山姆·奥特曼谈埃隆·马斯克、《人造》与人类未来（上）

来自《名利场》。这里是 Fair Game。我是 Mark Aduchi。每天，似乎都有新的科学突破或生存威胁，让世界对那个据说正迅速逼近的 AI 未来更加警觉、更加焦虑、也更加不信任。而作为人工智能——无论是作为技术、产业还是文明转型——的代言人，没有人比 OpenAI 联合创始人兼 CEO Sam Altman 更受瞩目。正因如此，Sam Altman 登上了本月《名利场》的封面，并在我们的新权势榜上位列第一。我在旧金山与 Altman 进行了两次访谈。第一次在 9 月 11 日，然后就是今天，10 月 2 日。在这几周里，关于 AI 的叙事似乎翻来覆去变了五六次。在我们的对话中，我们聊了从 Sam Altman 的生活、他的孩子、他的竞争关系、鼓励生育、为末日做准备、Luca Guadagnino 的《Artificial》，到他是否觉得自己被社会错误地塑造成了反派。我们谈了他的公司、他的新产品、推迟的 IPO、他的薪酬、OpenAI 高管近期的离职潮，也谈了技术本身、放缓发展、政府监管、行业自律、习近平主席的国事访问，以及保护人命，包括自杀案例。开始前我要说明，《名利场》的母公司康泰纳仕与 OpenAI 有内容授权协议。两次 Altman 访谈现在都已上线。这是第一部分。

From Vanity Fair. This is Fair Game. I'm Mark Aduchi. Every day, a new scientific breakthrough or existential threat seems to make the world more aware, more anxious, and more distrusting of an AI future that we're told is rapidly approaching. And nobody is more recognized as the face of artificial intelligence, as a technology, an industry, and as a civilizational shift than OpenAI co-founder and CEO Sam Altman. For that reason, Sam Altman is on Vanity Fair's cover this month and sits at number one on our new establishment list. I sat down with Altman for two interviews in San Francisco. The first on September 11th and then again today, October 2nd. In those weeks, the story about AI seemed to turn upside down about half a dozen times. And over the course of our conversations, we discussed everything from Sam Altman's life, his children, his rivalries, pronatalism, prepping for the apocalypse, Luca Guadagnino's Artificial, and whether he thinks he's been mischaracterized as a villain in society. We talked about his company, his new products, the delayed IPO, his compensation, the recent exodus of OpenAI executives, and we talked about the technology itself, slowing down, government regulation, industry self-policing, President Xi Jinping's state visit and protecting human life, including in the case of suicide. I should note before we begin that Condé Nast, Vanity Fair's parent company, has a content licensing agreement with OpenAI. Both interviews with Altman are live now. This is part one.

Sam，很高兴见到你。

Sam, it's good to see you.

我也是。

You too.

我们正处在一个非凡的时刻，我认为，对 AI 而言，无论是能力、风险还是公众对两者的认知。这感觉像是一个分水岭。我只想知道，Sam，过去这一个月对你来说是什么样的？你睡过觉吗？电话是不是响个不停？各国领导人是不是都在打电话找你寻求答案？这感觉像 2022 年你推出 ChatGPT 的那个时刻，有些东西已经改变，还是只是你生活中普通的一周？

We are speaking at an extraordinary time, I think, for AI in terms of the capabilities and the risks and the public awareness of both. This feels like a watershed moment. And I just want to know, Sam, what has the past month been like for you? Have you slept? Is your phone ringing off the hook? Are world leaders calling you asking for answers? Does this feel like the 2022 moment when you launched ChatGPT where something has changed or is this just an average week in the life?

说实话，两者都有一点。我想，从内部体验来说，某种意义上，过去十年我们一直在谈论这件事、为此做准备。我们一直看到这种指数级的进步，也一直困惑为什么世界上更多人没有像我们这样看到进步，因为对我们来说，这一刻会到来是非常可预测的。

A little bit of both, to be honest. I think the internal experience of this is we have been talking about this and preparing for this for the last decade in some sense. And we have been seeing this exponential progress and we have been confused about why more of the world has not seen progress the way that we've seen it because to us the fact that this moment would come has been very predictable.

你觉得你理解现在外面的人有多害怕吗？你觉得有必要安抚人们，或者帮助他们理解这件事吗？

Do you feel like you understand how frightened people are out there right now and do you feel a need to reassure people or help them understand it?

当然。两个都是肯定的。

Of course. Yes to both.

两个都是肯定的。

Yes to both.

如果你回到 10 年前，读读我们创办公司时谈的内容，和此刻非常吻合。区别在于，10 年前很少有人相信我们，现在大多数人都相信我们了。

If you go back 10 years and read what we talked about when we started the company it matches this moment very well. Now the difference is 10 years ago very few people believed us now most people believe us.

既相信 AGI 会到来，也

Both that AGI is coming and

它已经来了吗？

Is it here?

这是主观的，对吧？

It's subjective, right?

是主观的，对，很接近了。我想可以说，我们在最初设定的目标上非常成功，现在我们正经历一次重新校准，一个社会说“好，这要来了”的阶段，而社会大体上仍在照常运转，我们有了这个不可思议的新东西，而且我认为很公平地说，世界并没有改变那么多。你知道，我们曾受到、现在仍受到但最近少一些的一个批评是：你说过你要做这些了不起的事，那对世界的影响在哪？经济增长在哪？癌症的解药在哪？这个那个在哪？

It's subjective, yeah, it's close. I think you could say we've been very successful at kind of what we set out to do and now we're going through a recalibration and a kind of like a period where society is like okay this is coming now society is still continuing much the same we have this incredible new thing and I think it's very fair to say the world hasn't changed that much and you know I think a criticism that we got we still get but less recently is you said that you said you were going to do these amazing things and like where's the impact on the world. You know, where's the economic growth? Where's the cure for cancer? Where's the this?

然而确实感觉有些重要的东西正在改变。

And yet it does feel like something important is changing.

我们是众多参与者之一，但我们想，你知道，与社会中所有其他利益相关方合作。

We are one player of many, but we want to, you know, work with all of the other stakeholders in society.

你是众多参与者之一，但特别是你，Sam，在很多方面是人工智能的人类面孔。你先到了那里，你作为个人和公司成了品牌。我认为很公平地说，如果这是曼哈顿计划，你就是奥本海默。

You are one player of many, but you specifically, Sam, are in many ways the human face of artificial intelligence. You got there first and you became the brand as a person and as a company. And I think it's fair to say if it's the Manhattan Project, you're the Oppenheimer.

我不是第一个这么说的人。我很好奇那是什么感觉？

I wouldn't be the first person to say that. And I'm curious just what does that feel like?

我会回答。这不是我想搪塞过去，也不是假谦虚之类的，而是这在多大程度上是集体努力，但主要是，这是一个关于相对少数杰出科学家和工程师做了了不起事情的故事。而远不是我个人的故事，我认为这一点没有被很好地理解。我认为有很多人可以完成我做的这部分，那些关键的研究人员和工程师，我认为他们现在得到了认可，但这是我们在任何科学或工程学科中很长时间以来见过的最重要、最令人印象深刻、最具影响力的事情之一。Kevin Roo 的新书很好地呈现了一些配角部分。

I will answer that. This is not me trying to brush it aside or like false modesty or anything, but the degree to which this is a collective effort, but mostly that this is about a story of, you know, a relatively small number of brilliant scientists and engineers that did something amazing. And much less a story about me I think is not well understood. I think there are many people that could have done the part of this that I did and the kind of key researchers and engineers and I think they are recognized at this point but it is like it is one of the most important and impressive and impactful things that we've seen in a very long time in any in any scientific or engineering discipline. Kevin Roo's book does a good the new book does a good job of bringing out some of the supporting parts.

确实，你知道，我像任何人一样、甚至比任何人更多地成了这整件事的面孔。也成了很多焦虑、恐惧和其他一切的宣泄口。但至少可以说，这是一种奇怪的体验。

It is true that I got, you know, as much as anybody or more than anybody else like the face of this whole thing. And the sink for all the a lot of the anxiety and fear and everything else. But it is it is a strange experience to say the least.

我想你可以——你这么说，我要更进一步。我认为此刻很多人把 AI 视为一种邪恶力量，而因为你从一开始就与之关联，你被视为反派。你觉得你被错误地刻画了吗？这让你沮丧吗？

I think you could you're saying this, I will go further. I think a lot of people at this moment see AI as an evil force and because of your association with it from the jump, you're seen as a villain, do you feel you've been mischaracterized? Is that frustrating to you?

我确实觉得我被错误地刻画了。只是有点沮丧。

I do feel I've been mischaracterized. It's only somewhat frustrating.

好。

Okay.

我认为如果是别人做我的工作，人们也会有同样的感受、说同样的话，只是可能不会和那个人产生共鸣。我认为这其实不是关于具体某个人。而是人们需要一个焦虑的宣泄口、一个替罪羊，就像，天哪，这种变化正在降临世界。

I think if there were somebody else doing my job, people would have the same feelings and say the same things that probably wouldn't resonate with that person. I think it's not really about the specific person. It is that people need a sink for the anxiety, a scapegoat, a just like, man, this change is happening to the world.

