内发现自己身处《美丽新世界》。那么，我怎么知道你不相信这是真的？所以你会对今天成长中的年轻人说什么，关于他们应该如何思考 AI，以及他们应该如何参与 AI？

Okay. I have three. 12-year-old, 15-year-old, and 19-year-old. And my 15-year-old is just violently opposed to any use of AI whatsoever. I mean, he is just ardently opposed to any use of AI. And I think I'm AI agnostic. So I'll get in these debates with him about... I mean, I think it's hard to argue that there aren't positive benefits of AI in terms of the medical industry, climate initiatives, all sorts of things. But there's a big move now. We've all seen the clips of commencement speeches of people booing AI executives or whatnot. And I'm not minimizing that, but there's a sense amongst young people today, a lot of young people today, that they're just going to be squeezed out of a job and that they're not going to have any viable options in the future. And we're going to find ourselves in Brave New World in 5 years. So, how do I know you don't believe that's true? So what would you say to a young person today growing up, in terms of how they should think about AI and how they should get involved in AI?

你知道，我想和你分享 AI 正在对软件行业做什么，因为我认为它预示着其他行业也会发生同样的事情，我认为这会非常…… AI 已经学会了写得非常好的代码，对吧？所以 AI 编程助手可以非常高效地编写大量代码。

You know, I want to share with you what AI is doing to the software industry, because it is a harbinger of what I think will happen with other industries as well, which I think would be very... And so AI has learned to write code really, really well, right? So AI coding assistants can write tons of code really efficiently.

因为很多 AI 团队是由像我们这样的软件工程师建立的，所以我们构建了工具来帮助自己。事实证明，这对我来说是软件行业有史以来最激动人心的时刻，因为 AI 无法完成软件开发者 100% 的工作，但它可以完成很多枯燥的部分。所以现在构建软件和 AI 感觉比过去，甚至比两年前，有趣得多。此外，我正在尽可能多地招聘高度 AI 赋能的软件工程师。这些高度 AI 赋能的开发者比上一代开发者生产力高得多，包括我在内的领先企业都希望尽可能多地雇佣他们，但我们就是找不到足够多的人。所以悲剧在于，很多关于 AI 的叙事一直在让人们远离 AI。我们就是培养不出足够多懂得使用 AI、懂得 AI 工程的高技能软件开发者。

Because a lot of AI teams are built by people like us, software engineers, so we built tools to help ourselves. And it turns out that this feels to me like the most exciting time ever to be in the software industry, because AI can't do 100% of a software developer's job, but it can do a lot of the boring bits. So it feels way more fun now than it used to be, even two years ago, to be building software and AI. Moreover, I am hiring as many highly AI-enabled software engineers as I can find. These developers that are highly AI-enabled are far more productive than the last generation of developers, and leading businesses, including mine, would love to hire as many of them as we can find, but we just can't find enough of them. So the tragedy is that a lot of these AI narratives have been turning people away from AI. So we're just not producing enough highly skilled software developers that know how to use AI, that know AI engineering.

随着 AI 开始加速对其他行业的影响，我认为其他行业也会如此。如果年轻人学会拥抱这些工具，他们会比没有这些工具的人生产力高得多，能做更多的工作。平均而言，考虑到我之前提到的 1.4% 这个令人担忧的重要例外，我认为薪资会上升，会有更多的工作机会，以及做更令人兴奋的工作的能力。而现在的悲剧——需要说明的是，我认为我们确实有一些问题。疫情期间，在接近零利率和大量政府刺激下，许多企业过度招聘，我们仍在承受一些后遗症，导致一定程度的裁员。这非常令人担忧，所以年轻人的就业市场比理想中更艰难。但我认为 AI 被指责的比它真正应得的要多。

As AI starts to accelerate its impact on other industries, I think this will be true for other industries as well. If young people learn to embrace these tools, they'll be far more productive and be able to do far more work than people without these tools. And on average, with that 1.4% stat that I gave earlier as one concerning and important exception, I think that salaries will trend up and there'll be more job opportunities and the ability to do more exciting work. And the tragedy now—to be clear, I think we do have a few problems. During the pandemic, with near-zero interest rates and massive government stimulus, many businesses did overhire, and we're still suffering from some of the overhang of that, leading to some amount of layoffs. That's very concerning, so it's led to a tougher job market for young people than is ideal. But I think AI gets blamed for more things than it truly deserves.

