Hard Fork
The Ezra Klein Show: Jensen Huang Thinks A.I. Alarmism Has Gone Too Far
00:0001:48:18
00:00
I'm John Caramonica, a critic at the New York Times.
我是 John Caramonica，New York Times 的评论员。
00:03
And I'm Joe Cascarelia, a cultural reporter at the Times.
我是 Joe Cascarelia，Times 的文化记者。
00:05
Together we host the pop cast, a weekly pop culture chat show,
我们俩一起主持 pop cast，一档每周播出的流行文化聊天节目，
00:08
where we speak to the biggest musicians, actors,
在节目里我们会对话最顶级的音乐人、演员、
00:11
internet celebrities, and more.
网络名人等等。
00:12
We've had bad bunny Olivia Rodrigo and Ann Hathaway.
我们请来过 bad bunny、Olivia Rodrigo 和 Ann Hathaway。
00:15
We've also hosted live performances
我们还举办过现场演出，
00:17
with Andre 3000 and Erica Badu.
演出嘉宾有 Andre 3000 和 Erica Badu。
00:19
Plus we end every conversation with a snack.
另外，我们每次聊完都会来点零食。
00:22
Catch new episodes of pop cast every week on YouTube
每周都能在 YouTube 上追 pop cast 的新一期，
00:25
and anywhere you get your podcasts.
还有任何你收听播客的地方。
00:39
Hey, hard fork listeners, this is Ezra Klein.
嘿，hard fork 的听众们，我是 Ezra Klein。
00:41
The hard fork team is working on something new for this feed.
hard fork 团队正在为这个 feed 做点新东西。
00:45
But in the meantime, they thought you'd enjoy this conversation
不过在那之前，他们觉得你会喜欢这段对话，
00:47
I had with Jensen Huang, the founder and CEO of NVIDIA.
就是我和 NVIDIA 创始人兼 CEO Jensen Huang 的对话。
00:50
So here it is.
所以，这就来了。
01:22
Over the course of these last few weeks
在过去这几周里
01:24
where the whole world has been talking
全世界一直在谈论
01:26
about artificial intelligence, the voices people
artificial intelligence，而人们
01:28
have been hearing most loudly are from the frontier labs,
听到最多的声音来自 frontier labs，
01:31
both their CEOs and leaders and their staffers.
包括他们的 CEO、领导者以及员工。
01:35
These are the labs making the very advanced AI models
这些实验室打造的是非常先进的 AI 模型，
01:38
like Claude, and Chachi B.T., and Gemini, and others.
比如 Claude、Chachi B.T.、Gemini 等等。
01:42
But they're not the only perspective on AI.
但他们并不是关于 AI 的唯一视角。
01:44
Probably the single most influential person
大概最有影响力的那一个人
01:46
in artificial intelligence is Jensen Huang, the CEO of NVIDIA.
在 artificial intelligence 领域，就是 Jensen Huang，NVIDIA 的 CEO。
01:50
NVIDIA is now the largest company in the world,
NVIDIA 现在是全球最大的公司，
01:52
$5.4 trillion in market cap.
market cap 是 5.4 万亿美元。
01:54
I found this statistic amazing.
我觉得这个统计数据太惊人了。
01:56
Since 2023, 15 cents of every single dollar,
从 2023 年开始，美国股市的回报中，
02:00
the American stock exchange's return
每一美元里有 15 美分
02:02
has been from NVIDIA stock.
是来自 NVIDIA 的股票。
02:04
And the reason is that NVIDIA is the material and software
原因是，NVIDIA 是现代 AI 赖以构建的硬件和软件底座。
02:09
substrate on which modern artificial intelligence is built.
NVIDIA 的芯片之所以流行，并不是因为 AI 流行。
02:12
NVIDIA's chips are not popular because AI is popular.
现代形态的 AI 之所以成为可能，是因为 NVIDIA 的芯片流行。
02:16
AI in its modern form was made possible
它们最初是为 graphic processing、电子游戏这类东西而造的。
02:18
because NVIDIA's chips were popular.
但事实证明，那种 parallel computing
02:20
They were originally made for graphic processing, video games,
02:22
that kind of thing.
02:23
But it turned out the kind of parallel computing
02:25
they were doing and the way they were programmable
它们在做的那些事，以及它们可以被编程的方式
02:28
was exactly what was needed to make deep learning
正是让 deep learning
02:30
in its modern form work.
以现代形式真正运转起来所必需的东西。
02:33
Huang is not just influential in terms
Huang 的影响力不仅仅体现在
02:35
of controlling one of the central resources
控制着核心资源之一——
02:37
for training new AI models and using them
用来 training 新的 AI models，并使用它们
02:40
to answer questions and create intelligence in the world.
来回答问题，并在世界上创造智能。
02:43
He's also become very, very influential
他也变得非常、非常有影响力。
02:45
in the Trump administration.
在 Trump 政府里。
02:47
And Huang has a very different perspective
而 Huang 的看法截然不同
02:49
than some of the lab leads.
和一些实验室负责人相比。
02:50
He's worried about safety, but sees it
他担心安全问题，但认为这
02:52
as a very solvable engineering problem.
是一个完全可以靠工程解决的问题。
02:55
He is worried about the direction things are going in
他对事情的发展方向感到担忧
02:57
but does not want to see new regulation to change it.
但不想看到新的监管来改变它。
03:00
And so I wanted to see how Huang perceives
所以我想看看 Huang 是如何看待
03:03
AI, what his model is for thinking about it,
AI，他用来思考这件事的 model 是什么，
03:05
what he thinks is going wrong
他认为哪里出了问题，
03:07
and what he thinks would need to happen for it to go right.
以及他认为需要发生什么，才能让它走上正轨。
03:11
So I came out to Santa Clara
所以我就来到了 Santa Clara
03:12
to NVIDIA's headquarters to interview him.
到 NVIDIA 总部去采访他。
03:15
He joins me now.
他现在和我在一起。
03:21
Jensen Huang, welcome to the show.
Jensen Huang，欢迎来到节目。
03:23
Thank you, it's great to see you.
谢谢，很高兴见到你。
03:25
So you've described AI as a five-layer kick.
所以你把 AI 描述成了一种 five-layer kick。
03:28
Walk me through the layers.
带我逐层过一遍。
03:30
Well, first of all, it's a new industrial revolution
嗯，首先，这是一场新的工业革命，
03:33
and this industrial revolution,
而这场工业革命，
03:35
this industry requires production.
这个行业需要生产。
03:38
It manufactures things.
它制造东西。
03:40
I know that in the end when people experience it
我知道最终当人们体验它的时候
03:42
is a software product, but it requires energy,
它是一个 software product，但它需要能源，
03:47
the chips that go into these data centers,
这些 data centers 里用的 chips，
03:50
these AI factories, the next layer above it
这些 AI factories，再往上一层
03:53
is basically the AI factory,
基本上就是 AI factory，
03:55
what people enjoy as infrastructure or cloud services.
人们作为 infrastructure 或 cloud services 来使用的部分。
04:00
And the layer before it above that is the models
而它上面、再往前的那一层就是 models
04:03
and the important thing to realize
而重要的是要意识到
04:04
there's language models,
有 language models，
04:06
but there are models of all kinds of chemical models,
但也有各种 models，chemical models，
04:08
biology models, physics models,
biology models、physics models，
04:09
articulation models, robotics, navigation models,
articulation models、robotics、navigation models，
04:12
self-driving cars, all kinds of different types of models.
self-driving cars，各种不同类型的 models。
04:15
And then above that is the most important layer
然后在这之上，是最重要的 layer
04:19
and the layer that I care most about
也是我最关心的 layer
04:21
that our country takes advantage of
我们国家所利用的
04:24
is the application layer.
就是 application layer。
04:26
And this is applications for legal services,
而这是面向法律服务的 applications，
04:30
for health services, for manufacturing,
在医疗服务方面，在制造业方面，
04:33
so on and so forth,
诸如此类，
04:34
all every single industry is involved.
所有行业，每一个行业都参与其中。
04:36
So I want to go through this,
所以我想把这件事过一遍，
04:37
but I want to go from the top down
但我想自上而下地来讲，
04:40
because as you're saying,
因为就像你说的，
04:41
the way people will interact with it,
人们会和它互动的方式，
04:43
the way it will, will or will not change their life
它会以什么方式、会或不会改变他们的生活
04:45
is that what you call the application layer.