所以你可以把人和事分开。

So you can divorce a person.

你可以把它和你对自己的感受分开。

You can divorce it from how you feel about yourself.

我有点往心里去，但没那么往心里去。我理解这里的人类本能。在以前的技术时代，我也曾对别人有过这种感觉。

I take it somewhat personally, but not that personally. Like I I understand the human instinct here. I have felt this way towards other people in previous eras of technology.

我现在为此感到抱歉。但我想我理解这种人类本能。那是谁？

I feel bad about it now. But I think I get the human instinct. Who was that?

嗯，我亲自认识、愿意谈的一个例子是 Mark Zuckerberg。在 Meta 早期，我脑子里对他有一个非常邪恶的漫画式形象。当时我不认识他。我和他不是很熟，但我现在认识他。我想我当时是在投射人们投射到某个他们认为对世界拥有过多权力和影响力的人身上的东西。

Well, one example of someone that I know personally that I'll talk about is like Mark Zuckerberg. I had like a very evil caricature of him in my head in the early days of Meta. I didn't know him at the time. I know him not super well, but I know him now. I think I was like projecting whatever people project onto someone with what they perceive is too much power and influence over the world.

Sam，你是两个孩子的家长。我没有孩子，但我所有有孩子的朋友都说，有孩子会极大地改变他们对未来的思考方式。

Sam, you are a parent of two. I don't have kids, but all of my friends who do talk about how having kids dramatically changes how they think about the future.

我只能想象，对你这样可能比大多数人更了解我们的未来、又在塑造未来上有独特角色的人来说，这一点更是如此。有了孩子之后，你做工作的方式变了吗？

I can only imagine that's truer for you as someone who probably understands our future a lot better than most, and has a unique role in shaping it. Has having kids changed how you do your job?

变了，但可能不是你会预期的那种方式，或者也许你也能猜到。有孩子是我这辈子做过的最好的事，远远超过其他一切。我觉得自己非常幸运，人生中发生过很多美好的事，但有孩子是第一名，而且遥遥领先。人们告诉我的所有话，包括“你没法真正理解这有多好”，全都是真的。这是最奇妙、最大的福气、最有趣的事。在有孩子之前，我就非常坚定地不想毁掉世界。我不需要有孩子才会觉得，好吧，我真的不该拿文明的未来去赌。我有点惊讶，居然有那么多人说，哦，我很高兴你有孩子了，因为这样你就不会为我们所有人去冒那种疯狂的灾难性风险了。可我本来也不会那么做。但说到我为了做别的事而离开孩子所付出的代价，从“我真正在乎什么、想优先什么”的角度看，这反而让我非常专注。然后看着孩子体验世界的方式，去想：我们显然不想毁掉世界，但我们想保留什么、想建设什么样的世界、这些神奇的人类体验是什么，以及我们希望孩子如何使用这项技术，能够拥有能动性、发展创造力等等。

Yes, but probably not in the way that you would expect, or maybe you might expect. Having kids is by far the best thing I've ever done. I feel very blessed to have had many wonderful things happen to me in my life, but having kids is the number one by a huge margin. Everything people told me, including that you can't really understand how good it is, has all been true. It is the most amazing, greatest blessing, most fun thing ever. I was very intent on not destroying the world before having kids. I didn't need kids to be like, okay, I really shouldn't gamble with the future of civilization. And I am somewhat surprised about the number of people who have said, oh, I'm really happy you have kids because now you won't take crazy catastrophic risk for all of us. Like, I wasn't going to do that anyway. But in terms of the premium I put on being away from my kids to do anything else, it's been wonderfully focusing from the perspective of what I actually care about and want to prioritize. And then watching the way kids experience the world and thinking about, obviously we don't want to destroy the world, but what do we want to preserve and what kind of world do we want to build, and what are these magical human experiences, and how do we want kids to use this technology and be able to have agency and develop their creativity and everything else.

那你觉得你什么时候会开始让他们接触 AI？这些问题人们也会问自己，但放到自己孩子身上，共鸣就不一样了。

So when do you think you'll start to introduce them to AI? It brings up questions that one would ask for themselves, but takes on different resonance when it's for your children.

每个人都会为自己的孩子做自己的决定，但我非常赞成对孩子全面限制技术。我觉得孩子就该在外面跑来跑去、在大自然里玩，做孩子们长期以来一直在做的事。我不认为我们完全理解技术对幼儿的影响，无论是好是坏。但作为一个为自己孩子做决定的家长，我肯定会说，当我看到孩子们黏在某种触摸屏上、刷着那种“脑腐”短视频时，我强烈地觉得那是件坏事。

I mean, everyone's going to make their own decisions about their own kids, but I'm a big fan of limiting technology across the board for kids. I think it's good for kids to be running around outside and playing in nature and doing whatever kids have done for a long time. I don't think we understand all of the impacts of technology on small children, for good or for bad. But as a parent making decisions for my own kids, I will certainly say when I watch kids that are glued to a touchscreen of some sort, scrolling through brain rot short form video, I have a strong instinct that that's a bad thing.

嗯。

Mhm.

而且我的意思是，对成年人来说可能也是坏事。

And I mean, it may be a bad thing for adults too.

我正想说，成年人也这样。

I was going to say, adults do that too.

但发育中的大脑，似乎尤其需要保护，免受我们尚未完全理解的新技术影响。我对 AI 也是同样的感觉。显然，今天出生的孩子会在一个有 AI 的世界里长大，也需要会用它。我怀疑幼儿期是不是合适的起点。我还没有想好一个年龄，说到了那时我会让孩子跟 ChatGPT 对话，但肯定不是现在。

But something about a developing brain seems especially important to protect from new technology we don't fully understand. And I feel the same way about AI. Clearly kids born today are going to grow up in a world with AI and need to be able to use it. I doubt early childhood is the right point for that. I don't have an age in mind where I would let my kids talk to ChatGPT yet, but it is certainly not now.

你说你喜欢有孩子。我想知道，你打算再要更多吗？

You said that you like having kids. I wonder, do you plan to have more?

是的。是的。对。

Yes. Yes. Yeah.

多少个，你知道吗？

How many, do you know?

就算只是要两个孩子，我记得我和 Ollie 聊过这个。时间真的很难分配，因为你想参与每一件事，想花很多时间陪他们。所以如果有无限的时间，我会想生很多孩子，但我也想对每一个孩子都非常投入。我知道 Ali 也是这么想的，那就会是限制所在。但如果有无限时间，我会想要很多孩子；而时间有限的话，就不会是一支孩子大军。

Even going to two kids, I remember Ollie and I were talking about this. It's so hard to split time, because you want to be there for everything. You want to spend so much time with them. So with unlimited time, I would want to have so many kids, but I want to be so present for each one of them. And I know Ali feels the same, that that will be the limit. But given unlimited time, I would want to have a lot of kids, and given limited time, it will not be an army of kids.

硅谷这边有一种“鼓励生育”的趋势，说要多生孩子来对抗人口下降。那是……

There's a trend out here in Silicon Valley about pronatalism and about having more kids to combat population decline. Is that...

我只是觉得，作为一种趋势，这很难理解。我非常爱我们的孩子，我也知道我会再要孩子，但其中没有任何一部分是“我们得对抗人口下降”。我就是喜欢大家庭。真的很棒。

I just find that tough to understand as a trend. I love our kids so much and I know I'll have more kids, but there's no part of it which is like, we gotta fight against population decline. I just love a big family. It's really great.

那么 Sam，你即将在 Luca Guadagnino 的《Artificial》里被拍成传记片。

So Sam, you're about to get the biopic treatment in Luca Guadagnino's Artificial.

哦，对。

Oh yeah.

哦，对。你看过预告片吗？

Oh yeah. Have you seen the trailer?

我看过预告片。

I did see the trailer.

好。你觉得怎么样？

Okay. What'd you think?

他走路确实像我。其实我跟你说件事。他们在我家外面拍摄。我们那条街上贴了通知，说会有拍摄之类的。我当时没意识到是怎么回事，完全没在意。然后有一天早上，我沿街走去买咖啡，那里全是摄像机，还有一个人穿得跟我一模一样，背着和我一样的背包，走路也特别像我，以至于我觉得自己像是掉进了模拟世界里的某种诡异故障。非常奇怪。然后我立刻明白了发生了什么。我们挥了挥手就走了。但那真的很怪，看到有人把你那些古怪的小动作都抓得那么准，是种很奇怪的感觉。超现实。不过你还没跟 Andrew Garfield 说过话之类的吧。

He did walk like me. Actually, I'll tell you something. They filmed outside of our house. There were these notices on our street that there was going to be some filming or whatever. I didn't realize what was going on. I paid no attention. And then one morning I was walking down the street to get coffee and there were all these cameras and there was a guy dressed exactly like me with my same backpack, walking like me so much that I felt I was in some weird glitch of the simulation or something. It was very strange. And then I immediately figured out what was happening. We waved and went on. But it was really weird, and it's a strange thing to see someone that has nailed your quirky mannerisms or whatever. Surreal. But you haven't spoken to Andrew Garfield or anything.

