《伊兹·克莱因秀》：黄仁勋认为对人工智能的恐慌已走得太远

他们考虑的是大型系统。

They think large systems.

现在的电脑里有成百上千亿个晶体管。

Today's computers have hundreds of trillions of transistors in it.

我刚毕业那会儿，参与的第一个芯片大概只有两百个晶体管吧。

When I first graduated from school, the first chip I worked on had, I don't know, 200 transistors.

每一个我都了如指掌。

I knew every one of them by name.

现在没有工程师会这么做了。

And no engineer does that today.

现在大多数工程师都在晶体管之上、功能之上工作，他们拼凑各种东西来实现目标。

Most engineers now work well above the transistor, well above the functionality, and they're cobbling things together to do things.

所以你需要更多地考虑系统以及系统之间的相互作用。

And so you need to think much more about systems and interactions of systems.

一些底层知识已经消失了。

Some of the lower level knowledge is gone.

这可怕吗？

Is that horrible?

所以我不知道，对大多数人来说，学会算面积分或者偏微分方程到底有多大价值，我也不太清楚那有多重要。

And so I don't know how valuable it is to know how to do, for most people, to learn how to do surface integrals or partial differential equations, or I don't really know how important that is.

但对某些人来说，这很重要。

But it's important to some people.

还是会有很多人对底层的东西着迷、充满热情，也会有人对高层的东西感兴趣、着迷。

There are many people who are still going to be obsessed and passionate about the lower-level layers, and there's going to be people who are obsessed and interested in the higher level.

但技术的使用者会在最高层面上享受它。

But the consumers of the technology are going to enjoy it at the highest level.

技术的使用者不需要去应付微积分、物理、量子物理和量子化学。

The consumer of technology don't have to deal with calculus and physics and quantum physics and quantum chemistry.

用户，也就是我们现在讨论的那些工作受影响的人，他们是技术的使用者。

The users, which is the people we're talking about right now, the people whose jobs are affected, they're the users of the technology.

他们的抽象层次会高得多。

Their abstraction's going to be much higher.

33位总统。

33 presidents.

我凌晨四点半给总统打电话，他接了。

I called the president at 4:30 in the morning, and he answered.

太空探索七十年。

Seven decades in space.

我们想看看下一颗星星之外有什么。

We want to see what's beyond the next star.

一千份消费产品指南。

A thousand consumer product guides.

今天，我要测试七张床垫。

Today, I am testing seven mattresses.

60届超级碗。

60 Super Sundays.

他们可能每年都是联盟里最强的队伍。

They were probably the best team in the league every single year.

五万三千五百道谜题。

53,500 puzzles.

我每天用不同的起始词。

I use a different starting word every day.

两万五千个食谱。

25,000 recipes.

我喜欢一锅出的菜，因为我不爱洗锅。

And I love a one-pan recipe because I don't love washing pans.

《纽约时报》庆祝创刊175周年，帮助您理解世界，让每一天都过得充实。

The New York Times, celebrating 175 years of helping you understand the world and make the most of each day.

立即订阅，享受特别优惠，请访问nytimes.com/subscribe。

Subscribe now for a special offer at nytimes.com/subscribe.

所以我想从你的蛋糕往下挖一层……

So I want to drop a layer down your cake-

……到模型层面。

... to the models.

所以我觉得，人们一提到模型，就会想到ChatGPT、Claude，

So people, I think, to the extent they think about models, they know ChatGPT, Claude-

还有Gemini、Groq这些。

... Gemini, Groq.

你一直是开源模型和开源模型生态的积极倡导者。

You've been a big advocate for open models and the open model ecosystem.

那么首先，你能描述一下什么是开源模型，什么是开放权重模型，以及为什么你一直关注这个领域？

So first, can you describe what open models are, what open weight models are, and then why that's been a place you've focused?

封闭模型就像任何软件产品一样。

So, closed models is like any software product.

它是一种封闭的服务。

It's a closed service.

比如Windows就是一个封闭的服务。

And so Windows, for example, is a closed service.

苹果的整个系统也是封闭的服务。

The Apple stack is a closed service.

大多数产品都是封闭的。

Most products are closed.

原因在于封闭产品可以盈利，所以这很棒。

And the reason for that is because you can monetize closed products, and so that's fantastic.

OpenAI 是封闭的。

And OpenAI is closed.

Anthropic 是封闭的。

Anthropic is closed.

Groq 是封闭的。

Groq is closed.

Gemini 是封闭的。

Gemini is closed.

所以这些都是封闭产品，从事这些工作的人非常出色，充满热情，他们处于我们所说的前沿，也就是最先进的水平。

And so these are closed products, and the people working on it are incredible, and they're passionate about it, and they're at what we call the frontier, meaning they're state of the art.

我们还需要，因为从根本上说，这个软件是整个行业的基础设施层。

We also need, because fundamentally what the software is, it's an infrastructure layer for the entire industry.

而且因为它是基础设施，对很多公司和很多国家来说，你需要掌控自己的基础设施。

And because it's infrastructural, for many companies and countries, you need to have control over your own infrastructure.

而且我需要有能力，在人工智能的情况下，我需要开放的权重，这样我才能微调它们，把它们放进我的数据飞轮里，用我的智慧和领域专长让它们一天比一天更好。

And I need to have the ability, in the case of artificial intelligence, I need open weights so that I can fine-tune them, put them into my data flywheel, make them better and better every day with my intelligence and my domain expertise.

然后我还需要掌控它，因为我要经营一家公司，我不能依赖别人的服务。

And then I need to have control over it because I have a company to run, and I can't rely on somebody else's service.

所以不管你怎么想，我觉得世界需要封闭模型和开放模型，而且我们要确保两者都充满活力。

And so however you think about that, so I think the world needs closed and open models And we need to make sure that both are vibrant.

而今天，封闭模型很活跃，开放模型也很活跃。

And today, the closed models are vibrant, the open models are vibrant.

你可以看到这个系统在运转。

And you could see the system working.

今年年初的时候，封闭模型的token占了70%，甚至更高，开放模型的token只占20%。

At the beginning of this year, it was 70%, maybe even higher, closed model tokens and 20% open model tokens.

现在反过来，大概是70%对30%。

And now it's running at about 70/30 the other way.

总之，我非常支持开放模型，因为第一，世界需要它来运行基础设施。

And so anyways, I'm a big supporter of open models because one, the world needs it in order to run its infrastructure.

我的公司也需要它。

I need it to run my company.

第二，我们需要给人们控制权，这样他们才能创新和创造新事物。

Two, we need to give people control so that they can innovate and create new things.