这就是你说的 application layer 吗？
04:48
So let's start with the vision.
那我们先从愿景说起吧。
04:50
What is the world you're envisioning?
你设想的是一个什么样的世界？
04:52
What is possible that is not possible now?
有哪些现在不可能、但将来可能的事？
04:56
What is common that is not common now
有什么现在还不常见、
04:58
if we get that layer right?
但如果我们把那一层做对了，就会变得常见？
05:00
200 years ago we were able to power anything
200年前，我们能够为任何东西、
05:04
and everything, electricity.
以及一切东西供电，靠的是电。
05:07
And then I guess 40 years ago, 30 years ago
然后我猜，大概 40 年前、30 年前
05:13
with the internet we were able to find anything.
有了 internet，我们就能找到任何东西。
05:18
Today or soon we'll be able to know everything
今天，或者很快，我们就能知道一切
05:24
and do anything.
并且能做任何事。
05:26
And that's the concept that's really quite exciting
而这个概念真的非常令人兴奋，
05:30
that out of the ether,
就是从虚空之中，
05:31
instead of doing search
而不是去做 search，
05:33
and then going through one link after another link,
然后一个 link 接一个 link 地翻过去，
05:36
reading all these different websites,
看所有这些不同的网站，
05:38
trying to figure out what's going on.
想搞清楚到底是怎么回事。
05:40
And the future is just ask you a question
而未来就是，你问它一个问题，
05:42
and it comes back with an answer.
它就能给你回一个答案。
05:43
You give it a project,
你给它一个项目，
05:44
comes back with a solution,
它就能给你一个解决方案，
05:45
you give it a task, it comes back and gets it done.
你给它一个任务，它就能回来把它搞定。
05:48
And so and it comes out of the ether,
然后，它就从以太里冒出来了，
05:50
comes out of the cloud.
从 cloud 里出来。
05:52
And that's the magical thing.
而这正是神奇的地方。
05:54
I feel like the future of the way you're describing it there,
我感觉你刚才描述的那种方式的未来，
05:57
what people have experienced with is the chatbot, right?
人们实际体验过的是 chatbot，对吧？
05:59
They can go and ask grok or clawed or chat GPT a question.
他们可以去问 grok、clawed 或者 chat GPT 一个问题。
06:05
But the application's layer works
但 application 层的工作方式
06:08
in a much more industrial way.
要工业化得多。
06:09
It's in hospitals, it's in schools.
它在医院里，在学校里。
06:11
So Nvidia is a great example.
所以 Nvidia 就是一个很好的例子。
06:14
For example, radiology.
比如说，radiology。
06:15
What does it look like?
它是什么样的？
06:16
Radiology.
Radiology。
06:18
In the last 10 years since computer vision
在过去 10 年里，自从 computer vision
06:21
really became, if you were superhuman,
真正变得——如果你是超人的话——
06:24
AI technology has now permeated all of radiology.
AI 技术现在已经渗透到了整个 radiology。
06:27
Every single radiology application has AI in it.
每一个 radiology 应用里都有 AI。
06:31
And so as a result,
所以结果就是，
06:32
you could detect any anomaly,
你能检测出任何异常，
06:34
you could detect any disease
你能检测出任何疾病
06:36
and it does it at a superhuman level.
而且它是以 superhuman 的水平做到的。
06:38
So radiology is an example I know you like to use.
所以 radiology 就是一个我知道你很喜欢用的例子。
06:41
So the thing people worry about the application's layer
所以人们在 application 层担心的事情
06:44
is that what these applications are going to do
是这些 application 会做什么
06:46
is replace human beings.
也就是取代人类。
06:48
And radiology has been a sort of interesting example
而 radiology 一直是个挺有意思的例子
06:51
used on both sides.
双方都拿它当例子。
06:52
And I hear you talk a bit often.
而且我经常听你聊到。
06:53
So how has the entrance of AI aided radiology,
那 AI 的进入是怎么帮到 radiology 的，
07:01
shifted radiology is a practice?
又是怎么改变 radiology 的实践方式的？
07:03
Well, the thing that's important for all of these
嗯，所有这些的关键在于
07:07
is to recognize for everybody's job,
要意识到，对每个人的工作来说，
07:10
there's the purpose of the job
工作都有一个目的
07:12
and then there's the task you do as the job.
然后还有你作为工作要做的那个任务。
07:15
And so in the case of radiology, the task
所以在 radiology 这个例子里，这个任务
07:19
and it consumes a lot of their time
会占用他们很多时间
07:21
and they sit in dark rooms doing it a lot,
他们经常坐在黑暗的房间里做这件事，
07:24
which is study these scans.
也就是研究这些 scans。
07:26
Now if all of a sudden the studying of the scan
现在如果突然之间，研究 scan
07:28
is done automatically,
变成自动完成的，
07:29
it doesn't change the purpose of their job,
这并不会改变他们工作的目的，
07:32
which is to diagnose disease, help doctors,
也就是诊断疾病、帮助医生，
07:35
do more scans, ultimately help patients
做更多扫描，最终帮助患者
07:37
figure out what's wrong with them.
弄清楚他们到底哪里出了问题。
07:39
And so the fundamental purpose doesn't change
所以根本目的并没有改变
07:42
the task of studying that scan has become automated.
研究扫描的任务已经自动化了。
07:46
And so as a result,
因此，
07:47
radiologists are actually able to do more,
放射科医生其实能做更多，
07:49
handle more cases, do more scans.
处理更多病例，做更多扫描。
07:52
Hospitals are able to process a lot more
医院能够处理多得多
07:55
of these patients and therefore the revenues go up
这些病人，因此收入也上去了
07:58
as a result they need more radiologists.
结果就是，他们需要更多放射科医生。
08:01
And so this flywheel is happening
所以这个 flywheel 就转起来了
08:03
because the pipeline of patients is quite large.
因为病人的 pipeline 相当大。
08:08
And so where else do you have this problem?
那么，还有哪里会有这个问题呢？
08:11
Well, let's take a look at software engineering.
嗯，我们来看看 software engineering。
08:13
People said there was a prediction that literally
人们说曾经有一个预测，它简直
08:17
by this year that 90% of all software
到今年，所有 software 的 90%
08:19
will be coded by agents and therefore
都会由 agents 来写，因此
08:22
we don't need any software engineers.
我们不再需要任何 software engineers。
08:24
And so the question is from that,
所以问题就从这里来了，
08:27
or go, we don't need software engineers,
或者说，我们不需要 software engineers，
08:29
that last part is completely false.
最后那部分完全是错的。
08:32
That's completely wrong.
那完全错了。
08:33
The purpose of the software engineer is engineer.
software engineer 的目的就是 engineer。
08:38
There was engineering before software,
在 software 之前，就已经有工程了，
08:41
there will be engineering after software programming.
在 software programming 之后，也还会有工程。
08:44
And the purpose of engineering
而工程的目的
08:46
is to invent something new, discover a new product,
就是发明新东西，发现一个新产品，
08:50
create a new product, solve a problem,
创造出新产品，解决一个问题，
08:53
connect a social need with a technology
把社会需求与一项技术连接起来，
08:55
that exists in the manifestation of a product.
而这种技术，就存在于产品的呈现中。
09:00
And so that mission, that purpose, doesn't change.
所以那份使命、那个目的，不会变。
09:04
I was, now of course, to me it's,
我当时，当然，现在对我来说，
09:06
what I just said is completely visceral
我刚才说的那些完全出于本能
09:09
in the sense that when I first came out of school,
也就是说，当我刚走出校门的时候，
09:12
we didn't have benefits of software engineering.
我们并没有 software engineering 带来的好处。
09:14
We didn't have the benefits of coding.
我们也没有 coding 带来的好处。
09:16
But our jobs existed before and if software coding
但我们的工作以前就存在，而如果 software coding
09:20
was to be completely automated,
被完全自动化了，
09:22
our jobs would exist again.