我那次朝他挥了挥手。

I waved at him that one time.

他那次朝他挥了挥手。

He waved at him that one time.

那部电影是 Amazon 制作的，但他们在六月宣布不会发行它。所以它被卖给了 Neon，他们将在圣诞节上映。今年早些时候，Amazon 向 OpenAI 投资了 500 亿美元。你能看出这看起来会是什么样。Sam，你或你这边有没有任何人跟 Amazon 的人谈过这件事？

Amazon produced that movie, but back in June they announced that they wouldn't distribute it. And so it was sold to Neon and they're going to put it out on Christmas Day. Earlier this year, Amazon invested $50 billion into OpenAI. You can see how that could look. Sam, did you or anyone on your behalf speak to anyone at Amazon about it?

我没有，我也不知道还有谁谈过。

I did not, and I don't know of anyone else who did.

好。你从没跟 Jeff Bezos 谈过这件事？

Okay. You never spoke to Jeff Bezos about it?

好吧。我还没看过这部电影，但有人告诉我它是一幅阴暗的画像。被那样描绘，你不在意吗？感觉你并不想去纠正什么记录。我要说的是，从看过预告片来看，我不觉得他们在追求准确性。

All right. I haven't seen the film, but I'm told it's a dark portrait. That doesn't bother you, to be portrayed that way. It doesn't feel like you want to correct the record on something. I will say, from having watched the trailer, I don't think they're going for accuracy.

好。

Okay.

所以从这个意义上说，不，我不介意。我的意思是，他们是在努力拍一部好电影，也许它会不错。我不知道。

So in that sense, no, it doesn't bother me. I mean, they're trying to make a good movie, and maybe it will be good. I don't know.

好。作为电影来说不错。

Okay. Good as a movie.

作为准确的描绘，它不会好。

It will not be good as an accurate depiction.

好。除了走路。

Okay. Except for the walk.

除了那个看起来不错。我相信他们也会把其他一些部分弄对，至少是碰巧弄对。

Except that seemed good. I'm sure they'll get some other parts right, at least accidentally.

预告片里有一个时刻，Sam Altman 的角色走进一个类似詹姆斯·邦德式的掩体，墙上挂着枪械。

So there's a moment in the trailer where the character of Sam Altman walks into kind of a James Bond sort of bunker and there's firearms on the wall.

那就是一个不准确的例子，但看起来确实挺酷的。我的意思是，那是不准确的。

That's an example of an inaccuracy, but I did look kind of cool. I mean, so that's inaccurate.

并没有那种墙上挂着枪械的地堡。

There's no bunker with firearms on the wall.

我正想问，你是末日准备者吗？这是这里的另一个潮流，在纽约我们觉得挺好奇的。

I was going to ask, are you a prepper? That's another trend out here that back in New York we find curious.

我会说，我有一种生存主义的心态。

I have a survivalist mindset, I would say.

好。

Okay.

我在智识上对这类东西感兴趣，但我并没有那个预告片里描绘的那种配置。

And I am intellectually interested in that stuff, but I don't have a setup like is depicted in that trailer.

你有撤离计划吗？

Do you have a bugout plan?

我是说，比如在我们的牧场，我们可以离网生存，我猜对很多人来说这就算撤离计划了。

I mean, like at our ranch, we can survive off-grid, which I guess to many people would be considered a bugout plan.

好。但并不是说

Okay. But it's not like

我们没有潜艇

we don't have like a submarine

一个应急包，里面有直升机计划之类的

a go bag with like a plan for a helicopter and

没有直升机，没有潜艇

no helicopter, no submarine

因为这种准备的想法，我觉得对普通美国人来说很可怕，不仅对大多数人来说这不是一个选项，而且这感觉像是一种他们不知道自己本该有的心态。

because this idea of prepping I think feels terrifying to normal Americans where they're not only is that not an option for most people but it feels like a mindset that they didn't know they were supposed to have.

新冠来的时候，我有一大堆口罩准备好了，家里有足够的罐头食品，如果世界停摆，我们也能撑一阵子，我们还有净水片，我记得一个朋友说，我以前还拿这些事取笑你。那大概是很合理的做法。

When COVID came, I had like a big stack of masks ready to go and I had like enough canned food in the house that if like the world had shut down, we could have like had enough food for a while and we had like water purification tablets and I remember a friend of mine said like, I used to make fun of you for that stuff. That was like probably something that was like very reasonable to do.

撇开 AI 不谈，我确实认为我们对这类不寻常事件准备不足，比如新冠，新冠之后我本以为会有更多人说，好吧

AI aside, like we have I do think we underprepare for these sort of like unusual events like COVID and after COVID I would have expected more people to say like okay

也许值得在我生活中多花 0.1% 的精力，为下一次大流行之类的做好准备。我不认为大多数人更新了这一点，这对我来说有点奇怪。我公寓里现在没有口罩了。

maybe it is worth like 0.1% more effort of my like effort in my life to like be ready for the next pandemic or whatever. And I don't think most people updated on that and it's like slightly strange to me. I don't have masks in my apartment anymore.

也许你应该有。

And maybe you should.

当然。也许我应该有。

Sure. And maybe I should.

那么 Sam，你和你在 AI 领域的同行们，我认为正在承担超越商业领袖的角色。我知道你曾一度考虑竞选加州州长。那

So Sam, you and your peers in AI, I think, are assuming roles that go beyond business leaders. I know that at one point you were contemplating the idea of running for governor of California. That

有一点。我猜这被夸大了，但没错，有一点是真的。

a little bit. I guess it's like overstated, but yes, a little bit is true.

然后在基奇的传记里，她说你考虑过竞选总统。我只想正式问一下，你目前有任何竞选公职的计划吗？

And then in Keach's biography, she says that you contemplated the idea of the presidency. And I just want to ask for the record, do you have any current plans to run for office?

没有。

No.

总统职位仍然是你可能会考虑的事情吗？

Is the presidency still something you would ever consider?

不。我不认为我质疑那件事的准确性。州长那件事是真的。

No. And I don't think that I dispute the accuracy of that one. The governor thing is true.

好。

Okay.

我从不打算竞选公职。我想我在目前的角色中已经足够尝到了那会是什么样子，知道那不是我想要的。

I do not ever plan to run for office. I think I have gotten enough of a taste of what that would be like in my current role to know that it's not something I would want.

最初吸引你的是州长职位的哪一点？

What do you think about the governorship that attracted you in the first place?

嗯，我爱加州，我认为总体而言它是一个管理得很糟糕的州，我认为任何州长能做的可能都有限。

Well, I love California and I think that it is like a very badly managed state on the whole and I think that there's maybe a limited amount any governor could do about that.

是的。

Yeah.

但我觉得至少试一试会很有趣。我认为 AI 重要得多。

But I thought it would be fun to at least try. I think AI is way more important.

你之前谈过自己是一个有灵性的人。我想知道 AGI 与你对上帝的理解有何关系？

You've talked about being a spiritual person before. I'm wondering how does AGI relate to your understanding of God?

完全没关系。

Not at all.

完全没关系。

Not at all.

没有。

No.

详细说说。

Elaborate.

我认为 AI 是科学领域内的东西，已经被很好地理解，或者正在被更好地理解，它是一个，它就像一个

I think AI is something inside the realm of science and well understood or becoming more well understood and it is a it is like a

其实我就说到这里吧。

actually I'll leave it there.

你认为这是你的命运或天命吗？

Do you think that this is your fate or your destiny?

不，我认为我本可以做很多事情，而且我对其中任何一件都会挺满意的。

No, I think I could have done many things and I kind of been pretty happy with any of them.

好。

Okay.

但我非常，我感到非常感激和幸运能从事这项工作。这就像一个，这是一种巨大的荣幸。

But I'm very I feel very grateful and lucky to get to work on this. It's like a It's like a great privilege.

但 AI 的发生几乎是不可避免的，而你恰好成为引领它的人？

But it's is it almost an inevitability that AI was going to happen and you get to be the person that leads it?

我认为它的发生是不可避免的。我的意思是，这就像人类的努力和人类社会集体朝着技术进步前进。但我能成为参与其中的人之一，多少有点偶然和幸运。

I think it's an inevitability that it was going to happen. I mean it's just like human effort and human society collectively builds towards technological progress. But it is like kind of chancy and lucky that I got to be one of the people involved.

是的。在我之前提到的凯文·鲁斯那本书里，他谈到奥本海默在 50 年代的听证会上，他说即使当他看到核科学能做什么时，他们也觉得它在技术上很美妙，那是一种他觉得不得不做的事，当你看到它可能实现时，有一种人性的东西，就像把最后一块拼图放进拼图里，不管那幅画是什么。

Yeah. In that Kevin Roose book I mentioned before, he talks about how Oppenheimer in the 50s in hearings, he talked about how even when he saw what nuclear science could do that they saw it also as technically sweet and it was like something that he felt compelled to there was something human about when you can see it's possible, it's like putting the last puzzle piece into the puzzle no matter what the picture is.