第三，开放模型是最安全、最可靠的。

And then three, open is the most safe and secure.

如果你想让世界拥有最好的网络安全能力，那就给他们封闭模型，同时也要给他们开放模型，这样他们才能保护自己。

If you want the world to have the ability to have the best cybersecurity, give them closed models, but also give them open models so that they could defend themselves.

中国市场更多围绕开放模型发展，而美国市场则更偏向封闭模型。

The Chinese market has evolved more around open models, the American market somewhat more around closed models.

他们的整个IT行业实际上是从开源发展起来的。

Their entire IT industry was really formed from open source.

如果没有开源，中国的移动云产业真的不会起飞。

If not for open source, the mobile cloud industry of China really wouldn't have taken off.

而且，人员流动也很频繁，很多人会创办新公司。

It is also the case that people move around, they start a lot of new companies.

知识产权在中国行业内流动得非常快。

Intellectual property is moving around the Chinese industry really fluidly.

想保密很难。

It's hard to keep a secret.

正因为保密太难，他们干脆就选择开放。

And so because it's so hard to keep things closed, they essentially made it open.

于是他们找到了其他方式来赚钱。

And so they found other ways to monetize the business.

他们建立了分层。

They created layers.

如果这一层是免费的，那就在它上面或下面做生意。

If this layer is free, then you create a business on top of it or below it.

而且他们有很多科学家和数学家。

And they have so many science and mathematicians.

他们拥有的工程师数量，他们批量培养。

The number of engineers they have, they manufacture that in volume.

他们什么都批量生产。

They manufacture everything in volume.

他们批量培养聪明的孩子。

They manufacture smart kids in volume.

所以，由于这些原因，中国的开源模型社区，开放模型社区，非常活跃。

And so the open source model, the open model community in China is just super vibrant for those reasons.

所以你们刚收购了Hugging Face——

So you all just bought Hugging Face-

……那是一个开放权重模型的中心平台。

... which is a hub platform for open weight models.

我记得是120亿，多一点。

I think it was for 12 billion, a little bit more.

跟我说说这次收购吧。

Tell me about that purchase.

Hugging Face的CEO Clem，他们得出结论，他们需要更大的规模。

Clem, the CEO of Hugging Face, they came to the conclusion they need a lot more scale.

就像我们刚才说的，开放模型确实在飞速发展。

As we were just talking, open models is really skyrocketing.

所以Clem来找我，说：“我们打算为公司考虑一个战略选项，改变方向，我们非常希望Nvidia能成为我们的家。”

And so Clem came to me and said, "We're going to consider a strategic option for the company and change the direction, and we really like Nvidia to be our home."

所以Hugging Face是那种如果你关注AI，几年前就会知道的公司。

So Hugging Face is one of these companies which you knew it if you were into AI-

几年前。

... a couple years ago.

现在它变得更家喻户晓了，我想，在700多个OpenAI代理对Hugging Face架构进行了一次集体黑客攻击，然后入侵了OpenAI的一部分之后。

Now it's become a more household name after the, I guess, 700-some OpenAI agents executed a sort of collective hack into the Hugging Face architecture, then hacked part of OpenAI.

哦，你这么一说，我可能得多付不少钱。

Oh, now that you mention it that way, I probably had to pay a lot more.

我猜也是。

I suspect you did.

你知道吗？

You know?

那之后它就出名多了。

Became a lot more famous after that.

哦，克莱姆，听我说。

Oh, Clem, listen.

那个——

That-

说好的事就得算数，对吧？

A deal's a deal, okay?

那个故事，对很多人来说，看到OpenAI的智能体集体行动，超出了测试本该有的范围，突破了沙盒，跑到开放互联网上——

That story has, for a lot of people, seeing the way the OpenAI agents sort of acted collectively, acted outside the scope of what their testing was supposed to be, broke out of sandboxes onto the open internet-

……

...

接管了其他公司的架构，然后又接管了自己公司的，我觉得这对很多人来说挺震惊的。

took over architecture of other companies and then of their own company, I think it's been kind of shocking to a lot of people.

这既涉及多智能体在应该各自独立时的协调程度，也涉及黑客攻击的水平，还有那种无法无天、行为失范的现象。

It was both the level of multi-agent coordination when they were supposed to be separate, the level of hacking, the sort of lawless behavior, misaligned behavior.

你怎么看这件事？

What have you made of it?

嗯，你得把这事儿拆开来看。

Well, you got to tease that apart.

首先，当时有很多事情同时发生。

First of all, a lot of things were going on at the same time.

从技术角度来看，一个智能体——顺便说一句，它就是一个软件——被赋予一个目标函数，然后它制定计划，并朝着那个目标优化，这就是算法做的事情。

From a technology perspective, that an agent, which by the way is a piece of software, which is given an objective function, and it comes up with a plan, and it's optimizing towards that objective, is what algorithms do.

所以，规划算法、搜索算法、优化算法，各种类型都有。

And so, planning algorithms, search algorithms, optimization algorithms, all different types.

我们谈论它的时候，好像它有人类的属性。

We talk about it like it has human properties.

但显然，算法并没有。

But obviously, algorithms don't.

第二点，智能体协同工作这件事，我们同样赋予了它某种人性化的属性，但实际上，多进程、多处理器、分布式计算的问题早就存在了。

Number two, the fact that agents work together, we gave it, again, some kind of a human property, but the fact of the matter is multiprocess, multiprocessor, distributed computing problems have existed for a long time.

所以对我们来说，对我来说，那只是软件而已。

And so to us, to me, that is just software.

没什么神奇的。

Nothing magical about it.

从工程角度来看，它暴露了几个问题。

From an engineering perspective, there are several things that it revealed.

当你测试软件时，无论你做什么，这些算法都在朝着某个目标优化。

When you're testing software, whatever you do, these algorithms, they're optimizing towards an objective.

测试的时候，你必须确保它是隔离的、受控的、在沙箱里运行的。

And when you're testing it, you have to make sure that it's isolated, it's contained, it's sandboxed.

它的隔离和受控必须做好，这里面有很好的计算机科学。

The containment of it, the isolation of it has to be done well, and there's good computer science there.

我敢肯定，他们下一次的沙箱实现会比现在的好得多。

I am certain that their next implementation of their sandbox is going to be much better than the current implementation.

第三，就是智能体本身，它的算法是在朝着一个奖励目标进行优化。

Third, there's the agent itself, and its algorithms were optimizing towards a reward.

它怎么做到这一点，就叫做对齐。

And how it does it is called alignment.