而且我发现——我知道我在斯坦福教过书，现在还在教。我热爱高等教育。我们应该支持高等教育。同时，我觉得高等教育对 AI 的适应速度不够快，这就是为什么很多大学还在培养学生，不是为了 2026 年的工作，而是为了 2022 年的工作。而那些工作正在变化。事实上，我们不希望大学培养学生去从事 2026 年的工作。我们应该培养他们去从事 2028 年或 2030 年及以后的工作。不幸的是，高等教育调整课程的过程相对于我们应对这一时刻所需的变化速度来说非常缓慢。所以，带着这些重要的告诫，我们有工作要做，帮助年轻学生走上更有前途的职业道路。但这些工作就在那里，我们似乎就是找不到足够多的人来做这些令人兴奋的、不断增长的工作。

And I find that—I know I taught, and still teach, at Stanford. I love higher education. We should support higher education. At the same time, I feel like higher education has not adapted fast enough to AI, which is why a lot of universities are still training students not even for the jobs of 2026, but for the jobs of 2022. And those jobs are changing. And in fact, we don't want universities to train students for the jobs of 2026. We should be training them for the jobs of 2028 or 2030 and beyond. And unfortunately, higher education's process for adapting curricula is just very slow relative to the speed of change that we need to meet this moment. So with those important caveats, we have work to do to help young students get onto more promising career paths. But those jobs are out there, and we just can't seem to find enough people to do these exciting, growing set of jobs.

而且，如果我说的话听起来空洞，让我再分享一件事。我现在办公室的团队——我们有一群实习生。我们有一个高中生，五六个大学生。他们是实习生，都是高度 AI 赋能的，他们非常棒。我希望我能找到更多这样的实习生。今年夏天我一直在招实习生，我爱我的实习生。我希望他们中的一些人会回来做正式工作。但我招的所有实习生往往都是高度 AI 赋能的那一类。而如果有人拒绝 AI，那就更难了——他们能做的事情最终会非常不同。

And in case what I'm saying rings hollow, let me share one other thing. My team right now in my office—we have a bunch of interns. We have one high school student, five or six college students. They're interns, all highly AI-enabled, and they're fantastic. I wish I could find more interns like that. I've been hiring interns this summer, and I love my interns. I hope some of them will come back for permanent jobs. But all the interns I hire tend to be on the highly AI-enabled side. Whereas if someone kind of rejects AI, it's harder—what they can do ends up being very different.

好的，我想我们还有时间回答几个问题。那么，请讲。谁有问题？

Okay, I think we have time for a couple of questions. So, yes, go for it. Who has a question?

你好。

Hello.

请讲。

Yes.

太好了。Guyer 先生，我觉得我们没怎么听到您的发言。您提到您在研究中使用了 AI。我非常好奇您如何权衡来自 AI 的研究。我为这个复合问题道歉，但还有，您是否会因为某些 AI 平台的加权分布而更看重它们？

Great. Mr. Guyer, I feel like we didn't really get to hear from you so much. You mentioned that you used AI in your research. I'm very curious to understand how you weight your research from AI. I apologize for the compound question, but also like, do you weight certain AI platforms better than others because of their weighted distribution?

好的。所以，是的，每当我做研究时，你都必须检查几乎任何 AI 模型的工作。如果我做研究，我倾向于喜欢那些引用来源的模型，这样我可以点击进入原始链接。但是，是的，我总是双重检查，而且我也假设，如果我使用研究——顺便说一句，我也会采访人——这不是我写脚本时使用的唯一研究手段。但我只是假设其中某个地方会有错误，我必须在后期草稿中审查。所以，是的，只是假设那是错误的。我的意思是，显然当我输入自己的简历时，它说我得了胰腺癌。所以我个人使用大约六种不同的模型。我不认为有任何一种模型在所有方面都更好；不同的模型在不同的事情上更好。所以，而且我绝不是 AI 专家，但是，是的，我使用大约六种不同的模型，我尽量保持对正在发生的不同创新或即将推出的新模型半了解。希望这回答了你的问题。