我确实有那种感觉。我记得有过非常相似的感受，就是在 AI 早期，有那么多极其困难、来之不易的科学和工程片段必须汇聚在一起，才能让我们所有人能够在此刻做这下一件事。

I did feel that. I remember feeling something very similar to that which is that like early in AI it was like the number of incredible difficult like hardfought pieces of science and engineering that had to come together to let all of us be able to do this this next thing in this moment.

嗯。

Mhm.

那真的打动了我，但我不认为它是最后一块拼图，因为接下来还会有很多事情，人们会说，“哦，那是我做这件更了不起的事情所需的原料之一。”

And that really struck me, but I don't think of it as like a final puzzle piece because there's going to be many things that come next and people will be like, "Oh, that was one of the ingredients I needed to do this much more amazing thing."

现在，我想谈谈你的公司。OpenAI 众所周知最初是一个非营利组织，后来转为营利结构，以便筹集资金来获得保持竞争力所需的大量算力和昂贵人才。现在你们正在准备 IPO。当你们上市时，公司的激励结构会像任何上市公司一样，与利润最大化和股东价值挂钩。你是否担心上市会对公司的灵魂造成什么影响？

Now, I want to talk about your company. So, OpenAI was famously founded as a nonprofit and then converted to a for-profit structure so as to raise capital for the huge amounts of compute and expensive talent that it was required to stay competitive. Now you're preparing an IPO. When you go public, your company's incentive structure will, like any public company, be tied to profit maximization and shareholder value. Are you nervous about what going public will do to the soul of the company?

首先，我们不会像任何其他公司一样。即使我们上市——我们并不急于上市——我们仍会有一个非营利组织来管理我们。我们会提醒投资者以及任何考虑购买我们股票的人，我们会把使命放在第一位。我们相信这与我们的股东表现极好是兼容的。我真的这么认为。但如果它们发生冲突，使命将优先，如果我们必须做符合世界利益但不符合我们特定股东利益的事情，我们会很容易决定怎么做。我认为这让我们与大多数公司不同，人们会决定他们是否愿意接受这一点。

First of all, we will not be like any other company. Even when we go public, which we are not in a rush to do, we will still have a nonprofit that governs us. We will remind investors and anyone that thinks about buying our shares that we are going to put the mission first. We believe that that will be compatible with our shareholders doing super well. I genuinely do think that. But if they are ever in tension, the mission is going to come first and if we have to do something that is in the interest of the world but not our particular shareholders, it's going to be very easy about what we're going to decide to do there. And that makes us I think different than most companies and people will decide if they're willing to sign up for that.

所以，你不担心市场需求会影响产品安全吗？

So, you're not worried that market demands will impact product safety?

我不担心。我不。嗯，这么说不太公平。我总是担心这类事情，但我有信心我们会做出正确的决定，因为我们担心它。

I'm not. I don't. Well, that's not fair. I always worry about this kind of stuff, but I am confident we will make the right decision because we worry about it.

在你之前的科技公司在哪里失败了。

Where tech companies that came before you failed.

他们做对一些事情，也做错了很多事情。我们会尝试学习，但我确信我们也会犯自己的错误。但至少我们会努力不犯同样的错误。

They got some things right and a lot of things wrong. And we'll try to learn, but I'm sure we'll get our own things wrong, too. But we'll at least try not to make the same mistakes.

正如你在这次对话开头指出的，世界已经认识到这件事的利害关系和这个时刻的分量。我相信我们会得到包括股东、员工、投资者、合作伙伴在内的支持，去做出我们认为最符合人类共同利益的决定。即使这意味着我们的营收增长会比原本可能的速度更慢。我们一直在做这样的决定。我们放慢了脚步。

As you pointed out at the beginning of this conversation, the world recognizes the stakes and the magnitude of this time. And I'm confident that we will have the support of people including our shareholders, employees, investors, partners in making the decisions that we believe are best for humanity's shared interests. Even if it means we are going slower on revenue than we otherwise could. We've been making decisions like this. We've slowed down.

但如果因为达不到所需的利润基准，股价开始下跌，那又该怎么办？到时候会发生什么？

How does that work though if your stock price starts to fall because you're not hitting your profit benchmarks that you need to? What happens then?

那就跌吧。也许跌得很惨。我不知道。再说一次，我们会继续非常清楚地表明：如果有一天要在股价上涨和为人类未来做正确的事之间做选择，那一刻我们不会去关注股价。如果我们有耐心的投资者，我想长期来看他们会非常满意，但逐季来看可能会非常波动，我们将不得不做出一些艰难的决定。这也是我认为我们长期会非常成功的原因之一。但如果我们错了，如果出于某种原因，AI 的进步与打造一家极有价值的公司不相容——我真的不认为会这样。我真的认为这两者是相容的。我们显然会选择为人类做正确的事。

Then it falls. Maybe it falls a lot. I don't know. And again, we will try to continue to communicate very clearly that if there is ever a question about our share price going up or doing the right thing for the future of humanity, we are not going to pay attention to the share price in that moment. If we have patient investors, I think they'll be very happy over time, but quarter to quarter probably going to be very volatile and we're going to have to make some hard decisions. That is part of the reason I think we'll be very successful over time. But if we are wrong about that, and for some reason AI progress is incompatible with making an incredibly valuable company—I really don't think it's the case. I really think these are compatible. We are clearly going to pick doing the right thing for humanity.

好吧。如果你看看我们最早筹集的第一笔资金，我认为我们有着长期认真对待这件事的记录，我们为拥有一个极其复杂的结构付出了所有这些代价，基本上就是为了这一刻。

Okay. And I think we've got a long track record if you look at the very first capital we raised of treating this with seriousness and we've paid all of these prices for having a crazy complex structure basically for this moment.

Sam，你现在持有零股权，对吗？

Sam, is it right that you currently hold zero equity?

是的，我没有直接股权。我只有一笔来自旧 YC 投资的极小极小的一点敞口。

Yeah, I have no direct equity. I have this tiny tiny sliver of exposure from an old YC investment.

所以这对大多数人来说说不通，他们会认为如果你在世界上创造了这么多价值，你应该有权参与其中。那么你希望如何获得报酬？

So that doesn't make sense to most people who would think that if you're creating so much value in the world, you would deserve to participate in it. So how do you hope to be compensated?

如果我能回到过去重新来过，我以为我在做的这件事——当时没有人那么认真地对待 AI，对吧？所以他们当时看着我说：‘什么？这太疯狂了。你要拼命工作，创造所有这些价值，却不试图变得更富有。这简直不可想象。’现在我觉得突然之间有更多人理解了其中的道理。但人们仍然觉得，你知道，这感觉太不——

If I could go back and do things over again, I thought I was doing this thing that would—at the time, no one took AI that seriously, right? And so they looked at me then and said, 'What? This is crazy. You're going to work super hard and create all this value and not try to get richer. That's unimaginable.' Now I think all of a sudden more people sort of understand the rationale. And yet still people are like, you know, it just feels so out of—

道理是说你投资了其他公司，那些公司会——

The rationale being that you've invested in other companies that will—

哦不，道理是说，你知道，我们其实希望你只代表人类做决定，而不是为了财务利益。

Oh no, the rationale being like, you know, we actually want you to just be making decisions on behalf of humanity and not financial interest.

报酬。

Compensation.

是的。我认为这是一个错误，也是缺乏对大多数人如何看待世界的理解。而且奇怪的是，没有股权反而成了一件严重破坏信任的事。

Yeah. I think it was a mistake and a lack of understanding of how most people think about the world. And weirdly, this big trust-destroying thing to not have equity.

你怎么看待金钱？有些人把它看作奖励。有些人把它看作让事情发生的燃料。有些人把它看作他们囤积的安全网。

How do you think about money? Some people think of it as a reward. Some people think about it as fuel to make something happen. Some people think of it as a safety net that they hoard.

我曾有过非常成功的职业生涯。我过去极其受金钱驱动。所以我并不是想说‘哦，我只是不在乎。我像个僧侣。’我喜欢好东西。按任何合理的标准，我都有昂贵的品味。但我有过非常成功的职业生涯，并没有更多我想要的东西。现在我认为，用钱去做更多项目是合理的，如果我有巨额财富，不必为 OpenAI 融资，因为我可以自己出资，我想那会非常好。但我不知道。我想这就是为什么我希望我当初就拿了股票。你没法说‘我钱太多了，没有更多我想要的东西’还能让人觉得有共鸣或不像你——所以我希望我当初就拿了股权。我不知道。

I had a super successful career. I used to be extremely money motivated. So I'm not trying to say that I'm like, 'Oh, I just don't care. I'm like a monk.' I like nice stuff. I have expensive taste by any reasonable stance. But I had a super successful career and there's not more stuff that I want. Now I think it is fair that you can use money to do more projects and if I had had crazy amounts of money and hadn't had to raise money for OpenAI because I could have just self-funded it, I guess that would have been really nice. But I don't know. I think that this is why I wish I had just taken the stock. There's no way to say like 'I have so much money that there's not more stuff I want' that comes across as in any way relatable or not like you—so I wish I just had the equity. I don't know.