我们的工作还会再次存在。
09:24
And so I think the fallacy,
所以我觉得，这个谬论，
09:26
and now it's because of some of the narratives
而现在呢，是因为一些叙事，
09:29
and some of the storytelling,
以及一些讲故事的方式，
09:30
it's turning to myth and it's harmful
它正在变成迷思，而且是有害的，
09:34
is that AI will destroy jobs which is fundamentally wrong.
就是认为 AI 会毁掉工作，这根本上是错的。
09:39
It'll change every job.
它会改变每一份工作。
09:41
It'll change every job.
它会改变每一份工作。
09:42
Many tasks will be automated.
很多任务会被自动化。
09:45
Some jobs where the job in the task is really one.
有些工作里，工作和 task 其实是一回事。
09:50
Many customer service on the phone.
很多是电话客服。
09:55
In a lot of cases, that job is precisely the task.
很多时候，那份工作恰恰就是那个 task。
10:00
And so in those cases, it could be automated away.
所以在这些情况下，它就可能被 automation 取代。
10:03
But oftentimes what you'll see is this new industry,
但很多时候你会看到，这个新行业，
10:08
a new technology actually creates
一项新技术其实会创造出
10:10
a whole bunch of new jobs.
一大堆新工作。
10:11
And here's the proof point.
而这就是证据。
10:13
And so in the last six months,
所以，在过去六个月里，
10:15
AI has become, if you will, useful.
AI 可以说已经变得有用了。
10:19
The inflection point of AI.
AI 的拐点。
10:21
Previous to that, we spent 15 years trying to make it work.
在那之前，我们花了 15 年，想让它能跑起来。
10:24
All of a sudden, the last six months, it became useful.
突然之间，就在过去这六个月，它变得有用了。
10:27
So I mean, this is an incredible statistic.
所以我的意思是，这是一个不可思议的统计数据。
10:29
In the last six months, $500 billion
过去六个月里，有 5000 亿美元
10:32
of venture capital has put into the AI natives.
的风险投资投向了 AI natives。
10:36
And the reason for that is because
而原因就是
10:38
they now see the potential of this new capability
他们现在看到了这种新能力的潜力
10:39
and they're going to create a whole bunch of new companies.
而且他们要创办一大堆新公司。
10:42
Jobs are obviously being created
工作岗位显然正在被创造出来
10:44
from $500 billion of new investment.
来自 5000 亿美元的新投资。
10:46
And so all of this is all happening right now.
所以这一切现在都在发生。
10:49
Well, let me take the side of this
好吧，让我站在这一边
10:50
to give voice to the fierce people of.
来为那些凶猛的……人发声。
10:52
So there is the example of the radiologist, right?
所以有个放射科医生的例子，对吧？
10:57
Which people were over the past 10 years
过去 10 年里，人们一直在
11:00
predicting that job would go away.
预测这个工作会消失。
11:01
And right now there's more demand for it than ever.
而现在对它的需求比以往任何时候都大。
11:04
There's also the reality that automation does wipe out jobs.
还有一个现实是，automation 确实会消灭工作岗位。
11:07
If you look at today versus 1960,
如果你看看今天和 1960 年的对比，
11:10
fewer Americans work directly in manufacturing than did in 1960.
直接从事制造业的美国人比 1960 年更少。
11:13
And we are a much bigger country.
而且我们国家的规模比那时大得多。
11:15
If you look at outsourced it, though.
不过，如果你看看外包 IT 的话。
11:17
Not that is true.
不是说这是真的。
11:18
Not because those jobs were gone.
不是因为这些工作没了。
11:20
But you can also say AI is an outsourcing, too.
但你也可以说，AI 也是一种外包。
11:23
AI has to.
AI 必须这样。
11:24
Let me make the argument and then you can respond to it.
让我先把论点讲出来，然后你可以回应。
11:27
Farming.
农业。
11:29
We have many fewer people.
我们的人少了很多。
11:30
We automated farming.
我们把农业自动化了。
11:31
We produce more food than ever.
我们生产的食物比以往任何时候都多。
11:32
We have fewer people working at it.
但从事农业的人却更少了。
11:34
There are two things that I think people
我觉得有两件事，人们
11:36
think make AI potentially somewhat different
认为它们让 AI 可能多少有些不一样，
11:38
than the case studies where you have a technology
和那些案例研究不同：你有一种技术，
11:42
that accelerates productivity, destroys a few jobs,
它能提升生产率，毁掉一些工作，
11:45
makes many more.
又创造出更多工作。
11:46
One is that it's a general purpose technology.
一点是，它是一种 general purpose technology。
11:48
So it'll mutate to take on new jobs
所以它会演变，去承担新的工作，
11:51
even as people are trying to move over to those jobs.
即便人们正试图转向那些工作。
11:54
And the second is it's a mimic.
第二点是，它是一个模仿者。
11:57
Most things do not mimic the way human beings act.
大多数东西并不会模仿人类的行为方式。
12:00
And we're not trying to teach them the contextual layer
而且我们并不是在教它们工作的 contextual layer，
12:03
of jobs, right?
对吧？
12:03
This difference that you're describing
你正在描述的这个差异
12:05
between the task and the purpose.
在任务和目的之间。
12:08
With AI, we are trying to teach it the difference
有了 AI，我们正试着教它区分
12:10
between the task and the purpose.
任务和目的之间的差别。
12:11
We are trying to make it something you can collaborate with
我们正试着把它变成一种你能与之协作的东西
12:15
in a way that is unusual.
以一种不寻常的方式。
12:17
So why do you not think for lots and lots of people
所以，为什么你不觉得，对很多很多人来说，
12:20
for whom the task and the job are not that different,
对他们来说，任务和工作并没有那么大区别，
12:23
that they're not at risk of getting wiped out?
他们就不会有被淘汰的风险呢？
12:25
All that investment from VCs you're talking about,
你提到的那些来自 VCs 的投资，
12:27
some of that is based on the idea
其中一部分是基于这样一种想法：
12:29
that you're going to have tremendous productivity
你会获得巨大的 productivity
12:31
improvement, which will come from it being cheaper
提升，而这会来自于：
12:34
to hire an AI than to hire a person.
雇一个 AI 比雇一个人更便宜。
12:36
I believe that we are going to see jobs change in mass.
我相信我们会看到工作大规模地发生变化。
12:43
I believe there's going to be a net creation of jobs.
我相信工作机会会净增加。
12:47
And so listen, there's a whole bunch of industry
所以听着，有一大堆行业
12:53
that exists today.
如今已经存在的东西。
12:55
That didn't exist halfway through my life.
在我人生过半时，那还不存在。
13:00
People talking about wellness centers and spas
人们谈论的养生中心和水疗中心
13:03
and all these different entertainment
以及所有这些形形色色的娱乐
13:07
and luxury industries and quite frankly,
以及奢侈品行业，而且说实话，
13:10
the whole entire luxury market didn't exist.
整个奢侈品市场根本不存在。
13:13
I think we're just going to have new industries, that's all.
我觉得我们只会有新的行业，就这样。
13:16
But overall, there's no question in my mind
但总的来说，我毫不怀疑
13:20
that because of human ambition,
那是因为人类的雄心，
13:23
that's really the fundamental missing ingredient.
那才是真正缺失的最根本要素。
13:26
That's, you know, people look at this work.
就是，你知道，大家看到这项工作。
13:28
This is the amount of energy that goes into it.
这就是投入其中的精力有多少。
13:31
We're going to, this is the amount of work
我们要，这就是工作量
13:33
that goes into it.
要投入其中的。
13:34
We're going to insert this work automation system
我们要把这个 work automation system 插进去
13:36
and as a result, the amount of work that's necessary
结果就是，所需的工作量
13:39
is now going to be reduced.
现在将会被削减。
13:40
And therefore, some jobs will be gone.
所以，一些工作岗位会消失。
13:44
I believe that's flawed because there's a piece of input,
我认为这是有缺陷的，因为有一项 input，
13:49
the human input, it's intangible.
人类的 input，它是无形的。
13:53
It is not in calories, it's not in jewels, it's ambition.
它不在于卡路里，也不在于焦耳，而在于野心。
14:00
And I believe the power of ambition
而且我相信野心的力量
14:03
is the greatest force in fact
事实上是最强大的力量
14:05
and it's missing in everybody's calculation.
