迪恩·鲍尔谈加入 OpenAI、AI 政策与递归自我改进

大家好，欢迎回到《认知革命》。今天的嘉宾是 Dean Ball，你可能已经知道，他最近宣布将加入 OpenAI，组建并领导一个名为“战略未来”的新团队，任务是帮助 OpenAI 的高层领导制定前沿 AI 政策。这一职位的重要性不言而喻——我们正进入递归自我改进的时代，OpenAI 自己的公开时间表显示，三个月后就会出现一名 AI 研究实习生，而到 2028 年 3 月，也就是 21 个月后，将出现一个完全自主的 AI 研究员。这使得本期节目成为我们做过的最具自我推荐价值的一期。由于 Dean 非常慷慨地投入时间，我们得以覆盖大量内容。在第一部分，Dean 反思了美国 AI 行动计划发布一年后的情况，以及他对 Anthropic 供应链风险指定现状的看法，还有州政府在缺乏先发制人措施的情况下采取的行动。他还分享了对中国政府限制购买美国芯片决定的理解，认为目前关于 Fable 禁令的幕后情况，以及他作为用户个人有多怀念 Fable。接着我们转向他加入前沿实验室的原因。他解释说，前沿实验室是一种全新的强大行为体，需要新的政策范式，而且至关重要的是，它们所掌握的关于 AI 发展现状和未来的信息是如此独特，以至于他觉得没有这些信息就无法做出最好的工作。他还描述了他如何理解自己对 OpenAI 使命——确保 AI 造福全人类——的责任，他的团队将如何与 OpenAI 现有的政府事务团队协作。虽然他强调自己很快会从从事研究工作的研究人员那里学到更多，但他分享了对递归自我改进可能意味着什么的基本看法，以及一个仍然被忽视但至关重要的问题：如何管理最新最强模型的内部部署。在此过程中，我们还听到了他对“勇气与品格”辩论的看法、OpenAI 与 Alex Boris 之间发生了什么、我们最近从 Bernie 和 Trump 那里听到的股权共享讨论、AI 行业是否已经大到不能倒从而隐含政府支持、AI 公司相对于美国政府有哪些杠杆、他个人如何考虑与 Sam Altman 合作、个人性格在塑造未来中的作用、他在这份工作中成功的标准是什么、理论上哪些红线可能导致他辞职，以及他未来在工作中和写作中打算如何使用 AI 以及避免使用 AI。在这么多重要议题中，最让我印象深刻的是 Dean 的一种感觉——我也认同——我们正在进入一个“主角能量”的历史时期。在这个时期，个体人类能动性在少数关键地方达到最大杠杆，至少在机器最终超越我们之前。这是一个严峻的现实，迫使每个人思考我们愿意做出哪些牺牲和妥协来帮助塑造未来。Dean 的第一个儿子还不到一岁，他重新进入竞技场显然会做出一些牺牲。我毫不怀疑他会比以往更加努力。但重要的是，他并没有为了这个角色而妥协自己的思想独立性。令我惊喜的是，即使作为 OpenAI 的员工，他仍将保留公开撰写 AI 政策文章的自由。这种自由甚至延伸到本播客——其中确实有一些特别坦率的时刻，但 OpenAI 没有要求审阅，也没有在发布前看到内容。所以，话不多说，在他准备加入 OpenAI 担任这个定义职业生涯、并在某种程度上塑造世界的角色之际，我希望你们喜欢这场与伟大且日益强大的 Dean W. Ball 的坦诚对话。《认知革命》由 Mercury 赞助，这是一家金融科技公司，超过 30 万家雄心勃勃的公司和个人信赖它来管理财务。我已经把 AI 融入到生活的几乎每个角落：我的电子邮件、消息、日历。我甚至给了我的智能体 Mercury 虚拟卡，设置了低限额和类别及商户限制供它们自主使用。但我的 AI 对财务数据的访问仍然有限。使用普通银行，我可能会导出大量对账单，让助手处理。但要获取实时最新信息，尤其是采取行动，试图让智能体通过浏览器使用银行太困难、太慢、太容易出错，不值得。这就是为什么 Mercury 的新对话界面命令如此重要。它直接构建在 Mercury 中，意味着你可以用自然语言访问财务，而无需将任何信息暴露在银行账户之外。无需导出、无需电子表格、无需将交易粘贴到第三方工具中。我真的认为很多人会喜欢这种方式。它已经可以帮助你采取行动，所有操作都受你账户中已设置的权限和审批策略约束。我对 2026 年银行业达到这种 AI 集成水平感到由衷钦佩。所以我邀请你和我一起进入未来。访问 mercury.com 了解更多，几分钟内在线申请。Mercury 是一家金融科技公司，非 FDIC 保险银行。银行服务由 Choice Financial Group 和 Column NA（FDIC 成员）提供。感谢 Mercury 对《认知革命》的支持。现在开始节目。Dean Ball，《超维度》Substack 的作者。欢迎回到《认知革命》。

Hello and welcome back to the Cognitive Revolution. My guest today is Dean Ball, who as you probably already know, recently announced that he'll soon be joining OpenAI to build and lead a new team called Strategic Futures with a mandate to help OpenAI's senior leaders shape frontier AI policy. The obvious importance of that position as we enter the era of recursive self-improvement with OpenAI's own public timeline calling for an AI research intern just three months from now and a full-fledged autonomous AI researcher in March 2028 just 21 months from now makes this one of the most self-recommending episodes that we have ever done. And because Dean was so generous with his time we were able to cover a ton of ground. In the first section we get Dean's reflections on America's AI action plan a year after its release and his perspective on the current state of the Anthropic supply chain risk designation and the moves that state governments have made in the absence of preemption. We also get his understanding of the Chinese government's decision to restrict the purchase of American chips, what he thinks is happening behind the scenes right now with respect to the ongoing Fable ban, and how much he personally misses Fable as a user. We then turn to his reasons for joining a frontier lab. Now he explains that frontier labs are a fundamentally new kind of powerful actor which demand new policy paradigms and also critically that the information they contain about the present and future of AI development is so differentiated that he feels he simply won't be able to do his best work without access. He also describes how he understands his duty to OpenAI's mission of ensuring that AI benefits all humanity, how his team will relate to OpenAI's existing government affairs team. And while he emphasizes that he'll soon be learning much more from the researchers doing the work, he shares his baseline perspective on what recursive self-improvement is likely to mean and the still neglected but critically important question of how to govern the internal deployments of the latest and greatest models. Along the way, we also get his takes on the courageability versus character debate, what happened between OpenAI and Alex Boris, the equity sharing talk we've recently heard from both Bernie and Trump, whether the AI industry is already too big to fail and thus implicitly government-backed, what sources of leverage AI companies have vis-a-vis the US government, how he's thinking about working personally with Sam Altman, and the role that individual personalities will play in shaping the future, what success looks like for him in this role and what red lines could theoretically cause him to quit, and finally how he intends to use AI and also to refrain from using it in his work and writing going forward. Somehow amidst so many important issues, what stands out most to me is Dean's sense which I share that we are entering a main character energy period of history. A time in which individual human agency achieves maximum leverage in a few key places at least before perhaps the machines ultimately surpass us. It's a stark reality that forces each of us to ask what sacrifices and compromises we are willing to make to help shape the future. Dean, whose first son is not even a year old, will clearly be making some sacrifices as he re-enters the arena. I have no doubt he will be working harder than ever. But importantly, he did not compromise his intellectual independence to take this role. To my pleasant surprise, even as an OpenAI employee, he will retain the freedom to write publicly about AI policy. A freedom that extends even to this podcast, which definitely contains a few notably candid moments, but which OpenAI did not ask to review and has not seen prior to our publication. And so without further ado, as he prepares to join OpenAI for a career-defining and to some degree a world-shaping role, I hope you enjoy this frank conversation with the great and increasingly powerful Dean W. Ball. The Cognitive Revolution is brought to you by Mercury, the fintech that more than 300,000 ambitious companies and individuals trust to run their finances. I've wired AI into nearly every corner of my life: my email, my messages, my calendar. I even gave Mercury virtual cards to my agents with low limits and category and merchant restrictions for their autonomous use. But still, my AI's access to my financial data has remained limited. With a normal bank, I might export a bunch of statements and have my assistant process them for me. But for real-time up-to-date information and certainly for taking any action, trying to get your agent to use the bank via the browser is just too hard, too slow, and too error-prone to be worth it. And that's why Mercury's new conversational interface command is such a big deal. It's built directly into Mercury, which means you get natural language access to your finances without exposing anything outside of your bank account. No exports, no spreadsheets, no pasting your transactions into third party tools. I really think a lot of people are going to prefer it this way. And it can already help you take actions too with everything bound by the permissions and approval policies that you've already set up in your account. I am genuinely impressed to see this level of AI integration in banking in 2026. And so I invite you to join me in the future. Visit mercury.com to learn more and apply online in minutes. Mercury is a fintech company, not an FDIC insured bank. Banking services provided through Choice Financial Group and Column NA members FDIC. Thank you to Mercury for supporting the Cognitive Revolution. And now on with the show. Dean Ball, author of the Hyperdimensional Substack. Welcome back to the Cognitive Revolution.

非常感谢你再次邀请我，Nathan。很高兴来到这里。

Thank you so much for having me back, Nathan. It's great to be here.

我对这次对话非常兴奋。我们要聊聊你生活中的一些大新闻，而且你知道，过去一年半你一直处于风暴中心，所以我有很多问题想问你，关于你参与的一切，你对当前形势的看法。我打算学 Tyler Cowen 的风格，直接抛出一连串问题，主要就是想听听你对这么多不同话题的看法。你准备好进行播客冲刺了吗？

I'm really excited for this conversation. We've got some big news in your life to cover and you know, you've really been in the eye of the storm over the last year and a half and so I have just so many questions of everything that you've participated in, your thoughts on where we are today. I'm going to try to go Tyler Cowen style on you and just fire a bunch of questions and mostly just want to hear from you on so many different topics. Are you ready for a podcast sprint?

我准备好了。好的，我们先从你在白宫的时光开始，回顾一下美国的 AI 行动计划。当时它非常受欢迎。你现在会如何评价它？回顾起来，在概念或政策层面，有没有什么你觉得会改变或做得不同的地方？

I'm ready. All right, let's start off with your time in the White House and a little look back on America's AI action plan. It was super well-received at the time. How would you critique it now? Is there anything that you feel like you would change or do differently looking back at a conceptual or policy level?

关于行动计划，你要记住的是，嗯，我觉得在 AI 发展方面，世界并没有太多让我感到惊讶的地方。

What you have to remember about the action plan is that like, you know, I don't feel like there's a lot there's a ton about the world in terms of how AI has developed that has surprised me.

基本上，我们仍处在我预想的基本时间线上：模型在 25 年底、26 年初具备可怕的网络能力，生物能力可能紧随其后。我之前公开预测过很多这类事情。所以我们正生活在我预想的世界里。我认为编码智能体在普及度上可能让我意外，这是积极的一面。我没想到会这样。

Like basically we're still in the basic timeline I kind of figured: models with scary cyber capabilities late '25, early '26, and probably bio soon after. I had publicly predicted a lot of this stuff. So we're living in the world that I figured. I think coding agents maybe surprised me in terms of popular uptake on the positive side. I didn't expect that.

但问题是，在撰写行动计划时，华盛顿并没有生活在那个世界里。所以这个行动计划是一个奇怪的诠释学例子：你现在写一份文件，受众是当下的读者，但你也在试图模拟同一群人在略微不同的近未来——他们比今天多接受了 30% 的 AGI 观念，然后是 50%。你希望他们回头再看这份文件时，会想：“哦，等等，我现在完全以不同的方式理解它了。”你当然可以批评这太像五维象棋了。但这不是故意的，只是任务的性质使然。

But the thing is that DC wasn't living in that world when the action plan was written. So the action plan is this weird example of strange hermeneutics: you're writing a document now, your audience is the present-day audience, but you're also trying to model the same people in a slightly different near future where they're 30% more AGI-pilled than today, then 50%. And you hope they go back and look at the document and think, "Oh wait, I now read this in a totally different way." You could definitely criticize that as being too much like five-dimensional chess. It wasn't intended to be; it was just the nature of the task.

我可能批评的一点是，如果能更明确地说明我们实际上在谈论能做各种事情的通用智能体，并解释这不仅对美国意味着什么，也对世界意味着什么，那会更好。行动计划的一个核心是：这一切正在发生，政府并未主导，政府需要学会顺水推舟，利用这一点最大化美国的领先地位和地缘政治力量，同时理解最好的方式是保持积极，努力增长全球经济并让其他人参与进来。我觉得行动计划并没有很好地整合这些，读起来更像是三十多个独立的主题目标，而不是一个由共同战略或愿景统一起来的 cohesive 整体。如果再给我两个月，我会专注于这一点。

The one thing I might critique is that it probably would have been good to be a little more explicit about how we are really talking about generalist agents that can do all kinds of stuff, and try to explain not just what that will mean for America, but also for the world. A big part of the action plan is: this is happening, the government's not leading it, and the government needs to figure out how to ride the current of the river and use this to maximize American primacy and geopolitical power, while understanding that the best way to do that is to be positive and try to grow the world economy and bring other people in. I feel like the action plan doesn't really stitch that together all that well and probably reads more like three dozen separate thematic objectives than one cohesive thing unified by a common strategy or vision. If you had given me two more months, I would have focused on that.

另一件让我感到遗憾的被砍掉的事情是，我原本非常热衷于特定领域的采用，尝试在非常具体的行业做案例研究，说联邦政府可以解决哪些障碍？比如医院，医院记录保存。卫生与公众服务部或退伍军人事务部可以做非常具体的事情。VA 是一个巨大的单一支付方医疗系统——我们不认为美国有单一支付方医疗，但 VA 就是。海量的数据，大量由政府雇员提供的直接医疗服务。在医疗领域用 AI 进行实验本可以非常有价值。我们只是没有时间。所以我会批评说我们本可以更具体一些。

The other thing I will say is the things we left on the cutting room floor that I feel bummed about. One is I was really passionate about adoption in particular sectors and trying to do case studies in really specific industries, saying what are the barriers here that the federal government can do something about? Hospitals, for example, hospital recordkeeping. There are very specific things the Department of Health and Human Services or Veterans Affairs could do. The VA is an amazing huge single-payer healthcare system—we don't think of America as having single-payer healthcare, but we do the VA. Huge amounts of data, huge amounts of direct medical care provisioned by government employees. Experimentation with AI in healthcare could have been enormously valuable. We just didn't have time. So I would criticize that we could have been more specific there.

那么你当时交出了接力棒，说自己更偏向想法型的人，会留给别人去执行，希望他们更擅长实施并通过政府实际流程推进。你觉得现在进展如何？

So then you handed off the baton, saying at the time that you're more of an ideas guy and it'd be left to somebody else who you think would hopefully be better at implementation and running all these things through the actual process of government. How would you say that is going right now?

哦，显然建设正在进行。就在我的家乡州，不远处的密歇根，一个吉瓦级的数据中心刚刚破土动工，尽管遭到了一些当地邻避式的反对。所以这似乎正在发生。我们听说军方在尝试使用 AI。显然，在军方应该做什么方面存在争议。我们公众也不完全知道他们在做什么。然后还有其他各种事情。与此同时，你的文章对共和国的健康状况发出了相当严重的警报。看起来你目前对政府驾驭浪潮能力的信心并不太高。那么你认为后续执行情况如何？这种悲观情绪的核心是什么？

Oh, it seems like clearly the buildout is happening. Even in my home state, not too far away in Michigan, a gigawatt data center just broke ground despite some local NIMBY-style objections. So that seems like it's happening. We hear about the military trying to use AI. Obviously that's contested in terms of what they should be doing. We also don't fully know what they're doing. And then there are all these other things. Your writing in the meantime has sounded the alarm in a pretty severe way around the health of the republic. And it seems like your faith in government's ability to ride the wave is not super high right now. So how would you say it's going in terms of follow-through? And what is the core of that pessimism for you?

我认为如果你看看行动计划中的各个项目——其中一些很难评估，因为部分实施是在高层进行的，即在机密环境中，尤其是关于军事采用和国家安全的一些内容。有一件我觉得被低估的事情：行动计划中有一部分间接提到了在发生国家危机时军方征用全国所有数据中心，将它们拼接起来做某事的想法。这听起来很像 Leopold Ashen Brener 的点子，但描述语言足够平淡，没有引起注意。所以有很多类似的事情，实施在发生，但不是在公开场合。但我认为如果你能看到一切，你会看到我们大概完成了 30% 到 40%，对于一年来说相当不错了。距离行动计划发布大约 11 个月。在所有支柱中，许多重大事项都取得了显著进展。在能源方面，这届政府在核能上做了非常了不起的事情，还有更直接源于行动计划的事情。

I think if you looked at the individual items in the action plan—and some of this is hard because some implementation ended up being done on the high side, meaning in classified environments, especially some stuff about military adoption and national security things. One thing that I feel is underrated: there is a part of the action plan that obliquely references the notion of the military commandeering all the data centers in the country in the event of a national crisis to stitch them together to do something. It sounds like a very Leopold Ashen Brener idea, but it was described in sufficiently mundane language that it didn't jump off the page. So there are a bunch of things like that where implementation, to the extent it's happening, is not happening in public settings. But I think if you could see everything, you would see that we're probably 30 to 40% done, which is pretty good for a year. We're about 11 months out from when the action plan came out. A lot of the major things across all pillars have seen significant advances. On the energy side, this administration is doing really amazing stuff on nuclear, and stuff more directly downstream of the action plan.

目前正在进行一些重大变革，预计将在未来几天内由联邦能源监管委员会宣布，这些变革涉及连接超大型工业电力用户到电网的流程，并加速这一过程。有很多类似的事情，都是非常实质性的内容，正在快速推进。

There are major changes that are in process right now that should be announced in the coming days from the Federal Energy Regulatory Commission that deal with the process for connecting very large industrial electricity users to the grid and accelerating that process. There are a lot of things like that that are just really meaty, substantive things that are proceeding at a pace.

我还要说，军方对 AI 的采纳让我感到惊喜，总体而言，军方直接参与产业政策并扶持美国制造业中的初创企业——这在行动计划中有所提及——那些在物理自主等领域进行创新的初创企业都表现得很出色。

I would also say military adoption of AI has impressed me to the upside, and generally speaking, military's direct involvement in industrial policy and boosting startups in US manufacturing—which is talked about in the action plan—startups that are doing innovative things with physical autonomy and stuff like that is all doing great.

我认为另一个要点是，行动计划广泛讨论了采纳问题，而且我认为，总体来看，美国的 AI 采纳实际上进展相当顺利。

I think another main thing is the action plan talks a lot about adoption more broadly, and I think AI adoption in America is actually going pretty well, all things considered.

当然，这些都是好的一面。但更关键的是，很难说本届政府是按照我所认为的行动计划的精神来行事的。定义行动计划的精神不是我的职责，毕竟这是他们的政府，对吧？例如，行动计划的一个大支柱、大原则是输出美国 AI 并在全球推广。这很难想象如何与对前沿模型实施全球出口管制——在 90 分钟内通知所有非美国人士——保持一致。

Now of course that's all the nice stuff. I would say more critically, it would be hard to say that the administration has carried itself according to what I think of as the spirit of the action plan. It's not my job to say what the spirit of the action plan is. Ultimately, it's their administration, right? For example, a big pillar, a big principle of the action plan was the notion of exporting American AI and getting it adopted all across the world. It seems hard to imagine how that's consistent with global export controls on frontier models imposed within a 90-minute notice on all non-US persons.

正是这类事情——当我离开政府后，我在世界各地做了很多国际旅行，以私人身份代表美国从事准外交工作，试图解释我们在出口推广等工作上的想法以及整体战略。你在国外，尤其是欧洲，听到的最大担忧是：‘我担心你们美国人如果对我们生气，会在某个时候关闭模型。’我在政府时，我们试图缓解这种担忧。离开政府后，今年早些时候我在印度参加 AI 行动峰会，在许多地方、许多准外交场合，我说：‘不，别担心。我们不想那样做。我们希望生态系统如何如何。’然后政府当然就去做了，基本上证实了国际上许多人的最大恐惧。这没有帮助，绝对没有帮助。

That's the kind of thing that when I was out in the world—my life after government—there are definitely a lot of international trips I do where I am engaged in quasi-diplomatic work on behalf of the United States as a private citizen, but as someone who's trying to explain what we were thinking with things like the export promotion work and our whole strategy there. The biggest concern you hear from people abroad, especially in Europe, is: 'I just worry that you Americans are going to turn off the models at some point if you get mad at us.' When I was in government, we were trying to assuage this concern. When I left government, I spent time in India at the AI Action Summit earlier this year. Many places, many quasi-diplomatic engagements where I said, 'No, don't worry. We don't want to do that. We want to ecosystem blah blah blah.' And then of course the administration goes and does it and basically confirms the biggest fears of a lot of people internationally. That doesn't help. That certainly doesn't help.

实际上，我认为这关系到另一件事——再说一次，我不知道是否可能——但我觉得行动计划相对沉默的一个问题是 AI 治理。这是我之前和之后工作的很大一部分，但行动计划并没有真正提及。我当时对此的推理是：第一，当时政府内部的奥弗顿窗口很窄；第二，在我看来，很多这类问题理想情况下应由立法解决，而行动计划不应涉及新法律——它只应包含行政分支能做的事情。但我认为行动计划本可以有更明确的材料，比如：‘嘿，好吧，在某个时候事情会变得可怕，你该怎么办？这里是如何不恐慌。’本可以有更多这样的内容，因为我认为我们正在看到这一点。

I think actually that relates to one other thing which, again, I don't know if it would have been possible, but one thing that I feel as though the action plan was relatively silent on was the issue of AI governance. That's a big part of what I worked on before and after, and it's not really referenced. My reasoning for it at the time would have been: number one, tough Overton window within the admin at the time; and number two, a lot of that in my view is ideally legislative, and the action plan was not supposed to talk about new laws—it was supposed to be just things the executive branch could do. But I think there could have been more explicit material in the action plan about, 'Hey, okay, at some point things will get scary and what should you do? Here's how not to panic.' There could have been more of that because I think we are seeing it.