因为这让人难以相信。但我从中得到的是，我认为我拥有现在世界上任何人所拥有的最有趣、最酷、最有影响力的工作。所以并不是说我在做这件事却没有报酬。我得到了一个难以置信的位置和有趣的生活。如果有人要说：‘哦，所以是为了权力。’那会更接近真相。我认为有趣是——而且我认为我们所有人都有留下某种印记、产生某种积极影响、以某种方式做些酷或有用的事情的渴望。我不太擅长运动，但如果一个体育明星为了对比赛的热爱而拼命努力，或者一个艺术家为了创作伟大的艺术而拼命努力，那不知怎么地就完全可以。但你们中的一些人在做商业上等价的事情。人们却只希望你为了钱或其他什么而做。再说一次，我喜欢好东西，所以我没有评判。我希望每个人都变得超级富有，我认为 AI 是实现这一目标的最佳方式。但我确实从中得到了不可思议的东西。

Because it feels hard to believe. But the thing that I get out of doing this is I think I have the most interesting, coolest, most impactful job that anyone in the world has right now. And so it's not like I'm doing this for no compensation. I get an unbelievable seat and interesting life. If someone were to say, 'Oh, so it's for power.' That would be closer to true. I think interestingness is the—and also I think all of us have a desire to leave some sort of mark and have some sort of positive impact and somehow do something cool or useful. And I'm not very athletic, but if a sports star works super hard for love of the game or an artist works super hard to make great art, that's super okay somehow. But some of you are doing the business equivalent of that. People just want you to be in it for money or something. And again, I like nice stuff, so I have no judgment. I want everybody to get super rich and I think AI is the best way to make that happen. But I do get something incredible out of this.

好的，Sam，我想你还没有对 Jacob Coxin 在《华尔街日报》和 Twitter 上发表的言论做出评论。他说 OpenAI 和 Anthropic 都没有‘负责任地行事’，而且你的公司尤其没有‘深刻内化 AI 的文明利害关系’。你认为他是什么意思，你如何回应？

Okay, Sam, I don't think you've commented on Jacob Coxin's remarks that broke in the Wall Street Journal and on Twitter. He said that neither OpenAI nor Anthropic were quote acting responsibly and that your company in particular hasn't quote deeply internalized the civilizational stakes of AI. What do you think he means and how do you respond?

再说一次，如果你回顾我们公司十多年前的公告，我们显然感受到了文明层面的利害关系。显然，这个人有权发表他的观点。我们也很早之前就自愿暂停训练，以便在安全和对齐以及监控方面取得更多进展。我们还做了许多其他行动，并长期谈论这里的文明利害关系。我认为这个时刻的不同之处在于，人们过去不相信我们，或者不在乎，或者忙于其他事情，或者不想听，而现在人们听了，我其实很高兴人们听了。我认为随着我们提升到新的能力水平，我们现在确实拥有非凡的模型，随着我们不仅建立所需的技术保障、技术安全和对齐技术，还建立支持这一点的社会、政府和文明基础设施。我认为世界对此觉醒是件好事。我认为我们更有可能在公司之外、在我们所需的技术之外完成这些艰难的事情。

Again, I think if you go back to the announcement of our company from more than 10 years ago, we clearly feel the civilizational stakes. Obviously, this guy's entitled to his opinion. We have also, well before this, we were the company that voluntarily paused our training to make more progress on safety and alignment and monitor. We have done many other actions and been talking for a long time about the civilizational stakes here. I think what is different about this moment is that people didn't believe us or didn't care or busy with other things or didn't want to hear it and now people do and I'm actually very happy that people do. I think as we rise to this new level of capabilities and we really do have extraordinary models now, as we build up not just the technical safeguards and the technical safety and alignment techniques that we need but also the sort of society and governmental and sort of civilizational infrastructure to support this. I think it's great that the world is waking up to this. I think it's much more likely we'll be able to get these hard things done outside of the companies and outside of the technology that we need.

所以你欢迎人们的担忧，因为这表明他们理解其中的利害关系，不仅是理解，还愿意采取行动。也许在这样的时刻之前，很难就共同的安全标准达成国际协议，而现在也许可以了。我想到社交媒体公司，他们曾说，哦，我们不是媒体公司，我们是科技公司。我们建造了城镇广场，人们往广场里放什么是他们的事，那是言论自由问题，我们洗手不管。当你们公司的智能体失控时，你觉得你该负责吗？

So you welcome people's concern because it shows that they understand the stakes and not just understand but are willing to act. Maybe it would have been really hard to get international agreement on shared safety standards before a moment like this, and maybe now you can do it. I think about the social media companies and the way they talked about, oh, we're not media companies, we are technology companies. We build the town square. What people fill the town square with is up to them, and that's a free speech issue, and we wash our hands. Do you feel responsible for what happens when your company's agents break out?

那样的话，是的。

In that case, yes.

是的。

Yes.

因为如果杂志出了事，比如我们报道有误，或者以某种方式造成了损害，即使我没有碰那篇报道，我仍然要负责。我认为人们应该能够以我们可能不总是喜欢的方式使用我们的工具和技术。然而，如果我们的工具和技术没有按照我们期望的方式运行，或者有人要求它们那样做，那么我认为我们应该为此负责。

Because if something happens at the magazine, for instance, if we misreport something, if we have done damage in some way, even if I didn't touch the story, I still am responsible. I think people should be able to use our tools and technology in ways that we might not always like. However, if our tools and technology are not behaving in the way we expect them to, or someone is asking them to, then I think we should be responsible for that.

今年你们的伦理主管、安全系统主管、数据中心主管，以及运营总监、营收总监和其他高管都离开了公司。对旁观者来说，这似乎表明 OpenAI 内部正在发生一些奇怪的事情。发生了什么？

This year your head of ethics, head of safety systems, head of data centers, along with your operating chief, your revenue chief, and other executives all left the company. To a casual onlooker, that would seem to signal that something strange was happening inside OpenAI. What's going on?

我们正在以某种速度增长，整个行业都在以科技史上前所未有的速度增长。也许每 5 年的公司增长被压缩到一年左右。我认为，有些人在公司某个阶段很出色，但不适合下一个阶段，这是正常的，但通常这种情况发生得比这慢一些。

We are growing at a rate, the whole industry is growing at a rate that is sort of unprecedented in tech history. Every maybe 5 years of company growth is compressed into a year or something like that. I think it is normal that there are people who are great at one stage of a company that are not the right fit for the next, but normally it happens a little more slowly than this.

这符合“快速行动，打破陈规”吗？那是

Is that in line with move fast and break things? Is that kind of the

我们尽量不打破东西。

We try not to break things.

好的。

Okay.

但我认为，随着公司发展，不同的人在不同的增长规模下茁壮成长，在错误的时间用错误的人对他们或公司都没有好处。

But I think the fact that as a company grows, different people thrive at different scales of growth and it's not helpful to them or the company to have the wrong person at the wrong time.

对。这并不令人意外。

Right. That that's not surprising.

好的。我们来谈谈基金会。OpenAI 基金会。就资产而言，它是世界上最大的，甚至比盖茨基金会还大。你们最近宣布了多项举措，其中包括投入 1 亿美元用于阿尔茨海默病研究。我们谈论时间线情景，并且越来越频繁地听到关于 AGI 或噩梦般末日事件的时间线。治愈阿尔茨海默病的时间线是什么？

Okay. Let's talk about the foundation. The OpenAI Foundation. It is the largest in the world in terms of assets, even larger than the Gates Foundation. And you announced recently among other initiatives $100 million towards Alzheimer's research. We talk about timeline scenarios and we're starting to hear about them more often for AGI or nightmare apocalyptic events. What's the timeline for curing Alzheimer's?