Okay. So yeah, anytime I'm doing research, you have to check the work of almost any AI model. And if I'm doing research, I tend to prefer the models that cite the sources so that I can click through to the original links. But yeah, I always double-check, and I also assume that if I'm using research—and by the way, I also interview people—it's not the sole means of research that I use when I'm writing a script. But I just assume that there will be mistakes at some point in it, and that I will have to vet that when I'm in the later drafts of it. So, but yeah, it's a mistake to just assume that. I mean, obviously when I was typing in my own bio, it said I had pancreatic cancer. So I personally use about six different models. I don't think there's any one model that's better at everything; different models are better at different things. So, and I'm by no means an AI expert, but yeah, I use about six different models, and I try to keep semi up-to-date on the different innovations that are happening or new models that are coming out. Hopefully that answered the question.

感谢你们分享智慧，先生们。我的问题来自你们对失败营销活动的描述——六件不符合你想法的事情，但也许其中两件符合。你认为 AI 本质上是否只更适合 1 到 100 的 Scaling（规模扩张）或收敛过程，而不是我们人类使用的从 0 到 1？无论你是在构建应用、公司还是电影。这是问题的第一部分。第二部分是：你提到了“AI 赋能”这个短语。你认为我们能否有一天在职场或专业场合使用像“AI 增强”这样的短语？还有，对于第一部分，如果你同意 AI 只更适合 1 到 100，那么我们作为成长中的艺术家或企业家，如何找到从 0 到 1 的灵感？

Thank you for sharing your wisdom, gentlemen. My question comes from your description on the failed marketing campaign—the six things that didn't fit your ideas, but maybe two of them did. Do you think AI essentially will just be better for the one-to-100 scaling or convergence process, not very much the zero-to-one for us human beings to use, whether you're building an app or a company or a movie? And that's the first part of the question. The second part would be: you mentioned the phrase 'AI enabled.' Do you think we could get to a day where we could have phrases like someone is 'AI empowered' as a thing we throw around in workplace or professionally? And for the first part, also, if you agree that AI is better for one-to-100 only, how do we find the zero-to-one inspiration as growing artists or entrepreneurs out there?

让我想想。我认为 AI 很棒，但 AI 在从 0 到 1 的某些部分和从 1 到 100 的某些部分都很糟糕。也许对于从 0 到 1，有一小部分决策如此重要，以至于感觉人类干预对于引导这些高层框架问题绝对至关重要。但然后我发现，即使对于从 1 到 100，如果我理解正确的话——填充细节——它仍然经常出错。事实上，David 的例子，踢出六种重写独白的方式，这实际上是一个非常先进的 AI 使用模式，仍然会弄错一些细节。所以你然后从它的不同部分剪切和粘贴。然后，我不知道，这是一个好问题，看看什么术语合适。是“AI 赋能”还是“AI 增强”？或者，在硅谷，这是否有我不喜欢的 AI 术语？但也许我们应该尝试围绕这个想出一个好术语。我喜欢“AI 增强”。

Let's see. I think AI is wonderful, and AI sucks for parts of the zero-to-one and for parts of the one-to-100. Maybe for the zero-to-one, there's a small set of decisions that are so consequential that it feels like human intervention is absolutely critical to steer it for those high-level scaffolding questions. But then I find that even for the one-to-100, if I'm interpreting the term correctly—filling in the details—it still messes up frequently. And in fact, David's example of kicking out six ways of rewriting the monologue, which is actually a very decently advanced AI usage pattern, still messes up some details. So you then cut and paste from different parts of it. And then, I don't know, it's a good question to see what's the right terminology to use. Is it 'AI enabled' or 'AI empowered'? Or in Silicon Valley, does this have AI-tilled terminology that I don't love? But maybe we should try to come up with a good term around this. I like 'AI empowered.'