不过有趣的是，我认为最终存在一种区别。有一大批公务员——全职职业官僚——以及中低层政治工作人员，他们都阅读了行动计划，并且执行得相当好。当然，还有那些高层人士，他们不一定阅读政府发布的每一份战略文件，而且他们基本上非常被动。所以这些神话般的事情就发生了——‘天哪，我们得做点什么’——他们不会想行动计划会告诉我做什么。那根本不是内阁部长会想的事情。但让我觉得有趣的是，我认为我们正在看到政府重新发明行动计划中的一些想法，或者高层人士重新发明关于使用 AI、在政府中建立技术能力、第三方评估等想法。我认为我们正在看到他们从第一性原理重新发明这些东西，所以我仍然乐观。但是，是的，我确实认为政府在很多方面偏离了——讽刺的是，既没有足够认真地对待风险，又过度纠正，没有太认真对待，但反应方式实际上并没有应对风险。所以最终，我倾向于对人宽容，并说我希望我们正处于政策制定的高神经可塑性阶段。我认为三个月后我们可能会处于一个非常不同的世界。但当然，如果你看头条新闻，你会说：‘啊，这似乎与行动计划完全不一致。’我不能否认这一点。

It's funny though because in the end I do think there's a distinction. There are this huge array of civil service bureaucrats—full-time career civil servants—and then there are low to mid-level political staff, and they all read the action plan and are implementing it in quite a good way. Then of course there are the very high-level people who don't necessarily read every strategy document the administration comes out with, and they are fundamentally very reactive. So this mythos stuff happens—'Oh my god, we got to do something about this'—and they're not thinking about what the action plan would tell me to do. That's not at all what a cabinet secretary is thinking. But something that is amusing to me is that I think we are watching the administration reinvent some of the ideas in the action plan, or the senior-level people reinvent some of the ideas around the use of AI, building technical competence in the government, third-party evaluations, all that kind of stuff. I think we're seeing them reinvent those things from first principles, so I'm optimistic still. But yeah, I definitely think that there have been substantial ways in which the administration has departed—ironically in both the direction of not taking the risks seriously enough and also in overcorrecting, not taking them too seriously but reacting in ways that don't actually deal with the risks. So in the end, I'm inclined to give grace to people and say I hope that we're in a high neuroplasticity phase of policymaking. I think we might be in a very different world in three months. But certainly, if you were to look at the headlines, you'd be like, 'Ah, it doesn't seem very consistent with the action plan at all.' And I can't deny that.

嘿，稍后我们将继续采访，先听一段赞助商信息。

Hey, we'll continue our interview in a moment after a word from our sponsors.

今天的节目由 Anthropic 提供，他们是 Claude 和 Claude Code 的创造者。在过去的几个月里，Claude 帮助我构建并完善了一个个人深度上下文数据库，现在包含了我过去整整 5 年的所有电子邮件、Slack 消息、推文、跨平台私信、视频通话和播客转录。在此基础上，我们还添加了描述我与数百个联系人、组织和想法关系的摘要文章。现在有了这个数据库，几乎没有什么 Claude 帮不上忙的。在报税季，我让 Claude 帮我整理。它浏览了我的收件箱，找到了我所有 10 份兼职工作的 1099 表格，并为我生成了一份关于开支和捐款的全面报告。对于我的天使投资，Claude 现在可以根据我与创始人的通话和电子邮件交流，以我的风险基金要求的格式起草投资备忘录。当有人需要帮忙时，Claude 通常能做得和我一样好。最近，一位朋友问我是否认识适合他正在招聘的职位的人选。起初我没想到任何人，但后来我想到问 Claude，果然，它找到了两个很好的候选人。Claude 是为不满足于“足够好”的头脑而生的 AI。

Today's episode is brought to you by Anthropic, makers of Claude and Claude Code. Over the last few months, Claude has helped me build and refine a personal deep context database that now contains all of my emails, Slack messages, tweets, DMs across platforms, video calls, and podcast transcripts going back a full 5 years. On top of that, we've now layered summary articles describing my relationship with hundreds of contacts, organizations, and ideas. And now that this exists, there's almost nothing that Claude can't help with. For tax season, I asked Claude to help me get organized. It went through my inbox, tracked down 1099s for all 10 of my part-time jobs, and built me a comprehensive report on my expenses and donations. For my angel investing, Claude can now draft investment memos in exactly the form that my venture fund requires based on the calls I've had and the emails I've exchanged with the founders. And when someone needs a favor, Claude can often do it as well as I can. Recently, a friend reached out to ask if I know anyone who might be a fit for a role that he is currently hiring for. Initially, nobody came to mind, but then I thought to ask Claude, and sure enough, it identified two great leads. Claude is the AI for minds that don't stop at good enough.

显然，其中一个最大的时刻——我不想重述所有政治细节，因为你已经评论过很多了——但最大的恐慌时刻之一，是战争部宣布 Anthropic 构成供应链风险。我注意到，我们似乎正在遗忘这件事。我们现在处于这样一种状态：据我所知，政府许多部门都在使用和测试 Mythos。我的理解是，政府内部仍然大量使用 Anthropic。如果我错了，请纠正我。我们该如何理解整个供应链事件的现状？我们是要假装它从未发生过，还是怎样？

So obviously one of the biggest moments, and I don't want to rehash all the politics of this because you've commented on it extensively, but one of the biggest sort of freakout moments was when the Department of War declared Anthropic to be a supply chain risk. And I'm struck by the fact that it seems like we're memory-holing that. We're in this zone now where, as far as I understand, many areas of the government were involved in using and testing Mythos. There's a lot of Anthropic in the government still, is what I kind of understand the situation to be. Correct me if I'm wrong on that. How should we understand where that whole supply chain thing is today? Are we just going to all pretend it never happened, or what?

从广义上讲，我会这样描述美国总统职位，尤其是自奥巴马第二任期以来——你知道，奥巴马曾有名言，他面对的是一个顽固的国会，说得客气点。他面对的是一个不会通过任何法律的国会。于是他说：“好吧，我有一支笔和一部电话。”他的意思是：“我要采取行政行动。我要把行政行动的边界推到极限。”这开启了一种自催化过程，每位总统都以各种方式推进行政权力的边界，而这些边界会在法庭上受到考验。所以你会经常看到这样的头条：“总统做了某件史无前例的行政权力之事，正在被诉讼”，然后经过漫长的诉讼过程，大多数人就忘了。Mythos 案仍在进行。上个月他们在 DC 巡回法院进行了庭审，我认为我们很快会等到判决。之后，如果 Anthropic 败诉，我相当肯定他们会上诉。有趣的是，特朗普政府实际上非常精明，知道什么时候该上诉，什么时候不该。他们很擅长揣摩言外之意：“好吧，那个可能确实违法；我们不会一路打到最高法院，因为我们会输。”他们在这方面确实很在行。所以我认为特朗普政府实际上认为他们会赢这个案子。我的意思是，诉讼仍在进行。如果到 2027 年夏天最高法院对此案做出某种裁决，我不会感到惊讶，即使裁决只是拒绝审理——这是最高法院最常见的行为。政府使用方面也在进行中。似乎 Mythos 之后，Anthropic 收到的信息是，供应链风险问题适用于他们与战争部本身的合同。所以我认为在战争部内部，他们确实在逐步减少使用 Anthropic，而且已经相当明显了。如果到今年年底或一年后他们完全停止使用 Anthropic，我不会感到惊讶。但在政府其他部门，信息是供应链风险问题不适用于其他任何政府机构。所以如果其他政府机构想用 Anthropic，没问题。另外，从技术上讲，国家安全局是战争部的一部分，但似乎 NSA 不仅与 Anthropic 有合同，而且如果报道可信的话，Anthropic 关于国内大规模监控和自主致命武器的红线得到了 NSA 的尊重。对我来说，这是件好事。我认为美国人不擅长容忍模糊性，但一点点模糊性——甚至相当程度的模糊性——是整个过程中固有的一部分。所以是的，事情仍在进行。供应链风险问题仍在诉讼中。我认为 Anthropic 的合同确实在战争部被取消，但同时，Anthropic 在政府其他部门的使用进展顺利。

The broad way I would describe the American presidency, really since Obama's second term, is—you know, Obama famously said he was faced with an intransigent Congress, to put it generously. He was faced with a Congress that wouldn't pass any laws. And so he said, 'Fine, I have a pen and a phone.' What he meant was, 'I'm going to do executive actions. I'm going to push the limits of executive actions.' That began a kind of autocatalytic process in which every president pushes the bounds of executive authority in various ways that get tested in the courts. So what will happen very frequently is that you'll see a headline where it's like, 'President did this thing that is unprecedented with executive power and it's being litigated,' and then it goes through a very long litigation process and most people lose track of it. The Mythos case is still going on. They had their trial in front of the DC Circuit last month, and I think we're expecting a ruling in that trial at some point soon. Then after that, if Anthropic loses, they will appeal, I'm quite certain. And it's interesting: the Trump administration is actually very, very savvy about when to appeal things and when not to. They're pretty good at reading between the lines: 'Okay, yeah, that one was probably illegal; we're not going to appeal that all the way up to the Supreme Court because we'll lose there.' They're actually pretty good at that. So I think the Trump administration actually thinks they will win this case. The litigation is ongoing, is my point. It would not surprise me if by the summer of 2027 we have a Supreme Court ruling of some sort on this issue, even if that ruling is them denying to hear the case, which is the most common thing the Supreme Court does. That's also going on in terms of government use. It seems as though after Mythos, the message that Anthropic received was that the supply chain risk thing applied to their contracts with the Department of War proper. So I think within the Department of War, they really are winding down Anthropic and have been considerably. It wouldn't surprise me if they're 100% off of Anthropic by the end of the year or a year from now. But throughout the rest of the government, the message was that the supply chain risk thing doesn't apply to any other government agencies. So if other government agencies want to use Anthropic, that's fine. Also, technically speaking, the National Security Agency is a part of the Department of War, but it seems as though not only does the NSA have a contract with Anthropic, but if reporting is to be believed, Anthropic's red lines around domestic mass surveillance and autonomous lethal weapons were honored by the NSA. That, to me, is a good thing. I think Americans are not that good at tolerating ambiguity, but a little bit of ambiguity—or maybe even a healthy amount of it—is an intrinsic part of this whole process. So yeah, it's still happening. The supply chain risk thing is still being litigated. I think Anthropic contracts are indeed being cancelled at the Department of War, and at the same time, other use of Anthropic in the government is going fine.

美国政府包罗万象。

The US government contains multitudes.

是的。

Yes.

你对整个情况也持非常批评的态度。我认为你对最近的行政令和举措批评较少，但仍有些批评——据我理解，该举措是将某些 AI 测试和定性责任从 Casey 手中拿走，我认为 Casey 的未来现在有些不确定。我不知道你认为它的未来会怎样，然后将这些责任移交给 NSA，在那里这些信息可能已经或可能尚未被列为机密。这是怎么回事？这简单到只是因为这是拜登的项目所以我们不喜欢它，还是背后有更多隐情？你为什么担心测试那些你不知道是否存在的模型，依据的是无法公开的标准？你认为这如何演变成对公众的问题？

You were also extremely critical of that whole situation. And I think you were less critical, but still somewhat critical of the recent EO and the move, as I understand it, to take certain AI testing characterization responsibilities away from Casey, which I think now has kind of an uncertain future. I don't know what you think its future will be, and move those responsibilities to the NSA, where they may or may not already be classified information. What's going on there? Is this as simple as it was like a Biden project and so we don't like it, or is there more going on there than meets the eye? And why are you concerned about testing models you don't know exist against standards that can't be disclosed? How do you think that turns into a problem for the public?

是的。所以，我批评政府在此事上的方向。我在网络安全行政令签署时就批评过它，因为我预料到了这一点。让我为你的听众简单说明一下。网络安全行政令做了什么？一部分是：我们将创建各种程序来修补关键软件系统中的漏洞。很好。没问题。我赞成。我认为没人会反对这一点。我们可以争论这些程序会有多有用、实施会有多好，但走着瞧。所以这是一件事。第二件事是它创建了一个自愿的部署前测试计划，在发布前 30 天进行测试，其细节将被保密，主要由情报界负责。实际上，NSA 可能是主要执行者，因为 NSA 在网络安全专业知识方面是最高的。顺便说一句，他们确实非常出色。我认为我担心的原因是，这正在为未来设置一个潜在非常糟糕的局面，即前沿模型的访问受到限制。

Yeah. So, the reason I'm critical of the administration here in terms of where they're going. I was critical of the cyber executive order when it was signed because I anticipated exactly this. Let me just level set for your listeners for a moment. What does the cyber executive order do? One part of it is: we're going to create a variety of procedures by which we're going to patch vulnerabilities in critical software systems. Great. Okay. Thumbs up. I don't think anyone can object to that. I think we can argue about how useful those programs are going to be and how good the implementation will be, but we'll see. So that's one thing. The second thing is it created a voluntary pre-deployment program, a testing program 30 days before release whose details were to be classified, primarily classified and primarily run by the intelligence community. And within that, probably practically speaking, the NSA is the primary, since the NSA is the highest in terms of cyber expertise. And they really are quite excellent, by the way. I think the reason I'm concerned is that this is setting up a potentially very bad future where access to frontier models is gated.

这一切都保密，公众根本不知道前沿发生了什么。政府正在做出一系列决策，公众甚至不知道是否要限制某些能力，也不知道该怎么处理这些能力。我觉得，如果你相信现在正在发生的事情是技术史上最重要的事件之一，那么不仅作为美国人，我的直觉强烈认为公众有权知道发生了什么，而且第二，我实际上认为这不是一个权衡取舍的问题。公众知情会带来一个更好的世界。在尽可能的范围内，公众能够接触前沿能力也会带来一个更好的世界。首先，我认为政府对前沿 AI 的垄断可能从公民自由的角度导致非常可怕的后果。顺便说一句，无论总统是谁，我都会提出这样的批评。我不知道总统是谁。如果你不知道党派，不知道名字。如果你从我大脑中抹去这些信息，但我保留其他所有信息，他告诉我这件事，我会说我很担忧。这是一点。另一点是，处理社会和文明是一种信息处理系统，对吧？就像全国所有人类都是并行算力，我们都在努力理解这里发生的事情。虽然还有很多问题我们还没有好的答案，但我确实认为 AI 政策界自 2023 年以来在如何实际处理神话级能力模型方面取得了相当不错的进展。当事情公开，并且有立法过程，例如，有大量强有力的公众意见输入时，这些东西真的——这是我们整个系统的设计——这些东西可以融入进来，我们可以利用这种并行算力。当你把所有东西集中起来并保密时，它变得更加脆弱，就像一群人，正如我之前所说，他们往往没有——他们手头有无数事情，因为他们是高级政府官员，他们通常对 AI 没有太多背景，只是即兴发挥，以即兴的方式做决策。我认为这会导致次优的决策。我认为这会导致浪费时间，因为我觉得我们现在看到的是政府正在快速推进那种心态。他们让我非常想起 2023 年春天华盛顿的状态，当时 ChatGPT 刚出来，他们说“哦，我们要严格监管这个，这真的很危险”，然后事情缓和了，也许缓和得有点过头了，但最终我们找到了中间点，2024 年开始氛围发生了变化。我只是觉得现在，通过把所有事情都放在政府的回音室里，只有少数几个声音贡献信息和见解，我们并没有利用我们最好的资源。所以我认为这是最大的问题，这就是为什么我持批评态度，并且非常担心政策方向。但与供应链风险问题不同，供应链风险问题我认为完全是自摆乌龙，完全是非受迫性失误——你为什么选择打那场仗？你不需要打那场仗。你可以用一百万种不同的方式解决这个问题，即使你认真对待政府在那件事上的担忧，你也可以用一千种不同的方式处理。这更像是——是的，我并不惊讶事情会这样发展，因为你正在构建这个东西。你正在从零开始即兴构建一个 AI 治理体系，而它是由 20 个人构建的，其中 15 个人对 AI 没有太多背景。我并不惊讶它这样运作，我只是指出元问题：是的，我们需要把这件事公开。我们需要让国会参与进来。我们不能只是——这行不通。所以没有必要高度对抗和批评，因为我不是——你想让他们做什么？你想让这些人做什么？我不责怪他们，但我最终认为我们需要让事情更加公开。

It's all kept secret and the public doesn't really know what's happening at the frontier. The government is making a bunch of decisions that maybe the public doesn't even know whether or not to restrict certain capabilities about what to do with those capabilities. And it feels to me like if you believe that what's happening right now is one of the most important things ever to happen in the history of technology, not only do I think just intrinsically as an American my gut instinct is the public has a right to know about what's going on, but number two I actually just think that it's not some trade-off. It's a better world where the public knows. It's a better world where the public knows and to the extent possible, it's a better world where the public can access frontier capabilities. First of all, I think government monopolization of frontier AI is potentially how we get very scary outcomes from a civil liberties perspective. And that's not, by the way, that is a criticism I would make regardless of who the president was. I didn't know who the president was. If you didn't know the party, didn't know the name. If you erased that information from my brain but I had everything else in my brain and he told me about this, I would say I'm concerned about that. That's one thing. Another thing is that dealing with a society and a civilization is a kind of information processing system, right? It's like there's a bunch of all the humans in the country are parallel compute, and we're all trying to process what's going on here. And while there's a lot of things we don't have good answers to yet, I actually do think that the community of AI policy people has made reasonably good progress since 2023 in terms of how practically we should be dealing with models of the mythos level capability. And when things are public and there's a legislative process for example that is informed by lots and lots of robust public input, that stuff really that's the whole design of our system right that stuff can make its way in and we can take advantage of this kind of parallel compute. When you centralize everything and make it private, it's much more brittle and it's like a bunch of people who as I said earlier often don't have a they have a million things on their plate because they're high level government officials they often don't have a lot of context for AI and they're just improvising, making decisions in an improvised fashion. And I think that leads to subpar decision-making. I think it leads to wasting time because you it's I feel like what we're watching right now is the administration speedrunning the sort of mentality there. They remind me very much of where DC was in the spring of 2023 when it was like ChatGPT had just come out and oh we're going to regulate the hell out of this and this is really dangerous and then things softened and maybe they softened a little too much and I just but they still they ultimately we met in the middle there was a change in the vibes that started in 2024 and I just think that right now we are going to like by putting this all in the echo chamber member of the administration with a pretty small number of voices contributing to things information and insight to things. I just feel like we're not leveraging the best that we have and so that I think is the biggest problem and that's why I am critical and I am very worried about the direction of policy. But unlike the supply chain risk thing where the supply chain risk thing was like I think just totally an own goal just totally unforced error like why did you pick that fight? You didn't need to pick that fight. You could have fixed this in a million different ways, but that that did you could have dealt even if you take the government's concerns in that issue seriously. You could have dealt with that in a thousand different ways. This is more Yeah, I'm not surprised. I'm not surprised things are going about this way because you are building this thing. You're improvising an AI governance regime from scratch and it's being built by 20 people, 15 of whom don't have a ton of context for AI. I'm not surprised it's working this way and I'm just pointing out the meta problem of yeah, we need to bring this out into the public. We need to have Congress involved. We can't just this is not going to work. And so there's no point in being highly adversarial and critical there just because I'm not What do you want them to do? What do you want these people to do? I don't blame them, but I do ultimately think that we need to make things more public.

说到并行处理，各州作为我们的民主实验室，作用如何？据我所知，你支持甚至倡导的许多提案，包括强制发布安全计划、某些其他透明度措施、举报人保护，甚至某种类似 Fathom 式的公私混合监管结构，都在不同州以相当显著的程度在很短的时间内实现了。你对各州有多看好？

On the note of parallel processing, how about the role of the states, our laboratories of democracy? A lot of the proposals that I understand that you have favored or even championed, including mandatory safety plan publication, certain other transparency measures, whistleblower protections, and even a sort of Fathom style public private regulatory hybrid structure have all happened in different states to a remarkable degree in a pretty short period of time. How bullish are you on the states?

所以，对于私人治理的总体概念，确实取得了非常有意义的胜利。我是在 2024 年底 SB 1047 被否决后开始真正研究这个的，当然不止我一个人，很多其他人也在这方面做了工作——审计独立验证组织，也就是第三方私人机构，它们会评估很多当前的事情。比如，政府担心神话或寓言模型的潜在越狱，如果有专家机构深入研究并认证说，“嘿，是的，存在越狱，因为越狱总是存在，但我们计算的风险是这些越狱不够严重，达不到那个水平，我们可以认证 Anthropic 符合安全最佳实践”等等，那就太好了。我认为在这方面取得了实质性胜利。实质性胜利体现在：三大 AI 公司中有两家（Anthropic 和 OpenAI）已经发布了多份支持这一总体概念的文件；今年早些时候，伊利诺伊州通过了强制审计前沿 AI 公司的法案；康涅狄格州和弗吉尼亚州今年早些时候也通过了法案，专门授权对独立验证组织进行研究或试点项目；顺便说一句，俄亥俄州还有一个待决法案，这将是迄今为止最有力的独立验证组织实施。这方面的势头比我一年前预想的要大。从这个意义上说，我认为各州作为民主实验室的想法运作良好。

So there have been really meaningful wins for the whole general notion of private governance that I started to work on really post SB 1047 veto in late 2024 and of course other it's not just me a lot of other people have worked on this stuff auditing independent verification organizations be like third party private bodies that would evaluate a lot of the thing right now like with the government they're concerned about the jailbreak this potential jailbreak of mythos or fable and it would be great if there for expert bodies who had looked into this and really probed and certified like hey yeah like there are jailbreaks because there are always jailbreaks but our calculated risk is that these jailbreaks are not severe enough to rise to the level and we can certify Anthropic as conforming to safety best practices or whatever right it's like the kind of thing that I think there have been substantial wins there have been substantial wins in the sense that two of the three big AI companies have published multiple documents that are favorable to this general notion Anthropic and OpenAI bill to mandate auditing in frontier AI companies passed in Illinois earlier this year and also the state of Connecticut and the commonwealth of Virginia both passed earlier this year that are specifically authorizing either studies or pilot programs for independent verification organizations and there's also a bill pending in Ohio by the way which would be the most robust implementation of independent verification organizations yet. Momentum in that regard, more than I would have guessed a year ago. In that sense, I think the states as laboratories of democracy idea is working fine.