看，AI 让我兴奋的事情有很多。就我个人而言，可能是在未来几年内，一些疾病可以开始被治愈。我不知道阿尔茨海默病具体什么时候能治愈。那将是科学家们的问题，但我预计未来几年在一些疾病上会有真正的进展。我认为这太棒了。我听过真正了解 AI、健康和生物学的科学家说，十年后，我们可能会看到一个令人难以置信的截然不同的景象，关于生病和保持健康对人意味着什么。我认为这很棒。这显然将是我们基金会的重点领域之一。事实上，前两个是健康和疾病以及科学，然后是 AI 韧性，帮助社会为这种变化做好准备。但我认为我们将看到非凡的成果。你知道，模型终于足够聪明来做这些事情了。一个能证明最难的数学问题的模型也能做很多其他新知识的发现。即使在小方面，这也稍微超出了基金会的范围。现在任何人做科学和工程的能力真的令人难以置信。我最近和一些大学生在一起。他们给我看了一个黑客松项目，他们用 GPT 5.6 做了一个自动驾驶高尔夫球车，它在斯坦福校园里行驶。我上学时，黑客松项目就像你做了一个勉强能用的网站，在一个周末里那已经相当令人印象深刻了，而他们直接做了一个自动驾驶高尔夫球车。它在订购零件。它在 3D 打印执行器。它显然在编写软件和做 AI 部分。太不可思议了。

Look, there's many things I'm excited about with AI. Probably personally, the fact that it feels like within the next few years some diseases can start to be cured. I don't know exactly when Alzheimer's will come. That will be a question for the scientists, but I expect real progress in the next few years for some diseases. I think this is amazing. And I have heard scientists who really understand both AI and health and biology say that a decade from now, we could see just an unbelievably different landscape for what it means to get sick and stay healthy as a person. And I think this is awesome. This will obviously be one of the big areas of focus of our foundation. The top two in fact are health and disease and science, and then AI resilience and helping society prepare for this change. But I think we're going to see just extraordinary results. You know the models are finally smart enough to do this stuff. A model that can prove the hardest math problems can do a lot of other new knowledge discovery too. And even in smaller ways, this zooms out a little bit from the foundation. The capability that anyone has now to do science and engineering is really incredible. I was with some college students recently. They're showing me a hackathon project and they used GPT 5.6 to make a self-driving golf cart and it was driving around the Stanford campus. When I was in school, a hackathon project was like you made a website that barely worked and that was pretty impressive in a weekend, and they just made a self-driving golf cart. It was ordering the parts. It was 3D printing the actuators. It was obviously writing the software and doing the AI part of it. It's incredible.

但这种力量伴随着——我只是在想，不是你的公司，而是 Anthropic，他们如何关闭那些他们认为可能在使用他们的工具至少探索生物武器的科学家。你知道，如果那是一个邪恶的黑客松，那就太可怕了。这很难。我认为大多数人很难同时接受这两件事。技术可以做这些奇妙的事情。接受两者

But that power comes with—I'm just thinking it's not your company, but Anthropic and how they shut down scientists who they felt were possibly using their tools to at least explore biological weapons. You know, if that was an evil hackathon, that's terrifying. It's hard. I think most people find it hard to reconcile those two things at once. The technology can do these wonderful things. Accept both

心甘情愿地。我认为有很多人会说，让我们停止这项技术。任何能制造病毒的技术，即使它也能治愈癌症。我宁愿冒癌症的风险。我不想要有人制造病毒。我认为这就像

willingly. I think there are a lot of people who would just say like let's stop the technology. Any technology that can make a virus even if it can also cure cancer. I'll take my risk with cancer. I don't want someone to make a virus. And I think this is like

这是可以理解的人类本能。我认为这是错误的。我不想回到 500 或 1000 年前，没有现代医学。

it's an understandable human instinct. I think it's wrong. I would not want to go back in time 500 or thousand years and not have modern medicine.

但你知道，技术在此过程中也做了一些坏事。

But you know technology has done some bad things along the way too.

我相信正确的答案是应对挑战，这样我们才能获得好处。我不想停止技术进步。我想要癌症的治愈方法。我想要阿尔茨海默病的治愈方法。

I believe the right answer here is to navigate the challenges so we can get the upsides. And I do not want to stop technological progress. I want that cure for cancer. I want that cure for Alzheimer's.

是的。

Yes.

我希望，你知道，10 年后的孩子们能做比自动驾驶高尔夫球车更了不起的事情。但我理解那种本能，说我不想要改变。我不想要风险。就停下来吧。我们会应付我们现在的生活。我真的希望几百年后，人们回顾我们的生活，简直不敢相信我们曾经过得那么糟糕。

And I want, you know, the kids in 10 years to be doing much more amazing things than self-driving golf carts. But I understand the instinct to say, I don't want the change. I don't want the risk. Just stop it. We'll deal with like our current life. I really hope that hundreds of years from now, people look back at our lives and just can't believe how bad we had it.

是的。

Yeah.

我认为这就是文明和技术的故事。我相信它会继续下去。现在人们确实对少数公司拥有如此大的权力感到非常不安，你知道，一家公司决定关闭这个科学家而不是那个，关闭这个政府而不是那个，你知道，我们将决定你是否能以这种方式使用我们的技术，这真的很可怕。我认为我们感到很大的责任，要成为一个可靠、可信赖、公平的行动者。这并不意味着我们不会对我们的技术施加限制。比如我们会以我们不希望人们用它来制造新型病原体的方式构建，当然。

And I think this is the story of civilization and technology. And I'm sure it'll keep going. Now people do get very uneasy about a small number of companies having so much power and you know a company deciding turn off this scientist and not that one turn off this government and not that one you know we'll decide if you get to use our technology in this kind of way or not and that is really scary. And I think we feel a lot of responsibility to be a reliable, dependable, fair actor. Doesn't mean we won't put constraints on our technology. Like we will build in ways that we don't want people to use it to create novel pathogens of course.

但就像你的电力公司仍然必须遵守法律一样，电力公司并不是决定你能插什么电器、不能插什么电器的人。对我来说，电力公司的类比感觉就像社交媒体公司说的公共广场，或者感觉像是在推卸责任。

But in the same way that your electricity company still has to follow the laws, your electricity company is not the one making decisions about what kind of appliance you can plug in and not. The electricity company comparison feels to me like what the social media companies were saying about a public square, or that it feels like maybe abdicating responsibility.

嗯，那正是我想说的部分：我们仍然必须内置保障措施。好吧。并不是字面上就像，你知道，这里是电子，你可以做任何你想做的事。我们一直因为说这太严格或其他什么而受到批评，人们会去用开源模型，但我们会表达一些关于我们不想让人们做的非常糟糕的事情的意见。我不认为我们会采取整个“这是城镇广场，你想做什么就做什么，我们对此没有责任”的立场。另一方面，我们经常谈论 AI 韧性这个想法。会有其他不是我们的模型，公开可用，人们能够用它们做非常可怕的事情。我认为这是不可阻挡的，或者至少在不付出疯狂的自由代价的情况下是无法阻止的。所以，我们想做的部分事情也是提供工具，利用我们拥有的任何领先优势，帮助社会在坏事发生之前变得更有韧性。

Well, that was the part I was trying to get to: we still have to build in the safeguards. Okay. It's not literally just like, you know, here are electrons and you can do anything you want. We are criticized all the time for saying this is too restrictive or whatever, and people go to open models, but we will express some opinions about some really bad things we don't want people to do. I don't think we're going to take the whole 'it's the town square, do whatever you want, we have no responsibility for it' stance. On the other hand, we talk a lot about this idea of AI resilience. There are going to be other models out there that are not ours, openly available, that people will be able to do very horrible things with. And this, I think, is unstoppable, or at least it's not stoppable without a crazy price to liberty in society. And so also part of what we want to do is provide tools with whatever lead we have to help society get more resilient before the bad things come.

好。那么这就引出了放慢速度的想法。据我理解，你不能放慢，因为你的竞争对手不会放慢，中国也不会放慢。所以我们都在竞赛中。这个历史时代的曼哈顿计划部分可能接近尾声。那么核武器之后是什么？是多边防扩散协议。如果有一个大红按钮，Sam，你和你所有的美国竞争对手以及中国都可以按下它并说：“我们现在暂停，我们要确保安全。”你会按下那个按钮吗？

Okay. So that brings us to the idea of slowing down. As I understand, you can't slow down because your competitors won't slow down and China won't slow down. And so we're all in a race to get there. The Manhattan Project era section of this era of history is perhaps nearing its conclusion. And now what comes after for nukes? It was multilateral non-proliferation agreements. If there were a big red button, Sam, that you and all of your American competitors and China could hit and say, 'We're going to pause for now and we're going to get safe.' Would you hit that button?

是的。

Yeah.

你会按下那个按钮吗？

Would you hit that button?