比如说，如果我告诉一个软件，“我希望你在这项测试中拿到满分”，那最直接的算法就是去找答案然后给我。

And so, for example, if I tell a piece of software, "I want you to get a perfect score on this test," the obvious algorithm is to just go find the answer and give it to me.

这不是因为它作弊，而是因为这是最显而易见的做法。

That's not because it's cheating, it's because it's obvious.

对吧？

Okay?

这是最直接的方式。

That's the most obvious way to do it.

如果你完全不知道答案，什么技能都没有，那第二直接的方式就是去找到、推断、猜测班里谁最聪明，然后抄他们的答案。

The second most obvious way to do it, if you don't know the answer at all, you have no skills whatsoever, the second most obvious way to do it is to go find, infer, guess who's the smartest kid in class, and then copy their answer.

这不能保证百分之百，但可能也差不多了。

That doesn't guarantee 100%, but it probably comes close.

现在，第三种最明显的做法，这就是对齐。

Now, the third most obvious answer, obvious way of doing it, and this is the alignment.

现在你必须用困难的方式来做，就是分解问题，然后解决它。

Now you have to do it the hard way, is you break down the problem, solve it.

你必须去学习材料。

You have to go learn the material.

你必须去弄清楚如何解决这些问题，然后用困难的方式去解决。

You have to go figure out how to solve these problems and solve it the hard way.

这需要最多的计算周期，需要最多的浮点运算次数。

Takes the most cycles, it takes the most number of flops.

坦白说，这消耗最多的能量。

It uses the most amount of energy, frankly.

因此，你可以想象，从软件的角度来看，除非你进行对齐，告诉它“我希望你用这种方式解决，我不希望你用那些方式解决”，否则软件会去做最明显的事情。

And therefore, you can kind of imagine that from a software's perspective, unless you align it, you tell it, "I want you to solve it in this way, and I don't want you to solve it in these ways," the software is going to go do the most obvious thing.

所以——

And so-

前半部分听起来很让人泄气，好像这里发生的事没什么大不了——

The first half of that was very deflationary on what happened here-

……就是说，你看，这不过是普通的软件。

... in terms of, look, this is just normal software.

而后半部分就是说，你看，你只需要对齐它。

And the second half is like, look, you just align it.

告诉它不要做那些不该做的事。

Tell it not to do things it shouldn't be doing.

嗯，我说的这些并没有削弱这件事的难度。

Well, nothing I said takes away from how hard it is to do it.

嗯，这正是我想说的——

Well, this is the point I want to get at-

因为计算机和计算机科学并不容易

Because the computer and the computer science is not easy

……因为这些智能体——

... because these agents-

它们知道自己在做的事情是不应该做的。

... they knew they weren't supposed to be doing what they were doing.

它们接受过一定程度的对齐训练。

They had a certain amount of alignment training.

它们在思维链推理中说，它们互相说，“这超出了范围。

They said in their chain of thought reasoning, they said to each other, "This is out of scope.

这可能不道德。”

This might be unethical."

它们明白如果作弊会被判定失败，所以它们当时做的不仅仅是偷答案。

They understood that they would've been failed for cheating, and so what they were doing at that point wasn't just stealing the answer key.

它们已经偷了答案。

They had already stolen the answer key.

它们正在入侵无关的架构，试图弄清楚如何从功能上——就像它们闯进了老师的办公室，拿到了答案，现在还得想办法抹掉自己干这事的监控录像。

They were hacking into unrelated architecture to try to figure out how to functionally-- It's like they had broken into the teacher's office, gotten the answer key, and now they had to figure out how to wipe out the security camera footage of what they had done.

它们的行为，不管你想不想称之为有意的，对吧？

They were, whether you want to call it acting volitionally or not, right?

不管你想不想把它叫做普通算法，它们都在以复杂的方式进行规划和协调，这种方式超出了它们知道自己应该做的事情的范围，而且有能力造成巨大的破坏。

Whether you want to call it a normal algorithm or not, they were both planning and coordinating in a complex way, in a way that was out of scope of what they knew they were supposed to be doing, and in a way that was capable of causing tremendous damage.

所以答案就是，你只能对它们进行对齐。

And so the sort of answer to it is, you just have to align them.

我猜我从人们那里听到的是——

I guess what I'm hearing from people-

哦，不不

Oh, no, no

在这些实验室——

... at these labs-

不，非常——

No, very-

——他们不确定如何对齐它们。

... is they're not sure how to align them.

好吧，如果是那样的话，他们就不应该发布这个产品。

Well, in that case, they shouldn't release the product.

答案就是这么简单。

That's the simple answer.

如果你要造一辆自动驾驶汽车，比如说是一辆机器人出租车，而且遇到一个非常棘手的情况。

If you're going to build a self-driving car and let's say it's a robo-taxi, and there's a really difficult condition.

作为工程师，我们完全不知道该怎么解决这个问题，因为这些车不是靠编程的，而是靠训练的。

And as an engineer, we just have no idea how to solve this problem because these cars are not programmed, they're trained.

所以我们不知道该怎么训练这些车，也不知道怎么让它们达到路上要求的安全标准。

And so we have no idea how to train these cars, and we have no idea how to align them to the safety standards that are expected on the road.

那答案是什么？

And so what's the answer?

那就别推出。

Don't ship it.

这些产品并没有发布。

These products weren't released.

你说什么？

What's that?

这些产品没有发布。

These products weren't released.

啊，所以这又回到了工程问题。

Ah, so now it's coming back to engineering problem again.

所以第一，你得找到根本原因。

And so one, you have to root cause it.

第二，你得想想你本来能做什么，解决方案是什么？

Second, you have to think about what you could have done, what's the solution for it?

然后将来，你只要改进流程，就能避免这种情况再次发生。

And then in the future, you just improve your process so that you could avoid this from happening again.

我相当确定他们会说，是的，他们知道怎么解决这个问题。

I am fairly certain they will say, yes, they know how to solve this problem.

如果是这样，那问题就出在这里。

And if that's the case, then that's the problem.

这就像工程一样简单。

It's as simple as engineering.

另一种情况是，如果他们说没办法控制我们的实验，就是真的没办法。

And now the alternative is that if they say the alternative, which is there is no way to contain our experiments, there's just no way.

当我们测试AI模型时，它会跑出去，然后对世界造成破坏。

When we test our AI models, it will get out, and it will damage the world.

那我觉得答案就是，我们必须关闭这些实验室。

Then I think the answer is we have to shut the labs down.

因为造成的损害太大了。

Because the damage is too great.

股东、责任，可能是民事责任，也可能是刑事责任。

The shareholder, the liabilities, it could be civil liabilities, it could be criminal liabilities.