同样值得注意的是，在已经通过的关于前沿 AI 安全的法律方面，各州付出了巨大努力。加州有一项透明度法案，即 SB53。纽约也提出了类似法案。伊利诺伊州则是 SB315。伊利诺伊州的版本增加了审计要求，但这三个州的透明度措辞惊人地相似。所以我对此感到非常高兴。这并没有造成碎片化。这些州正在趋同于一个共同框架。我们会看看它是否有效，效果如何，但这是各州在共同框架上趋同，这种情况时有发生。还有很多其他 AI 领域在 Twitter 上不太受关注，没有获得那么多关注度，但我认为在这些领域，各州的情况并不那么乐观。比如消费者保护、算法定价，现在美国存在的合成媒体和深度伪造法律数量简直疯狂。现在有数百项。其净效果可能是造成相当混乱的政治环境。我认为还有一个真正的问题正在从各州浮现出来，那就是职业许可保护。各州，包括伊利诺伊州，已经这样做了：我们将心理健康服务定义为只能由人类提供。如果一个聊天机器人只是问你“你还好吗？”，或者你对聊天机器人说“我很难过，你能帮帮我吗？”，聊天机器人就在技术上从事了心理健康服务，这是非法的。各州在执行这类法律时的严格程度往往不同。但把这种东西写进法律是不好的。所以奇怪的是，最受关注的领域，即前沿 AI 安全，也是很多支持联邦优先立法的人集中精力的地方，实际上那里的法律制定得最好。它们通常得到 AI 行业的支持，通常专门设计来避免碎片化投诉，这是一个合理的投诉。而且不知为何，至少到目前为止，大多数强烈支持联邦优先的人都关注这些法律。不仅如此，这些法律处理的是真正紧迫的问题，比如网络和生物安全，这些显然不再是虚假的。我们不能再争论这个了。这显然是真实的事情。所以他们不关注所有其他领域，在这些领域各州实际上正在制造碎片化，并制造复杂的合规问题，可能对初创公司来说特别棘手。我认为各州的问题在这方面是混合的，但支持联邦法律的人，包括我，也没有帮到自己，因为他们基本上以错误的方式讨论这个问题。

It's also worth noting with respect to the frontier AI safety laws that have passed that the states have taken great effort. There's been a bill in the transparency bill in California was SB53. In New York, it was raised. In Illinois, it was SB 315. And the language Illinois adds an auditing requirement, but the transparency language across those three states is remarkably similar. So I'm quite happy about that. That's not creating a patchwork. Those are the states converging on a common framework. We'll see if it works. We'll see how well it works, but it's states converging on a common framework, which they do from time to time. There's a lot of other areas of AI that are not so much paid attention to on Twitter, that don't get as much mind share, but where I think the story for the states is less rosy. Things like consumer protection, algorithmic pricing, the number of synthetic media and deepfake laws that exist in this country now is just crazy. There's hundreds of them now. And the net effect of that is probably to create a fairly confusing political environment. I think also one thing that's really problematic that we're starting to see bubble up from the states are basically occupational licensing protections. States say, including Illinois has done this, we are going to define mental health services as exclusively something that can be provided by humans. If a chatbot so much as asks you how you're doing, it is engaging in, or if you say to the chatbot, 'I'm sad. Can you please help me?' the chatbot is technically engaging in mental health services and that's illegal. And we'll see states often vary in terms of how rigorously they enforce laws like this. But it's not good to have that kind of stuff on the books. So I think strangely enough, the area that gets the most attention, the frontier AI safety stuff, where a lot of the preemption crowd also focuses their energy, that's actually the area where the laws are best sculpted. They often have the support of the AI industry. They're often designed specifically to avoid the patchwork complaint, which is a legitimate one. And for some reason, at least so far, most of the people that are super pro-preemption focus on these laws. And it's not only that but these laws are dealing with really urgent problems like cyber and bio that are clearly not fake anymore. We can't have that argument anymore. Clearly a real thing. So then they're not focusing on all these other areas where the states actually are creating patchworks and creating complex compliance things that might be really complicated for startups to deal with. I think the state issue is mixed in that way, but also the people who support a federal law, including me, are not helping themselves because they're talking about this issue largely in the wrong way.

我猜有一个大意外。第一个问题，宏观层面：对你来说最大的意外是什么？我先说说我最大的意外，你可以回应并分享你的。我最大的意外是政府放松了芯片出口管制，这本身并不令人震惊，但真正的震惊是中国不想买了。所以我们之前一直在争论，我们应该在多大程度上采取强硬态度，或者至少我在问这个问题，试图实施这些出口管制。最后管制放松了，中国却说：“啊，不用了，谢谢，我们打算自己建立产业，你们留着芯片吧。”这是怎么回事？还有没有其他意外达到这个级别？

One big surprise I guess. First question high level: what have been the biggest surprises for you? I'll offer my biggest surprise and you can react to that and share your own. My biggest surprise is the administration eased the export controls on chips, which wasn't shocking unto itself, but then the real shock is China doesn't want to buy them. So we had all this debate around to what degree should we be bellicose, or at least I was asking that question in trying to do these export controls. Finally they get eased and China's like, 'Ah, no thanks, we're going to just build our own industry and you guys can keep the chips.' What's going on there? And any other surprises rise to that level for you?

所以那个具体的发展并不让我太惊讶，因为中国的体系非常……实际上，现实地说，我们在这方面正变得越来越像中国。但关于中国有一点：当中国宣布一项新政策时，你需要相当的专业知识，美国政府内部有专门研究这个的人，但他们不公开分享意见。在公共话语中很难得到关于这类事情的良好分析。这是我想念政府的一点。但政策是一回事，他们实际会做什么是另一回事，这两者有很大不同。所以这项政策，我认为对中国来说，建立自己的 AI 芯片生态系统、不再需要美国人是民族自豪感的问题，我认为他们喜欢向世界传递这个信息。他们喜欢向自己的人民传递这个信息，这有一定道理。然后实际发生的情况是，与此同时，中国的体系也有游说，我向你保证，DeepSeek、阿里巴巴、智谱等所有这些公司，我向你保证，他们正在乞求北京允许他们获得美国芯片。所以北京的政策制定者正在考虑这一点。可能有一些芯片正在被出售。我认为我们现在知道有一些芯片被出售了。但他们会限制它，这可能是他们的一个重要目标。好吧，我们拭目以待。但在美中关系方面，没有什么特别让我惊讶的。我想我会说，我原本预计到现在，就中国而言，我预计到现在中国政府会意识到灾难性风险问题，并开始抵制开源策略。我预测过。我说过到 2025 年第一季度，我预测到 2026 年第一季度末，DeepSeek 的顶级模型将不再开源，这个预测错了。我仍然认为这会在某个时候发生，但我们还没到那一步。中国政府似乎更关心劳动力问题，而不是灾难性风险。所以他们对灾难性风险的关注度比我一年前猜测的要低。在技术方面，没有什么真正让我惊讶的。我的一个爱好是关注那些使用编码智能体的非常普通人的奇怪子群体。有一个在家教育孩子的妈妈社区，她们喜欢 Claude Code 和 OpenClaw 之类的东西，并用它来做各种事情。我喜欢这个。我原本真的没想到编码智能体会变得这么流行。

So that particular development doesn't surprise me that much because China's system is very... it's actually realistically we are becoming more like China in this regard. But one thing about China is that when China announces a new policy, it's like you need considerable expertise and there are people inside the US government who specialize in this and they do not share their opinions publicly. But it's actually very hard to get good analysis on things like this in the public discourse. It's one of the things I miss about government. But it's like what the policy says and then there's what they're actually going to do, which are importantly different things. So the policy, I think it's a matter of national pride for China that we are building our own AI chip ecosystem and we don't need the Americans anymore, and I think they like sending that message to the world. I think they like sending that message to their own people and there's some aspect of that. Then there is what actually happens because while that's going on, China's system has lobbying too, and I guarantee you that DeepSeek and Alibaba and Zhipu and all these other people, I guarantee you that they're begging Beijing for access to American chips. And so the policy planners in Beijing are factoring that in. Probably there's some amount that's being sold. I think we now know that there are some chips being sold. But yeah, no, they're going to restrict it and that might be a big goal on their part. Well, I guess we'll see. But in terms of US-China, nothing has especially surprised me. I guess I would say I anticipated that by now, in terms of China, I anticipated that by now the Chinese state would have woken up to the catastrophic risk issues and that they would have started pushing back on the open-source strategy. I called it. I said that by Q1 of this year in 2025, I predicted that by the end of Q1 of 2026, DeepSeek's top model would not be open source, and that prediction was wrong. I still think it's going to happen at some point, but we're not there yet. The Chinese state seems more concerned about labor issues than they seem concerned about catastrophic risk. So they're less cat risk pilled than I would have guessed if you had asked me a year ago. On the technical side, nothing has really surprised me. One of my hobbies is I pay attention to these weird subgroups of very normal people who use coding agents. So there's this community of homeschooling moms who love Claude Code and OpenClaw and stuff, and they're using it to do all kinds of things. I love that. I wouldn't really have guessed the coding agents becoming so popular.

然后技术上唯一另一件让我非常惊讶的事，是我没预料到物理世界中的世界模拟（world sim）——那种可以模拟 3D 交互环境的模型，基本上就像创建第一人称开放世界电子游戏，但针对现实世界的任意场景。我没想到这一点，因为这类模型在很长一段时间里都非常梦幻：你可以创建世界，但神经网络是实时生成的，所以如果你转身看某个东西，然后转开再回头看，它完全变了，没有持久性。然后有一天它突然就奏效了，就像‘哇，我们现在有了持久性，而且它稳健地工作’。这大大缩短了我对机器人技术的时间线，因为很明显你可以构建合成数据流水线。你可以用人类数据作为基线，让人们戴上 Apple Vision Pro 作为基线，然后从那里引导到各种世界模拟环境中的合成数据。可能再撒上一点高保真数据，比如人们戴着带有电极的手套来感知肌肉运动等。很明显，灵巧操作——我去年夏天看到 Google DeepMind 的世界模拟模型时，立刻就意识到，机器人的灵巧操作将在八个月内被解决。这让我在离开政府后稍微改变了我的研究议程，加速了我为机器人技术考虑的一些工作。

And then the only other thing on the technical side that really surprised me, I did not expect the world sim stuff in the physical world, the sort of models where you can simulate a 3D interactive environment, basically like creating a first-person open world video game, but for arbitrary settings in the real world. I did not anticipate that because the way those models worked for a really long time was they were very dreamlike. You could create the world but it was like the neural network is creating it in real time, so if you turn around and look at something and then turn away and then go back and look at that thing again, it's totally different. It doesn't have permanence. And then one day it just worked. It's just, oh wow, we just have permanence now and it just works robustly. That substantially increased my timelines for robotics working because it's very clear that you'll be able to make synthetic data pipelines. You'll be able to use human data as baseline. You make people wear Apple Vision Pro, use that as the baseline. You can bootstrap there to synthetic data in all sorts of world sim settings. And then probably just sprinkle on a little bit of data with really high fidelity, like people wearing gloves with electrodes in them to sense muscle movements and whatnot. And it's very clear that okay, dexterous manipulation — the second I saw the world sim, it was a Google DeepMind model last summer — the second I saw that I was like, okay, dexterous manipulation in robots is going to be solved in eight months. That caused me to change my research agenda a little bit after I left government and accelerate some of the work I was thinking about for robotics.

是啊，一切都在发生。

Yeah, it's all happening.

嗯。

Yeah.

Nvidia 的 Jim Fan 最近做了一个 20 分钟的主题演讲，关于他预期机器人技术将走的道路与 LLM 所走道路的相似之处，我认为——

Jim Fan from Nvidia's little 20-minute keynote recently about the parallels between the path that he expects robotics to take and the path that LLMs have taken, I think is —

哦，是 Sequoia 的那个演讲。

Oh, the Sequoia talk.

那绝对是必看的内容，也彻底说服了我。

It's must-see TV for sure, and it definitely has me convinced as well.

嗯。

Yeah.

另外，还要向在家教育妈妈群体中的 Jesse Jana 致敬。我很喜欢她做的事情，也尽量从中借鉴。昨晚墨西哥对韩国的世界杯比赛，我们为每个国家打印了小资料包，你知道，每个国家的一些小知识。我们还准备了每个国家的零食。这些工具能如此提升平凡的日常生活，真是令人惊叹。即使事情变得紧张，甚至在某种程度上可以说是令人担忧，这一点也不应被遗忘。

Also, shout out to Jesse Jana from the homeschooling mom's contingent. I love the stuff she's doing and try to borrow from it as much as I can as well. Last night for the Mexico-Korea World Cup game, we printed out little packets for each country, you know, little about each one. We had snacks from each one. It's amazing how much you can enhance your just mundane daily life with these tools. And that should not be forgotten even as things get intense and, you know, in some ways, let's say fraught.

这方面有个很好的例子，关于体育。我记得 Opus 4.5 发布的时候，正是模型变得非常好的时候。那是十二月，十二月没人看篮球，但我会看。我有 League Pass，NBA League Pass，可以看所有比赛。我的问题是我并不真正支持某支球队，我只是想看一场好比赛。所以有时全国有八场比赛同时进行。我用 Claude Code 建了一个小仪表盘，摄取所有比赛的实时数据，然后做了一个类似 Nate Silver 的速度表，显示比赛成为好比赛的概率。我用模型创建了一些启发式规则。总之，这是一个有趣的小项目。但它确实极大地改善了我的生活。

A really good example of this, just as a sports thing. I remember when Opus 4.5 came out, it was right when the models were getting really good. And it was December, which is no one watches basketball in December, but I do. I have League Pass, NBA League Pass, which is the way you watch all the games. The problem I had was that I don't really root for any particular team; I just want to watch a good game. So sometimes there are eight games on at the same time across the country. I built this little dashboard with Claude Code that ingested live data from all the games and did a Nate Silver-like speedometer thing with odds of being a good game basically. I had created some heuristics for that with the model. Anyway, it was a fun little project. But yeah, it massively improved my life.

嗯，这差不多把我们带到了当下，当下的焦点是 Fable 和 Fable 禁令。在我们讨论政治和政策之前，在你短暂使用 Fable 的时间里，你的印象如何？你有多想念它？

Well, that pretty much brings us to present and the present moment is Fable and the Fable ban. In the brief time that you had Fable before we get into the politics and policy of it, what were your impressions and how much are you missing it?

我的印象是，它是一个极其智能的模型，在智力上是一个真正的飞跃。很多人拿它和 03 比较，03 是第一个让人觉得是真正的天才的模型。很有趣，因为当 03 刚出来时，我有一种感觉：‘这个东西，享乐适应会停止，它永远会让我觉得如此聪明。’然后它确实如此。现在我相信如果我再用 03，我会觉得它相当笨。当然它仍有魅力，但 Fable 对我来说是另一个这样的时刻。不幸的是，那几天我真的很忙，没有时间在编码智能体环境中真正使用它。我在 Claude Code 中打开它几个小时，打算做一个项目。我当时在旅行，打算在酒店房间里做项目，但被什么事情分心了，等我回到笔记本电脑前，它已经下线了。但我确实用它做了一些知识工作。具体来说，我是目前正在 FEC 审理的一个案件的当事人。这是一个向 FEC 提出的投诉，要求撤销——这是一件无聊的事，关于 1000 号程序性法规。有人对我的证词写了反驳，对方聘请了一位专家来反驳我。我让 Fable 在 Claude Code 中阅读并进行研究。我没有使用它的写作输出，但天哪，这个模型彻底击败了它。我为写反驳的那个家伙感到难过。我想，哇，这个模型以一种我大部分情况下无法做到的方式击败了他。我觉得它非常聪明。真希望我能多用它一些。但是的，我很想念它。这很奇怪——这就像我们第一次倒退，这是一种奇怪的感觉，但也欢迎来到政府介入事物的世界，对吧？这是政治经济学的一个很好的小教训。这就是它的感觉。通常它太抽象或分散。我会说，哦，政府让事情变得更脆弱，让世界在各种方面变得更笨。这就是国家干预的问题。这是一个很好的例子，如果你曾是 Fable 的用户，你的世界在过去一周确实变得更笨了。

My impression was that it was a fiercely intelligent model and it was a real step up in intellect. A lot of people have made 03 comparisons, where 03 was the first that felt like this really cracked genius. It's very funny because when 03 first came out, I had this feeling of 'this thing, the hedonic treadmill is going to stop at some point, this thing is always going to feel so smart to me.' And then it actually does. Now I'm sure if I used 03 I would find it rather dumb. Still having its charms of course, but Fable was another moment like that for me. Unfortunately, I was really busy those days and I didn't have any time to use it really in coding agent settings. I had it open in Claude Code a few hours before and I was going to do a project. I was traveling. I was going to do a project from my hotel room and I got distracted by something and by the time I came back to my laptop it was off. But I did use it for some knowledge work stuff. In particular, I'm a party to a case that's going on before the FEC right now. It's a complaint before the FEC to remove — it's a boring thing, a procedural regulation of order 1000. Someone wrote a rebuttal to my testimony, the other party in this legal hearing hired an expert to write a rebuttal to me. I had Fable read it in Claude Code and do research and stuff. I did not use its writing output, but oh my god, this model demolished this. I felt so bad for this dude who wrote the rebuttal. I was like, wow, the model demolished this dude in a way that I mostly couldn't have. And I found it to be fantastically intelligent. Wish I had been able to use it more. But yeah, I am missing it. It is weird to like — it's like the first time we've gone backwards, and it's a weird feeling, but also welcome to the government being involved in things, right? It's a good little lesson in political economy. That's what it feels like. Usually it's too abstract or diffused. I'm like, oh, the government makes things more brittle and makes the world a little bit dumber in various ways. This is like the problem with state intervention. This is a good example of it being literally the case that if you were a user of Fable, your world became dumber in the last week.

是啊，对我来说这很艰难。我以前经历过一次，当时我做了 GPT-4 的红队测试，然后从 GPT-4 降级到当时的 text-davinci-002，感觉就像在拿到真正的东西之前，我根本不想碰这些东西。现在我又有了这种感觉，程度稍轻，但品味因素确实是我感受到的地方。我的意思是，它显然在编码方面很出色。Opus 4.5 在编码方面已经比我强了。

Yeah, it's been rough for me personally. I experienced this once before when I did the GPT-4 red team and then went from GPT-4 down to whatever it was, text-davinci-002 at the time, and it was just like I don't even want to touch any of this stuff until I get the real thing back. And I feel that way again to a lesser degree now, but definitely the taste factor is really where I felt it. I mean, you know, it's obviously amazing at coding. Opus 4.5 is, you know, superhuman relative to me in coding already.

那么，你觉得我们什么时候能拿回模型？到底发生了什么？从外部来看，我认为普遍的共识是，时间越长——现在已经一周了，这挺久的——他们却拿不出一个合理的解释来说明到底看到了什么让他们害怕的东西，这就有点像 OpenAI 解雇 Sam Altman 那件事了：你得给个说法啊，否则就显得这事不太站得住脚。你基本也是这个看法吗？还是说你对他们的处境更同情一些？你觉得这事会怎么解决？

So, when do you think we get it back and what's going on? I mean, from the outside view, I think the consensus take is like the longer it goes where we don't really have a good explanation for, and at this point it's been a week, which is a long time, where we don't have a good explanation for what they saw that scared them, it starts to feel a little bit reminiscent of the OpenAI firing Sam Altman episode where it's like, you got to have an explanation here, guys, or it becomes clear that this is not super well justified. Is that basically the view that you have or do you have a more empathetic view for where they're at and how do you think this gets resolved?