嗯，我认为这个对话有三个层面。有我们单方面会做的事情，即使假设没有人愿意与我们合作。好的。有我认为我们作为一个行业应该被允许做的事情，我认为我们需要像反垄断豁免这样的东西，以便能够谈论，你知道，一起放慢速度。然后是在其他美国公司之间，然后是需要在国际上发生的事情。在所有这些三个层面中，我还要说，我认为有很多人只是说停下来，就像暂停。这太令人压力大了，无法思考。我真的不知道我想要什么，但我不希望它再进一步。我对此有很多同理心，但我认为它忽略了细微差别。经过多年的思考，我们得出的结论是，我们真正想要的是，在每个新能力阶段，我们在对齐、责任、安全、理解社会和技术如何共同进化方面有足够的进展，以便能够安全地进入下一个阶段，对吧？当我们在这个单方面层面上没有这些时，即使那样我们也会暂停。例如，我们最近停止了一个模型训练运行，暂停了一个，因为我们认为鉴于能力进步的速度，我们需要在对齐和责任方面取得更多进展。

Um, I think there are three levels to have this conversation at. There's what we will do unilaterally even assuming no one was willing to cooperate with us. Okay. There is what I believe we should be allowed to do as an industry and where I think we'll need things like antitrust exemptions to be able to talk about, you know, slowing down together. And then among other American companies and then there's what needs to happen internationally. Across all three of these I'll also say I think there's a lot of people that just say just stop, like just pause. It's too stressful to think about. I don't really know what I want but I don't want it to go any further. Again I have a lot of empathy to this but I think it misses the nuance. What we have come to after years of thinking about this is what we really want is that at each stage of new capability we have enough progress on alignment, on liability, safety, understanding how society and the technology are co-evolving to be able to safely go to the next level, right? And when we don't have that at this unilateral level even then we do pause. So for example we recently stopped a model training run, paused one because we thought we needed to make more progress on alignment and liability given how quickly the capabilities were advancing.

所以我们必须放慢速度。现在，已经有很多次你只是没有注意，就像媒体在我们谈论这个之前没有注意一样。这是另一个能力快速发展的时刻，我们一直在，你知道，过去几个月谈论我们需要自己这样做，而且多年来我们一直说行业需要这样做，但我认为这是一个会有更多公众支持这样做的时候。

So it is necessary that we slow down. Right now there have been many times already you just haven't paid attention, like the media have not paid attention before we've talked about this. This is another one where capabilities are moving quickly and we have been, you know, for the last few months talking about how we need to do that on our own and also for years we've been saying the industry will need to do this but I think this is a moment where there will be more public support for this to happen.

我认为当然。我认为现在当然。

I think for sure. I think for sure right now.

你能告诉我们你现在正在做什么来实现这种合作吗？因为从你所说的、我们读到的来看，合作和协作对于安全、对于真正的安全是必要的。

Can you tell us what you're doing right now to make that kind of cooperation a reality? Because it seems like from what you're saying, what we're reading, that cooperation and collaboration is necessary for safety, for real safety.

我们正在与其他实验室的安全团队和领导者交谈。我们正在与政府交谈。我认为这是一个重要的时刻。如果美国和中国能够就共同的安全标准和要求达成一致，以验证我们应要求达到的门槛，我认为那将是一个不可思议的步骤。我认为他们可以做到。

We are talking to safety teams and leaders at other labs. We are talking to the government. And I think this is an important time to do it. And if the US and China could agree on shared safety standards and requirements to proceed to verify compliance with the thresholds that we should demand to proceed, I think that would be an incredible step. I think they could do it.

我们需要领导力，无论是来自我们的政府——在我们历史的这个时刻，政府往往不提供领导力——还是来自你个人和你的同行。我们在要求它。我认为这是公众的情绪。我想我只能代表我自己说话，但我觉得公众在要求你们所有人一起合作。

We need leadership whether it's from our government who quite often doesn't provide it at this time in our history or from you specifically and your peers. We're asking for it. I think that's the public sentiment. I can only speak for myself, I guess, but I feel that the public is asking for that from you all to work together.

看，就我自己而言，我认为我在这个问题上的声明在过去十年中一直非常一致。现在不同的事情是，我认为更多的人感受到并相信它，我们可能有一个公众支持的窗口来做这件事。

Look, speaking for myself, my statements on this I think have been unbelievably consistent over the last decade. The thing that is different now is I think more people feel it and believe it and we may have a window of public support to do it.

除了你在推特上贬低埃隆，以及 AI 行业内部的行为方式。你认为在那段关系中谁对谁更友好？

Except when you're negging Elon on Twitter and the way that the AI industry behaves amongst itself. Who do you think has been nicer to each other in that relationship?

我不会称自己为圣人。

I won't call myself a saint.

不。

No.

但我不确定在埃隆和我之间，你是否应该来找我。

But I'm not sure if I'm the one that you're supposed to come after on between Elon and me.

我知道。我只是说作为一个——这不是直接把它推给你。这是行业的问题，就像现在是以某种方式成熟起来的时候了。

I know. I'm just saying as a—it's not putting it on you directly. It's on the industry that like now it's time to grow up in a way.

看，我不喜欢恶霸，我认为埃隆是个恶霸。所以我认为偶尔你得回击恶霸的鼻子。我也认为像埃隆，像许多恶霸一样，只尊重那些这样做的人。所以如果你希望我能够与他合作，也许你应该高兴我偶尔回击。但我会与埃隆合作。

And look, I don't like bullies and I think Elon's a bully. And so I think once in a while you got to punch a bully back in the nose. And I also think like Elon, like many bullies, only respects people who do. So if you want me to be able to work with him, maybe you should be happy I hit back once in a while. But I will work with Elon.

当然，美国公众，美国心理，我认为倾向于末日论，因为许多人成长的文化。他们看过《终结者》、《黑客帝国》、《2001 太空漫游》，甚至《沙丘》，在那里计算机被禁止。所以，很容易想象一个世界，这变得非常错误。你能描绘一幅它变得正确的图景吗？就像，帮助我想象它并具体一点。政府的角色是什么？这些公司的角色是什么？并思考中期和长期。

Of course, the American public, the American psyche, I think, tends towards doomerism because of the culture that many people grew up in. They've seen The Terminator, The Matrix, 2001: A Space Odyssey, even Dune, where computers are outlawed. And so, it's very easy to imagine a world where this goes very wrong. Can you paint a picture in which it goes right? Like, help me imagine it and be specific. What's the role of the government? What's the role of these companies? And think mid and long term.

是的。好的。首先，我用一点玩笑回答了上一个问题，但我们真的需要——我们需要公司之间足够和睦，以便为我们的未来做出良好的集体决策。当然，公司之间会有竞争。这在许多方面可能是好的。

Yeah. Okay. First of all, I answered the last question with like a little bit of a joke, but we really do need to get—we need the companies to get along with each other enough to make good collective decisions for our future. And of course, there's going to be competition between the companies. There can be many ways in which that's good.

这显然已经奏效，带来了令人难以置信的产品和服务。但我们需要即使彼此不喜欢的人也能在公司研究人员和领导层之间做出良好的集体决策。你知道，无论你对我或 OpenAI 有什么批评，我认为我们非常擅长寻求共识和共同点，并与他人合作。所以这是我认为我们需要并且会去做的一件事。另一件事是我们需要政府挺身而出。私营公司的权力应该有限制，人民的意愿应该反映在社会必须做出的一些极其重大的决定中。所以我认为，看起来通过公司和政府的结合，我们已经找到了一套共同的安全标准。这样就不会出现类似竞态条件的情况，即能力会超越保障措施，我们将承担某种社会规模的风险。然后在这个约束下，我们将找到一种方法，公平公正地分配不仅是这项技术的好处——疾病的治愈等等——还有它的访问权和权力，以便社会能够集体继续创造未来。人们可以建造各种奇妙的东西。公司——我认为我们将看到的创业热潮将完全令人惊叹。

That clearly has worked to deliver incredible products and services. But we need even people who don't like each other to make good collective decisions among the company's researchers and leadership. And you know, whatever criticisms you might have of me or OpenAI, I think we are very good about trying to find consensus and common ground and working with other people. So that's one thing I think we need and that we will do. Another thing is we need our governments to step up. There should be limits to the amount of company power that private companies have, and there should be the will of the people reflected in some of these incredibly consequential decisions society has got to make. And so I think it looks like with a combination of the companies and the governments, we have figured out a shared set of safety standards. So there is not something that feels like race conditions where capabilities are going to get ahead of safeguards and we're going to take sort of societal scale risk. And then with subject to that constraint, we will have figured out a way to fairly and equitably distribute not just the benefits of this technology—the cures for disease, everything else—but the access and the power of it so that society can collectively keep inventing the future. People can build all sorts of amazing things. Companies—I think the boom in entrepreneurship that we're going to see here is going to be completely amazing.

那么那些不一定想成为企业家的人呢？你知道，有些人想在餐馆或农场工作，或者别的什么。

What about the people that don't want to be entrepreneurs though necessarily? You know, some people want to work in a restaurant or on a farm or whatever.

我很想在农场工作。去农场工作吧。我认为我们会——我还认为部分在于，如果这成功了，社会集体会变得如此富有，我相信我们必须想办法稍微改变社会契约和经济运作方式，但社会变得如此富有，以至于人们能做的事情、谋生方式、用于爱好、家庭等的时间都会大大增加。

I would love to work on a farm. Go work on a farm. I think we will—I also think part of this is that if this works, society collectively gets so rich and I'm sure we'll have to figure out how to change the social contract and the way the economy works a little bit, but society gets so rich that the aperture of what people can do and the ways you can make a living, the amount of time you can have for hobbies and family and whatever will go way, way up.