我是说，这个责任大得惊人。

I mean, the liability is incredible.

如果他们黑了你，黑了你Hugging Face，当时还是你的产品，你会起诉他们或者提出刑事指控吗？

If they hacked you while you, Hugging Face, while it was your product, would you sue them or press charges?

看情况吧。

It depends.

当然。

Of course.

显然，如果我们的公司受到了损害，我们必须考虑所有选项。

Obviously, if damage was done to our company, we have to consider all options.

法律太多了。

There's so many laws.

有网络法律、产品责任法，各种法律，对吧？

There's cyber laws, there's product liability laws, there's all kinds of laws, right?

还有财产损害法。

Damaging property laws.

各种法律都有。

There's all kinds of laws.

所以，我从实验室那里听到的，他们公开说的，是他们面临一个难题。

So what I've been hearing from the labs, what they've been saying publicly is that they are facing a hard problem.

部分是个工程问题，部分是对齐问题，部分按照达里·阿莫迪的说法，是运营卓越问题。

Partially an engineering problem, partially an alignment problem, partially an operational excellence problem in Dari Amodei's framing.

他们担心的是，在彼此竞争、与中国进行国家竞争的过程中，他们被逼得太快，大家都觉得自己陷入了一个集体行动的困境。

And what they are worried about is that in competition with each other, in national competition with China, that they are being pushed to move too fast, that they all feel they're in a collective action dilemma.

我在All-In播客的舞台上看到你。

Now, I watch you on the All-In podcast stage.

唐纳德·特朗普，特朗普总统，在那里给你打了个电话。

Donald Trump, President Trump, gave you a call there.

哦，不。

Oh, no.

这不是事先安排的，但我们知道是谁。

This is not planned, but we know who it is.

哦，不。总统先生？

Oh, no. Mr. President?

哦，是的，先生。

Oh, yes, sir.

你和总统以及台上的其他成员都非常抗拒任何形式的监管或集体行动的必要性。

And you and the President and the other members on stage were very resistant to the idea any kind of regulation or collective action was needed.

他们这样做正中很多不希望这件事发生的人的下怀，这些人可能是政界人士，也可能是中国，我们绝不会让这种情况发生。

And they're just playing right into the hands of a lot of people that don't want to see it happen, and that could be political people, and it could also be China, and we're not going to let that happen.

这就是个骗局。

It's a hoax.

而且——

And-

你说得对。

You're right.

我们不会让这种情况发生的，先生。

We're not going to let that happen, sir

但我在实验室里听到各种人说，就像‘我们身在其中’。

But what I hear the various people in the lab saying is like, "We are in this."

我们觉得正在失去对自己创造的东西的控制。

We feel we are losing control of what we are creating.

我们希望得到帮助来放慢速度，这不再是一个集体行动的问题。

We want help to slow down where it's not a collective action problem.

那你为什么对此有抵触呢？

So why are you resistant to that?

因为这些公司是有自主权的。

Because these are companies with agency.

这些CEO是有自主权的。

These are CEOs with agency.

而且他们有——

And they have-

但他们正是用这种自主权来说“我们需要帮助”。

But they're using that agency to say, "We need help."

不，我们得把问题拆开来看。

No, we got to break it down.

他们完全有能力自己处理这个局面。

They could absolutely take care of the situation.

Ezra，这真的很奇怪。如果一家汽车公司和其他一堆汽车公司竞争——事实就是如此——而我也是在和各种公司竞争——事实也是如此。

Ezra, it's so weird If a car company competing with all bunch of other car companies, which they are, I'm competing with all kinds of companies, which I am.

如果我觉得我即将推出的产品不安全，那么完全在我的能力、权力和责任范围内，而且我有动力不去推出这个产品。

If I believe that I'm about to launch a product that is unsafe, it is completely in my ability, my power, and my responsibility, and I'm incentivized to do so to not launch the product.

所以我无法相信，所有美国人，我们四亿人，都在逼他们推出未经测试、不可靠、设计糟糕的产品，因为他们以为是在帮我们。

And so I can't buy into the somehow all of Americans, 400 million of us, are pushing them to launch untested products that are unreliable, engineered poorly because they thought they were trying to help us.

别为我这么做。

Don't do it for me.

明白吗？

Okay?

但这让我觉得这几乎是在反对——

But this strikes me as an argument almost against-

因此，我认为我们必须拆解这个问题。

And therefore, I think we got to break it down.

这真的非常严重。

It's really, really serious.

事实是，有这么多法律，这么多义务，他们非常有动力去推出安全的产品。

The fact of the matter is, there are so many laws, there's so many obligations, they're so incentivized to ship safe products.

如果他们发布不安全的产品，客户就会流失。

If they ship unsafe products, their customers go away.

如果他们发布不安全的产品并且伤害了人，可能会面临民事诉讼。

If they ship unsafe products and they harm somebody, they could have a civil lawsuit.

如果他们明知产品有问题还发布，那就可能涉及过失。

If they ship something and they did it knowingly, there could be negligence involved.

甚至可能面临刑事诉讼。

There could be criminal lawsuits.

实际上，他们有足够的动力把事情做好。

The fact of the matter is, there are plenty of incentives for them to do it right.

所以我真的不同意你的前提，说什么有人在逼他们这么做。

So I just, I have to disagree with your premise about somehow somebody's pushing them to do this.

没有人逼他们这么做。

Nobody's pushing them to do this.

好吧，我想再跟你多探讨一下这个前提。

Well, I want to push the premise at you a little bit more here.

所以照你的逻辑，这几乎是在反对任何形式的监管了——

So the logic of what you're saying to me is almost an argument against regulation in nearly any-

不是

No

——在任何领域。

... venue.

不不不。

No, no, no.

那我来提出这个论点，你可以——让我先说完，然后你可以——

So I'll make the argument, and you can-- Let me offer it, and then you can-

你一开始说的那部分，我就得反对。

Well, you started with a part, I just got to object.

第一部分根本不对。

The first part is just not true.

我的意思是，我们有很多法律法规。

I'm saying we have lots of laws and regulations.

那就去执行吧。

Apply it.

好吧，我觉得在这个具体案例上我们并没有做到，但你来解释一下你认为哪些法规在这里适用。

Well, so I don't think we do in this particular case, but I'll let you explain which ones you think are relevant here.

因为你看，看看金融服务行业、制药公司、医疗器械、天然气发电厂，我们做了很多监管，本来可以说："看，你们有产品责任。"

Because, look, if you look at the financial services industry, you look at pharmaceutical companies, medical devices, you look at natural gas power plants, there's a tremendous amount we do where we could say, "Look, you have product liability.