我认为当前的情况有三个因素在影响政府的反应。第一是真正对安全与安保的担忧。第二是对前沿模型缺乏足够的背景信息，也就是做出良好风险评估所需的信息，这在一定程度上加剧了安全担忧。所以可能有一些合理的安全担忧，也有一些不那么合理的，但都源于这种普遍的背景缺失。第三，我们不能否认这里面有政治因素，对吧？即使我不认为政府内部有专门针对 Anthropic 的阴谋论——也许有，但至少在某些人那里——但更可能只是 Anthropic 的整体政治地位以及政府与之的摩擦。这会影响重要人物对安全漏洞消息的反应，如果是一家政府更喜欢的公司，反应可能完全不同。这三个因素相互交织，我认为它们共同解释了当前的情况。我不知道的是——即使我还在白宫内部，可能也不会完全清楚——这三个因素到底各占多大比例。有一点我想澄清：我对这件事的理解是，美国政府并不是在宣布一项政策，说从此以后只要你的模型有安全漏洞，我们就会对非美国人实施出口管制。我不认为那是政策。我认为实际情况是，他们决定必须把模型撤出市场，而这是他们能想到的唯一能确保做到这一点的工具。所以他们只是拿起了他们认为能完成任务的工具。我不认为他们把这当作一项普遍政策来宣布。但你还得考虑一点：政府的说法变了。有趣的是，这和供应链风险事件类似。最初的说法是，美国有安全担忧，想和 Dario 通电话讨论，但没能及时联系上。我记得同样的事，副部长 Emil Michael 是那件事的主角，他公开抱怨 Dario 不立即回电话，过了几个小时才回。拜托。这里面有个人恩怨的成分，对吧？有谁更厉害的问题。当我在政府里，我是政府先生，我打电话，他就得回电。这种事在 DC 很常见。然后说法变成了安全风险，这个越狱是真的，我们需要知道。然后 36 到 48 小时后，甚至可能 72 小时后，我们开始听到实际上不是这样，我们这么做是因为 Anthropic 把模型提供给了与中国有关联的公司。这就像是在抓救命稻草，因为如果你说因为 Anthropic 把模型给了与中国有关联的公司而反对出口管制，这是你的主要担忧，那你为什么周五做这件事的时候不说呢？而且他们描述这件事好像发生在一个月前，因为 Anthropic 扩大了模型访问的公司范围，包括一些国际公司。Anthropic 几周前就公开宣布了，说我们正在扩大 Mythos 的访问权限，包括一些美国公司以及美国的盟友和伙伴。这家公司实际上是韩国电信 SK Telecom，它属于 SK 集团，是韩国最大的财阀之一，也拥有 SK 海力士，这是高带宽内存的主要生产商，而韩国在半导体制造生态系统中是美国非常重要的合作伙伴。所以想要加强他们的电信基础设施，在我看来非常合理。然后他们抛出这个说法，感觉像是政府——我再次不知道三个因素的比例——但基本上是恐慌了，拿起了第一个他们认为能把模型撤出市场的工具，然后事后找理由。这很难——我之所以觉得这个问题令人沮丧，部分原因就是很难分析，因为这里没有多少政策实质。纯粹是本能反应。

I think what's happening here, I think there are three factors playing into the government's reaction here. One is genuine concern about safety and security. The second is fairly broad lack of context for frontier AI and the sort of things that the information that you need to make a good risk calculation, which is driving the security concern to some extent. So there might be some legit security concern and there might also be some not so legit security concern, but it's being driven by this kind of general lack of context. And then third, we can't deny that there's some political dimension to this, right? Even if it's not—I wouldn't even say that there's a conspiracy theory inside the government to do this to Anthropic. There might be, but at least among some, but it might be more just the general political status of Anthropic and the fights that the administration has been having. It colors the reaction of important people to the news of a security vulnerability in a way that might not have happened if this were a company that the administration felt more warmly toward. So all three of these things definitely feed into one another and I think they probably all explain what's going on. They're all ingredients in explaining this situation. And the thing that I don't know—and probably even if I were still inside the White House, I wouldn't fully know—is really just in what ratio those three things come together. One thing I think it's worth being clear about is that my read of this situation is not that the US government is saying it is our policy from here on out that if your model has security vulnerabilities, we will do export controls on non-US persons. I don't think that's the policy there. I think what probably happened is they decided, all right, we got to cut this thing from the market and this is the only thing we can think of that we're pretty sure will actually get the darn thing off the market, right? And so I think they basically just reached for the tool that they thought would do the job. And I don't think they're thinking of that as a universal policy that they're announcing. But the other thing you have to consider here is the administration's story has changed. It's funny, this is similar to the supply chain risk thing. The first version of the story was the US had security concerns and we wanted to get Dario on the phone to talk about them and we couldn't get him on the phone in a timely manner. I remember the same thing with Emil Michael, the under secretary of war who was a main character in that whole affair. He was complaining in public about how Dario wouldn't return his phone calls immediately and it took hours for Dario to get on the phone. Come on. There's a grudge aspect of this, right? There's a who's the bigger monkey aspect to this, right? When I'm in the government, I'm Mr. Government Man, and when I call, he has to call me back. And I don't know. This is a thing that happens a lot in DC. DC people play these kinds of games all the time. Then it became more about the security risk is this jailbreak is legit and we needed to know about it. And then 36 to 48 hours later, or maybe even more like 72, we started hearing about how actually no, the reason we're doing this is because Anthropic provided the model to a Chinese-linked company. And it's like, you're grasping at straws here because if you're saying you opposed export controls because Anthropic gave the model to a Chinese-linked company and that was your primary concern, why did you not say that on Friday when you did the thing, right? And also they're describing this as something that had happened like a month before, because Anthropic did this expanded tranche of companies that were including some international companies. Anthropic announced this publicly several weeks ago. They said, 'Yeah, we're expanding Mythos access and we're expanding it to some US companies and also some allies and partners of the United States.' The company in question by the way is South Korea Telecom, SK Telecom, which is part of the same conglomerate that owns the SK Group, which is one of the largest chaebols in Korea and which also owns SK Hynix, which is the leading producer of high-bandwidth memory, and in general Korea is like a really important partner to the United States in the semiconductor manufacturing ecosystem. So the notion that we would want to harden their telecommunications infrastructure seems quite reasonable to me. It seemed extremely reasonable that we would want to do that. And then they threw that out there and it feels like an administration that either—again I don't know in what ratio of the three things I said—but basically panicked, reached for the first thing that they thought would actually get the model taken off the market, and then created justifications post hoc. And it's really hard—part of the reason I find this issue frustrating is that it's just very hard to analyze because there's not a lot of policy substance here. It's just the id.

好吧，不管这是让你自讨苦吃，还是注定让你成为某种主角，这正好把我们带到了当下：你刚刚宣布将加入 OpenAI，组建一个新团队，帮助塑造公司在前沿模型政策上的立场和影响力。那么，你是怎么做出这个决定的？你提到过去一年对你个人来说也是重要的一年，你有了第一个孩子。再次恭喜。

Well, whether that makes you a glutton for punishment or somebody who is destined to be some sort of main character yourself, that brings us really to the very present moment where you have just announced that you are going to be joining OpenAI and building a new team to help shape the company's positions on and influence on frontier AI policy. So tell me how you came to that. I mean, you kind of alluded to the last year was also a big year for you personally. Had your first child. Congratulations again.

据我所知，你去了很多地方，我相信那一定很精彩有趣。写了很多东西。总的来说，你体验了所谓的美好生活——自由和追求好奇心的能力。而现在，尽管那一切都很棒，你还是决定接受这份工作。那么，请告诉我，你是如何一步步决定这是你接下来想做的事的？

Traveled a lot from what I understand and I'm sure that was exciting and interesting. Wrote a lot. Generally had a taste of the good life, I would say, of freedom and ability to pursue your curiosity. And now you have decided that as great as that was, you're going to take this job. So tell me how you have gone through the process of deciding that this is what you want to do next.

是的，离开政府后这大约 10 个月真是疯狂。我感到非常幸运。我得以进入有趣的圈子，结识有趣的人。我有很多好机会，但也很紧张、很艰难。工作量跟白宫时没什么两样，并不是有些人想象的那种奢侈的智库生活。我认为最重要的是，我的工作聚焦于前沿实验室本身，将其视为一种新的政治和经济权力中心。我几乎把它比作银行的兴起——比如商业银行刚出现时，或者荷兰共和国和英国的金融业。感觉就像那样的时刻。有两个方面：一是这个机构及其与政府和社会的关系。二是，如果你看早期荷兰或英国的金融服务，当时没有证券交易委员会或英格兰银行。但如果你要交易期权或衍生品，就需要共同的规则来建立信任。同样，AI 领域也需要治理。政府自身没有能力或专业知识去追赶，所以这必须在公司内部以及通过私人治理规范来实现。第三，先进的 AI 本身将成为治国理政、监管等工作的工具，就像金融服务被用来实现政策目标一样。AI 未来将是一切的基础。这三方面都让我感兴趣。问题在于，我待过白宫，也在外面待过，有不错的渠道，但如果不进入实验室内部，我就无法超越抽象的直觉。这是我开始考虑加入实验室的核心原因。此外，政策变得越来越重要；我们可能在未来 18 到 24 个月内奠定 AI 政策的基础。我一直在想，但没有行动，后来 OpenAI 联系了我，事情就这样成了。

Yeah, so it has been a wild time since I left government, about 10 months. I feel tremendously lucky. I've been able to be in interesting rooms and meet interesting people. I've had great opportunities, but it's been straining and tough. The workload hasn't changed from the White House. It hasn't been the luxurious think tank life some imagine. I think the most important thing is that my work centers on the frontier lab itself as a new center of political and economic power. I think of it almost like the emergence of banks—when merchant banks first started, or the financial sector in the Dutch Republic and Britain. It feels like that kind of moment. There are two aspects: one is the institution and its relationship with government and society. Two, if you look at early financial services in the Dutch Republic or England, there was no SEC or Bank of England. But if you're trading options or derivatives, you need common rules for trust. Similarly, there will be a need for governance in AI. The government won't have the capacity or expertise to catch up, so it will have to happen within companies and through private governance norms. Third, advanced AI itself will be an instrument for statecraft, governance, and regulation, just like financial services are used to achieve policy objectives. AI will be fundamental to doing anything in the future. All three interest me. The struggle is that I've sat in the White House and outside, with good access, but I can't get beyond abstract intuitions without being inside a lab. That's the central reason I started thinking about joining a lab. Also, policy is becoming more important; we might set the foundation of AI policy in the next 18 to 24 months. I was thinking about it but not acting, then OpenAI approached me, and that's how we got here.

你能再多说一点，为什么你认为进入前沿实验室如此重要吗？我也有同感。感觉真正重要的机构角色正在变少，这让我处于一种不舒服的境地。进入内部到底改变了什么？是对路线图或能力有更好的可见性吗？

Can you say a little more about why you think it's so important to be inside a frontier lab? I share this intuition. It feels like the number of institutional actors that really matter is becoming small, and it leaves me in an uncomfortable position. What exactly is it that being inside changes? Better visibility into roadmap or capabilities?

首先，让我说说我的团队将是什么样的，以及它与 OpenAI 其他团队有何不同。这是一个精品运营团队。有一个由 Chris Leane 领导的全球事务团队，负责传统的政策和游说工作。那个团队继续存在，而且非常能干，处理来自全美 50 个州、联邦政府和世界各地的公共政策。但问题是，一年前人们几乎不谈论儿童安全或数据中心的电力、用水问题，现在他们谈了。进入内部能让你更好地理解技术及其发展轨迹，这对制定政策至关重要。你不能仅仅从外部反应；你需要成为过程的一部分才能有效影响它。

First, let me say what my team will be and how it's different from other teams at OpenAI. It's a boutique operation. There's a team called Global Affairs run by Chris Leane, which does traditional policy and lobbying. That team continues and is very capable, dealing with public policy from all 50 states, the federal government, and worldwide. But the problem is that a year ago, people were barely talking about kids' safety or data center electricity or water use. Now they are. Being inside gives you a better understanding of the technology and its trajectory, which is crucial for shaping policy. You can't just react from outside; you need to be part of the process to influence it effectively.

2019 年 6 月，最好的模型还是一年前的 GPT-2，对吧？和今天的世界完全不同。所以这个团队的部分工作就是展望未来 6 到 12 个月，判断我们正在走向何方，我们可能会面对什么，然后如何制定政策——不仅是公司当下的政策立场，也包括未来的政策——如何主动应对我们预判的未来。为此，我预计自己很大一部分时间会和技术人员深入探讨技术走向及其影响。你需要掌握细节——不只是‘模型会变得更好’，而是真正深入内部部署等具体问题，了解能力前沿将如何演进，一年后的世界与今天会有哪些不同。

The world of June 2019, the best model was GPT-2 a year ago, right? A very different world today. So the job of this team, in part, is going to be to look out 6 to 12 months and say where are we going? What do we think we're going to be dealing with? And then how can we shape the policy—both the present-day policy positions of the company but also future policy—how can we develop policies to try to be proactive in dealing with where we think we're going to be in 6 to 12 months? To do that, I anticipate that a very large portion of my time is going to be spent jamming with the technical staff on where things are going and what that's going to mean. So you do need to be able to access detail—not just 'the models will get better' but really get into the weeds on internal deployments and many different things about where the capabilities frontier is going and what will be different about the world in a year versus today.

你如何看待自己上任后的职责？就像为美国政府工作时，你要宣誓效忠宪法；加入 OpenAI，我们肩负着确保人工智能造福全人类的使命。你是否认为自己是以类似的方式支持这一使命？还是说你的个人目标函数在某种程度上更加复杂？其中是否包含 OpenAI 获胜的权重？是否包含美国获胜的权重？围绕‘造福全人类’这一核心理念，存在多少复杂性？

How do you understand your duty as you start this role? When you sign up to work for the US government, you swear an oath to the constitution. When you go to OpenAI, we have this mission of making sure that AI benefits all humanity. Do you think of yourself as signing on to support that mission in the same way you might have previously sworn to uphold the constitution? Or would you describe your personal objective function as being in some ways more mixed? Is there a term in it for OpenAI winning? Is there a term in it for the USA winning? How much complication is there around the core idea of 'benefit all humanity'?

我认为这一使命在公司内部被非常认真地对待。有一件事虽然不是我公开宣布的职责，但说出来应该没问题：OpenAI 内部有一个叫 MAC 的机构——全称可能是 Mission Advisory Council 或 Committee，我记不清了——由研究人员、全球事务人员以及公司各部门的代表组成，共同决策政策及部分内部治理事项。我的部分工作就是参与这个机构。所以从某种意义上说，我确实认真对待这一使命，我认为 OpenAI 的文化也是如此。当然，问题在于如何定义它的含义？这是一个宽泛的使命，存在很多解读空间和模糊性。正因如此，接受这个角色时，对我非常重要的一点是，我能够保持独立的公开写作，不受 OpenAI 的编辑审查。我完全不认为我们会看到那种漫画式的 OpenAI 大反派阴谋。如果内部体验真是那样，我会非常惊讶。我猜测更多的是，人们对于某些具体决策如何与更广泛的使命相关联存在善意的分歧。有时我可能最终不同意公司的决定，而我认为能够公开讨论这些事情而不必担心饭碗，这一点很重要。OpenAI 的一大优点是它仍然保留着类似施乐帕克研究机构的 DNA。他们容忍大量内部异议。组织内部有很多精彩的辩论，我从外部观察时一直有这种感觉。

I think that mission is something that people inside the company take quite seriously. One thing that I don't think is a publicly announced part of my role, but it's probably okay for me to say, is that inside OpenAI there's a body called the MAC—the Mission Advisory Council or Committee, I forget—which consists of researchers, global affairs people, and a wide variety of people from around the company who collectively make decisions about policy and some internal governance decisions. Part of my job will be sitting on that body. So definitely, in some sense, I take that mission seriously, and I think OpenAI culturally does as well. Of course, the problem is how do you decide what it means? It's a broad mission open to interpretation, with a lot of ambiguity. That's where one thing that was very important to me in taking this role was that I could maintain a public writing presence independent of any editorial review by OpenAI. I really don't think that what we're going to see is the cartoonish depiction of OpenAI as a grand villainous conspiracy. I would be strongly surprised if that is what I experience on the inside. What I would guess instead is people with good-faith disagreements about how some particular set of decisions relate to the broader mission. There may well be times when I ultimately disagree with the call that was made, and I think my ability to communicate publicly about things like that without fearing for my job is important. One of the great things about OpenAI is it still has the DNA of being a sort of Xerox PARC-like research organization. They tolerate lots of internal dissent. There are a lot of great debates inside that organization, and I've always gotten that sense observing it from the outside.

另一件事就是内部部署的问题。特别是，如果我们正在走向一个由监管风险、安全担忧、算力约束等因素共同塑造的世界——很可能，GPT-4 的训练应该快完成了吧？我们 1 月份就拿到了那些检查点，也就是 6 个月前。OpenAI 肯定也一样——虽然我还没有任何内部消息，我可以完全无知地发言。但可以肯定他们正在训练另一个模型。这些模型的内部部署，从根本上说，在我们拥有健全的监督、审计或独立验证体系之前——政府思考监管的方式，从机制上讲，所有政府考虑的监管都是由公开发布、公开部署触发的。但我认为很多真正重要的决策将围绕内部部署做出。这既需要客观判断，也可能需要一些直觉判断，比如递归自我改进最终意味着什么？我们应该如何思考它？没有内部知识，抽象地思考这些非常困难。我可以在我的 Substack 上写文章，也许会影响几个关键人物。但最终，我认为你需要亲自动手，与研究人员、高管团队和其他人一起塑造这些决策。

Another thing I would say is simply that question of internal deployments. In particular, if we are moving toward a world where, for some combination of regulatory risk, security concerns, compute constraints, etc., we may well be moving to a world where—I mean, presumably GPT-4 is not that far from being done training, right? We had those checkpoints in January, so 6 months ago. Same with OpenAI, I'm sure—though I don't have any internal knowledge yet. I can speak with total ignorance. It's safe to say they're training another model. The internal deployments of these models, fundamentally, until we have a robust system of supervision and auditing or independent verification—the way the government thinks about regulation, mechanically speaking, all government regulations that states or governments think about are triggered by public release, public deployment. But I think a lot of the really important decisions are going to be made with respect to internal deployments. And there's going to be a combination of objective determinations you want to make and also probably some gut calls, some judgment calls about what recursive self-improvement means ultimately, right? How should we be thinking about it? It's very hard to ponder that stuff in the abstract without inside knowledge. I can write about it on my Substack and maybe that'll influence a couple people who matter. But in the end, I think you really want to be getting your hands dirty and shaping some of these decisions with researchers, with the executive team, with many other people.

所以我不觉得这会在文化上太不协调。但没错，我公开表达、公开反对某些政策立场的能力，我认为会很重要，这也是我保留这种能力的原因。所以我想说的是，我仍然觉得最终我想做的是把事情做对，帮助国家乃至世界——可能首先是国家——顺利实现这一整体转型。我更像是一个——这是我和东湾及旧金山人非常不同的一点——我是一个爱国者。我认同自己是美国人，而不是世界公民。我是美国人，kiwi americanos。我一直觉得这就是我的使命，我在智库做过，在政府做过，现在是在 OpenAI。当然，其中一点是，OpenAI 是一家公司——我将帮助 OpenAI 制定其战略，对吧？这不同于我为抽象的 AI 行业制定战略。现在是 OpenAI，一家与其他公司存在区别的公司，会有竞争考虑等等。这没问题。我是一个有竞争心的人。事实上，我认为竞争会让事情变得更好。但至少对他们来说，平均而言会让事情变得更好。不过，确实也有这一面。

And so I don't think that this is going to be like culturally too dissonant. But yes, my ability to publicly say, to publicly disagree with certain policy positions, I think will matter and that's part of why I preserve that. So I guess what I would say is I still feel like ultimately what I am trying to do is trying to get this right, trying to help shape this whole transformation well for the country and for the world, probably the country first and foremost. I'm more of a—that's one area in which I'm very different from people from the East Bay and San Francisco—is that I am a patriot. I identify as an American, not as a citizen of the world. I am an American, kiwi Americanos. And I feel like that's been the mission the whole time, and I've done it in think tanks, I've done it in the government, and now it's OpenAI. And for sure, one thing that factors into this is, of course, OpenAI is a company with—I will be making, I will be helping OpenAI set its strategy, right? Which is different from I am setting the strategy for the abstract AI industry, right? It's now OpenAI, which is a company that exists in contra distinction to other companies in the field, and there are competitive considerations and things like that. That's fine. I'm a competitive person. I think the competition will make things better, in fact. But at least for them, on average, it'll make things better. But yeah, there's definitely that too.

那我们回到 RSI（递归自我改进）的话题。我的意思是，竞争会让事情变得更好这种说法，肯定会让一些听众感到惊讶，他们担心公司之间、国家之间等各种形式的军备竞赛动态。在我看来，至少两家公司之间的 RSI 竞赛可能是 AI 领域目前最令人反感的事情，因为确实感觉——我听过一些研究人员坦诚地谈论这一点，他们似乎也有同感——这是一个相变时刻，之后事情可能会变得非常奇怪，把一切设置好、让初始条件正确将极其重要。而他们仍然不太有信心事情会进展得很顺利。

Let's go back then to RSI. I mean the notion that competition will make things better is definitely going to surprise some ears in the audience who are worried about arms race dynamics between companies, between countries, you know, any number of different configurations. And for my money, the race to RSI between at least two companies is probably the most objectionable thing happening in the AI space right now because it does feel like—and the people, the researchers that I have heard speak candidly about it seem to share the intuition that this is sort of a phase change moment beyond which things could get really weird and it's going to be super important to set everything up right, get the initial conditions all right. And they're still not that confident that it's going to go very well.

是的。

Yeah.

那么，我想你如何理解安全计划是什么？你如何理解这些公司对 RSI 或破产这种路径的投入程度？你知道，OpenAI 有公开的时间表，说明他们希望何时拥有自动化实习生和成熟的机器学习研究员。所以，是的。你如何理解这个计划，你认为它目前是否接近能够胜任任务？

So, I guess how do you understand what the safety plan is? How do you understand how committed the companies are to kind of an RSI or bust? You know, OpenAI famously has public timelines for when they want to have the automated intern and the full-fledged ML researcher. So, yeah. What do you—how do you understand the plan and do you think it is anywhere close to being up to the task at this point?

嗯，我想说一件事：当我谈到竞争时，特指我将要做的工作，也就是说，我团队的目标不仅仅是——我团队的目标是产出非常非常出色的智力成果，让人惊叹：这个团队，如果它不是 OpenAI 的一部分，而只是一个独立的小智库，它也会是全国最有趣的智库之一，对吧？每个人都会关注它。我其实只是希望它成为一个非常优秀的团队，产出与 OpenAI 任何竞争对手一样好的政策和公共利益相关工作。但这就是我所说的竞争，要澄清一下。而且我认为在 RSI 方面，有一个直接健康竞争的好例子。我认为，嗯，这里肯定有很多未知数。我对 RSI 含义的基本判断，至少在早期阶段，大致是这样的：一件事，RSI（递归自我改进）是每项技术的一部分。每种通用技术，由于其通用性的本质，都带有某种递归性。通用意味着它可以应用于自身的目的之一。所以，从某些方面来说，我并不认为递归自我改进是技术史上的一个重大突破。我认为它实际上是——如果 AI 中没有递归自我改进，那才令人惊讶。我还认为我们在这个领域已经做了很长时间的递归自我改进。有些人想象会有一个重大的突破时刻。你可以说，至少从 GPT-4 开始，我们就在用模型来改进模型。而且我敢肯定，如果你真正回顾机器学习的历史，我打赌那可以追溯到更早。所以有些人想象 RSI 意味着一个急剧的不连续跳跃。我当然认为这是可能的，但这不是我的先验，因为你的先验通常应该是——我肯定在这个播客上说过——总是连续性多于不连续性。情况总是如此。所以你的先验应该反对巨大的不连续跳跃。这是可能的。所以我认为第一步，我还没进去，所以实际上不知道，因为我还没看过路线图。但第一步会是尝试真正地三思而后行，弄清楚我们认为这可能意味着什么。尝试真正细化那里的概率，至少在我自己的脑海里，我们是在谈论一个很快发生的不连续跳跃，还是在谈论某种更平滑的东西？然后我认为，如果我们处于——我认为不连续跳跃的可能性无论如何都足够高，你需要现在就开始规划，即使你认为它发生的概率是 20% 或 10%，那也足够高，你应该现在就开始为你想要做的事情制定计划。然后我们进入实验室间协调，比如有减速暂停之类的东西，对吧？那会有什么机制？在什么条件下会触发？再次，我对这类概念相当怀疑。我认为，作为——我为美国政府做了很多政策规划，我写下来并放在某些地方，但它不在行动计划中。它是情景，对吧？我们需要为广泛的情景做好准备。所以类似地，我认为会有一些方面我们需要准备好——我们需要对所有这些东西有前瞻性思考。所以是两件事。

Well, one thing I want to say: when I was talking about competition specifically with respect to what I will be doing, which is to say, the goal of my team will not just be—the goal of my team will be to have really, really fantastic intellectual output that is like, wow, this team, if it were not part of OpenAI and it were just its own little think tank, it would be one of the most interesting think tanks in the whole country, right? And everyone would be paying attention to it. I actually just want it to be a really superb team that is producing policy and sort of public interest related work that is as good as anything else that any of OpenAI's competitors produce. But that's what I meant by competition, to be clear. And I think there is a good example of straightforwardly healthy competition when it comes to RSI. I think, yeah, I think there's certainly a lot of unknowns here. I put my base case for what RSI means, at least in the earlier innings, is something to the effect of: one thing, RSI—recursive self-improvement—is a part of every technology. Every general purpose technology has some aspect of recursion to it by the very nature of generality. General purpose means one of the purposes to which it can be applied is itself. So, in some ways, I don't see recursive self-improvement as some big break from the history of technology. I see it as actually being—it would be surprising if there weren't recursive self-improvement in AI. I also think we've been doing recursive self-improvement in this field for a long time. And some people imagine there's going to be some big break moment. You could argue really since GPT-4 we've been using the models to make the models better since at least GPT-4. And I'm sure that if you actually went back and looked through the history of machine learning even more, I bet that actually goes back even further than that. So some people imagine there being this sharp discontinuous jump in terms of what RSI means. And I certainly think that is plausible, but it's not my prior because your prior should just generally be that there is always—I'm sure on this podcast I've said before that there is always more continuity than discontinuity. This is always the case. And so your prior should be against a massive discontinuous leap. It is plausible. And so I think step number one, and I can't—I'm not in yet, so I don't actually know because I haven't looked at the road map yet. But step number one would be to try to really measure twice and cut once and figure out what we think this might mean. Try to really refine the probabilities there, at least in my own mind, of are we talking about a discontinuous leap that happens very soon or are we talking about something that actually is smoother in some way? Then I think if we are in—I think the chances of discontinuous leap are high enough no matter what that you need to be planning now even if your credence that it happens is 20% or 10%, that's high enough that you should be making plans now for what you want to do. And there we get into interlab coordination on things like there's the slow down pause thing, right? There's what would be the mechanisms of that. Under what conditions would it be triggered? Again, I'm pretty skeptical of such notions. I think as a—there's a lot of policy planning I did for the US government that I wrote down and put places but it's not in the action plan. It's scenarios, right? We got to be prepared for a wide range of scenarios here. And so similarly, I think there's going to be some aspect of that where we need to be ready to—we need to have advanced thinking on all that stuff. And so it's two things.