我个人会区分给图像和给文本的提示词。我发现给图像提示词时，有时它会生成极其怪异的东西，我百万年都想不出来。我们在制作我参与的剧集《基地》第二季时，刚开始用 Midjourney。我和我的视觉特效总监用它做概念设计的早期迭代，最终我们会得到一些有趣的东西，然后交给人类艺术家。但有时它生成的玩意儿就是那么怪异，以最好的方式怪异，我们百万年都想不出来。所以我发现，给文本到图像的提示词时，幻觉是有价值的，因为我得一直嗑药才能想出那种东西，太神奇了。我会说：“天哪，太神奇了”，然后人类艺术家再迭代。但反过来，当我给它文本提示词时，我试过一些实验，比如“给我写一个关于 X 的短片”。如果你只给它最基础的提示词，它往往会吐出最老套、最平庸的版本。所以你不能——但很多人很懒。我确实读过 AI 生成的剧本，你能看出来。所以我不认为，至少在散文方面——我不是说写十四行诗之类的——但像剧本，我不认为从零到一已经实现了。我认为你得依靠自己的个人人生经验。

I will say personally that I would split it when I'm giving prompts for images versus prompts for texts. I find when I'm giving prompts for images personally that sometimes it will generate something so bizarre that in a million years I never would have thought of that. We were using Midjourney when it first came out, when we were on the second season of a show I did called Foundation. My VFX supervisor and I were using it for early iteration for concept design, and then eventually we would come up with something interesting and then we would feed it to human artists. But some of the times it would come up with something that was just so bizarre, in the best possible way, that we never in a million years would have come up with that. So I found when I was giving text-to-image prompting, the hallucinations were valuable because I would have to be on acid all the time to come up with that kind of stuff, and it was amazing. I would say, 'Holy, this is amazing,' and then human artists would iterate on this. But conversely, when I'm prompting it for text, I've tried some experiments where I'm trying to come up with, 'Oh, write me a short film about X.' If you just give it the most basic prompt, it will tend to spit out the most cliched, banal version possible. So you can't—but a lot of people are lazy. I have definitely read scripts that were generated by AI. You can tell. So I don't think, at least in terms of prose—and I'm not talking about writing a sonnet or something like that—but like a script, I don't think the zero-to-one is there. I think you've got to rely on your own human experience personally.

你好，我叫 Ted Sim。这个问题是问 Andrew 的。作为 AI 领域的思想领袖，外面有那么多行业，特别是我们这里在谈好莱坞。所以我想问问你对好莱坞的特殊兴趣，以及现在好莱坞和 AI 之间的冲突。你认为好莱坞有必要赶紧靠拢、变得更 AI 化吗？如果有，你能描绘一下如果我们不参与会是什么样子吗？

Hi there, my name is Ted Sim. This question is for Andrew. As a thought leader in AI, there are so many industries out there, and in particular we're talking about Hollywood here. So I want to ask about your particular interest in Hollywood and the conflict right now between Hollywood and AI. Do you see any urgency for Hollywood to need to lean in and become more AI enabled? And if so, can you paint that picture of what that could look like if we don't participate?

是的，谢谢。我确实得到了你的数据。我希望好莱坞能多讲一些正面的 AI 故事。我觉得好莱坞——很多好莱坞人都担心 AI 会导致失业。我认为是 1.4%，对吧？我不知道好莱坞的数字会是多少。我不是说完全不用担心，但不会像大多数人担心的那么糟。但由于各种原因，包括对工作的担忧，我们看到好莱坞出了一连串非常负面的 AI 故事。我们知道我们讲的故事会影响文化。所以我们在美国建立的是一个国家，绝大多数孩子甚至很多成年人都不喜欢 AI，不想拥抱它，想阻止它，不想建数据中心。这对我们的国家非常有害。相比之下，在中国，你听到很多故事，讲某个年轻企业家非常努力，建了一个太阳能农场，用它来给数据中心供电，因为他们做得很好，遇到了真爱。你猜孩子们因此被激励去做什么。我深受感动，因为我一个做 K12 教育的团队遇到一个高中生，她说：“我听说 AI 和人类灭绝有关，我不想和它有任何关系。”所以她自我选择退出了任何与 AI 相关的职业，而此刻 AI 正处于前所未有的最有前景的时期。所以她做了一个糟糕的——我认为非常糟糕的职业选择，或者说不算好的职业选择。所以如果好莱坞，除了我们讲的《终结者》那种可怕的 AI 故事之外，能给出一个更平衡的观点，即 AI 也能解决很多问题，我认为那会以更健康的方式塑造国家关于 AI 的叙事。而且这很紧迫。