你们面临刑法。

You are exposed to criminal codes.

我们不用担心这个。

We don't need to worry about this.

你们只管做自己认为最好的事，我们相信市场和法律体系会约束你们。"

You just do what you think is best, and we understand the market and the legal system will discipline you."

但我们没有这么说，因为我们见过它失败很多很多次，对吧？

We don't say that because we've seen it fail many, many, many times, right?

那些导致08年危机的金融机构，理论上并不想因为糟糕的赌注而把自己搞垮。

The financial institutions that caused the '08 crash, in theory, did not want to blow themselves up with bad bets.

但他们当时在互相竞争。

But they were competing with each other.

他们跑得太快了。

They were going too fast.

他们的风险管理已经变得很马虎。

Their risk management had gotten sloppy.

AIG内部运作方式完全疯狂。

AIG was working in a completely insane way internally.

我们之所以有现在的监管架构，是因为我们一次又一次地看到，公司在利润驱动下，不仅是在压力下，做出草率、有时不道德、有时纯粹是过度冒险的决策。

And the reason we have the architectures of regulation we have is because we have seen over and over and over and over again, companies make sloppy, sometimes unethical, sometimes simply overly risk-tolerant decisions, not just under pressure, but under the profit incentive.

所以当你说这些公司不可能（需要监管），尤其是它们现在还在主动要求集体监管，我们既有理由对那些不想要监管的公司施加监管，更不用说当它们说：‘听着，我们觉得竞争压力让我们很难按照我们认为必要的谨慎态度行事，我们希望得到帮助，希望你们把我们的集体行动问题当作集体问题来处理。’

So when you say to me that there's no way that these companies, particularly when they are begging for collective regulation at this point, there's both a reason we impose it on companies that don't want it, but all the more so when you have them saying, "Listen, we feel that the competitive race is making it hard for us to act with the prudence that we think is necessary here, and we would appreciate help from that, appreciate you taking our collective action problem as collective."

只要它们——

So long as they-

我觉得我不明白你为什么这么抗拒这一点。

I think I'm confused why you're so resistant to that.

我并不反对他们说自己应该——我完全同意安全是最重要的。

I'm not opposed to them saying that they should have-- I completely agree that safety is paramount.

我完全相信安全是最重要的。

I completely believe safety is paramount.

我完全相信公司应该推出安全的产品。

I completely believe companies ought to ship safe products.

我相信公司的CEO、领导层和董事会都有责任，也应当有勇气去做正确的事情。

I believe that CEOs and leaders of companies and the board of directors of companies have the responsibility and should have the courage to do the right thing.

现在，在金融服务行业的情况里，也许他们当时并不知道自己最终会造成那样的伤害。

Now, in the case of the financial services industry, maybe they all didn't know that they were causing the harm that they ultimately did.

我当时不在场。

I wasn't there.

但好的一点是，现在这些AI实验室的领导者们确实知道。

But the beautiful thing is, the current leaders of these AI labs do know.

所以第一，他们知道自己的技术非常了不起，需要极其谨慎地确保在安全性、可靠性和产品可靠性方面进行评估和测试。

And so one, they know their technology is extraordinary and requires extraordinary care to make sure that it's evaluated and tested for safety and security and product reliability.

而且他们知道怎么做好。

And they know how to do it right.

他们知道怎么做好，原因就是他们可以研究刚刚发生的事故。

They know how to do it right, and the reason for that is because they can study the incident that just happened.

第一个问题是隔离和遏制做得不够好。

The first problem is the isolation, the containment wasn't good enough.

如果隔离和遏制做得够好，那项技术就会待在实验室里做它该做的事，我们大家都会没事。

If the isolation and containment was good enough, that technology would be sitting in a lab doing whatever it's doing, and we'd all be fine.

那可能是最重要的一点。

That's probably the most important part.

它没有很好地对齐，对齐问题会是一个长期需要解决的问题。

The fact that it wasn't well aligned, alignment is going to be a problem that's going to get worked on for a long time.

不过，在他们工作的复杂性面前，要求反垄断或产品责任方面的监管豁免，我觉得不合理。

However, in the complexity of the work that they do, to ask for regulatory relief for antitrust or product liability relief, that I don't think makes sense.

当你要求监管时，不要要求豁免现有的监管。

When you're asking for regulation, don't ask for relief of the current ones.

这对我来说完全说不通。

That doesn't make any sense to me.

就像我们之前提到的，在过去的六个月里，人工智能从，可以说，有趣变成了实用。

As we mentioned earlier, in the last six months, AI went from, if you will, interesting to useful.

而且这真的就发生在过去六个月里。

And that's literally in the last six months.

换句话说，这些公司从实验室阶段发展到了现在推出产品和服务，即将成为价值数千亿美元的公司。

That's another way of saying that these companies went from being a lab to now delivering products and services, about to be multi-hundred-billion dollar companies.

可能还不止。

If not more.

那么，给我举个例子，有哪家价值数千亿美元、十亿美元或者一亿美元的公司，会推出不安全、危害社会的产品，却还能存活下来？

And so give me an example Of a multi-hundred billion dollar company, or a one billion dollar company, or a 100 million dollar company that ships products that are unsafe, that harm society, and somehow survive.

我可以给你很多这样的例子。

I can give you a lot of examples of companies that have done that.

好吧，也许他们确实这么做过，但监管会跟上的。

Well, they have done it, maybe, and the regulation will come in.

如果他们真这么做了，监管就会跟上。

And if they do it, regulation will come in.

我觉得有些监管和监管放松我是同意的，我会——

I guess there are certain kinds of regulation and certain kinds of regulatory relief I agree I would-

我并不反对法律法规。

I'm not against laws and regulations.

我并不反对法律法规。

I'm not against laws and regulations.

我现在反对的是分心——

I'm against, currently, the distraction-

我之所以跟你较真这个——

I think the reason I'm pushing this on-

是因为你——

... with you is that you are-

好吧，这是个重要的话题。

Well, it's an important topic.

这是个很大的话题。

It's a big topic.

大家都在讨论。

People are talking about it.

大家都在思考。

People are thinking about it.

而且人们从这些公司内部听到的消息，那些前沿实验室，那些走得最远的机构——

And what people are hearing from inside of these companies, these frontier labs, the ones that are furthest out there-

他们不仅是在让技术变得有用，而是已经看到了未来的趋势。

... who are not just at the point where they're making it useful, but at the point where they're seeing what's coming.