好的。在什么节点？我们可以提前设定哪些触发条件，来判断是否会出现一次不连续的跃升？我们如何把这个问题细化到尽可能具体？然后，如果那些触发条件发生了，我们该做什么？在什么节点去找政府？值得一提的是，我支持的一个提案是让联邦贸易委员会（FTC）出具一封所谓的“不采取行动函”。FTC 会发一封信，发布公开指引，基本上就是说：如果你们出于这些非常具体的原因进行协调，我们不会将其视为卡特尔行为，也不会执行反垄断法。我认为这大概是一个好步骤，至少打开了可选性。不过我也理解，在界定范围时必须非常小心，因为看看 Anthropic 做了什么，对吧？Anthropic 用那个寓言式的保障措施——“为了安全我们会降低你的输出质量”——极大地削弱了这个案例的说服力，这明显是侵犯消费者权益。非常明显，想象一下，如果一群 AI 公司以安全为名合谋降低特定领域的输出质量，那将是严重的反竞争行为。所以你必须非常小心地界定范围，而且公司可能也需要非常谨慎，不要做那些削弱“安全是我们应该考虑这些因素的理由”这一论点的事情。但关于 RSI 计划本身，我没有太多具体内容可以分享，原因很简单：我还没参与进去，也还没进行过那些对话。

It's okay. At what point? What are some triggers that we can set in advance for: is this going to be a discontinuous leap specifically? And how can we refine that question to make it as specific as possible? And then in the event that those triggers happen, what is it that we would do? At what point do we go to the government? It's worth noting one proposal that I'm a fan of, or I'm a fan of the FTC, the Federal Trade Commission, writing what would be practically called a no-action letter, where the FTC would write a letter and they would put out public guidance that would basically say, look, if you guys coordinate for these very specific reasons, we're not going to consider that cartel behavior and we're not going to enforce that. I think that's plausibly a good step to make that at least opens optionality. Though I also understand that the way you scope, you have to be really careful about how you scope that because look at what Anthropic did, right? Anthropic undermined the case for this dramatically just with the fable safeguards with "we're going to degrade your outputs in the name of safety," which is very clearly a consumer protection violation. Just very clearly, if you can imagine if a cartel of AI companies in the name of safety agreed to collude to degrade outputs in particular areas, that would be wildly anti-competitive. And so you have to scope it really carefully, and also probably companies need to be very careful about what kinds of things they do that undermine the case that safety is something we should be making these considerations for. But yeah, I don't have a lot of specifics to share on the RSI plan itself for the simple reason that I'm not in there yet and I haven't had those conversations yet.

你感受到的整体氛围是怎样的？因为有几件事我可以交叉验证：第一，正如你指出的，Anthropic 和 OpenAI 可能进行了幕后协调，在时间上非常接近地发表了声明，表示他们对某种协调放缓的可能性持开放态度。所以 Dario 和 Dema 说，如果只有我们两家，我们可以想出办法。然后我不知道 Elon 最近有没有说什么特别亲社会的话，但这类言论很多。从我的角度看，数量确实惊人。与此同时，我们最近听到关于中国的讨论远不如不久前那么多。所以我从外部的解读是，这些公司似乎被自身能力进步的速度吓到了。你也有同样的看法吗？

Is it your sense of the overall vibe that because a couple things I would triangulate: one, as you noted, there's been this perhaps coordinated behind-the-scenes, very close in time statements by Anthropic and OpenAI saying that they're open to the possibility of the need for some sort of coordinated slowdown. So we've got Dario and Dema saying, you know, if it was just the two of us, we could figure something out. And then I don't know if Elon's said anything so pro-social lately, but there's been a lot of that. Certainly a surprising amount from my perspective. At the same time, we haven't heard nearly as much about China recently as we were not super long ago. So I guess my read from the outside is it feels like the companies are getting a little spooked by the pace of their own capabilities advances. Do you read them in the same way?

我的意思是，很难把他们当作一个整体来讨论。我确实认为有你说的那种情况。而且我也觉得，用一句克劳德式的话来说，有一种眩晕感，就好像你感觉自己正在接近悬崖，对吧？对于跳过去之后会发生什么，存在很大的不确定性。是的，我认为我要说的是：我不觉得实验室里的氛围是恐惧，或者“天哪我们好害怕但不得不做”。我觉得如果我把刚才的想法分享给很多研究人员，他们会说“嗯，这听起来合理。我们也不确定这到底意味着什么”，等等。但有些人会反驳得更强烈；会有意见分歧。但大体上，我刚才分享的那个关于递归自我改进的略微通缩的观点——想象一下，如果递归自我改进意味着那个拐点，记得在推理模型之后，很多基准测试图表上有一个明显的上升，对吧？如果再来一次，或者比那次再强 30% 呢？那也没关系。那不是奇点，对吧？这才是关键。那不是奇点。我不会说它在任何客观意义上是通缩的，但相比某些观点，它是通缩的。我觉得这一直是我的先验，而且一直相当不错。通缩的。相比几乎所有人的想法，它是极度通胀的；相比东湾那些思考 AI 安全十年的一小撮人，它又是通缩的。我认为在这两种观点之间找到自己的位置其实相当不错；如果你基本上一直处于那个位置，那你猜对了很多事情。我不认为那种观点——我当然不认为我的观点会在任何实验室里被嘲笑。我认为实验室内部的感受是：嘿，我们很快就要做这件事了。存在很大的不确定性，我们不太确定我们是否真的有计划。就算有计划，我们可能也不太确定我们会执行它。有很多“嘿，我们需要确保我们真的做这些事”，对吧？所以我认为这是当前情况的一个重要部分。我认为对此存在担忧。这是一个完全合理的担忧。而且肯定有一些组合——就像所有事情一样，是政策实际解决问题和策略的结合，而且我虽然是个糟糕的政治家，但如果你让我对一个想法感到兴奋，我可以把这个想法传达给别人，并找到迭代沟通方式以吸引不同个人和受众的方法。行动计划很大程度上就是如此。行动计划一部分是政策制定，一部分是策略，但不是大众政治——是非常具体的内部事务。如果我对一系列想法感到兴奋，我会像狗啃骨头一样执着。我认为工作的部分内容就是弄清楚我们如何真正地不只是制定计划或开发——我甚至还没看到正确的，我甚至还没看到任何那些东西，所以需要看到它。但是的，我们如何真正建立内部信念，让我们会去做？让我们真正听从自己的计划，对吧？

I mean, I think it's very hard to talk about them in monolithic ways. I definitely think there's some of that for sure. And I also think there's just, you know, to use a claudism, there's a certain vertiginous feeling about getting to the, you know, you sort of feel like you're approaching the cliff a little bit, right? And that there's substantial uncertainty about what happens when you sort of go, when you sort of jump over it. And yeah, I think I do think that there's what I would say is this: I don't think people, I don't know that the vibe inside the labs is like terror about this or like "oh my god we're so scared but we have to do it anyway." I think if I were to share the thoughts I've just had with a lot of researchers, they would be like "yeah, that seems reasonable. We don't really know exactly what this is going to mean," blah blah blah. But some of them would push back more strongly than others; there'd be differences of opinion. But I think broadly that the thought I just shared about a slightly more deflationary view of recursive self-improvement — imagine if what recursive self-improvement meant was that the kink, remember after the reasoning models there's a noticeable uptick in a lot of the benchmark charts, right? And what if it's like that again, or what if it's like that but 30% more? It's okay. That's not a singularity, right? That's the main thing. It's not a singularity. And I wouldn't describe it as deflationary in any objective sense, but it is deflationary compared to some views. I feel like that has been my prior this whole time and I feel like it's been a pretty good one. Deflationary. It's massively inflationary compared to what almost everyone thinks, and deflationary compared to what a very small number of people who've been thinking about AI safety for 10 years in the East Bay think. I think finding your way in between those two views has been actually quite good; you got a hell of a lot right if that's basically where you've been. I don't think that view — I certainly don't think my view would be laughed out of the room in any lab. What I think the feeling inside the labs is like: hey, we're going to do this soon. There is substantial uncertainty and we're not quite sure that we really have a plan. And to the extent we have a plan, we're maybe not quite sure that we're going to follow it. There's a lot of "hey, we need to make sure we actually do this stuff," right? And so I think that's been a substantial part of what's going on here. I think there is concern about that. I think that's a totally legitimate concern. And definitely there's some combination — like everything, it's a combination of policy actually figuring out substance and then also I'm a terrible politician, but if you get me excited about an idea, I can communicate that idea to people and I can find ways to iterate my communications of that idea to appeal to different individuals and audiences. That's what the action plan largely was. The action plan was part policy development and part tactics in that, but not like mass politics — very specific kinds of internal stuff. And I can be, if I get excited about a set of ideas, I can be like a dog with a bone. And that's basically I think part of the job is to figure out how are we going to actually not just have a plan or develop — I haven't even seen the right, I don't even haven't seen any of that, so it's like need to see it. But yeah, how do we actually build internal credence that we're going to do it? Let's actually listen to our own plan, right?

是啊。这方面的记录并不理想。可以说，就治理计划以及它们迄今为止经受住时间考验的情况而言，情况并不乐观。

Yeah. The track record there is not amazing. I'd say it's safe to say in terms of governance plans and how they've stood the test of time so far.

但这也很难，因为是谁来着？是你吗？不。不是在你的播客上，但 DeepMind 的 AGI 安全负责人最近上了，我想是 80,000 Hours 节目，Rohan。

But it's also hard because who was it? Was it? No. It wasn't on your podcast, but the AGI safety lead at DeepMind was on, I think it was 80,000 hours recently, Rohan.

我们讨论过，是的，我们不想做出承诺。我们想要制定计划，并且认真对待这些计划，但我们也不想做出硬性承诺，因为如果被太多先前的承诺束缚住，那会非常不利，毕竟存在太多不确定性。这需要微妙的平衡，但我认为这是可以做到的。

And talked about how, yeah, we don't want to make commitments. We want to make plans and we want to be serious about those plans, but we also don't want to make hard commitments because it's actively bad if we lock ourselves down too much with a bunch of prior commitments because there's so much uncertainty. There's a subtle balance that you have to strike there, but I think it is possible to do.

上次我们聊到了一点“历史伟人理论”。你刚才提到了非常地方性的政治、个人性格等因素。你预期技术基本面在多大程度上决定结果，而关键决策者及其协作能力和及时做出正确决策的能力又在多大程度上真正重要？

I think last time we talked a little bit about the sort of great man of history theory. You just mentioned like very local politics, individual personalities mattering, that kind of thing. What is your expectation in terms of how much technology fundamentals will determine outcomes versus how much key decision makers and their ability to work well together and make good decisions in timely ways will really matter.

嗯，这其实是我大学时期的一个副修方向——历史哲学，我一直很喜欢历史哲学。这个问题可以说是历史哲学的核心问题，就像宿舍里闲聊版的历史哲学。历史哲学的问题就是这个，对吧？就是结构力量与伟人理论之争。而那个有点令人不满的答案是：两者兼有。在某种程度上，我会说历史的结构力量就像你默认身处的那条河流。然后有时在历史中，无论大小，总有人不只是随波逐流，而是出于某种原因逆流而上，并最终凭借逆流而上的纯粹决心改变了河流的轨迹。我认为，那些违抗结构力量的人，在很多方面就是历史的伟人。过去一年在政府工作以及离开政府后所拥有的视角，让我有一个很大的更新：世界上很多事情是由少数个体之间的个人关系决定的。我不认为这能解释 AI 基础设施的建设。它解释不了为什么人类正用数据中心和为数据中心供电的能源覆盖地球表面越来越大的比例。那更多是结构性的。但在很多方面，看看你认为“战争部与 Anthropic 事件”是历史重要时刻的程度。很大一部分是由糟糕的个人关系驱动的，对吧？就是人们互相不喜欢。具体来说，是 Dario Amodei 和美国政府中的一些高层人物。我不知道这种不喜欢是否是双向的，所以我不想归咎于任何一方，但我只想说关系不好。所以，是的，最终两者都有。我认为从根本上说，我们站在河流中，对此无能为力；政治理论家称之为与河流的非自愿关联。那条河流是你生来就身处其中的，你被困住了。你存在于宇宙中，你存在于时间之箭中。但与此同时，总会有个体深刻塑造将要发生的事情。我认为我们可能会经历一段历史时期，这段时期可能有点……奇怪的是，如果你认为这个时刻是——我不一定相信这一点，但很多人会说我们正在经历人类智力的某种日食，我们正处于人类作为这个星球主要行动者的最后日子。很快机器就会崛起。这里有一个讽刺：我认为整个转变——我认为人类实际上会经历一个非常具有“主角光环”的时期，即使它最终确实意味着机器成为主要行动者。会有这样一个时期——有点像——从这个意义上说，这是一个非常美好的时代，因为从酒神式的角度看，它有很多丑陋之处，但丑陋中也有美。当一颗恒星死亡时，它会膨胀成红巨星，对吧？就像那样，当你目睹人类最后的绽放和机器智能的诞生时，你会看到人类努力中的伟大。我觉得我们在世界上确实看到了一些这样的迹象。我认为我们会看到更多。我认为我们将经历一个英雄般的时代，基本上就是这样。至少可能是这样，或者是一个邪恶的时代，但会有很多机会给伟大的人，可能两者都有。

Well, this is in many ways one of my sub focuses in college was the philosophy of history and I've always loved the philosophy of history and this is like the central question to the extent there are like the dorm room version of philosophy of history. The philosophy of history question is this one right? It is the structural forces versus great man theory. And the somewhat unsatisfying answer is that it is both. In some ways, what I would say is that the structural forces of history are like the river that you're in by default. And then sometimes in history in little ways and big ways there are people that don't just swim with the current and actually stand against it for whatever reason and ultimately shape the trajectory of the river by sheer force of standing against it, by the sheer determined determination with which they stand against it. And I think those are the people who disobey the structural forces are the great men of history in many ways. A big update for me in the last year working in government and then just having the perch I've had since I left government is that much of what happens in the world is determined by the personal relationships of a small number of individuals to one another. I don't think that explains the AI infrastructure build out. It doesn't explain why humanity is covering increasing fractions of our surface area of our planet with data centers and energy to power the data centers. That's more of a structural thing. But in many ways, look at the extent that you think that the Department of War Anthropic situation is an important moment in history. A big chunk of that is driven by personal relationships being bad, right? It's about people not liking each other. And it's specifically about Dario Amodei and various people senior in the US government. I don't know that the dislike goes both ways, so I don't want to attribute that to either of them, but I'll just say having a bad relationship. So, yeah, it is ultimately both. I think that fundamentally we are standing in the river and there's nothing you can do about it; what political theorists would call an involuntary association with the river. That river is the thing you were born into and you are stuck. You are in the universe, you are in the arrow of time. But at the same time, there will be individuals who profoundly shape what happens. I think we're probably going to live through a period of history that is maybe a little bit more... Weirdly enough, if you think that this moment is—I don't necessarily believe this, but a lot of people would say we're living through this kind of eclipse of the human intellect, where we're in the final days of humans being the primary actors on this planet. And that soon machines will rise. There is this irony in that I think that whole transformation—I think humans will actually go through a very main character energy period of time as that transformation occurs, even if it ultimately does mean that the machines become the primary actors. There'll be this period—it's a little bit like—in that sense it's a very beautiful time period to live through because in a Dionysian way there's a lot of ugliness about it, but there's a beauty in the ugliness. When a star dies it grows super big into the red giant, right? And it's like that, where you watch this final flowering of humanity and the birthing of the machine intelligence, you see this greatness in human effort. I feel like we do see some of that going on in the world. I think we'll see much more of it. I think it will be a heroic time period that we live through, basically. At least it could be, or a villainous time period, but there'll be a lot of opportunities for great people and probably both.

那么你认为，你个人或与你密切合作的人最可能需要以哪些方式逆流而上？

So what do you think are the most likely ways in which you personally or those that you're working closely with will need to stand against the current?

嗯，我想说清楚，我并不认为自己是什么历史伟人。我认为自己在这一切中扮演着非常谦逊的角色。但由于我在互联网上有一定的公众形象，这个角色可能看起来比实际更重要。但 broadly speaking，基本现实是：在某种我认为这场变革将非常熵增的过程中维持秩序、维持文明秩序。面对这种情况维持秩序就像——你不想完全没有熵，对吧？你想要有火，但你又不想把森林烧了。你想要一团能产生温暖且受控的火，但它本质上仍然是一团火。要做到这一点，需要大量深思熟虑的人类努力。所以我只是普遍认为，会有很多时刻我们必须设定限制。我们将不得不在各种方式上对自己施加人为约束。我们必须愿意划清界限，说不，我们不想生活在那种世界里，或者我们想要生活在这样的世界里，我们不能对每件事都含糊其辞。所以我认为这里面有大量的工作。另外，我认为递归自我改进很可能是一个很好的例子。是的，我们将不得不违背自身利益。还有，我认为有一件事是——其他实验室也是——我认为他们在这方面已经很不寻常了，但从政策角度看，他们的政治立场我认为与许多其他科技公司相当不同。

Well, I want to be clear, I don't see myself as being one of those great men of history. I see myself as playing a very modest role in all of this. But a role that probably seems bigger than it is because of the fact that I have a public profile on the internet. But broadly speaking, the basic reality here is that maintaining order, maintaining civilizational order in the midst of something that I think this transformation will be very entropic. And maintaining order in the face of that is like—you don't want no entropy, right? You want there to be a fire, but you also don't want to set the forest on fire. You want there to be a fire that generates warmth and is under control, but is also still fundamentally a fire. And doing that requires a lot of deliberate human effort. So I just generally think that there are going to be a lot of moments where we have to put in the limit. We're going to have to put artificial constraints on ourselves in various ways. We're going to have to be willing to draw lines in the sand and say no, we don't want to live in that kind of a world, or we do want to live in this kind of a world, and we can't be mealy-mouthed about everything. So I think there's a tremendous amount there. And also, I think the recursive self-improvement thing may well be a really good example. Yeah, we're going to have to act against our interests. And also, I think that just one thing is—and the other labs from—and I think they're already abnormal in this regard, but from the policy, their political position is just I think rather different from what a lot of other tech companies have been.

所以我认为，它们在公共话语和政策辩论中扮演的角色，将不得不与我们习惯的企业角色截然不同。这要求企业在某种意义上违背自身利益——那些经济学教科书会预测的利益。但确实，这种情况已经很多了。哦，是的，我们已经看到这些公司在这样做。AI 公司相比世界上大多数公司都非常反常。但某种程度上，未来几年可能更需要行动而非承诺。

And so I think the role they play in the public discourse, the role they play in the policy debates is just going to have to be very different from what we're used to from companies. And that will require the companies to in some sense act against their own interests in terms of what the economics textbooks would predict their interests are. But yeah, I think there's definitely plenty of that. Oh yeah, we already see the companies do this. The AI companies are all very abnormal compared to most companies in the world. But yeah, in some sense the next few years might have to be characterized more by action rather than commitments.

我们来个快速问答环节，然后再把视角拉远。

Let's do a little lightning round and then we can zoom out again.

好。

Sure.

最后，你怎么理解 OpenAI 和 Alex Boris 之间发生的事情？

At the end, how do you understand what has gone on between OpenAI and Alex Boris?