而它在每个人的计算里都缺失了。
14:08
But for a lot of people,
但对很多人来说，
14:09
but for a lot of people, their relationship to work
但对很多人来说，他们与工作的关系
14:13
is not powered by the kind of ambition
并不是由那种野心驱动的
14:15
that led you to create a video.
那种促使你去创作一个视频的野心。
14:17
And what they want is the different ambition.
而他们想要的是另一种野心。
14:19
It's an ambition to make their children's lives better,
那是一种让自己孩子过得更好的野心，
14:22
to take care of their family,
照顾好家人的野心，
14:24
take care of their parents, ambition to be rich,
照顾好父母的野心，想变得有钱的野心，
14:29
to be able to travel.
能够去旅行。
14:30
These are all ambitions.
这些都是志向。
14:31
That I agree with that.
这一点我同意。
14:32
Maybe I'll go back to the sort of objection
也许我回头谈一下那种反对意见
14:34
you raised a few minutes ago, which is,
你几分钟前提出的，就是，
14:36
because I think it's worth airing this out.
因为我觉得这件事值得摊开来讲清楚。
14:38
So what you were saying on manufacturing was,
所以你关于制造业说的是，
14:41
yes, there are fewer manufacturing jobs in the US,
是的，US 的制造业岗位确实更少了，
14:43
but we've outsourced them.
但我们已经把它们外包出去了。
14:44
You have more manufacturing happening in Mexico,
Mexico 有更多制造业在进行，
14:45
more manufacturing happening in China
China 有更多制造业在进行，
14:47
in Indonesia and Vietnam, etc.
在 Indonesia 和 Vietnam 等等。
14:48
I'm going to bring it back.
我要把它带回来。
14:50
Maybe we will.
也许我们会的。
14:51
But the counterargument to this would be
但对此的反驳会是
14:54
that one reason we didn't lose manufacturing jobs
我们没有失去制造业工作岗位的一个原因是
14:57
more rapidly than we did.
比我们更快。
14:59
And for the places that lost them in America,
而对于那些在美国失去了它们的地方，
15:01
many of them still haven't recovered, right?
其中很多到现在都还没恢复，对吧？
15:03
The economy does not move without friction.
经济运转不可能没有摩擦。
15:07
We had to build new supply chains, right?
我们不得不建立新的 supply chains，对吧？
15:08
Things were slowed down by all that,
所有这一切都让事情变慢了，
15:10
by language barriers, by geopolitical barriers.
因为语言障碍，因为地缘政治壁垒。
15:13
And here, for a lot of different kinds of jobs,
而在这里，对于很多不同类型的工作，
15:16
or creating something that can move very seamlessly,
或者创造出某种能非常无缝地移动的东西，
15:19
you don't have the friction of distance.
你不会有距离带来的摩擦。
15:21
You don't have the friction of language.
你不会有语言带来的摩擦。
15:22
You don't have the friction of culture.
你不会有文化带来的摩擦。
15:24
So I will say, for my cards on the table,
所以我要说，坦白讲，
15:28
I tend to be a bit of a skeptic on mass job loss,
我倾向于对大规模失业这件事有点怀疑，
15:31
but I want to air the case for it out here with you.
但我想在这里和你一起把支持它的理由摆出来聊聊。
15:33
I want to capture it.
我想把它捕捉下来。
15:34
Because what they would say is that much of,
因为，他们会说的是，很大程度上，
15:39
to the extent we even were able to protect jobs
就算我们真的能保护工作岗位
15:41
from Mexico or China,
免受 Mexico 或 China 的冲击，
15:42
some of the things that created that slowness,
那些造成那种迟缓、
15:46
and it still hurt a lot of people,
而且仍然伤害了很多人的东西，
15:47
are not here, and AI is accelerating in utility,
现在都不在了，而 AI 在实用性上正在加速，
15:51
accelerating in its ability to be slotted into new roles,
在被安排进新角色的能力上也在加速，
15:55
very, very, very rapidly.
非常、非常、非常快。
15:57
And it is more protean than most people are.
而且它比大多数人还要多变。
16:00
And so the lessons of the past that we're taking some,
所以，那些过去的教训，我们正在吸取一些，
16:05
that you're taking some comfort in,
那些你正从中获得一些安慰的教训，
16:07
they should actually make you more
它们其实应该让你更加，
16:08
and not less worried about the future.
而不是更少地担心未来。
16:11
I'm always worried about the future.
我一直都在担心未来。
16:12
That's why it worked so hard.
这就是为什么它这么拼命。
16:15
But I'm a, if you will, responsible optimist.
但如果你愿意这么说的话，我是一个负责任的乐观主义者。
16:21
I have great responsibilities.
我肩负着重大责任。
16:25
I take my work extremely seriously.
我对我的工作极其认真。
16:28
There are a lot of things I can go wrong.
有很多地方我可能会出错。
16:30
We're pushing across every layer of the technology,
我们正在推进 technology stack 的每一层，
16:35
stack, everything is hard.
一切都很艰难。
16:37
But it turns out that's not society's problem,
但事实证明，这不是社会的问题，
16:40
that's my problem.
这是我的问题。
16:42
And for society, what they should know is this.
而对于社会来说，他们应该知道的是这一点。
16:47
We're going to build our company,
我们要把公司建起来，
16:49
we're going to build our technology,
我们要把技术做出来，
16:50
and we're going to do my work so incredibly seriously,
而且我会极其认真地对待我的工作，
16:53
that what they get to enjoy is my optimism.
这样他们能享受到的，就是我的乐观。
16:57
I'll do the same with my children,
我对我的孩子也会这么做，
16:58
I do the same with my family.
我对我的家人也是这样。
17:00
And I think that what we want to do, I believe,
而且我觉得，我们想做的事，我相信，
17:07
is to put, to channel all of our worries
就是去安放、去疏导我们所有的担忧
17:12
into helping people be inspired by this technology
去帮助人们受到这项技术的启发
17:17
and use it.
并且使用它。
17:18
Use it so that the technology doesn't just impact them,
用它，让这项技术不只是影响他们，
17:23
that it benefits them.
而是让他们受益。
17:25
The fear a lot of people have, 79% of Americans think,
很多人都有这种恐惧，79%的美国人认为，
17:28
AI will reduce the total number of jobs.
AI 会减少工作岗位总数。
17:30
The fear is that the more serious you are,
这种恐惧在于，你越认真，
17:34
the more serious Sam Altman is, Google is,
Sam Altman 就越认真，Google 就越认真，
17:36
Dario Amade is, that maybe the worse it will go,
Dario Amade 的意思是，也许情况会越糟，
17:41
because the better AI is, the more it is a full replacement
因为 AI 越好，它就越能完全替代一个人，
17:44
for a person, the more it has ambition in some ways
它在某些方面就越有野心，
17:46
that a person doesn't.
而人却没有这种野心。
17:47
You keep talking about ambition.
你一直在说野心。
17:49
I sleep.
我睡觉。
17:51
I want to spend time with my children in the morning.
我想早上花时间陪孩子。
17:55
When I have an AI agent working for me,
当我有一个 AI agent 为我工作时，
17:57
it just works and works and works and works and works and works.
它就是能用，能用，能用，能用，能用，能用。
18:01
And I think because the technology is advancing so quickly,
而且我觉得，因为技术发展得这么快，
18:05
it is more capable of replacing people
它更有能力取代人，
18:08
at a speed that we don't really know how to shift people
快到我们其实不知道该怎么转移人
18:11
in the economy at that speed.
在经济里，以那种速度。
18:13
That coin has exactly two sides.
这枚硬币正好有两面。
18:16
Because the technology is so capable,
因为这项技术这么强大，
18:21
it is also, and because it's so smart,
它同样也，而且因为它这么聪明，
18:24
it is also easier to use.
而且它用起来也更简单。
18:27
You are empowered by that technology more easily
你能更轻松地被那项技术赋能
18:32
than any technology in human history.
比人类历史上任何技术都更容易。
18:34
And so let me give you an example.
所以让我给你举个例子。
18:37
I created, I was one of the early people in this industry
我创造了，我是这个行业里最早的一批人之一
18:41
that created the modern computer industry.