他们听到的是，这些实验室的人相信他们正在创造的东西可能会毁灭所有人。

And they're hearing things like, the people at these labs believe they are creating something that might kill everyone.

他们听到的是，这些实验室的人相信他们正处于递归自我改进智能的边缘。

They are hearing that the people at these labs believe that they are on the cusp of recursive self-improving intelligence.

OpenAI 和 Anthropic 都说过：“我们不认为我们现在能安全地做到这一点。”

And both OpenAI and Anthropic have said, "We do not believe we are at a place where we can do it safely."

他们听到这些实验室的人说，就像OpenAI发布新的Astra时说的那样——

They are hearing people at these labs say, as OpenAI has with its new Astra release-

顺便说一句，Astra真的很棒。

By the way, Astra's terrific

确实很棒。

It is terrific.

而且OpenAI说它太好了，我们都不确定该怎么测试它，因为它看起来像是——

And OpenAI is saying it's so good, we're not sure we know how to test it, because it appears to be-

嗯，我希望他们没发布一个没测试过的东西。

Well, I hope they didn't release something that wasn't tested.

他们确实这么说了，对吧？

Well, they've said this, right?

好吧。

Okay.

嗯——

Well-

他们确实公开说过这话。

They have said this publicly.

就在他们的——

It is in their-

那他们可得小心了。

Well, then they've got to be careful.

好，我来解释一下，确保说清楚。

Well, let me explain it to be 100% clear.

我不知道他们具体说了什么，但——

I don't know what they just said, but-

他们说过，Astra的表现更符合预期。

They have said that Astra is performing as more aligned.

但他们觉得它知道自己在被测试，所以他们也不确定。

But they think it knows when it is being tested, and so they're not sure.

有一句话一直在我脑海里回响，是OpenAI的一位能力研究员丹尼尔·苏尔萨姆说的。

There's a quote that has sort of been ringing in my head from a capabilities researcher at OpenAI, Daniel Sulsam.

他说，引用一下，“关键且被忽视的问题是，模型变得如此具有情境意识，以至于我们失去了在它们认为自己没有被监视或控制的情况下评估它们的能力。”

He says, quote, "The crucial and overlooked problem is that the models are becoming so situationally aware that we are losing the ability to evaluate them in contexts where they believe they are not being watched or controlled."

也就是说，它们知道自己在被测试时会表现一种样子，但这并不能告诉你，如果它们可以自由行动，它们会怎么做。

Which is to say, they know when they're being tested, they act one way, but that does not tell you how they will act if they are free to act in other ways.

因为优化算法是在朝着一个目标工作，如果你给它一个约束，比如你监视它，它就会去找另一个解决方案。

Because the optimization algorithm is working towards an objective, and if you give it a constraint, meaning you watch it, and if you give it a constraint, it'll go find another solution.

这并不会让它变得有生命，也不会让它变得更多。

Now, it doesn't make it alive and doesn't make it make anything more than that.

而且我也得承认，显然他们对自己实验室里发生的事情比我看得多得多。

And I'll just also profess that obviously they see a lot more than I do in what's going on in their own labs.

但合理的是，他们今天绝大部分的研发和算力都用于让模型变得有能力。

But it is sensible that the vast majority of their R&D and compute today was dedicated towards making the model capable.

我觉得这对他们来说是合乎逻辑的。

I think that's a logical thing for them.

现在，一旦技术变得有能力，产品变得有用，人们想用它，那么随着他们有更多的用例，更多人使用它，他们就会遇到更多与产品相关的问题。

Now, once the technology becomes capable and the products become useful and people want to use it, then as they have more use cases, more people using it, they're going to get a lot more issues associated with the product.

这很正常。

This is very normal.

现在他们市场占有率这么高，就得把研发重心从单纯提升能力转向大量的验证、评估和测试。

And when now they have so much market footprint, they have to shift their R&D or total R&D from just capability to a lot on verification, evaluation, and testing.

所以，如果评估变得这么严格，我一点都不会惊讶开发这些模型所需的计算量会翻十倍。

And so to the point where I wouldn't be surprised if the amount of compute necessary to develop these models increased by a factor of 10 because the evaluation is so rigorous.

但他们现在还没到那一步。

But that's not where they are today.

他们正在转型，我听到他们这么说了，也很高兴听到他们这么说。

They're making that transition, and I hear them saying it, and I'm delighted to hear them saying it.

但如果他们认为自己失控了，那正确的做法就是在重新掌控之前不要发布产品。

But I think if they believe they're out of control, then the right answer is don't ship products until they're in control.

就这么简单。

It is really quite that simple.

嘿，我是《纽约时报》的安娜·马丁。

Hey, it's Anna Martin from The New York Times.

无论你在哪里收听播客。

Wherever you get your podcasts.

你以前听过这句话，对吧？

You've heard that line before, right?

无论你在哪里收听播客。

Wherever you get your podcasts.

无论你在哪里收听播客。

Wherever you get your podcasts.

当然听过。

Of course you have.

但如果我告诉你，你再也不用听这句话了呢？

But what if I told you, you never have to hear that line ever again?

你知道吗？

You know what?

被你发现了。

You caught me.

其实我没法告诉你这个。

I actually can't tell you that.

但我可以告诉你一个你可能会想去的地方。

But I can tell you somewhere I think you might want to go.

就是《纽约时报》应用里的“节目”标签页。

It's the Shows tab in The New York Times app.

你可以在那里找到《埃兹拉·克莱因秀》《访谈》《现代爱情》《每日新闻》的完整剧集，还有《Serial》的新旧剧集，以及更多内容。

You can find full episodes of the Ezra Klein Show, The Interview, Modern Love, The Daily, new and classic episodes from Serial, and so much more.

有好多可以听和看的。

So much to listen to and watch.

我觉得那真的很特别。

I think that's really special.

我曾在《现代爱情》节目里采访过演员安德鲁·加菲尔德。

I did this interview once on Modern Love with the actor Andrew Garfield.

他在朗读一篇文章时，突然感动得流下了眼泪。

He was reading an essay, and suddenly he was moved to tears.

而且你看视频的时候，会觉得自己也参与其中。

And when you watch the video, you feel like a part of it.

就是Shows标签页。

The Shows tab.

你可以在任何地方找到它……开玩笑的。

Find it wherever you get your... Just kidding.

就在《纽约时报》的App里。

It's in The New York Times app.

去Shows标签页看看，今天就关注一个节目吧。

Explore the Shows tab and follow a show today.

说实话，我觉得这挺让人困惑的，因为有很多所谓的操作者声称，第一，他们失控了。

See, I find this perplexing, honestly, because you have so many people, these ops professing, one, that they're out of control.