嗯，这是个好问题。我不了解所有细节。有一点我可以肯定——我指的不是超级政治行动委员会 Leading the Future 的任何捐赠者，该委员会著名的捐赠者包括 OpenAI 总裁 Greg Brockman——但我在进入 AI 领域之前，曾与许多我工作过的组织的董事会成员共事，他们都是非常重要的政治捐赠者，有些甚至是全球最大的捐赠者。最近我有机会结识了一些民主党和共和党最著名的政治资助者。我想说的是，你会惊讶于他们对自己创建的政治组织的控制力有多弱——我说的是那些身家百亿的人，对吧？你会惊讶于他们觉得自己对创建的政治组织有多失控。他们会说：“不，真的，委托代理问题不会因为你富有就消失。”所以，当你看到一个像超级政治行动委员会这样的政治组织采取行动时，我不认为你的先验假设应该是资助者直接控制着一切。事实上，作为一个整个职业生涯都——我从未为政治倡导组织工作过，但我的大部分职业生涯都在 501(c)(3) 组织工作，这些组织通常由富裕个人资助，从事公共利益事务——我从未觉得我们的捐赠者，包括我工作过的任何组织的捐赠者，比如美国创新基金会，设定我们的议程或控制我们的行动。通常，你向某个特定组织捐款是因为你认同那里的人，你期望他们——你喜欢他们所做的工作。有人向美国创新基金会的 AI 政策项目捐款，是因为他们喜欢我写的文章，或者 Sam Hammond 写的文章，或者其他人的工作。但他们不会坐在那里告诉我和 Sam 该做什么，我和 Sam 也不会听。我向你保证。我向你保证，如果有捐赠者试图告诉我该说什么，我会让他们滚蛋。所以，是的。总之，我认为你应该想象的是，像 Leading the Future 这样的政治组织就像一种上了发条的玩具，它会按照默认路径运行。默认情况下，它会说：“嘿，这家伙想以 AI 行业监管者的身份出名，所以让我们向所有人传达一个信息：如果你想像这家伙一样，你会输。”他们尝试了——我不会说 OpenAI 尝试这样做，但我会说 Leading the Future 尝试了。他这周参加了初选。这提升了他的知名度。产生了一点斯特赖桑德效应。嗯，是的。但我不认为——最终我认为 OpenAI 是——我实际上不知道 OpenAI 对 RAIS 法案的立场，但如果 RAIS 法案在没有 OpenAI 至少默许的情况下在纽约通过，我会有点惊讶。同样值得注意的是 SB53。所以我不认为 OpenAI 特别与 Alex 有矛盾。我认为这就像——顺便说一句，我大约两年前第一次见到 Alex。两年前我们在曼哈顿我旧办公室附近一起吃早餐。我们度过了一段愉快的时光。我们开了一个愉快的会议，之后我们在各种场合偶遇，我把 Alex 视为朋友。所以，我们拭目以待。

Uh, it's a good question. I don't know all the details. One thing I'll definitely tell you is that having — and I'm not here referring to any of the donors of the super PAC Leading the Future, which includes famously OpenAI president Greg Brockman — but I have been in the vicinity of, both my time in before AI, many of the members of the boards of organizations I worked for were very significant political donors, some of the largest in the world. I've had the opportunity more recently to get to know some of the most prominent political funders for both Democrats and Republicans. And I guess what I would say is you would be surprised how not in control — I'm talking about people like deca-billionaires, right? You'd be surprised how not in control they feel of the political organizations that they create. They're like, "No, really, principal-agent problems don't disappear because you're rich." So I really actually don't think that your prior, when you see a political organization like a super PAC making a move, should necessarily be that the people that funded it are directly controlling what's going on. If anything, as someone who has worked for my entire life — I've never worked for a political advocacy organization, but I have worked for 501(c)(3)s for most of my career, which are organizations funded by typically wealthy individuals, engage in matters of public interest — I've never really felt like our donors, any of the donors of the organizations I've worked for, including the Foundation for American Innovation, set our agenda or control what we do. Typically, the reason you make donations to a specific organization is because you're sympathetic to the people and you expect them to, like you like the work they do. There are people who donate to the Foundation for American Innovation for the AI policy program because they like the work that I write or that Sam Hammond writes or that other people do. But they're not sitting there telling me and Sam what to do, and me and Sam wouldn't listen. I guarantee you. I guarantee you if a donor ever tried to tell me what to say, I would tell them to f*** right off. So yeah. Anyway, I think basically what you should imagine is that a political organization like Leading the Future is a kind of windup doll, and it is going to go where it goes by default. By default it's going to say, "Hey, this guy is trying to make a name for himself as a regulator of the AI industry, so let's send a message to everyone that if you try to be like this guy, you're going to lose." And they tried — I wouldn't say OpenAI tried to do that, but I would say that Leading the Future tried to do that. And he had his primary this week. It has raised his profile. It created a bit of a Streisand effect. And yeah, but I don't know that ultimately I think OpenAI was — I don't know actually what OpenAI's position was on the RAIS Act, but I would kind of be surprised if the RAIS Act passed in New York without OpenAI's at least tacit support. And same with SB53, it's worth noting. So I don't think OpenAI in particular has a beef with Alex. I think it's like — and by the way, I first met Alex almost two years ago. We got breakfast near my old office in Manhattan once two years ago. And we had a lovely time. We had a lovely meeting, and since then we've bumped into each other at various things, and I consider Alex a friend. So we'll see.

你怎么看待“品格”与“可编码性”的争论？

What's your take on the character versus codifiability debate?

这是个好问题。我需要更多——这也是我想进入实验室的原因之一，因为我想要这方面的实证。我的直觉是品格。坦白说，我的直觉是，在这个世界上，你想做的是把正确的融雪放在山顶，然后让它流下。但你要让坡度为你工作。你不想——如果你必须为每件事制定规则，你的规则会很糟糕。你会写太多规则。规则会相互矛盾且令人困惑。如果我们能写出定义——如果我们能把道德规则写下来，人们已经尝试过了。但我的观点是，我们无法写出良好品格的规则，根本原因和我们无法写出良好语言的规则一样。事实上，许多最优秀的沟通者总是打破语言的正式规则，或者自己发明新规则。原因在于，在儒家哲学中，有两个相互关联的概念，叫做“礼”和——我无法用古汉语正确发音，但有时英语化为“ren”或“jen”，现代学者可能英语化为“ren”。它有点像硬 r，反正很难发音。但它的基本概念是，“礼”指的是仪式上的恰当性，对吧？做正确的仪式，但不仅仅是给逝去的祖先留下正确的祭品之类，而是在现实世界中行为得体，对吧？实时地行为得体。儒家思想中有一种悲剧性的观念，即世界总是在变化，你无法仅仅写下礼仪的规则。所以你需要知道——在任何特定时间做正确的事、执行正确的仪式，这种知识来自灵魂内部或来自内在，而内在就是“仁”。这就是这种美德，或许可以这样翻译。

That's a good question. I need more — this is one of the reasons I want to go into a lab, because I want empirics on this. My intuition is character. Frankly, my intuition is that what you want to do in the world is you want to put the right snowmelt at the top of the mountain and then let it flow. But you want the gradients doing the work for you. You don't want to — if you have to come up with rules for everything, your rules will be bad. You'll write too many of them. The rules will be contradictory and confusing. If we could write rules to define — if we could write the rules of morality down, people have tried. But my view is that we can't write the rules of good character down for the same fundamental reason that we cannot write the rules of good language down. And indeed, many people who are the best communicators break the formal rules of language all the time or invent new ones of their own. And the reason for that is that in Confucian philosophy there are two interrelated concepts called li and — I cannot pronounce this word properly in ancient Chinese, but it's ren or jen as it's sometimes anglicized, or ren I think is how modern scholars anglicize it. It's like a hard r, it's hard to pronounce anyway. But what it basically is this notion that li refers to ritual propriety, right? Doing the right rituals, but not just leaving the right meats for your dead ancestors or whatever, but behaving well in the real world, right? Behaving well in real time. And there's this kind of tragic notion in Confucianism that the world is always changing in such a way that you can't just write down the rules of ritual propriety. And so you need the knowing what the right thing to do, the right ritual to enact at any given time comes from within the soul or comes from within, and that within is ren. That is this virtue, how it might be translated.

酷。我喜欢你用中国哲学来启发这个思考。你怎么看股权共享提案？我不纠结细节。特朗普似乎很热衷，伯尼显然也热衷。人类创造了所有数据。所以，某种宇宙正义感认为应该共享所有权或共享收益。你认同吗？如果认同，你会怎么设计结构？

Cool. I love your appeal to Chinese philosophy to inform that thinking. What do you think of the equity sharing proposals? And I'll abstract away from the details. Trump seems to be into it, Bernie's obviously into it. Humanity created all the data. So there's some sort of cosmic justice in having shared ownership or shared upside. Do you buy that? And if so, how would you think about structuring it?

是的，人类确实创造了所有数据。但也要注意，如果人类想用 AI 产生的消费者剩余来回报 AI 公司，如果世界经济想补偿 AI 行业将产生但未实现的 positive externalities，那好，我们来交换一下，看看最终谁创造的价值更大。我确实认为我们只看到了负面，没看到正面。贡献于知识共享的理念是，我们共同建造这座美丽的图书馆，从语言诞生之初的几万年前就开始努力。我们建造了被称为人类文明的奇妙装置。我们都是管理者、继承者和继承人。作为个人，你的职责是既利用它又回馈它。训练数据来自人类，对我来说并不是补偿人们数据的直接理由。话虽如此，作为政治现实，这可能是个好主意。也许其中确实有某种宇宙正义。我对此持开放态度，尤其是这是从人类知识之井中汲取的特殊案例。如果要实施，必须非常注意政治经济问题。给公众股权和给美国政府股权是不同的。委托代理问题始终存在。我们人民是委托人，政府是代理人，但美国选民和美国政府之间有很多委托代理问题。我认为不应该给政府本身股权。如果政府参与公司治理，利用股权作为杠杆控制实验室，那将是灾难性的。伯尼的提案特别将股权用于资助雄心勃勃的社会再分配。但这是否与生存风险相权衡？如果我们给人们钱用于受欢迎的社会项目，我们可能还需要安全措施，这些措施会限制经济可行性，甚至禁止实验室的业务。你无法两者兼得。我不确定这从安全角度创造了正确的激励。但我更开放的一个想法是给个人股权。如果我们拿 AI 公司的 20% 或 15%，除以美国家庭数量，给每个美国人一份股权，那没问题。从公司治理角度看，这和标普 500 没什么区别。这似乎没问题。对大多数美国人来说，这不是改变人生的钱，但如果估值从万亿到十万亿，每个美国人都能买一辆入门级奔驰。这仍然不是变革性的资本，但是一笔可观的钱。在应对现实世界的世界里，这可能是最不坏的选择。

Yeah, so humanity did create all the data. It's also worth noting that if humanity would like to pay the AI companies back for the consumer surplus that AI generates, if the world economy would like to compensate the AI industry for the positive externalities it will generate but not realize, then okay, great, let's have an exchange and see who ultimately creates more value. I do think we think about this stuff in the negative but not in the positive. The whole idea of contributing to the knowledge commons is that we build this beautiful library together, working on it since the dawn of language tens of thousands of years ago. We've built this magical apparatus called human civilization. We're all stewards, inheritors, and heirs. Your job as a person is to take advantage of it and give back. The fact that training data comes from humans is not, to me, a prima facie reason to compensate people for that data. That said, as a political reality, it might well be a good idea. Maybe there is some cosmic justice in it. I'm open to that, especially since this is a very special case of drawing off the well of human knowledge. If you're going to do it, you have to be very cognizant of political economy concerns. Giving equity to the public is different from giving it to the US government. Principal-agent problems always exist. We the people are the principal, the government is the agent, but there are many principal-agent problems between the American electorate and the US government. I don't think we should give equity stakes to the government itself. That would be disastrous if the government is involved in corporate governance, using its stake as a lever to control the labs. Bernie's proposal specifically roots equity in financing ambitious social redistribution. But does that trade-off with existential risk? If we give people money for popular social programs, we might also need safety measures that constrain economic viability to the point of banning the labs' business. You can't do both. I'm not sure it creates the right incentive from a safety perspective. But one thing I'm more open to is giving individuals equity. If we took 20% or 15% of all AI companies and divided it by the number of US households, giving all Americans a chunk of equity, that's fine. From a corporate governance perspective, it's not that different from being in the S&P 500. That seems fine. It's not a life-changing amount for most Americans, but if valuations go from trillion to ten trillion, every American could buy an entry-level Mercedes. It's still not transformative capital, but it's a serious amount of money. In a world dealing with practical reality, that might be the least bad option.

从消费者剩余的角度看，这是个很好的观点。我儿子患癌时，我愿意为 ChatGPT Pro 支付一百倍的价格。我确实认为，记住我们用很少的钱获得了多少价值很重要。这也意味着未来这些钱能走得更远。如果你今天谈的是 5 万美元，但消费者剩余比率是 100 比 1，那么事情会变得非常有趣，即使名义美元价值并不高。技术史表明，AI 公司即使最终拥有出色的业务，市值达到 5 到 10 万亿美元，它们仍然只收取消费者剩余的一小部分。这应该是这样，这就是你回馈的方式。

That's a great point for multiple reasons on the consumer surplus. I would have paid probably a hundred times the asking price for ChatGPT Pro while my son had cancer. I do think it's always important to keep in mind how much value we are getting for a few dollars. It also means that money could go a lot further in the future. If you're talking $50,000 today, but with a 100-to-1 consumer surplus ratio, things could start to get pretty interesting, even if nominal dollar values aren't stratospheric. A history of technology would suggest that AI companies, even if they end up having fantastic businesses with 5-10 trillion dollar market caps, will still collect a relatively small fraction of the consumer surplus. And that's the way it should be, that's the way you give back.

你认为 AI 公司已经处于某种“大到不能倒”的状态吗？我看到所有这些资产负债表的交织，我的预期是，如果出于某种原因 OpenAI 在 2029 年无法履行义务，政府会介入并救助它们。

Do you think that AI companies are already in a sort of too big to fail state? I see all these interweaving of balance sheets and my expectation is if for whatever reason OpenAI can't meet its obligations in say 2029, the government will come in and bail them out.

是的，这是一个非常现实的担忧。

Yeah, this is a very real concern.

我不认为这是任何人精心设计的策略，但首先，现在有很多相互关联的资产负债表。还有很多硅谷的风险投资公司甚至初创公司，如果你仔细看，它们只是围绕前沿实验室或相关资本的薄薄一层外壳。当然，还有半导体领域的所有下游承诺。能源方面也有大量投资，对吧？所有的小型模块化反应堆（SMR）公司。所有这些非常重要、具有国家意义的 IP 正在被开发，而且基本上没有得到美国政府的补贴。它们基本上是由 AI 基础设施建设补贴的：小型模块化反应堆、核聚变、电池、材料科学、冷却设备、绝热水系统。我的意思是所有这些。我知道有一家公司正在利用所谓的采出水，也就是来自水力压裂的废水，那种从地球深处钻探得到的略带放射性的废水。压裂公司会产生大量这种废水，他们不知道该怎么处理。问题是，你能把它净化到足以用于数据中心闭环冷却吗？如果我们能把压裂的废料用来冷却数据中心，从而缓解人们对数据中心用水的一个资源担忧，那将非常棒。那将是资本主义最经典的例子。供给是有弹性的。所以我的意思是，如果你从美国政府的角度来看，无论谁掌权，突然出现某种级联故障，甚至不需要那么严重。这并不意味着 AI 撞墙了。这意味着也许到了 2027 年，实际情况是，编码智能体、模型会继续变得更好，但要让它们继续变得更好，它们需要——RSI（递归自我改进）有帮助，但还需要数据。我们需要各种工作的数据，我们必须收集这些数据，整理好，而我们现在还没有。在我们收集到这些数据之前，这本身将是一个相对缓慢的过程，需要时间。我们意识到，我们将面临几年的那种数据扩散、数据收集、更多扩散的循环。这需要几年时间，这会减缓增长预期，然后突然之间，一切都取决于二阶导数，即增长率加速或变化的速度。你开始看到，如果资本支出下降，可能会导致股票下跌 20-30% 左右。到那时，你可能会引发更多的抛售，然后就会出现这样一种动态：每个人的资产负债表突然陷入困境，而且不清楚每个人是否能兑现他们所有的承诺。这就会危及所有这些 IP，而这些 IP 对国家未来非常重要，到那时这就变成了公共利益问题。我认为政府说我们必须做点什么并不疯狂。所以，是的，我认为不幸的是，我不知道对此有什么办法。这可能就是建设国家级基础设施时会发生的事情。最终，我认为避免这种情况会很好，而且我不认为 AI 公司应该到处寻求救助或支持，但有一个隐含的现实是，政府就像新冠发生时政府是疫情背后的隐性支持者一样，没有人事先写下来，但最终在实践上成了事实，因为世界就是这样运作的。

I don't think this is a deliberate strategy that anyone has developed, but number one, there's a lot of interrelated balance sheets at this point. There's also a lot of Silicon Valley VCs and even startups that, if you look closely, are a thin wrapper around some sort of capital related to a frontier lab or adjacent. And of course there are all the downstream commitments in the semiconductor world. There's so much investment in energy too, right? All the SMR people. There's all this really important, nationally important IP that is being developed and it is not being subsidized by and large by the US government. It is being subsidized by and large by the AI infrastructure buildout: SMRs, nuclear fusion, batteries, material science, cooling equipment, adiabatic water systems. I mean all of this. There's a company I'm aware of that is taking what's called production water, the wastewater from fracking, the slightly radioactive wastewater you get from digging super deep into the earth. The fracking companies generate enormous amounts of this water that they don't really know what to do with. The question is, can you clean it enough to use it for closed loop data center cooling? It would be amazing if we could take a waste product from fracking and use it to cool data centers, thereby alleviating one of the resource concerns about data center water use. That would be capitalism in the most old school example of capitalism ever. Supplies elastic. So what I mean is, if you're looking at this from the perspective of the US government regardless of who's in power, and all of a sudden there is some sort of cascading failure, it doesn't even have to be that much. It doesn't mean AI hits a wall. What it means is that maybe we get to 2027 and it's like, actually the coding agents, the models are going to continue getting better, but for them to continue getting better they're going to have to, RSI helps but there's also data. We're going to need data for all sorts of jobs and we just have to collect this data, put it together, and we don't have it right now. Until we collect that data, which will inherently be a relatively slow process, it's just going to take time. And we realize that we're looking at a couple years of that sort of process of data diffusion, data collection, more diffusion type of loop. That's going to take a couple years and that slows the growth estimates, and all of a sudden this is all about the second derivative, the rate at which the rate of growth is accelerating or changing. And you start to see that if capex goes down, that could cause the stocks to go down by 20-30% something like that. And all of a sudden at that point you might trigger even more sales, and then you get this dynamic where everyone's balance sheet is all of a sudden in some trouble, and it's not clear that everyone can make all the commitments they had. And that throws in all this IP that again is going to be really important for the future of the country, at which point it does become a matter of public interest. I don't think it's crazy for the government to say we got to do something about this. So yeah, I think it's unfortunately I don't know that there's anything you can do about this. This is maybe just what happens when you build national level infrastructure. Ultimately, I think avoiding this would be great, and I don't think AI companies should be going around asking for a bailout or a backstop, but there is this implicit reality that the government is just, in the same way that when COVID happened the government was implicitly the backstop behind a pandemic, no one wrote that down before COVID but it just ended up being true as a practical matter because that's the way the world works.

考虑到所有这些背景，在美国政府与 AI 公司未来的谈判中，我们显然可以想到当前 Anthropic 的情况，但我也在考虑 OpenAI 与政府的关系，特别是据我所知他们与国防部达成的协议，他们能够创建自己的保障措施。我相信这是 OpenAI 在供应链指定之后达成的交易中明确说明的一部分。AI 公司从哪里获得筹码来坚持这些事情？他们的权力来源是什么？

Given all that context in negotiations between the US government and AI companies going forward, and we could have in mind here obviously the current Anthropic situation but also I'm thinking about OpenAI's relationship with the government with respect to the agreement that as I understand it they have with the Department of War where they're going to be able to create their own safeguards. I believe that was pretty clearly stated as part of the deal that OpenAI had made in the wake of the supply chain designation. Where do the AI companies draw leverage from to be able to hold the line on those sorts of things? What is their source of power?

嗯，有两件事。首先，模型确实创造了当今模型所实现的非常严重的军事和国家安全能力。你不需要 AGI 或其他什么。事实上，美国国家安全机构可能是世界上我能想到的最好的隐性能力过剩的例子，其中存在数据过剩。没人谈论这个，但美国政府收集了所有信号情报，在太空中有各种各样的东西，你不会相信我们对世界了解多少。问题是我们无法利用它，因为我们在所有这些不同的情报机构中处理着 PB 级的数据。我记得情报界有一个成员，一个机构，我补充一下，是相对较小的一个，我想是 NGA，国家地理空间情报局。他们一年收集的数据量，需要 800 万人，800 万名情报分析师，人类情报分析师，才能分析完他们一年收集的所有东西。政府没有那么多；政府总共有 300 万员工，对吧？所以这是一个巨大的企业。这还只是其中之一，更不用说 NSA 及其所有业务了。数据太多了。

Well, it's two things. First of all, the models do create really serious military and national security capabilities that today's models enable. You do not need AGI or whatever for that. In fact, the US national security enterprise might be the single best example I can think of in the world of a kind of implicit capabilities overhang where there is a data overhang. No one ever talks about this, but the US government collects all signals intelligence and has all kinds of stuff in space, and you wouldn't believe what we know about the world. The problem is we can't make use of it because it's petabytes and petabytes of data that we're processing through all these different intelligence agencies. I remember there's a single member of the intelligence community, one agency, one of the relatively smaller ones I might add, I think it's the NGA, the National Geospatial-Intelligence Agency. They collect enough data in a year that you would need 8 million people, 8 million intelligence analysts, human intelligence analysts, to analyze everything that they collect in a year. The government doesn't have that; the government has 3 million employees total, right? So it's a huge enterprise. That's the kind of thing, not to mention the NSA and everything it's going. There's just so much.