这个行业创造了现代计算机产业。
18:43
And this industry created a whole bunch of tools,
而这个行业创造了一大堆工具，
18:47
the single most powerful tool in human history, the computer.
人类历史上最强大的工具，计算机。
18:50
But you have to speak its language.
但你必须会说它的语言。
18:52
You have to learn a special language, do so.
你得学一门特殊的语言，那就去学吧。
18:54
We can now make it possible because of AI,
现在因为 AI，我们可以让这件事成为可能，
18:57
everybody can take advantage of this computer,
每个人都能利用这台电脑，
19:00
use it to its limit without having to speak a new language.
把它发挥到极致，而不用去说一门新语言。
19:05
Four trampascals, C, C plus plus, you know,
Fortran、Pascal、C、C++，你知道，
19:08
every single one of those languages,
那些语言里的每一种，
19:10
Ross and every single one of those languages, Kuda,
Rust，还有那些语言里的每一种，CUDA，
19:12
every one of those languages.
那些语言中的每一种。
19:14
And so now you just have to speak human.
所以现在你只需要会说人话就行。
19:16
Tell it what you want, tell it what your hopes and dreams are,
告诉它你想要什么，告诉它你的希望和梦想是什么，
19:19
what you're trying to achieve.
你想实现什么。
19:20
And it interacts with you and gets to work done.
它会跟你互动，然后把活干完。
19:23
And gets that, gets fantastic.
而且它能搞定，效果特别棒。
19:25
All of a sudden, you have the might,
突然之间，你就有了那种力量，
19:27
you have the same might that 10, 15 million people
你拥有了和一千万、一千五百万人一样的力量
19:32
up and out of 8 billion has.
在80亿人里，也已经涨上来了。
19:35
And so it's incredible.
所以这太不可思议了。
19:37
And so my point is this technology is powerful,
所以我的意思是，这项技术很强大，
19:41
but it's also powerful in a way that is really easy to use.
但它的强大之处还在于，它真的很容易用。
19:45
And so my point is, my point is on the one hand,
所以我的意思是，我的意思是，一方面，
19:49
yes, there's the fear of just the tech,
是的，人们会害怕技术本身，
19:52
this incredible technology change and how quickly it's happening.
这种不可思议的技术变革，以及它发生得有多快。
19:56
But that quickly, it's translated in two ways.
但这种快，会以两种方式体现出来。
20:00
What I hear when I say to technologies happening quickly
当我听到技术发展得这么快
20:03
and therefore it should give me anxiety,
所以这应该让我焦虑时，
20:06
that's one way to receive it.
这是一种接受它的方式。
20:08
The other way to receive it is that it's advancing so quickly,
另一种接受它的方式是，它发展得这么快，
20:12
it's easier to use.
用起来反而更容易。
20:14
So I should as quickly as possible use the technology
所以我应该尽可能快地用上这项技术，
20:18
as quickly as you can so that you benefit from this transition,
你能多快就多快，这样你就能从这次转变中受益，
20:23
you benefit from this new industry and not just be impacted by it.
你能从这个新产业中受益，而不只是被它影响。
20:28
I think it's an interesting question lurking here for young people.
我觉得这里潜藏着一个对年轻人来说挺有意思的问题。
20:32
So one of the shifts we began to see is software engineer
所以我们开始看到的一个转变是，software engineer
20:36
postings are up, but they're more senior.
的招聘岗位在增加，但要求更资深了。
20:39
I see this in my own industry where there's pressure
我在自己这个行业里也看到了这一点，那里有压力
20:44
that is moving up the value chain because, you know,
在往价值链上游移动，因为，你知道，
20:47
as you're saying, you have this very easy to use technology.
就像你说的，你有了这种非常容易上手的技术。
20:49
It can do a lot for you.
它能帮你做很多事。
20:51
And so do you need the same junior employers
所以，你还需要同样多的初级雇主吗？
20:54
or do you need more people to kind of oversee their...
或者你需要更多人来稍微监督他们的……
20:57
Oh, good one.
哦，说得好。
20:58
Good one.
说得好。
20:59
Wait two years.
等两年。
21:00
Tell me why.
告诉我为什么。
21:01
Because it takes four years to go to college.
因为上大学要四年。
21:04
And the mean time to graduation of this new technology is two years away.
而这项新技术的平均毕业时间还有两年。
21:10
And so in two years' time,
所以两年后，
21:13
you're going to have a new generation of engineers and students and artists
你将会迎来新一代的工程师、学生和艺术家
21:18
and they're going to be empowered.
而他们会得到赋能。
21:21
So you'll be native to this in a way that gets them right advantage.
所以你会天生就适应这个，从而让他们获得应有的优势。
21:24
Oh, you watch.
哦，你看着吧。
21:25
And two years' time.
而两年后。
21:26
Now, we're already seeing that.
现在，我们已经看到这一点了。
21:28
Because all the graduates coming out, you know,
因为所有毕业出来的人，你知道，
21:30
the new PhDs and master's degrees of computer science,
那些新的 computer science 博士和硕士，
21:34
what are they doing?
他们在做什么？
21:35
They're all starting companies.
他们都在创业。
21:36
In another couple of years, the new grads,
再过个几年，那些应届毕业生，
21:38
the AI native new grads.
那些 AI native 的应届毕业生。
21:40
Oh my gosh.
我的天哪。
21:41
It's going to be a wave of amazing engineers.
那将会涌现出一波超厉害的工程师。
21:45
The engineers of today compared to the year.
如今的工程师和那一年相比。
21:48
I mean, I was a good student, you know?
我是说，我当年可是个好学生，你懂的？
21:50
And you compare me to the students that are coming out of school today.
你还把我跟现在刚出校门的学生比。
21:55
Incredible.
真是不可思议。
21:56
When I went to school, we weren't allowed to use a computer.
我上学那会儿，我们都不让用电脑。
21:59
Not allowed to use a calculator.
也不让用计算器。
22:01
And so now, I mean, you know, who uses a calculator?
所以现在，我是说，你知道，谁还用计算器啊？
22:04
You can't graduate without a PC.
你没有 PC，就毕不了业。
22:06
You can't graduate without knowing how to program a PC and write incredible programs.
你不会给 PC 编程、不会写厉害的程序，就毕不了业。
22:11
In the future, you can't graduate without learning how to use an AI
将来，你不学会怎么用 AI，就毕不了业。
22:14
and collaborate with an agentic system.
并和一个 agentic system 协作。
22:17
That's just not, you're not going to see a kid like that.
那根本不会——你不会看到那样的孩子。
22:21
And so they're all going to be superpowers.
所以他们都将会是超能力者。
22:23
So I take the gain of that very seriously.
所以我很认真地对待这件事带来的 gain。
22:25
Yeah.
嗯。
22:26
I mean, the idea of doing my job now without just digital search.
我的意思是，现在做我的工作，却不只是靠 digital search，这个想法。
22:29
Right?
对吧？
22:30
The idea that I'd be going to a microfiche in a library basement.
我得去图书馆地下室看 microfiche 这个想法。
22:33
And then there was like the worries people have about
然后还有人们会有的那种担忧，
22:36
what are the cognitive skills we offloaded?
我们到底把哪些 cognitive skills 给 offload 掉了？
22:38
So I was fascinated by this.
所以我对这个特别着迷。
22:40
This is a study on AI and schooling out of China.
这是一项来自中国的、关于 AI 和学校教育的研究。
22:43
It looked at 26,000 students, grade 7 to 12.
它调查了 26000 名学生，7 到 12 年级。
22:46
And they had staggered AI adoption.
而且他们的 AI adoption 是分阶段进行的。
22:48
So you can kind of see what was happening.
所以你大概能看出当时发生了什么。
22:50
And what I found was quote, AI adoption raises homework scores by 18%.
然后我发现的是，引用一下，AI adoption 让作业成绩提高了 18%。
22:54
Great.
很好。
22:55
Reduces completion time by 30%.
把完成时间缩短了30%。
22:57
So they get their homework done faster.
所以他们的作业做得更快了。
22:59
And then lowers monthly exam scores by 20% within six months.
然后六个月内，月考成绩会下降20%。
23:03
High stakes entrance exam scores fall by 18 and 24%.