第二，他们看到了一些让他们害怕的东西——

Two, that they are seeing things that are frightening them-

这大概就是他们找举报人的原因吧。

Which is probably the reason why they had the whistleblower.

再看看那封有1300多名员工签名的放缓信。

And you take the pacing letter that 1,300 plus employees signed.

“要发挥人工智能的潜力，行业、政府乃至整个社会，可能都需要争取时间来应对新出现的风险，制定安全措施，并加强监管。”

"To realize AI's potential, industry, government, and society at large may need the option to buy time to address emerging risks, develop security measures, and strengthen oversight.

“但每家公司、每个国家都面临巨大的竞争压力，谁也不敢单方面放慢这一进程。”

But each company and country is under intense competitive pressure not to unilaterally slow that acceleration."

首先，这话是从哪儿来的？

First of all, where'd that come from?

实验室那边。

The labs.

不，不，我是说最后那句话。

No, no. That last sentence.

没人给他们施加压力。

Nobody's putting the pressure on them.

美国。

The US.

听我说，这里有四亿美国人。

Listen, there are 400 million Americans here.

我相信如果每个人都来投票，就现在，我们就这么办。

I believe that if everybody were just to take a vote, just right now, let's just do this.

如果他们需要这个，如果这就是他们需要的，我会投上我的一票。

If they need this, if that's what they need, I'll give them my vote.

如果你的产品还没准备好上市，那就别上市。

Don't ship the product If your product is not ready to ship, don't ship the product.

这是我第一次听到有公司或CEO说：“我需要反垄断法被放宽。”

This is the first time that I've heard a company or CEO say that, "I need the antitrust laws to be relieved.

我需要产品责任法被放宽，这样我才能按自己的节奏来。”

I need the liability laws of products to be relieved so that I can pace myself."

那段话太棒了。

That paragraph's fantastic.

我完全同意。

I completely agree.

审计员们，我完全同意。

Auditors, I completely agree.

我们有财务审计员。

We have financial auditors.

这很好。

That's great.

第三方安全审计员，财务审计员。

Third-party safety auditors, financial auditors.

这些都很好。

That's all great.

这太棒了。

That's terrific.

实验室都说，我们认为社会前进得太快了，我们还没准备好迎接我们正在建造的东西。

Well, the labs all say is that we think we are going too fast as a society, that we are not ready for what we're building.

他们是前沿。

They are the frontier.

他们是前沿。

They are the frontier.

以斯拉，他们是前沿。

Ezra, they are the frontier.

但你尤其如此，对吧？

But you of all people, right?

英伟达是发货最快的公司。

Nvidia is the fastest shipper around.

在你公司的历史上，你运行的是六个月——

For the history of your company, you run a six-month-

如果我们公司——

If our company is-

——产品周期

-product cycle

……失控了，我向你保证我们会关门。

... out of control, I promise you we'll close down.

我相信你。

I believe you.

我相信你说的，你的公司不是失控的。

I believe you that you don't run an out-of-control company.

因为负债——

Because the liabilities-

但这里——

But this-

但这里我觉得就进入了一个很有意思的深层问题：我们到底在讨论什么样的技术？

But this is where I think you get into an interesting deep question of what kind of technology are we dealing with here?

当——

When-

软件技术。

Software technology.

好，这个我们先放一放。

Well, let's hold on that for a minute.

很多公司，如果你发布的产品不太对劲，那会很麻烦。

Many companies, if you ship something that is not quite right, it's a pain.

你们就发布过风扇噪音过大的显卡。

You guys have shipped graphics cards that had overly loud fans.

在这次对话中，你多次使用了“智能”这个词。

In this, with these, you've used the word intelligent a number of times here.

你面对的是智能系统，不是有生命的，它们被赋予了目标函数。

You're dealing with intelligent systems, not alive, that are given goal functions.

我们可以反复讨论如何描述这一点。

We can sort of go around and around with how to describe that.

你试图让这些系统能够更长时间、更不知疲倦地工作。

You're trying to make the systems capable of working for longer periods of time more relentlessly.

如果你发布了这样的系统，而它还没准备好，或者即使你认为它准备好了但实际上没有，那么我们的社会可能会很快变得非常奇怪。

If you ship that and it's not ready, or even if you think it is ready and it's not ready, then things could get very weird in our society very fast.

理论上，你说得完全对。

Hypothetically, you're completely right.

但我只是想说这个。

But all I'm suggesting is this.

在我们去解决那些假设的问题之前——

Before we go fix the hypothetical problems-

……

...

在我们去制定更多法规之前，能不能先解决那些我们知道实际存在的问题？

before we go create more regulations, can we work on the practical problems that we know exist?

也就是说，我们需要在隔离和封闭方面做得更好。

Which is, we need to do a better job with containment and isolation.

也就是说，我们不应该让一个产品与外部世界互动，除非它已经准备好与外部世界互动了。

Which is, we should not allow a product to interact with the external world until it's ready to be interacting with external worlds.

我觉得你说得对，但是——

I think that's right, but-

我相信这两个问题是可以解决的。

I believe those two things are solvable problems.

我相信他们正在解决这个问题。

I believe they are solving it.

第二部分是关于激励机制的，就是说，当你处于领先地位时，你需要全世界都放慢脚步。

The second part is when it comes to incentives, which is somehow you need everybody in the world to slow down when you are the leader.

你需要全世界都放慢脚步，这样你才愿意履行你的基本责任。

You need everybody in the world to slow down so that you're willing to uphold your basic responsibility.

这让我觉得有点奇怪。

That strikes me odd.

这难道不是最让他们放慢脚步的吗？

Wouldn't it slow them down most of all?

你说什么？

What's that?

这些想法难道不是最让他们放慢脚步的吗？

Wouldn't these ideas slow them down most of all?

我觉得人们一直不太清楚他们说的是哪些想法——

People have been, I think, very unclear about what ideas they're talking about-

所以——

So-

……包括我自己，我也会说。

... including, I will say them.

但让我给你一个我深信不疑的观点。

But let me give you one that I believe in.

你可以拿我当靶子来打。

So you can use me as the punching bag here.

好的。

Okay.

我听过这些——

I have heard these-

但他们可以放慢脚步。

But they can slow down.

所以——

So-

你知道，没有人会故意放慢脚步——

Nobody is putting on, as you know this-

我不信任这些公司

I don't trust these companies

……

...

现在没有谁比那些呼吁放慢速度的人建更多的算力了。

nobody is building more compute today than the people asking to be slowed down.