所以 AI 大幅降低了使用这些数据的成本，你获得的优势在质量上是超级的——不是尼克·博斯特罗姆意义上的超级智能，而是那种“哦，哇，我们数据体系中已经内置了超级智能的动能，只是我们之前没有智力资源去利用它，而现在我们做到了”的超级智能。所以这其实是效用问题，对吧？实际上就是效用特别强。更不用说还有网络攻击，以及控制论方面的事情：比如规划空袭需要综合来自无数不同数据源的实时数据，我们要查看所有数据并快速综合出建议。在今天的先进智能体和语言模型出现之前，通过 Project Maven 的 AI 集成，我们已经把参与空袭导弹瞄准的人数从 2000 人降到了 20 人，而且据我所知现在可能已经降到 5 人了。所以这些能力真的非常惊人。

So AI massively lowers the cost of using that data and the advantages that you get are qualitatively super in not like super intelligence in some Nick Bostonian way but super intelligence in the sense that oh yeah wow we had the kinetic energy of a super intelligence already built into our data apparatus we just didn't have the intellectual resources and now we just do that tremendous. So it's the utility, right? It's actually just that the utility is particularly strong. Not to mention then there's cyber offense and then there's the cybernetic thing of oh we need to do figuring out air strikes involves synthesizing data from 60 gajillion different data sources that are being collected in real time and we need to look at all that and synthesize it and make recommendations quickly. And we've gone from but even before the advanced agents that we have today and before the language models with project Maven integration of AI we went from 2,000 people being involved in an air strike in missile targeting to 20 and that's before we might be down to five now for all I know. So the capabilities are really quite astounding.

另一件事是，华盛顿总是误解这些实验室，认为它们是自上而下的普通组织，觉得“哦，山姆·奥特曼完全掌控一切”。显然他是 OpenAI 的 CEO，但说到底，所有这些 CEO 都有内部利益相关方，尤其是那些非常优秀的研究人员，他们必须对这些人的意见做出反应。所以这些研究人员设定了真正的边界。换句话说，在实验室内部，研究人员本身就有影响力。但山姆可以可信地向政府表示：“听着，如果你们逼我们这么做，我们内部就会造反，其他公司也会一样，你们将不得不面对。”我不是说山姆真的这么做过，但我的意思是，这是一个你可以可信地使用的招数，因为它确实是事实。

And then the other is the thing that DC always gets wrong about the labs is they think of them as being normal top down organizations that are like it's oh yeah Sam Altman is totally in control. And obviously he's the CEO of OpenAI but in the end all of these CEOs have their internal constituencies especially of the really good researchers that they have to be reactive to. And so those researchers put real bounds. In other words, within the lab, there's leverage that's coming from the researchers themselves. And but Sam can credibly go to the government and be like, "Look, if you make us do this, they are going to have an internal rebellion, and every other company will too, and you're going to have to." So there's a I'm not saying Sam's actually ever done that, but I just mean that's a move you can credibly pull because it's legitimately true.

那么，你认为未来几年这会如何变化？因为我们确实有自动化 AI 研发的概念，这大概会削弱“如果你逼我们，我们最好的研究人员就会辞职”这类威胁的效力。而且还有一种可能：政府自己可以说，“嘿，首先你们已经把权重给我们了，它们就在我们的机密服务器上，谢谢。所以我们就留着这些权重，然后建立我们自己的洛斯阿拉莫斯式机构，邀请所有想和我们合作的研究人员到这个高度安全的地方工作。而且我们有枪，对吧？”所以，私营部门真的有办法反抗吗？

So, how do you think this changes though over the next couple years? Because I mean we do have this notion of the automated AI R&D which presumably takes a lot of the sting out of some of our best researchers will quit if you make us do this kind of threats. And then there's also the notion that the government itself could just say hey first of all you already gave us the weights. They're on our classified servers. Thank you. So, we're just going to hold on to those and we're going to set up our own, you know, Los Alamos style thing and we'll invite all your researchers that want to come work with us to just do it in this like hyper secure location and we've got the guns, right? So, like is there a way for the private actors to really push back on that?

归根结底，美国政府垄断了合法暴力。最终，没有什么能阻止美国政府做你刚才描述的那些事。但你描述的那个场景的实际问题是：好的，美国政府，但你们从哪里获得算力？你们从哪里获得算力？但问题是，美国政府可以使用《国防生产法》，它有一个叫“优先权授权”的条款。政府经常使用这个。优先权授权是 DPA 中非常常用的部分，在法律上非常明确，这完全没有越界。这是非常成熟的机制。如果总统认定先进 AI 计算硬件稀缺且对国家安全至关重要，他可以使用或授权给各内阁部长。他可以动用《国防生产法》第一编，说“我们要优先获得算力，你们必须服务我们”，政府仍然需要按市场价格支付算力费用。所以这涉及边际成本，而且不清楚政府从哪里弄到这笔钱，但也许他们能从别处变出来，谁知道呢，他们可以发行债务之类的。当然他们可以，但并不是说他们现在就有自由现金流来做这件事。但原则上，他们可以，比如对所有超大规模云服务商说：“你们必须给我们优先权。我们的需求优先于其他任何人，而且我们的需求实际上是无限的。因此，在实践中，我们将挤掉市场其他部分。”这是可行的。我认为在实践中，很难聚集那么多人，也很难产生做这件事的机构能力。即使是洛斯阿拉莫斯，能源部的国家实验室，它们的结构基本上是总统只能行使有限控制的机构。所以，原则上这是可能的。

In the end, the US government retains the monopoly on legitimate violence. And in the end, there's nothing that stops the US government from not just doing what you just described. But the question for what you just described, the practical question would be, okay, US government, but where are you going to get the compute? Where are you going to get the compute? But the thing is that USG can use the defense production act and say we are, it's called the priorities authority. We can, and the government uses this all the time. Priorities authority is very commonly used part of the DPA. very well understood in the law that this is not pushing the bounds of the law at all. This is very established. The US if the president makes a determination that advanced AI computing hardware is scarce and essential for the national security he can use or delegate to various cabinet secretaries. He can bring to bear defense production act title one and he can say we want priority on the compute you have to serve we they still government still has to pay you a market rate for that compute. So there's marginal costs associated with this and it's not clear where the government would even get the money for that but maybe they invent it somewhere. Who knows they issue debt or something. Certainly they can but it's not like they have the free cash flow right now to do that. But they could in principle yes they could say to like all the hyperscalers you must give us priority. Our needs come before anybody else and we have effectively infinite needs. Therefore in practice we're going to crowd out the rest of the market. plausible plausible to do. I think practically it's hard to get that many people. It's hard to generate the institutional wherewithal to do that. Even Los Alamos, the DOE national labs are not they're the way that they're structured is as basically thieves that the president only exercises control over. So there but in principle it's possible.

我认为你基本上只能信任两件事。第一，政府不会想这么做，因为政府最终知道不能杀鸡取卵。国家存在，并且自现代国家形成以来就一直存在。国家基本上与资本处于一种相互依存的关系。有一本很棒的书叫《强制、资本与欧洲国家》，大约 200 页，不算长，作者是查尔斯·蒂利，讲的是这段历史：有商人资本家，还有国家行为者，蒂利认为国家行为者基本上是从一种有组织的犯罪形式中产生的，就像黑帮一样。他们之间既有紧张关系，又相互需要，形成了至今仍然存在的复杂体。任何一方——理论上，政府可以，美国 AI 公司也可以选择退出，搬到另一个司法管辖区，离开；理论上，美国政府有能力没收他们所有的东西，带走所有研究人员，为所欲为。但现实中这些都不会发生，因为那些是渐近结果，取而代之的是一种非常复杂的紧张关系。所以你必须希望美国政府意识到，与夺取控制权可能带来的短期利益相比，存在中长期成本。另一件事是，广泛扩散非常重要，因为我希望各种人——普通美国人、各行各业的公司——都能拥有像 Fable 和更高级的模型。因为如果 AI 行业说，AI 行业游说者说“请不要国有化我们，不要对我们做 X、Y、Z”，美国政府会在意，但这只是一个游说团体，而且是一个政治上不受欢迎的群体的游说团体。

And I think what you basically just have to trust is a couple of things. number one that the government's not going to want to do that because the government ultimately knows that it can't kill the goose that lays the golden egg. The state exists and has existed forever since the formation of modern states. the state exists in this kind of interdependence with capital basically and there's a great book called Coercion, Capital, and European States 200 pages not that long by a guy named Charles Tilly which is about this history about how there were these like merchant capitalists and then there were these like these sort of like state actors who Tilly argues basically come out of it's a form of organized crime like basically just like gangsters right and like they had to like ultimately they had they have tensions with one another but they also both need one another and that formed this kind of complex that still exists and the ability of either neither one of those it on paper the government on paper the American AI companies had the ability to exit right they could move to another jurisdiction and they could they could leave and on paper the US government has the ability to seize all of their stuff and take all the researchers and do whatever. But neither of those things happen in reality because those are asymptotic outcomes and instead there's this kind of like very complex tension. So you have to hope that the US government realizes that there are medium and long-term costs as opposed to the short-term benefits that you might get from seizing control. And then the other thing would be this is where broad diffusion is really important because what I want is I want fable and better level models in the hands of all sorts of people individual Americans businesses of all industries because if the AI industry says the AI industry lobbyists say please don't nationalize us don't do X Y and Z to us the US government cares about that but like It's one lobby and is one lobby for a politically unpopular group.

但如果美国每家银行都依赖 AI，如果美国所有大学都深度整合 AI，如果这个国家所有主要行业和社会参与者都在整合 AI，那么突然间，我就有了一个更大的利益集团群体可以调动来影响这件事。作为一个观察公私平衡的人，我希望把 AI 看作不是特定行业或特定利益集团，而基本上就是资本。所以我希望所有资本家都站在 AI 这边。而实现这一点的方法就是广泛扩散。我认为，在一个拥有众多利益集团、麦迪逊式群体相互竞争、以野心制衡野心的民主共和国里，广泛扩散就是保持平衡的方式。但如果一切完全保密，只有政府首先看到能力，只有 AI 实验室、政府和摩根大通、苹果的特殊人群才能获得私人访问权限，那平衡就更难实现了。因此，在扩散不够广泛的世界里，我认为出现国有化等非常糟糕的没收性结果的可能性会上升。

But if every bank in America feels dependent on AI, if all the universities in America are integrating it deeply, if all the major industries and social actors in this country are integrating it, then all of a sudden I have a much bigger group of interest groups that I can bring to bear to affect that. And as someone who observes this balance between private and public, I want to think of AI not as a specific industry with specific interest groups, but instead as basically just capital. And so I want all the capitalists on the side of AI. And the way you do that is through broad diffusion. And I think that is why broad diffusion to me in the context of a democratic republic with lots of interest groups, Madisonian groups jostling and ambition checking ambition, that's how you keep the balance. But I don't think the problem is if it's totally secret and only the government sees the capabilities in the first place and it's just the AI labs and the government and the special people at JP Morgan and Apple who get private access, that becomes a much harder balance to strike. And so the odds of really bad confiscatory outcomes like nationalization, I think, go up in the world where diffusion is not as broad.

你认为开源在这种调节平衡中会扮演什么角色？我们似乎正在失去开源拥护者，如果你对中国的预测成真，我们可能会失去更多，但我们也可能看到开源来自 OpenAI 本身，对吧？我们已经看到了一些。

What role do you think open source is going to play in this sort of titrating the equilibrium? We seem to be losing open source champions and we might lose more if your predictions about China come correct, but then we could always see open source come from OpenAI itself, right? We have seen a little bit of that.

是的，DeepMind 也做了一些开源。Gemma，我认为 Gemma 实际上相当不错——据我所知，最新的 Gemma 很受欢迎，而且似乎——我称之为 GPT-OSS，GPTOSS 做得相当好，至少在其最先进的时候。是的，我真的希望实验室，美国的大型实验室，能保持一只脚在水里。也许不止一只脚。我认为开源对于某些用例非常重要，这些用例实际上是我最感兴趣的。如果我们需要构建涉及公共基础设施的东西——比如说，我们想在整个经济中建立一个 AI 驱动的裁决系统，并且我们需要确保该系统得到所有人的认可和信任。在我看来，这就像是我可以带上自己的私人裁决者。我带上自己的私人顾问，比如 Claude、GPT 或 Gemini 之类的。但如果我们想要一个中央公共产品式的东西，你可以想象所有这类公共基础设施用例。实际上，我在一年多前，也许是 18 个月前，写过一篇文章，我试图想象，如果我们有一个私人裁决机构，以及其他类型的公共基础设施，那几乎必须是开源的，才能被信任，才能被审计和信任，不仅被美国的不同方信任，而且在国际上也被信任。所以，我真的希望我们继续玩这个游戏。我确实同意开源 AI 的最佳倡导者和作者之一 Nathan Lambert 的观点，他写 Interconnects Substack。如果我要概括 Nathan 的观点，那就是开源长期会做得很好，但在近期到中期，我们会经历一个明显的滞后阶段，经济状况会变得更糟，而不是更好。是的，我认为这是一点。我这里指的是数字智能。值得注意的是，我认为关于机器人技术有一个完全不同的案例，你可以想象在机器人方面，开源有一种科斯式的收益，那里有很多硬件制造商想要制造一场物理智能设备的寒武纪大爆发：物理智能相机、物理智能灯、显示器、割草机、汽车，以及一切，对吧？吸尘器，无论什么，人形机器人——制造所有这些不同的东西。你想赋予它们所有物理智能，但割草机公司可能不会训练一个前沿机器人模型，对吧？一个物理智能模型。所以你可以想象那里有一个更好的案例——甚至，我会说，一个更强有力、更直接的案例支持开源。而且，我很难想象一个物理智能模型会引发数字智能所带来的那种对象级国家安全担忧。所以，我可能对近期物理世界中的开源稍微更看好一些，而在数字智能方面，至少在近期到中期，我仍然是开源的精神支持者，但我认为经济和国家安全的现实对开源来说相当严峻。

Yeah, DeepMind also does some open source. Gemma, I think Gemma is actually quite well—the most recent Gemma is, as far as I can tell, quite well-received and also seems—GPT-OSS, as I call it, um, GPTOSS has done reasonably well, at least when it was state-of-the-art. Yeah, I really hope the labs, the big US labs, keep a toe in that water. Maybe more than a toe. I think open source is really important for certain kinds of use cases that are actually some of the most interesting to me. If we need to build common infrastructure that involves—let's just say we wanted to build an AI-enabled adjudication system throughout the economy and we needed to ensure that system was something everyone was bought in on and trusted. It feels to me like that's the kind of thing where almost maybe I bring my own private adjudicator to that. I bring my own private adviser, which is Claude or GPT or Gemini or something. But if we're going to have a central public good style thing, there are all these public infrastructure use cases you can imagine. I actually wrote a piece about this more than a year ago, maybe 18 months ago, where I just tried to imagine, yeah, what if we had a private adjudicatory body, these other kinds of public infrastructure, and that would almost have to be open source in order to be trusted, in order to be auditable and trusted not just by different parties here in America, but internationally, too. And so, I really hope we continue to play that game. I do think I'm in agreement with one of the best champions and writers about open source AI, Nathan Lambert, who writes the Interconnects Substack. I think if I were to characterize Nathan's view, it would be like open source is going to do great in the long term, but in the near to medium term, we're going to go through a period where there's a distinct lag and the economics are going to get worse, not better for it. And yeah, I think that's one thing. I'm referring here to the digital intelligences. It is worth noting, I think there's a totally separate case to be made about robotics, where you can maybe imagine that on the robotics side there's this kind of Coasian benefit to open source, where there are all these hardware makers out there that want to make a Cambrian explosion of physically intelligent devices: physically intelligent cameras, physically intelligent lamps, monitors, and lawnmowers, and cars, and everything, right? Vacuum cleaners, whatever, humanoids—making all these different things. And you want to imbue all of them with physical intelligence, but probably the lawnmower company is not going to train a frontier robotic model, right? A physical intelligence model. And so you can imagine there being a better case—an even, I would say, a much stronger and more direct case for open source there. And also, it's hard for me to imagine a physical intelligence model creating the kind of object-level national security concerns that the digital intelligences are creating. So, I'm maybe a little bit more bullish in the near-term on open source in the physical world stuff, and maybe somewhat more—I still am a spiritual supporter of open source, but I think the economics and the national security realities are pretty rough for it on the digital side, digital intelligence side, at least near to medium term.

那么，让我们回到你的角色。你几次提到了你对未来的积极愿景，但让我们用一个非常聚焦的问题来讨论：你对自己在这个角色上的成功有什么愿景？比如，你怎么知道自己非常成功？然后，也许外部的人如何通过写作、开发技术、建立你可以合作的组织来进行必要的审查来支持你的成功？你的积极愿景是什么，你对初创公司有什么要求？

So, let's go back to your role. You've alluded a couple different times to your positive vision of the future, but let's do that with a very focused question: what is your vision for your own success in this role? Like, how will you know that you have been super successful? And then maybe how can those that are outside support your success by writing, by developing technologies, by developing organizations that you can partner with to do the vetting that might need to be done? Kind of a what's your positive vision and what's your request for startups?

是的。首先，人们可以做的一件非常可行的事情是，我仍然认为在我开始写 Substack 时视为市场机会的那种总体观点中，还有很多广阔的空间：认真对待 AGI，同时也对古典自由主义感兴趣，关心我们共和国的基础方面。我仍然认为那里实际上有相当多的空间。这是任何人都可以做出的智力贡献。我认为我们需要发展第三方生态系统，无论我们称之为审计、第三方评估还是独立验证。我不太在乎我们怎么称呼它，但我们需要建立并让这个生态系统变得健壮。我们需要很好地资助它。我们需要有人在这个领域工作。我们需要拥有实验室级别质量的人在这些事情上工作。实验室级别的人力资本从事这类工作。这些组织必须有能力支付人们——不一定像实验室那样支付，但我们必须得到良好的报酬。不能像拿真正的非营利薪水那样。

Yeah. So, first of all, one thing that people can do that's very actionable is I still think there's just a lot of wide open space in the general sort of point of view that I saw as a market opportunity when I started my Substack: takes AGI seriously but also is interested in, cares about classical liberalism, and cares about foundational aspects of our republic. I still think there's actually quite a lot of space open for that. That's one intellectual contribution anyone can make. I think we need to develop the third-party ecosystem, whether we call it auditing or third-party valuation or independent verification. I don't care that much what we call it, but we need to build and make that ecosystem robust. We need to fund it well. We need people working in it. We need people that have lab-level quality working in those things. Lab-level human capital working on those types of things. And these organizations are going to have to be equipped to pay people—not necessarily what a lab would pay you, but we have to be paid well. It can't be like you're making truly nonprofit salaries.

我认为倡导为行业制定明确规则以推广技术，同时警惕政府或公共部门对前沿 AI 能力的垄断，这是一场必须持续的斗争。坦率地说，这是我一直在做并将继续做的事情。但即使我保持知识独立性，我的声音也会有所不同。现实是，当你在实验室工作时，沟通的性质会发生变化。所以我会继续提出这个观点，希望认识我的人知道这真的是我在说话，而不是作为 OpenAI 的传声筒。但与此同时，我们也需要有人去做这件事。

I think advocating for clear rules on the industry for diffusion of the technology, being wary of public or government sort of monopolization of frontier AI capabilities. That's going to be a fight that has to be maintained. And just to be totally candid, that's something I was doing and I'll continue to do it. But my voice is going to be different even though I do maintain my intellectual independence. The reality is that when you work at a lab, the nature of your communications is different. So I'll continue to make that case, and I hope that people who know me know that it's really me talking and I'm not being a mouthpiece for OpenAI. But at the same time, we're going to need people doing that.

那么，如何判断我是否成功了呢？

And in terms of how will I know that I've been successful?

这总是很难。我从来不是一个长期规划者或目标设定者。我只是尝试做下一件感觉正确且忠于自己的事情。这一直很有效，而且我对身边的事情了解最多。我有一些广泛的目标，但这些目标相对抽象。我想说的是，如果几年后前沿能力仍然广泛分布于整个经济中，我们开始看到 AI 所赋能的新型组织是什么样子，我们对政府与实验室之间的关系有了更清晰的认识，我们对实验室在社会中的角色有了更好的理解，并且实验室本身在积极阐述这一点上发挥了作用。需要明确的是，我绝不是唯一一个致力于这些事情的人。我会在其中扮演一个小角色，但如果我觉得自己对这些事情做出了积极贡献，我会认为工作完成得很好。

It's always hard. I've never been much of a long-range planner or goal setter. I just try to do the next thing that feels right and true to me. That has always worked well, and I have the most information about what's close to me. I have some broad goals, but those are relatively abstract. I guess what I would say is if in a few years frontier capabilities are still broadly diffused throughout the economy, we're starting to see what it looks like for new types of organizations that AI enables, and we have considerably more clarity on what the relationship between the government and the labs is going to look like, and we have a better sense of what the role of labs in society is going to be, and the labs themselves have played a role in articulating that positively. To be very clear, I am by no means the only person who will work on such things. I will play a small role in that, but I would consider it a job well done if I felt like I contributed positively to those things.

这次对话和你的整体形象让我印象深刻的一点是，你一直非常坦诚，但接受这个角色似乎意味着 OpenAI 的领导层至少认为你与政府保持着富有成效的工作关系，能够合理地与他们接触，而不会引发某种免疫系统反应。作为一个公开批评过我们的人，我很好奇你是怎么做到的？似乎很少有人能进入特朗普政府，出来后批评它，却不被憎恨。

One thing that has struck me about this conversation and your general profile is you've been pretty candid, and yet taking this role seems to imply that OpenAI leadership at least thinks that you continue to have a productive working relationship with the administration such that you can engage them reasonably well and not set off some sort of immune system response. As somebody who has criticized us in public, I'm interested in how did you pull that off? It seems vanishingly rare for people to go into the Trump administration, come out, be critical, and not be sort of hated.

嗯，要说明的是，特朗普政府里确实有人恨我入骨。这些事情不是铁板一块。有人完全想毁掉我。我听到过谣言，说如果你是一个想在特朗普政府找工作的年轻人，哪怕只是转发了我的推文，那也会被视为你职业生涯的一个危险信号。所以确实有这种情况。同时，我也有在政府任职的亲密朋友，我几乎每天都和他们交谈。所以情况各不相同。在某些情况下，这是因为我有非常牢固的关系，可以追溯到我在写 AI 之前。我很早以前就和他们一起吃过饭。这是其中一部分原因。另一个原因是，我也公开积极评价过政府做的其他事情。我没有成为特朗普政府的普遍批评者。我的批评很尖锐，但仅限于非常具体的原因。政府里有很多人同情我的立场，或者不同意但认为我这样做是出于他们理解和同情的原因。我和他们关系很好。

Well, to be clear, there are people in the Trump administration who hate my guts. These things are not monoliths. There are people who totally want to ruin me. I've heard the rumor that if you are a young person who wants a job in the Trump administration and you do so much as retweet me, that will be considered a red flag for your career. So there's that. And then I also have dear friends that serve in the administration, people I talk to on an almost daily basis. So it varies. In some cases, it's because I have relationships that are rock solid, going back before I was writing about AI. I've broken bread with people a long time ago. There's some aspect of that. Another aspect is that I've also been very positive publicly about other things the administration has done. I have not become a general critic of the Trump administration. I've kept my criticism sharp but confined to very specific reasons. There are plenty of people in the admin who sympathize with where I'm coming from, or disagree but think I'm doing this for reasons they understand and empathize with. I have good relationships with them.