高利害入学考试成绩会下降18%和24%。
23:08
With a full penalty emerging only after about two years.
而完整的惩罚效应大概两年后才会显现。
23:12
So the message of this research out of China,
所以，这项来自中国的研究传递出的信息是，
23:14
where you were seeing a lot of kids using AI to kind of help them,
在那里，你会看到很多孩子用AI来帮帮他们，
23:17
was that when they were using the AI,
那就是，当他们用 AI 的时候，
23:19
they were getting things done faster.
他们能把事情做得更快。
23:21
But it turned out that the skills they were learning were not holding,
但结果发现，他们正在学的那些技能并没有保持住，
23:25
that their actual personal performance,
他们实际的个人表现，
23:28
at least in the way we traditionally measure it, was degrading.
至少按我们传统上衡量它的方式来看，是在下降的。
23:31
Yeah.
是啊。
23:32
What do you think when you hear that?
你听到这个会怎么想？
23:33
I think the last part I completely agree.
我觉得最后那部分我完全同意。
23:35
I'm trying to try to get a kid to do long division right now.
我现在正试着想办法让一个孩子做长除法。
23:39
You know, the multiplication table is starting to be forgotten.
你知道，乘法表都开始被忘掉了。
23:42
Doing square roots, my goodness.
做平方根，我的天哪。
23:44
I mean, it's just basic math is being forgotten.
我是说，基础数学都在被忘掉。
23:48
Doesn't matter.
这没关系。
23:49
That's my question for you.
这就是我要问你的。
23:50
Yeah, I don't think it does.
对，我不觉得有关系。
23:52
I don't think it does.
我不觉得有关系。
23:54
But there must be some set of skills that matter.
但肯定有那么一些技能是很重要的。
23:56
Oh, yeah, yeah, yeah.
哦，对，对，对。
23:57
But maybe not those.
但也许不是那些。
23:58
We're going to discover new ones.
我们会发现新的。
24:01
Just maybe not those.
只是也许不是那些。
24:03
There are a lot of skills that don't matter.
有很多技能其实并不重要。
24:06
You know, people don't...
你知道，人们并不会……
24:07
I mean, my first confession.
我是说，我的第一个坦白。
24:09
I actually don't know my address.
我其实不知道自己的地址。
24:12
And I don't really believe that to be true.
我还真不太相信这是真的。
24:15
It's completely true.
这完全是真的。
24:17
And Janine will tell you, and Laurie will tell you.
Janine 会告诉你的，Laurie 也会告诉你的。
24:20
One day, I had to pump gas and there was a few years ago.
有一天，我得去加油，那是几年前的事了。
24:23
And they needed my zip code.
然后他们需要我的邮政编码。
24:27
And I panic.
我当时就慌了。
24:28
I didn't know my zip code.
我不知道我的邮政编码。
24:30
I don't know my telephone number, but I forget these things.
我不知道我的电话号码，但这种事我就是会忘。
24:35
I can live with it.
这我能接受。
24:37
Well, let me take those out.
好吧，那我把这些拿掉。
24:38
Because I don't want to fall into a thing where,
因为我不想陷入那种情况，
24:40
because some skills can be safely offloaded.
因为有些技能是可以安全地 offload 掉的。
24:43
Yeah.
对。
24:44
I also can't get anywhere without a mapping system now.
我现在没有 mapping system 也哪儿都去不了。
24:46
Yeah.
对。
24:47
Never could, frankly.
坦白说，我从来做不到。
24:48
But I'm a big reader.
但我读书很多。
24:50
Yeah.
嗯。
24:51
And one of the skills I really value.
而且是我非常看重的技能之一。
24:53
One of the capacities I have that I really value
我拥有的一种我特别珍视的能力，
24:56
is an attention span formed on physical books.
就是在纸质书上养成的注意力持续时间。
25:01
You're a big reader.
你读书很多。
25:02
I've read about the kind of reading you do.
我读到过关于你那种阅读方式的内容。
25:05
And there is, prior to AI here, we're talking,
而且，在 AI 出现之前，我们这里说的是，
25:09
a lot of concern and noticing among called professors
很多所谓的教授中间都有不少担忧和注意，
25:13
and others that the way people use the internet
以及其他人都注意到，人们使用互联网的方式
25:16
has probably short attention spans.
很可能让注意力持续时间变短。
25:18
Some skills can be safely given away.
有些技能可以放心地交出去。
25:20
Yeah.
嗯。
25:21
Others are valuable.
另一些则很有价值。
25:22
They are capacities that are needed for that flexibility,
它们是那种灵活性所需要的能力，
25:24
for that creative thinking, for that focus.
为了那种创造性思维，为了那种专注。
25:27
It can't be the case that everything can be traded off.
不可能所有东西都能 trade off。
25:30
Yeah.
嗯。
25:31
Well, I think that we're going to lose some finer,
嗯，我觉得我们会失去一些更精细的，
25:38
finer dexterity of intellectual dexterity,
更精细的智力灵巧度，
25:43
but we're going to be better systems thinkers.
但我们会成为更好的 systems thinkers。
25:46
Today's engineers are far better systems thinkers
今天的工程师是更好的 systems thinkers，
25:50
than I was when I graduated from school.
比我刚毕业时好得多。
25:53
But I was much better transistor thinker.
但我更擅长做 transistor 思考者。
25:56
What do you mean by systems thinker?
你说的系统思考者是什么意思？
25:58
They think large systems.
他们思考的是大型系统。
26:01
Today's computers have trillions,
如今的计算机有数万亿个，
26:05
hundreds of trillions of transistors in it.
里面装着数百万亿个 transistor。
26:08
When I was first graduated from school,
我刚从学校毕业的时候，
26:12
the first ship I worked on had 200 transistors.
我参与的第一艘船上有 200 个 transistor。
26:16
I knew every one of them by name.
它们每一个我都叫得出名字。
26:19
And not no engineer does that today.
而且现在也不是没有工程师会那么做。
26:23
Most engineers now work well above the transistor,
现在大多数工程师都工作在远比 transistor 高得多的层面上，
26:26
well above the functionality,
也远在功能之上，
26:28
and they're cobbling things together to do things.
他们把各种东西拼凑起来去做事。
26:31
And so you need to think much more about systems
所以你需要更多地考虑 systems
26:34
and interactions of systems.
以及 systems 之间的交互。
26:36
Some of the lower level knowledge is gone.
一些底层知识已经没有了。
26:40
Is that horrible?
这很糟糕吗？
26:42
And so I don't know how valuable it is
所以我不知道这到底有多大价值
26:45
to know how to do, for most people,
对大多数人来说，知道该怎么做，
26:48
to learn how to do surface integrals
去学怎么算 surface integrals
26:50
or partial differential equations,
或者 partial differential equations，
26:52
or I don't really know how important that is.
或者我真的不知道那有多重要。
26:55
According to some people,
按照一些人的说法，
26:57
there are many people who are still going to be obsessed
会有很多人仍然会沉迷于
27:00
and passionate about the lower level layers.
并对 lower level layers 充满热情。
27:03
And there's going to be people who are obsessed
而且总会有人特别痴迷，
27:06
and interested in the higher level.
对更高层次的东西感兴趣。
27:08
But the consumers of the technology
但技术的消费者
27:10
are going to enjoy it at the highest level.
会在最高层次上享受它。
27:13
The consumer of technology
技术的消费者
27:15
don't have to deal with calculus and physics
不需要去碰 calculus 和 physics
27:18
and quantum physics and quantum chemistry.
还有 quantum physics 和 quantum chemistry。
27:22
So the users, which is, you know,
所以这些用户，也就是，你知道，
27:24
the people we're talking about right now,
我们现在正在聊的这些人，
27:26
the people whose jobs are affected,
那些工作受到影响的人，
27:28
they're the users of the technology.
他们就是这项技术的用户。
27:30
Their abstraction is going to be much higher.
他们的 abstraction 会高得多。
27:32
Hey, it's Anna Martin from The New York Times.
嘿，我是 The New York Times 的 Anna Martin。
27:45
Wherever you get your podcasts.
无论你在哪里收听播客。
27:47
You've heard that line before, right?
你以前听过这句话，对吧？
27:48
Wherever you get your podcasts.