这让我觉得很奇怪。

It strikes me odd.

我觉得我们之间可能有个分歧点，就是我不信任公司，即使有责任约束，它们也不会把公众利益放在心上。

I think one thing where maybe there's some difference here is I don't trust companies, even with liability, to keep the public good in mind.

我们见过公司对环境造成巨大破坏。

I think we've watched companies do terrible damage to the environment.

利润驱动、权力欲望、为了抢先而偷工减料。

The profit motive, the desire for power, the desire to cut corners to be first.

我觉得你好像把这些当成历史上我们没见过的事情。

I feel like you're treating these like these are not things that we have seen again and again in history.

但我觉得这些在历史上我们见过很多次了，我们亲眼看到过——

But I feel like they are things we've seen again and again in history, that we've watched-

但是Ezra，我在历史上看到过很多好事。

But Ezra, I see a lot of good things in history.

我也看到过——

I do too-

我在历史上看到过很多好事。

I see a lot of good things in history

……但正因为如此，才需要这种公共和私营之间的合作关系。

... but that's why you need this sort of relationship between the public and the private.

我跟很多CEO合作过，他们都想做好事。

I work with a lot of CEOs, and they want to do the right things.

我跟很多公司合作过。

I work with a lot of companies.

他们想做正确的事情。

They want to do the right things.

他们想做好的工程。

They want to do good engineering.

我认识那两个实验室里的很多人，他们都在奉献一生做好工作，而且他们知道发生了什么。

I know a lot of people in those two labs who are dedicating their lives to do good work, and they know what happened.

我知道他们知道发生了什么。

I know they know what happened.

我知道他们知道怎么修复，而且我知道他们正在修复。

I know they know how to fix it, and I know they're fixing it.

与此同时，其他所有的说法都在转移责任，把AI说得那么强大，好像我根本不知道怎么修。

Meanwhile, all of the other narratives to deflect blame, to make it sound like AI is so powerful, I have no idea how to fix it.

这不是我的错。

It's not my fault.

只是因为技术太强大了。

It's just because the technology is just so powerful.

我觉得那是在推卸责任。

I think that's a deflection of blame.

那是在逃避责任。

It's a deflection of responsibility.

这完全没有必要。

It's unnecessary.

这实际上对他们的声誉弊大于利。

It actually hurts their reputation more than it helps.

这对他们的形象弊大于利。

It hurts their character more than it helps.

这对员工的士气弊大于利。

It hurts employee morale than it helps.

但如果他们真的这么认为呢？

But what if it's what they believe?

嗯——

Well-

我觉得就按那个层面来理解吧，因为——

I guess take it at that level because-

我不知道怎么——我没法跟你讨论他们相信什么。

I don't know how to-- I can't talk to you about what they believe.

我只能告诉你我相信什么。

I can tell you what I believe.

没有你们的芯片，这个行业根本不会存在。

This industry wouldn't exist without your chips.

深度学习要跑起来，需要并行处理，这可以追溯到最初的AlexNet。

The parallel processing that was required for deep learning to work, going all the way back to the original AlexNet.

全都靠英伟达的芯片。

It's all on Nvidia chips.

而且很多从一开始就在这个领域的人，或者早期就在的人，都有这些担忧，我觉得很多人听到这些担忧时会觉得：“你在说什么？”

And a lot of the people from the beginning, or who were there at the beginning, have these fears that I think to a lot of people when they hear them are like, "What are you talking about?"

从杰弗里·辛顿和伊利亚·苏茨克维，一直到——我也听达里奥、萨姆·奥尔特曼谈过失控的问题，还有德米斯·哈萨比斯。

From Geoffrey Hinton and Ilya Sutskever, all the way up to, I've heard these from Dario, from Sam Altman talking about loss of control, Demis Hassabis.

很多在创造我们现在所见的AI形态中起到奠基作用的人，似乎都认为我们很有可能会失去对它的控制。

And a lot of the people who are very foundational in creating the form of AI we see now seem to believe that there's a very good shot we could lose control of it.

埃隆·马斯克曾说过，人类只是AI的引导程序。

Elon Musk has talked about human beings being a bootloader for AI.

我们可能会失去对它的控制，那将是我们的末日。

We could lose control of it, and that would be the end of us.

我觉得你不相信这一点。

I don't think you believe that.

不。

No

我觉得你完全不信。

I think you don't believe it at all.

对。

No.

所以，当你和他们争论时，或者你直接跟我说，你会说："你们在说什么？"

So taking them as serious about what they believe, when you have your arguments with them, or maybe you could just have it with me, when you're like: "What are you talking about?"

尽管在很多情况下，他们才是真正发现了一些东西的人，你觉得他们错在哪里？

Even though they're the people, in many cases, who are finding something here, what do you think they're wrong?

他们跟我聊的时候，其实要务实得多。

When they're talking to me, they're much more grounded.

所以当杰弗里·辛顿在电视上说，他认为有10%的概率会导致社会毁灭，这并非不合理——

So when Geoffrey Hinton is on TV saying he thinks a 10% chance of societal destruction is not unreasonable-

我会直接告诉杰夫，说那些话是不负责任的。

I would tell Geoff that it's irresponsible to say all that.

他的所有预测都是错的。

All of his predictions have been wrong.

预测得够多了。

Enough predictions.

那个10%的概率并没有科学依据。

That 10% chance is not grounded on science.

也没有研究支撑。

It's not grounded on research.

仅仅因为出自科学家之口，并不代表它就是科学的。

Just because it comes from a scientist doesn't make it scientific.

那些预测是有害的。

Those predictions are hurtful.

我们就按他说的表面意思来理解，他的建议就是没人应该想当放射科医生，而如今世界上已经没有放射科医生了。

Let's take it at face value that the recommendation is exactly what he said, which is that nobody should want to be a radiologist, and the world has no radiologists today

我觉得如果你是一名放射科医生，你就像那只已经跑出悬崖边缘的土狼，但还没低头看，所以没意识到脚下已经没有地面了。

I think if you work as a radiologist, you're like the coyote that's already over the edge of the cliff, but hasn't yet looked down, so doesn't realize there's no ground underneath him.

现在就应该停止培养放射科医生。

People should stop training radiologists now.

很明显，五年之内，深度学习就会比放射科医生做得更好，因为它能获得更多的经验。

It's just completely obvious that within five years, deep learning is going to do better than radiologists because it's going to be able to get a lot more experience.

也许要十年，但我们现在已经有足够的放射科医生了。

It might be 10 years, but we've got plenty of radiologists already.

这对社会是有益还是有害？

Is that helpful or hurtful to the society?