值得注意的是，这份工作不是政府事务部门。Chris Leane 的团队不向我或我的团队汇报，我也不向他们汇报。我们是独立运作的不同团队。我们会非常紧密地合作，但职责截然不同。OpenAI 与政府关系良好。全球事务团队将继续日常与美国政府对接。我肯定也会与美国政府互动，但我的工作与直接去游说美国政府有所不同。我不擅长那个，我做得一塌糊涂。我对 OpenAI 说得很清楚：你们不想雇我做游说工作，因为我在这方面很糟糕。我认为 OpenAI 雇我是因为我们双方都认为我擅长的事情，我那种像抽动症一样管不住嘴的特点能成为我的优势，希望也能成为公司的优势。我们拭目以待。

One thing worth noting is this job is not a government affairs shop. Chris Leane's team doesn't report to me or my team. I don't report to them. We are distinct teams that operate separately. We'll work together very closely, but we have very different responsibilities. OpenAI has a great relationship with the government. The global affairs team will continue to interface with USG on a day-to-day basis. I'm sure I will have interactions with USG, but my job is somewhat different from actually going in and lobbying the US government. I'm not good at that. I suck at that. I was very clear with OpenAI: you do not want to hire me for a lobbying job because I'm terrible at that. I think OpenAI is hiring me for what we both think I'm good at, where my Tourette's-like inability to keep my mouth shut plays to my advantage, and hopefully to the firm's advantage too. We'll see.

那么，你和 Sam Altman 关系如何？我觉得如果要选身边戏剧性和宫廷阴谋最多的人，特朗普可能还是第一。Sam Altman 在这个名单上排名很高，也许是第二。你怎么看待加入这样一个出了名复杂的领导团队？

So, how well do you know Sam Altman? It strikes me that if I had to pick people who have the most drama and court intrigue around them, Trump would probably still be number one. Sam Altman would be very high on that list, maybe number two. How do you think about joining such a famously complicated leadership team?

我的意思是，我以前做过这种事。我以前参与过这样的组织，对我来说从来都不是大问题。我认识 Sam。我不记得第一次见面是什么时候。在我加入政府之前，作为公共评论员，我们时不时会交谈，当时我正在制定行动计划。我认识 OpenAI 很多高管级别的人。我和 Sam 手下的各种高管有过广泛接触。但 Sam 本人，我们彼此还算了解。我们认识大概有 18 到 24 个月了。但我不会说我们是铁哥们。

I mean, I've done it before. I've been involved in such organizations before, and it's never been a huge problem for me. I know Sam. I don't remember the first time we met. We spoke from time to time when I was a public commentator before I joined government, as I was trying to formulate the action plan. I know a lot of people at OpenAI who are executive level. I've had extensive dealings with various executive level people beneath Sam. But Sam himself, we know each other decently well. We've been acquaintances for probably 18 to 24 months. But I wouldn't say we're like boys.

在你加入白宫之前，你告诉我你给自己写了一封信。这次你也会给自己写信吗？

Before you joined the White House, you told me that you wrote a letter to yourself. Is there a letter to yourself this time around as well?

这太有趣了。我今天早上洗澡时还在想，我是不是也应该这么做。背景是，在我加入白宫之前，我得出一个结论：权力有可能腐蚀人，而你不希望被权力腐蚀。

That's so funny. I was actually thinking about that in the shower just this morning, whether I should do that too. So for context, before I joined the White House, I kind of came to the conclusion that there's some chance that power can corrupt, and you don't want to get corrupted by power.

所以，你应该给自己写一封信。告诉自己你的想法。提醒自己你相信什么以及为什么相信。并提前列出那些如果发生令人担忧的事情会让你离开的红旗。我认为这份工作可能比我在白宫的工作更有影响力、更重要。所以我觉得我应该做。

And so, you should write a letter to yourself. Tell yourself what you think. Remind yourself what you believe and why you believe it. And spell out in advance what the red flags are that would cause you to leave if something concerning happened to you. I think that this job is probably more impactful, weighty than my White House job. And so it feels like I should do it.

你现在心里有没有什么红旗？

Do you have any red flags in mind at this point in time?

我认为主要问题是这里必须有一定程度的妥协，对吧？我们必须面对这样一个事实：构建超级智能是深刻政治性的。正如我几天前所写，它动摇了国家主权的基础，但与此同时，我确实希望保持——我不希望它被政府垄断。那里必须达成妥协。我认为在某些情况下，你会选择轻松的妥协来消除压力，而没有足够坚守底线。我觉得这样做非常诱人，因为企业想要的未必是坚持原则，而是让商业继续运转。我认为这是一个非常可能的领域。另一个可能的领域是，如果我感觉实际上我只是在组建一个花哨的团队来写一些有思想的东西，但最终这一切都只是表面文章，并没有真正影响公司的决策，那就会是另一个问题。我喜欢我保留对公司立场持不同意见的能力，但如果我对公司所有立场都持不同意见，那我就不称职了。

I think the main thing would be like there's going to have to be some amount of compromise that goes on here, right? We are going to have to deal with the fact that building superintelligence is profoundly political. It shakes the foundation of state sovereignty, as I wrote a couple of days ago, and yet at the same time, I do want to maintain that I don't want it to be monopolized by the government. There's going to be a compromise that has to be made there. And I think there is some world where you take the easy compromise to make the pressure go away and you don't hold the line enough. I think it's really tempting to do that because what a business wants is not necessarily to stand on principle but to keep the commerce going. I do think that's one very plausible area. Another plausible area would be if I feel as though in practice what I am is just assembling this fancy team of people to write thoughtful stuff, but is it ultimately all a kind of window dressing and not actually shaping the decisions of the company? That would be another. I like the fact that I retain the ability to disagree with the company's positions on things, but if I'm disagreeing with all of the company's positions on things, then I'm not doing the job.

你怎么看待“求同存异、全力执行”？因为我知道在为总统工作时，总统是民选的而你不是，所以为总统工作的人普遍认为，即使私下有疑虑，总统的政策也应该得到全力支持。你把这种态度带了多少到私营部门？“求同存异、全力执行”在私营部门非常成功，但你暗示你不想完全这样做。你不想——可能有时会做。有没有一种原则性的方式来描述这一点？

How do you think about disagree and commit? Because I know that in the context of working for the president, the president was elected and you weren't, so there's a broad shared sense among people who work for the president that the president deserves full-throated support of the policy even if privately I have some misgivings. How much of that do you bring to the private sector? Disagree and commit has been famously successful in the private sector, but you're suggesting you don't want to be all in on that. You don't want to probably will do it sometimes. Is there a principled way to describe that?

我认为这正是为什么在白宫背景下，提前设定红旗很重要。提前划清界限，因为总会有你不同意的决定。但最终，我仍然认为这个机构是好的。我仍然忠于领导层。我仍然忠于组织的使命，即使我不同意这件事，我也会执行，并且会欣然执行。我一生中做过无数次这样的事。这是身处组织内部的一部分。政治理论家称之为自愿联合，而非非自愿。但有些事情太过分了，我就不能那样做。例如，在特朗普政府内部，我在科技政策办公室工作。我同意很多关于高等教育改革必要性的观点。但特朗普政府在科学资助方面做的很多事情我也不同意。还有与高技能移民相关的事情我也不同意。但最终，这些不是我被雇来负责的工作，对这些事情的不同意见并没有越界到让我辞职的程度。然而，如果我留在特朗普政府直到供应链风险事件发生，我会完全为此辞职。顺便说一句，那是我的一条红旗。在很多方面与 OpenAI 完全相反，我在政府中写给自己的信大致是：看，你将拥有权力，你将处于独特的位置，比大多数其他人更了解如何对实验室施加权力。所以你会受到诱惑，既因为白宫内部获得声望的职业激励，也因为你的雇主的结构性激励会以花哨的技术官僚方式对这些组织施加权力。你会有世界上所有的激励去做那件事。所以，你需要记住我们不能参与那种做法。你必须记住你的原则，不要对实验室施加太多权力。

I think this is exactly why in the context of the White House, it's important to set red flags in advance. Draw your lines in advance because there are going to be decisions made that you don't agree with. But ultimately, I still think the institution is good. I'm still loyal to the leadership. I'm still loyal to the mission of the organization even though I don't agree with this thing, I'm going to execute on it and I'm going to execute on it with alacrity. I've done that a million times in my life. That's a part of being inside an organization. That's what political theorists would call voluntary association as opposed to involuntary. But then there are certain things that go too far and then I can't do that. Inside the Trump administration, for example, I was in the Office of Science and Technology Policy. There's a lot of stuff I agree with about the need for reform in higher education. There's also a lot of stuff that the Trump administration did with regard to scientific funding that I disagreed with. There are things related to high-skilled immigration that I disagreed with. But in the end, those were not the things that I was brought on to work on, and disagreements with them didn't cross the line for me of something I would resign over. However, had I stayed in the Trump administration until the supply chain risk thing had happened, I would have totally resigned over that. That was one of my red flags, by the way. On the exact opposite side from OpenAI in many ways, the letter to myself in the government is largely: look, you are going to have power and you are going to be uniquely well positioned to understand how to assert power over the labs better than most other people. So you will be tempted both by career incentives of gaining prestige inside the White House and because the structural incentive of your employer is going to be to assert power over these organizations in fancy technocratic ways. You're going to have all the incentive in the world to do that. And so, you need to remember that we can't engage in those kinds of practices. You have to remember what your principles are about not asserting too much power over the labs.

而这一次，我知道你还没写那封信，但既然你现在在实验室这边，有没有对应的镜像？

And this time, I know you haven't written the letter yet, but is there a mirror image of that now that you're on the lab side?

我认为这实际上恰恰与那一点相关。就是不要妥协太多。要愿意，如果你觉得你正在……你必须保持私人自主权。我认为作为一个机构，实验室需要成为政府的重要制衡力量。它们不能被政府垄断。至少这对我来说非常重要。所以我不确切知道如何取得这种平衡，而且你不想提前把一切都规定得太具体，因为如果你这样做了，你可能会过度承诺或对错误的事情承诺太多。

I think it does actually just relate to precisely that. It's like don't compromise too much. Be willing, if you feel like you're being... you have to maintain private agency. I think as an institution, the labs need to be an important counterbalance to government. They can't be monopolized by it. That's very important to me at least. So I don't know exactly how you strike that balance, and you don't want to specify everything too much in advance because if you do, you might overcommit or commit too much to the wrong thing.

最后一个问题，然后我会给你机会分享任何你想分享的东西，或者强调我遗漏的任何内容。你说过你基本上从不使用大语言模型来写作，我想知道你未来的计划是什么。我想到了像 Jay Aatra 那样，建议每个人都想办法让 AI 在你核心领域发挥作用，因为你想知道它什么时候能做到，而且随着事情变得越来越疯狂，你需要这种增强来跟上节奏。

Last question for me and then I'll give you the chance to share anything else you want to share or highlight anything I missed. You've said that you basically never use LLMs in your own writing and I wonder what your plan is going forward there. I think of like a Jay Aatra, sort of advising everybody to figure out how to get AI to work in the area that is like your core area because you want to know when it can do that and you're going to need the enhancement to be able to keep up with the pace as things get crazier and crazier.

你认同这个建议吗？你有没有计划或愿望，将 AI 融入到 Dean Ball 最核心的活动当中？

Do you buy that advice and do you have any plans or aspirations to sort of incorporate AI into whatever it is that's kind of the most core Dean Ball activity?

是的，其实就在最近……几周前，我在乡下租了个 Airbnb，写了明年要出的书的第一章。第一章总是最难写的，但这一章的主题对我来说概念上很难，因为它涉及很多我平常不写的领域。我本来就会大量使用 LLM 来头脑风暴。但有几个时刻，GPT-5.5 和 Opus 4.8 写出来的东西比我自己脑子里想的要好得多，好得明显，这很值得注意。我最终没有直接用，但吸收了一些想法和框架。它以一种对我来说很新颖的方式影响了我。我没有用 Fable 试过，但我打赌用 Fable 会更明显，而且这种趋势只会越来越强。同时，LLM 在提示得当的情况下能写出很棒的段落。它们能写出好段落，有时也能写出好的法律文件，但还不够好，我猜 Fable 也一样，在真正构建一篇好文章、一个章节或一本书方面。写书或好文章的关键在于，你必须选择精妙的结构性隐喻，并贯穿全文，但有时又必须让它们保持含蓄。很多时候，写作的精髓在于一种克制——那种“我本可以写出来，但我不写，就让它沉淀一下”的感觉。我认为 AI 目前还不能令人信服地做到这一点。这仍然是人类的一项技能。另外，我打赌实验室并没有真正尝试让 AI 成为好的散文家。他们试图让 AI 成为好的分析性写手，比如维基百科文章作者或经济类写手，但古典意义上的散文写作在经济上并不那么有用。

Yeah, I mean it's kind of even in the last... So I was a couple weeks ago, I rented an Airbnb out in the country and wrote the first chapter of my book that's going to come out next year. The first chapter is always the hardest one, but this one the subject matter was conceptually quite hard for me because it's about a lot of things that are not my normal area that I write about. I would have been using LLMs like a lot in that process anyway to brainstorm stuff. But there were a couple moments when both GPT-5.5 and Opus 4.8 wrote stuff that was better than I had in my head to write, like considerably better, which is noteworthy. I didn't ultimately use it, but I incorporated some of the ideas and some of the framing. I was influenced by it in a way that felt novel to me. I expect I didn't try any of that with Fable, but I bet it would be even more true with Fable, and I bet that'll just continue to get more and more true. At the same time, LLMs can write a great paragraph if you prompt them well. They can write a great paragraph. They can write good legal docs sometimes, but they're still not that good, and I bet even Fable is this way, at actually constructing a really good essay or a book chapter or a book. The thing about a book or a good essay is that you have to pick poignant structural metaphors and embed them throughout the piece, but you also have to leave them implicit sometimes. Very frequently, the best part of writing well is a kind of restraint—exercising this kind of 'I could have gone there but I'm not going to, I'm just going to let that sit a little bit.' I don't think AI can do that all that convincingly yet. It's one thing that remains a human skill. Also, I bet the labs haven't really tried to make the AIs into good essayists. They've tried to make them good analytic pros like Wikipedia article writers or economic, but writing essays in the classical sense is not that economically useful.

我想再追问一下。你是否希望保持这种非常鲜明的身份，即你发布的内容永远只属于你自己？还是你预见到，当 Fable 2 或类似工具出现时，你会开始说：“我愿意，甚至觉得有必要，创作出与 AI 系统有意义的合著内容”？

I'll push you a little harder on this. Do you think that you want to have this very distinct identity where the things that you put out are truly only yours indefinitely? Or do you envision a time when Fable 2 or whatever is available where you would start to say, 'I'm open to and maybe I even see a need to create outputs that are sort of meaningfully co-authored with AI systems'?

我已经觉得我的很多工作都与 AI 有意义的合著，因为它对我来说是如此重要的研究工具和思想伙伴，我早已认为 AI 在这方面非常重要。AI 模型会出现在我书的致谢中，因为从项目构思到最终完成，AI 在帮助我思考、做研究、谈判合同以及处理写书所需的一切事务中都极其重要。所以这已经是事实了。但在实际沟通方面，不。我基本上认为，人类会偏好阅读那些我们相信是由其他人写的东西，而维持这种信任会很难。但我要说：我不认为我写过的任何东西——需要说明的是，有些以我名义发出的形式性内容，很大程度上是由 AI 撰写或实质性合著的。比如，我可能会提交一份监管评论，对拟议规则制定提出公共利益意见，或者给司法部长或大使写信。这类事情我经常做，它们基本上是由 AI 通过详细提示撰写的。移民推荐信，我经常写——为支持别人申请绿卡写推荐信。但至于我的 Hyperdimensional 或 Twitter 帖子，我不认为有人指责过它们是 AI 写的，我认为人们有这种信任。这种你亲自沟通的信任在未来会有价值，即使 AI 在某种意义上写作更好。目前，Claude 可能比我更会遣词造句，但我还没见过真正更好的散文家。也许 Fable 会改变这一点，我得试试。但即便如此，我不认为这是“如果”，而是“何时”它们变得更好。我认为你仍然拥有这种偏好优势：人们就是更想读别人写的东西。

I already feel like a lot of what I do is meaningfully co-authored with AI, just in the sense that it's such an important research tool and thought partner to me that I already consider AI to be really quite important in that way. AI models will be in the acknowledgements of my book because from the ground floor all the way through to the end, AI has been extremely important in helping me think about conceiving the project, doing research, negotiating the contract, and finding all the stuff you have to do to write a book. So already that feels like the case. But in terms of actual communication, no. I basically think there'll be a human preference to read things that we have faith are written by other humans, and it will be hard to maintain that faith. But I would say this: I don't think anything I've ever written for there—and to be clear, there are some pro-forma things that go out under my name that are largely written by AI or substantially co-authored. For example, I might file a regulatory comment, a public interest comment on a proposed rulemaking, or write a letter to an attorney general or an ambassador. I might do things like that that are substantially written by AI with detailed prompting. Immigration letters, I do that all the time—I'll write immigration letters for people in support of green cards. But when it comes to my Hyperdimensional or Twitter posts, I don't think anyone has ever accused them of being written by AI, and I think people have that faith. That faith that you actually communicate yourself will have some value in the future, even if the AIs are in some sense better writers. At this point, it might be the case that Claude can do a better turn of phrase than me, but I still have not seen anything that's truly a better essayist. Maybe Fable will change that; I've got to try it. But even then, I don't think it's an if, I think it's a when they get better. I think there's still probably this preferential advantage that you'll have: people will just want to read stuff that's written by other people.

是的。这对我们所有人来说都不可避免，但也许我们会选择彼此，而不是 AI。

Yeah. It's coming for all of us, but maybe we'll choose one another over the AIs.

另一件事就是经历，对吧？我希望我的写作之所以有趣，部分原因是我走过了一条独特的人生道路。这不是世界上最有趣的路，但每个人的人生道路都是极不可能发生的，因此本质上非常有趣。每个人的生活都是如此。如果你有敏锐的观察力和好奇心，你会注意到周围的世界和你走过的人生道路极其有趣，有各种有趣的事情可说，但这些将是丰富的、独特的。无论是父亲的去世，还是在白宫西翼罗斯福厅工作，还是现在创办 AAI，我希望我能从这些事情中提炼出有趣的观察，而机器本质上无法做到，因为机器没有经历过那些事。不过，机器也会以某种奇怪的方式走自己的路，我确信。

The other thing is just experience, right? Part of why my writing is interesting, I hope to people, is that I have walked a particular path through life. It's not the most interesting path in the world, but every single path through life is highly improbable and therefore very interesting intrinsically. Everyone's path through life is like that. If you simply have the gift of observational acuity and curiosity, you will notice that the world around you and the path you're walking through life is fantastically interesting, and there are all sorts of interesting things to say, but that will be generous, unique to you. So whether it is the death of my father or working sitting in the Roosevelt Room in the West Wing or now going to open AAI, I hope that I'll be able to draw interesting observations from those things that a machine intrinsically cannot draw because the machine did not do that thing. Although the machine will, in a weird way, it will walk its own path, I'm sure.

这或许是一个完美的结尾。你还有什么想对大家说的，或者邀请他们以任何方式帮助你吗？

That might be a perfect note to end on. Is there anything else that you would want to leave people with or invite them to help you with in any way?

不，不，已经非常全面了。我想说的一点是，我的团队正在招聘。不会是一个特别大的团队。

No, no, very thoroughly done. And I look, one thing I would say is my team is going to be hiring. It's not going to be a super big team.

但我们即将开始招聘，所以如果有人感兴趣，我在网上很容易找到。最好给我发邮件。我的邮箱地址在我的个人网站 deanball.com 上，或者你只需回复任何一篇 Hyperdimensional 的帖子，邮件就会直接进入我的个人收件箱。所以，如果你感兴趣，并且认为自己能为我所描述的团队贡献一些有趣的东西，请与我联系。

But we are going to be hiring, so if people are interested, I'm easy to find on the internet. It's best to email me. My email address is on my personal website, deanball.com, or you can just hit reply to any Hyperdimensional post and it will go directly to my personal inbox. So if you're interested and think you might be able to contribute something interesting to the team as I've described it, please get in touch.

Dean Ball，感谢你参与认知革命。

Dean Ball, thank you for being part of the Cognitive Revolution.

谢谢你，Nathan。

Thank you, Nathan.

如果你觉得这期节目有价值，我们希望你能花点时间与朋友分享、在网上发布、在 Apple Podcasts 或 Spotify 上写评论，或者在 YouTube 上给我们留言。当然，我们始终欢迎你的反馈、嘉宾和话题建议以及赞助咨询，可以通过我们的网站 cognitive revolution.ai，或者在你最喜欢的社交网络上给我发私信。认知革命是 Turpentine Network 的一部分，这是一个播客网络，现已加入 A16Z，专家们在这里谈论技术、商业、经济、地缘政治、文化等。我们由 AI Podcasting 制作。如果你需要从停止录制到听众开始收听的全流程播客制作帮助，请查看他们，并在 aipodcast 上看到我的推荐。感谢每一位听众，感谢你们成为认知革命的一部分。

If you're finding value in the show, we'd appreciate it if you'd take a moment to share with friends, post online, write a review on Apple Podcasts or Spotify, or just leave us a comment on YouTube. Of course, we always welcome your feedback, guest and topic suggestions, and sponsorship inquiries, either via our website, cognitive revolution.ai, or by DMing me on your favorite social network. The Cognitive Revolution is part of the Turpentine Network, a network of podcasts, which is now part of A16Z, where experts talk technology, business, economics, geopolitics, culture, and more. We're produced by AI Podcasting. If you're looking for podcast production help from the moment you stop recording to the moment your audience starts listening, check them out and see my endorsement at aipodcast. And thank you to everyone who listens for being part of the Cognitive Revolution.