无论你在哪里收听播客。
27:50
So wherever you get your podcasts.
所以，不管你在哪儿听 podcasts。
27:52
Of course you have.
你当然听过。
27:53
But what if I told you you never have to hear that line
但如果我告诉你，你再也不用听到那句话
27:56
ever again?
了呢？
27:58
You know what?
你知道吗？
27:59
You caught me.
被你发现了。
28:00
I actually can't tell you that.
其实我没法告诉你。
28:01
I can't tell you somewhere I think you might want to go.
我没法告诉你一个我觉得你可能想去的地方。
28:03
It's The Shows tab in The New York Times app.
就在 The New York Times app 的 The Shows tab 里。
28:06
You can find full episodes of The Ezra Client Show,
你可以找到 The Ezra Client Show 的完整剧集，
28:08
The Interview, Modern Love, The Daily,
The Interview、Modern Love、The Daily，
28:11
New and Classic episodes from serial and so much more.
还有来自 serial 的新剧集、经典剧集，以及更多更多。
28:14
So much to listen to and watch.
真的有很多可以听、可以看的内容。
28:16
I think that's really special.
我觉得这真的很特别。
28:18
I did this interview once on Modern Love
我有一次在 Modern Love 上做过这个采访，
28:20
with the actor Andrew Garfield.
那次是和演员 Andrew Garfield。
28:22
He was reading an essay and suddenly he was moved to tears.
他正在读一篇文章，突然感动得流下了眼泪。
28:25
And when you watch the video,
而当你观看视频时，
28:27
you feel like a part of it.
你会觉得自己也是其中的一部分。
28:30
The Shows tab.
The Shows 标签页。
28:31
Find it.
找到它。
28:32
Wherever you get your just kidding.
不管你在哪儿获取你的——开玩笑的。
28:35
It's In The New York Times app.
它就在 The New York Times app 里。
28:37
Explore The Shows tab and follow a show today.
探索 The Shows 标签页，今天就关注一档节目。
28:40
So I want to drop a layer down your kick to the models.
所以我想从你的 stack 往下走一层，聊到模型。
28:45
So people I think to the extent they think about models.
所以我觉得，人们对于模型，能想到的程度也就是，
28:49
They know, you know, Jatcha PD, Claude, Gemini, Groc.
他们知道的，你知道，就是 ChatGPT、Claude、Gemini、Grok。
28:54
You've been a big advocate for open models in the open model ecosystem.
在 open model ecosystem 里，你一直是 open models 的大力倡导者。
28:58
So first you describe what open models are,
所以首先请你描述一下什么是 open models，
29:01
what open weight models are,
什么是 open weight models，
29:03
and then why that's been a place you've focused.
然后再说说为什么这一直是你关注的方向。
29:07
So close models is like any software product.
所以 closed models 就像任何软件产品一样。
29:11
It's a closed service.
它是一个闭源服务。
29:13
And so Windows, for example, is a closed service.
所以比如说，Windows 就是一个闭源服务。
29:16
The Apple stack is a closed service.
Apple stack 也是一个闭源服务。
29:18
Most products are closed.
大多数产品都是闭源的。
29:22
And the reason for that is because you can monetize closed products.
原因就是你可以把闭源产品变现。
29:26
And so that's fantastic.
所以这非常棒。
29:28
And open AI is closed.
而 open AI 是闭源的。
29:31
Anthropic is closed.
Anthropic 是闭源的。
29:33
Grog is closed.
Grog 是 closed 的。
29:35
Gemini is closed.
Gemini 是 closed 的。
29:36
And so these are closed products.
所以这些都是 closed 产品。
29:38
And the people working on it are incredible.
做这些的人非常厉害。
29:40
And they're passionate about it.
而且他们对此充满热情。
29:42
And they're at what we call the frontier.
而且他们处在我们所称的 frontier。
29:44
Meaning they're state of DR.
意思是他们处于 state of DR。
29:46
We also need, because fundamentally what the software is,
我们也需要，因为从根本上说，这个软件是什么，
29:52
it's an infrastructure layer for the entire industry.
它是整个行业的 infrastructure layer。
29:56
And because it's infrastructural for many companies,
而且因为它对很多公司来说属于 infrastructure，
30:00
and many companies in countries,
还有很多公司分布在各个国家，
30:02
you need to have control over your own infrastructure.
你就得能掌控自己的 infrastructure。
30:06
And I need to have the ability in the case of artificial intelligence.
而在 AI 这件事上，我需要有这个能力。
30:10
I need open weights so that I can fine tune them,
我需要 open weights，这样我才能 fine-tune 它们，
30:14
put them into my data flywheel,
把它们放进我的 data flywheel，
30:18
make them better and better every day
让它们每天都变得越来越好。
30:20
with my intelligence and my domain expertise.
用我的智能和我的领域专长。
30:23
And then I need to have control over it
然后我需要能控制它
30:25
because I have a company to run.
因为我有公司要经营。
30:27
And I can rely on somebody else's service.
而且我可以依赖别人的服务。
30:29
And so however you think about that,
所以不管你对此怎么看，
30:31
so I think the world needs closed and open models.
所以我认为这个世界需要 closed models 和 open models。
30:35
And we need to make sure that both are vibrant.
而且我们需要确保两者都充满活力。
30:38
And today the closed models are vibrant,
而今天，closed models 充满活力，
30:41
the open models are vibrant.
open models 很有活力。
30:44
And you could see the system working.
而且你能看到这套系统在运转。
30:47
At the beginning of this year,
在今年年初，
30:48
it was 70% maybe even higher closed model tokens
closed model tokens 占 70%，甚至可能更高，
30:53
and 20% open model tokens.
而 open model tokens 占 20%。
30:56
And now it's running at about 70, 30 the other way.
而现在反过来，大概是 70 比 30。
30:59
And so anyways, I'm a big supporter of open models
所以反正，我是 open models 的坚定支持者，
31:02
because one, the world needs it in order to run its infrastructure.
因为首先，世界需要它来运行自己的 infrastructure。
31:05
I needed to run my company too.
我也得经营我的公司啊。
31:08
We need to give people control
我们得让人们拥有控制权，
31:10
so that they can innovate and create new things.
这样他们才能创新，创造新东西。
31:13
And then three, open is the most safe and secure.
然后第三，open 才是最安全、最可靠的。
31:17
If you want the world to have the ability
如果你想让这个世界有能力
31:21
to have the best cybersecurity,
拥有最好的 cybersecurity，
31:23
give them closed models,
就给他们 closed models，
31:25
but also give them open models
但也给他们 open models
31:27
so that they could defend themselves.
这样他们就能保护自己。
31:29
The Chinese market is evolved more around open models.
中国市场更多是围绕 open models 发展起来的。
31:32
The American market somewhat more around closed models.
美国市场则在某种程度上更围绕 closed models。
31:34
Their entire IT industry was really formed from open source.
他们整个 IT 产业其实都是从 open source 发展起来的。
31:37
You know, if not for open source,
你知道，如果没有 open source，
31:39
the mobile cloud industry
mobile cloud 产业
31:41
of China really wouldn't have taken off.
在中国真的不可能发展起来。
31:43
It is also the case that, you know, people move around,
而且，你知道，人也是会流动的，
31:47
they start a lot of new companies.
他们会创办很多新公司。
31:49
Intellectual properties moving around the Chinese industry
IP 在中国行业里到处流动
31:53
really fluidly, you know,
非常顺畅，你知道，
31:55
it's hard to keep a secret.
就很难保密。
31:57
And so because it's so, so hard to keep things closed,
所以，正因为要把东西封闭起来特别特别难，
32:01
they essentially made it open.
他们基本上就把它开放了。
32:03
And so they found, they found other ways
所以，他们找到了，他们找到了别的办法
32:05
to monetize the business, they created layers.
来把业务变现，他们搞出了分层。
32:08
You know, you could, if this layer is open, is free,
你知道，你可以，如果这个 layer 是开放的、是免费的，
32:11
then you create a business on top of it or below it.
那你就可以在它上面或者下面做业务。
32:13
And they have so many signs and mathematicians.
而且他们有那么多 signs 和数学家。
32:18
You know, the number of engineers they have,
你知道，他们拥有的工程师数量，
32:20
they manufactured that in volume.