Yeah, thank you. I really got your stats. One of the things I wish Hollywood would tell more positive AI stories. I feel like Hollywood—many people in Hollywood have been worried about job displacement from AI. I think 1.4%, right? I don't know what the number in Hollywood will be. I don't say don't worry about it at all, but it's not going to be nearly as bad as most people fear. But for various reasons including the job worries, we see a sequence of very negative stories about AI coming out of Hollywood. And we know that the stories we tell influence culture. So what we've built in America is a country where the vast majority of our kids and even many grown-ups do not like AI and do not want to embrace it and want to put a stop to it and do not want to build data centers. And this is very damaging for our nation. In comparison, in China, you hear a lot of stories being told about some young entrepreneur that worked really hard and built a solar farm and used that to power data centers, and because they did so well, they met the true love. And guess what the young kids therefore are motivated to do. I was really moved when one of my teams working in K12 education met this high school student, and she said, 'I heard AI has something to do with human extinction, and I don't want anything to do with that.' So she self-selected out of a career that had anything to do with AI at this moment when AI is the most promising it has ever been. So she made a terrible—I think really terrible career move, or maybe not a great career move. So if Hollywood, in addition to all the Terminator scary AI stories we tell, can give a more balanced view that AI could also solve a lot of problems, I think that would shape the national narrative on AI in a much more healthy way. And it is urgent.

你好，我叫 Michael Lander。我是编剧、导演。我用 AI 的方式和你很像。我发现，在我的创作圈子里，作为一个共情者，我觉得当我说我用 AI 时，我有点像你儿子，你 15 岁的儿子。你知道，AI 是建立在偷来的财产上的。有时候我对我那些这么说的人有自己的回答。我很想听听对这个问题的技术性回答，还有艺术家的回答。

Hi, my name is Michael Lander. I'm a writer, director. I use AI very similar to you. I have found in my creative circles, being an empath, I feel like when I say I use AI, I kind of get like your son, your 15-year-old. You know, it's AI is built on stolen property. And there are times where I have my own sort of answer to those folks who say that to me. I would love to hear sort of the technical answer to that and also the artist answer to that.

我来做艺术家的回答，你可以做技术性的回答。如果我们互换会很有趣。有意思。我儿子就像，他说：“我讨厌你那样做。”我认为首先，AI 的输出只取决于 AI 的输入。好吧。我知道我个人——对于在座创意社区的人来说——我确实收到过开发主管用 AI 生成的反馈意见。我是说 100%。而且那些意见很糟糕，因为有些——可能甚至不是主管，而是实习生或助理，就输入“我们怎么让这个剧本更好？”句号，对吧？然后它吐出了你能得到的最通用的意见。在某些情况下，我收到的意见里，他们显然在给我们之前都没读过，对吧？因为里面全是各种错误。所以关键在于，它只取决于输入，对吧？如果我们作为艺术家只是给它提示词，说“给我写首热门歌曲”或“给我拍部热门电影”，而这是我们做的唯一干预，那它就会制造垃圾。它会制造平庸的垃圾。但如果我们把它当作回音板，或者如果我们说：“好，我写了这个场景。现在给我 10 个角色试图摆脱那次约会的惊人方式。”同样，它会给 7 个可能不太好，但就我而言，也许 3 个会很有趣，或者有趣到能把我带到我以前没想到的路上。所以我认为，再说一次，它是一个工具。我知道当我们从胶片转向数码相机时，很多人——我和克里斯托弗·诺兰合作很多。他绝对拒绝这么做。吉尔莫·德尔·托罗立刻接受了。现在我已经转向数码，我再也不会回到胶片了，因为它有更多用途。所以在这方面我把 AI 视为工具，我们都看过糟糕的数码拍摄电影，也看过精彩的数码拍摄电影。所以我再次认为它只是一个工具，我们不能在使用上偷懒。