在去年的一次采访中，你谈到每周有 15,000 人自杀。世界上大约 10% 的人在使用 ChatGPT，所以每周有 1500 人，你说这让你夜不能寐。相比之下，我们想到车祸，我们痴迷于安全带、安全气囊和交通法规，有一个完整的跨国基础设施来保证我们在车内的安全，更不用说飞机了。我知道这对你来说是一个大问题，我很好奇你现在正在做什么来保护那些考虑自残的人。

In an interview last year, you talked about how there are 15,000 people a week who commit suicide. That about 10% of the world is talking to ChatGPT and so that's 1500 people a week and you said that keeps you up at night. And by comparison we think about car accidents and we're obsessed with seat belts and airbags and traffic laws and there's a whole multinational infrastructure built around keeping us safe in cars, to say nothing about planes. I know this is a big concern for you and I'm curious what you're doing right now to protect people who are considering self harm.

我们在产品上投入了大量工作来减少自残，并确保我们的产品是巨大的净善。我相信，而且你知道我们与许多专家合作，他们同意有可能让 ChatGPT 和类似产品成为有史以来最伟大的心理健康工具之一。我们可以帮助和影响的人数你知道是巨大的，因为很多人确实经历了非常艰难的时期。

We have put a huge amount of work into the product to minimize self harm and make sure that our product is a huge net good. I believe and you know we work with a lot of experts who agree that it is possible to make ChatGPT and similar products into one of the greatest mental health tools ever. And the amount of people you know that we can help and impact here cuz a lot of people do go through very hard times is huge.

你会包括强制报告吗？

Would you include mandated reporting?

我的意思是，显然专家们对强制报告是否好有极其不同的看法。我们在不同地方遵循不同的法律。我希望有更多专家共识关于在那里做什么和何时做，显然有时候这是一件非常糟糕的事情。但总的来说，我们当然相信，像人类专家帮助提供给处于危机中的人是件好事。

I mean obviously experts have extremely different views on whether mandated reporting is good or not. We follow different laws in different places. I wish there were more expert consensus on what to do there and when and there's clearly times where it's a really bad thing. But on the whole we of course believe that like human expert help being made available to people in crisis is a good thing.

那么你知道劳拉·莱利是谁吗？

So do you know who Laura Riley is?

我不知道。

I don't.

她是一名记者。她去年在《泰晤士报》上写了一篇文章。她的女儿在与 ChatGPT 交谈后自杀了。它没有告诉她自杀。

She is a journalist. She wrote an essay in the Times last year. Her daughter committed suicide after speaking to ChatGPT. It did not tell her to kill herself.

是的。

Yes.

还剩两分钟。我想尊重你，但我想继续。我们谈了一点关于未来和即将到来的事情。我们能在还有时间的时候稍微谈谈那个吗？

Two minutes left. I want to be respectful, but I'd like to move on. And we talked a little bit about talking about the future and what's coming. Can we go a little bit there while we have time?

我只需要几分钟超时来完成这个问题，然后我们会回到我们正在谈论的内容，然后结束。

I'm just going to finish this question a few minutes over and then we'll go back to what we were talking about and then wrap.

不，我们真的必须在一两分钟内出门。

No, we really actually have to walk out the door in one or two minutes.

好的。好吧。让我回答那个。

Okay. All right. Let me answer that one.

好的，谢谢。所以，劳拉是一名记者。她为《纽约时报》写了一篇文章，讲述她 29 岁的女儿索菲在与 ChatGPT 长时间交谈后自杀身亡。ChatGPT 没有指示她自杀，但同意充当她的治疗师。她称聊天机器人为哈利。它帮助编辑了她的自杀笔记。所以，我和劳拉谈过，她谈到 AI 聊天记录，以及它们如何几乎像实时路线图，展示人们如何决定结束自己的生命，以及这些信息如何影响政策、治疗方法、心理健康教育。你知道，我知道这对你意义重大。所以，我很好奇你认为这些信息有多大用处，以及是否有任何计划利用这些数据做些什么。

Okay, thanks. So, Laura is a journalist. She wrote an essay for the New York Times about how her 29-year-old daughter Sophie died by suicide after speaking at length to ChatGPT about her decision. ChatGPT did not instruct her to kill herself, but did agree to act as her therapist. She called the chatbot Harry. And it helped edit her suicide notes. So, I've spoken to Laura and she was talking about AI chat logs and how they could be almost like real-time road maps of how people come to the decision to take their own lives and how that information could shape policy and therapeutic approaches, mental health education. You know, I know this means a lot to you. So, I was curious what you think of how useful that information could be and if there's any plans to do anything with that data.

我的意思是，首先，这些是我们或任何开发这项技术的人面临的一些最难的问题。我们尽可能依赖专家意见，但专家也不确定该怎么做。这就像这是新的，事实上，有人在需要帮助的时刻，也许无法获得帮助或不愿意与治疗师交谈，可以在这里得到一些帮助。显然人们认为有好的版本。也有明显的坏版本。你知道，我们应该在人们最艰难的时刻将他们的私人数据发布给研究人员吗？我认为可能不应该未经他们同意，或者我应该说未经他们同意。我们是否想找到与社区合作解决世界心理健康危机的方法，并让我们的产品在这方面有所帮助？当然。是的。

I mean first of all like these are some of the hardest questions that we face or that anybody developing this technology faces. We try to rely on expert opinion as much as we can but experts also aren't sure what to do. This is just like this is new and the fact that someone in a moment of need that maybe can't get help or isn't willing to talk to a therapist can get some help here. Clearly people think there's a version of this that is good. There are also clearly versions of it that are bad. You know should we release people's private data in some of their hardest moments to researchers? I think probably not without their consent or I should say not without their consent. Do we want to find ways where we can work with the community to address the mental health crisis in the world and for our products to be helpful there? Of course. Yes.

好的。那么，你认为人们不应该用 AI 做的一件事是什么？

Okay. So, what's one thing that you think people should not use AI for?

取代人类联系。

Replacing human connection.

除了智能，还有什么品质使我们成为人类？

And beyond intelligence, what's a quality that makes us human?

情感似乎是 AI 还没有真正拥有的东西。

Emotions seem like something that AI doesn't really have yet.

你不认为它会发展出情感吗？

And you don't think that it'll develop emotions?

不会以同样的方式。

Not in the same way.

好的。我们问 GPT，如果你能问山姆·奥特曼一个他以前没有回答过的问题，会是什么？它说：“山姆，你私下里相信的关于 AI 未来的哪一件事你没有公开说过？不是因为它是机密，而是因为你认为人们会误解它的含义。”

Okay. We asked GPT if you could ask Sam Altman one question that he hasn't answered before, what would it be? And it said, "Sam, what is one thing you privately believe about the future of AI that you've not said publicly? Not because it's classified, but because you think people will misunderstand its implications."

一切都会很棒。

That it's all going to be great.

我认为，公开说这些之所以困难，是因为当然存在这些非常可怕的事情，我们必须把它们处理好，人们想听到我的声音，我真的很在意我们能否把所有这些都处理好，我们每天都在思考这个问题。但我对未来极其乐观。

I think that the reason that that's hard to say publicly is there are of course these very scary things and we have to get those right and people want to hear from me and I really do care that we're going to get all those right and we think about that every day. But I'm incredibly optimistic for the future.

在我们结束之前，我想和你做个自由联想。我会说一个名字，然后你回应你脑海中浮现的第一个词。不是他们的职位，可以吗？

Before we finish, I want to do a bit of a free association with you. So I'm gonna say a name and then you respond with the first words that come to mind. Not their job title, but okay.

你还想要什么？好的。

Anything else you want? Okay.

唐纳德·特朗普，

Donald Trump,

强大。

powerful.

达里奥·阿莫代伊，

Dario Amodei,

过。

pass.

阿里·穆尔哈林，

Ali Mullharin,

世界上最好的人。

best person in the world.

瑞秋小姐，

Miss Rachel,

呃，乐队。

uh, band.

彼得·蒂尔，

Peter Thiel,

很棒。

great.

皮特·赫格塞斯，

Pete Hegseth,

了解不够，无法评价。

don't know enough to say.

格温妮丝·帕特洛，

Gwyneth Paltrow,

有趣。

interesting.

伯尼·桑德斯，

Bernie Sanders,

始终如一。

consistent.

伊利亚·苏茨克维，

Ilya Sutskever,

才华横溢、关心他人、温暖而美好的人。

brilliant and caring and warm and wonderful person.

罗南·法罗，

Ronan Farrow,

呃，有很多话可以说。

uh, many things that could be said.

埃隆·马斯克，

Elon Musk,

不友善。

not nice.

好的。你知道，即使我说我不想对达里奥发表评论，或者我认为埃隆不友善，我真的想强调我多么愿意，以及我认为我们整个公司多么愿意与他们合作。嗯，我们感受到这件事的分量。

Okay. You know, like even though I would say like I don't want to make a comment about Dario or I think Elon is not nice, I really want to emphasize how willing I am and how willing I think our whole company is to work with them. Um like we feel the weight of this.

山姆·奥特曼，感谢你参加我们的节目。

Sam Altman, thank you for joining us.

谢谢。

Thank you.