I'll do the artist answer and you could do the technical answer. It'd be funny if we swapped. It's funny. My son is like, he's like, 'I hate that you do that.' I think that first of all, the output of AI is only as good as the input of AI. Okay. I know that I personally—and for those of you who are in the creative community here—I've definitely received notes from development executives that were generated by AI. I mean 100%. And the notes are terrible because some—it probably isn't even the executives, an intern or an assistant just typed in 'How do we make this script better?' period, right? And it spat out the most generic set of notes you could get. And in some cases, I've been given notes where they obviously didn't even read them before they gave them to us, right? Because there were all sorts of mistakes in them. So it's only as good as the input, right? If we as artists are just giving it prompts and saying 'Make me a hit song' or 'Make me a hit movie,' and those are the only interventions that we're doing, it's going to create crap. It's going to create generic crap. But if we use it as a sounding board, or if we say, 'Okay, I've written this scene. Now give me 10 really surprising ways that the character could try to back out of that date,' again, it will give seven of them probably won't be very good, but in my case, maybe three of them will be interesting or interesting enough that it'll take me down a path that I hadn't thought of before. So I think that again, it's a tool. I know that when we were switching from film to digital cameras, so many—I've worked with Chris Nolan a lot. He absolutely refuses to do it. Guillermo del Toro embraced it immediately. Now that I've switched to digital, I would never go back to film because there are so many more uses for it. So I regard AI as a tool in that regard, and we've all seen terrible digitally shot films and we've seen amazing digitally shot films. So I again just think it's a tool, and we just can't be lazy about its use.

你知道，这涉及到人类必须不断地——我们听过“人在回路”这个说法，但还有一个说法，当然不是我发明的，我觉得必须是“人在主导”，你必须保持人在主导。如果你想创作出任何有分量的作品，就不能把创造力外包出去。而且我确实认为——最后一点，然后安德鲁，请你接着讲——我确实认为我们正在迅速走向一个阶段，无论是 Spotify 上的歌曲、短视频内容，还是最终的长视频内容，艺术真的会分成两个层级。一类基本上只是勉强有人类在回路中，另一类则是人类在主导。

You know, it's got that the human has to constantly—we've heard this phrase 'human in the loop,' but there's another phrase that I certainly didn't coin, where I just think it has to be 'human in the lead,' and you have to remain human in the lead. You can't outsource your creativity if you want to create anything of any note. And I do think—last point, and then Andrew, please take it from here—I do think we're quickly moving to a place, whether it be songs on Spotify or short-form content or eventually long-form content, where there's really going to be two tiers of art. There's going to be the stuff that's basically just barely human in the loop, and then there's going to be stuff that's human in the lead.

但“人类精英”这个概念非常有趣，对吧？不知何故，AI 从互联网上学习，而 AI 技术倾向于把所有东西平均化，这就是为什么我们最终得到的是平庸的垃圾或平均化的东西，而不是非凡的作品，除非我们以某种方式推动技术发展。我觉得作为技术专家，我对技术很了解，但我远不如你们几乎所有人那样会讲故事。所以我作为技术专家，很想找到与好莱坞合作更多项目的方式。我实际上正在和一个好莱坞团队合作一个项目，但现在还不能说，但如果能有机会与你们所有人合作更多项目，理想情况下讲述更多关于 AI 的正面故事，那就太好了。我会去硅谷，然后我们——

But the 'human elite' is a very interesting concept, right? Somehow AI has learned from the internet, and the AI technology tends to average all the stuff, which is why we end up with this middling slop or milling average rather than something extraordinary, unless we push the technologies in certain ways. And I feel like as a technologist, I know the technology well, but I'm not nearly the storyteller that almost all of you are. So I'd love to find ways as a technologist to do more projects with Hollywood. I'm actually working on one that I'm not allowed to talk about yet with a Hollywood team, but it'd be great if there's a way to do more projects with all of you to ideally tell more positive stories about AI. I'm gonna come up to Silicon Valley and we'll—

哦，是的，我们会谈这个的。对，我们应该达成这个合作。

Oh yes, we'll talk about that. Yeah, we should do that deal.

为了回应“AI 建立在窃取作品之上”的说法，我想分享我的个人观点。今天早些时候我在布罗德博物馆，庞德也在场，我很喜欢盖蒂博物馆。当我参观这些博物馆时，有时会看到年轻艺术家坐在地板上，用画笔临摹大师的作品——不是在布罗德，也许是在盖蒂或史密森尼等。所以我们社会已经决定，我们喜欢年轻艺术家坐在地板上临摹这些百年画作，作为他们掌握技艺的一部分。他们不允许逐字重绘某人的画作并当作原创作品——对此有某些限制。但我们鼓励年轻艺术家向大师学习。今天，我们也有年轻作者会阅读大量书籍、阅读互联网上的内容，并综合这些形成自己的创作风格或绘画风格。如果我们允许年轻艺术家这样做，那么作为一个非艺术家，我很乐意被允许派 AI 为我做这件事。

Just to answer the 'AI built on stolen works' remark, I want to share my personal view. Earlier today I was at the Broad Museum, ponder who's also here, and I love the Getty Museum. When I visit some of these museums, sometimes I see young artists sitting on the floor using paintbrushes to copy the work of the great masters—not at the Broad, but maybe at the Getty or the Smithsonian, and so on. So we as a society have decided that we like it when young artists sit on the floor to copy these century-old paintings as part of how they master their craft. And they're not allowed to repaint someone's painting verbatim and pass it on as an original—there are certain barriers to that. But we encourage our young artists to learn from the great masters. Today we also have young authors that will read tons of books, read things on the internet, and synthesize that to form their own creative style of writing or painting. And if we allow young artists to do this, I would quite like—as a non-artist—to be allowed to send AI to do this for me.

我认为有一条线不应该被跨越。有一桩针对 Anthropic 的诉讼，Anthropic 被发现未经许可使用了受版权保护的书籍——

I think there was a line that should not be crossed. There was a lawsuit against Anthropic where Anthropic was found to have used copyrighted books without permission—

包括我的三本书。

Including three of mine.

是的。所以——

Yes. So—

那是那起诉讼的一部分。

That was part of that lawsuit.

所以我认为那确实跨越了某条线，但是——

So I think that did cross a certain line, but—

嗯，他们越界了，因为他们故意拿走了未经许可数字化的书籍库，而且他们知道自己在做什么。所以是的，那绝对越界了。但抱歉，请继续。

Well, they crossed the line because they knowingly took a trove of books that were digitized without permission, and they knew they were doing that. So yeah, it was absolutely crossing the line. But sorry, keep going.

说得有道理。我实际上认为那样使用受版权保护的作品是有问题的。与此同时，如果我可以阅读某人选择在互联网上免费发布的网页，我不能逐字复制——那会侵犯版权——但我可以阅读大量网页，从中学习，汲取灵感，综合出自己的观点，写一篇全新的文章，不逐字复制任何内容。对我来说，与其亲自去做，也许我可以派 AI 替我做这件事，这感觉是可以的。因为如果我们认为人类可以做到这一点，我们能否让人类派 AI 为他们做同样的事情，而不逐字复制？现在我意识到，还有一个重要的补偿和公平问题需要解决，因为 AI 现在可以以人类无法企及的规模做到这一点——这似乎是件好事——但从长远来看，我们如何找出一种经济模式来确保一切仍然有效？我认为最终会解决的，但我也意识到，那些担心财务模式的出版商也有很多焦虑。

Fair point. I actually think using copyrighted works that way is problematic. And at the same time, if I'm allowed to read a web page that someone chose to freely publish on the internet, and I'm not allowed to copy it verbatim—that would violate copyright—but I'm allowed to read tons of web pages and learn from that and take inspiration to synthesize my own view to write a brand new article that doesn't copy anything verbatim. It feels okay to me to say that instead of me doing it personally, maybe I can send AI to do it for me. Because if we think humans can do this, can we let humans send AI to do the same thing for them without verbatim copying? Now I realize that there's an important compensation and fairness question that needs to be answered too, because AI can now do this at a scale that no human can—which seems like a positive thing—but longer term, how do we figure out an economic model to make sure that it all still works? I think it'll actually work out okay, but I realize that there's a lot of angst for the publishers that worry about the financial models as well.

好的。我希望我们有更多时间，但我想——

Okay. I wish we had time for more, but I think—

是的，不幸的是，我想我们的时间到了，但让我们再次为我们的杰出演讲者鼓掌吧？

Yes, unfortunately, I think we're all out of time, but can we hear it again for our incredible speakers?

非常感谢两位。

Thank you both so much.