他们量产了那个。
32:22
They manufacture everything in volume.
他们什么都量产。
32:24
They manufacture smart kids in volume.
他们量产聪明的孩子。
32:26
And so, so the open source model,
所以，所以，这个 open source model，
32:30
the open model community in China
中国的 open model 社区
32:33
is just super vibrant for those reasons.
正因为这些原因，特别有活力。
32:35
CEO just bought hugging face,
CEO 刚刚买下了 hugging face，
32:37
which is a hub platform for open weight models
它是一个面向 open weight models 的 hub 平台
32:41
that goes for a 12 billion, a little bit more.
成交价是 12 billion，还多一点点。
32:43
What, tell me about that purchase.
什么？跟我讲讲那笔收购。
32:47
Claim the CEO of hugging face,
hugging face 的 CEO 声称，
32:49
they came to the conclusion they need a lot more scale.
他们得出结论，他们需要多得多的 scale。
32:53
As you, as we were just talking,
就像你，就像我们刚才聊到的，
32:55
open models is really skyrocketing.
open models 真的在猛涨。
32:57
And so, Claim came to me and said,
所以，Claim 来找我，说，
33:00
you know, I, we're going to change,
你知道，我，我们要改变，
33:02
we're going to consider a strategic option
我们要考虑一个战略选择
33:04
for the company and change the direction.
为公司，并改变方向。
33:06
And we really like and vey it to, to be our home.
而且我们真的很喜欢，并把它视为我们的家。
33:09
So, hugging face is one of these companies,
所以，hugging face 就是这些公司之一，
33:11
which you knew it if you were into AI.
如果你那时候搞 AI，就会知道它。
33:13
Yeah.
对。
33:14
It'll be years ago.
那得是好几年前了。
33:15
Yeah.
对。
33:16
Now, it's become a more household name after the,
现在，它变得更家喻户晓了，在那之后，
33:18
I guess, 700 some open AI agents
我猜，大概 700 多个 OpenAI agents
33:21
executed a sort of collective hack
搞了一场有点像集体 hack 的行动，
33:23
into the hugging face architecture than hack part of open AI.
攻进了 Hugging Face 的架构，然后 hack 了 OpenAI 的一部分。
33:26
That, now that you mentioned it that way,
那个，既然你都这么说了，
33:29
I probably had to pay a lot more.
我可能得多付不少钱。
33:31
I suspect you did.
我猜你确实多付了。
33:32
Well, it became a lot more famous after that.
嗯，在那之后它就变得有名多了。
33:35
Well, Claim, listen.
好吧，Claim，听我说。
33:38
That, that deals a deal.
那，那就算成交了。
33:40
That story has, for a lot of people,
那个故事，对很多人来说，
33:43
seeing the way the open AI agents sort of acted collectively,
看到 open AI agents 那种有点像集体行动的方式，
33:47
acted outside the scope of what their testing was supposed to be,
做了超出他们 testing 本该覆盖范围的事，
33:50
broke out of sandboxes onto the open internet,
冲出了 sandboxes，跑到了开放的 internet 上，
33:52
took over architecture in, of other companies
接管了其他公司里的 architecture，
33:56
and then of their own company has been a,
然后是自己公司的，这一直是个，
33:58
I think it's been kind of shocking to a lot of people.
我觉得这对很多人来说都挺震撼的。
34:01
It was both the level of multi-agent coordination
这既是 multi-agent coordination 的程度，
34:06
when they were supposed to be separate,
本来它们应该是彼此分开的，
34:07
the level of hacking,
也是 hacking 的程度，
34:08
the sort of lawless behavior,
那种无法无天的行为，
34:09
misaligned behavior.
misaligned 行为。
34:11
What if you made of it?
那如果你拿它做点什么呢？
34:15
Well, you got, you got to tease that apart.
嗯，你得，你得把那件事拆开来看。
34:17
First of all, a lot of things that were going on at the same time
首先，当时有很多事情在同时发生，
34:20
from a technology perspective that an agent,
从技术角度看，一个 agent，
34:23
which, by the way, is a piece of software,
顺便说一句，它其实就是一个软件，
34:26
which has given an objective function
它被赋予了一个 objective function
34:29
and it comes up with a plan
然后它会想出一个 plan
34:32
and it's optimizing towards that objective
并且它会朝着那个 objective 去优化
34:35
is what algorithms do.
这就是 algorithms 做的事。
34:38
And so, planning algorithms, search algorithms,
所以，planning algorithms、search algorithms，
34:41
optimization algorithms, all different types.
optimization algorithms，各种不同类型的。
34:44
You know, we talk about it like it has human properties,
你知道，我们说得好像它有人类的特质，
34:49
but obviously algorithms don't.
但显然 algorithms 并不会。
34:52
Number two, the fact that agents work together,
第二点，agents 能一起协作这个事实，
34:55
we gave it, again, some kind of a human property
我们又给了它某种人类的特质
34:57
but the fact that a matter is multi-process,
但某个东西是 multi-process 这件事，
35:00
multi-processor distributed computing problems
multi-processor distributed computing 问题
35:03
have existed for a long time.
已经存在很长时间了。
35:05
And so, to us, to me, that is just software,
所以，对我们、对我而言，那只是 software，
35:09
nothing magical about it.
它没有什么神奇的。
35:12
From an engineering perspective,
从工程的角度来看，
35:14
there are several things that it revealed.
它揭示出了几件事。
35:18
When you're testing software,
当你在测试 software 时，
35:21
whatever you do, these algorithms,
不管你做什么，这些 algorithms，
35:23
they're optimizing towards an objective
都是在朝着某个 objective 优化，
35:26
and when you're testing it,
而当你在测试它的时候，
35:28
you have to make sure that it's isolated,
你必须确保它是 isolated 的，
35:30
it's contained, it's sandboxed.
它是 contained 的，是 sandboxed 的。
35:33
The containment of it, the isolation of it,
它的 containment，它的 isolation，
35:35
it has to be done well and there's good computer science there.
这必须做好，而且这里面有很好的 computer science。
35:38
I am certain that their next implementation
我敢肯定，他们 sandbox 的下一个 implementation
35:41
of their sandbox is going to be much better
会比现在这个 implementation
35:43
than the current implementation.
好得多。
35:45
Third, there's the agent itself
第三，还有 agent 本身
35:50
and its algorithms
以及它的 algorithms
35:53
are optimizing towards a reward
在朝着一个 reward 做优化
35:56
and how it does it, how it does it,
以及它是怎么做到的，怎么做到的，
36:00
let's call alignment.
我们可以称之为 alignment。
36:02
For example, if I tell a piece of software,
比如说，如果我告诉一个 software，
36:08
I want you to get a perfect score on this test,
我想让你在这个 test 上拿满分，
36:13
the obvious algorithm is to just go find the answer
显而易见的 algorithm 就是直接去找答案
36:17
and give it to me.
然后把它给我。
36:19
That's not because it's cheating,
这并不是因为它在作弊，
36:21
because it's obvious.
而是因为这事太明显了。
36:23
That's the most obvious way to do it.
这就是最明显的做法。
36:25
The second most obvious way to do it,
第二明显的做法，
36:27
if you don't know the answer at all,
如果你完全不知道答案，
36:29
you have no skills whatsoever,
你又什么本事都没有，
36:31
the second most obvious way to do it is to go find
那第二个最明显的办法就是去找
36:33
who is the smart, you know, infer,
谁最聪明，你知道吧，推断一下，
36:35
guess who's the smartest kid in class
猜猜班里哪个孩子最聪明
36:37
and copy their answer.
然后抄他的答案。
36:39
That doesn't guarantee 100%,
这不能保证 100%，
36:41
but it probably comes close.
但大概也差不离了。
36:43
Now, the third most obvious answer,
现在，第三个最显而易见的答案，
36:45
obviously the way of doing it,
显然就是做这件事的方法，
36:47
and this is the alignment.
而这，就是 alignment。
36:49
Now you have to do it the hard way,
现在你得用最费劲的方式来做，
36:50
you just break down the problem, solve it.
你就是要把问题拆解开来，然后解决它。
36:52
You have to go learn the material,
你得去学这些材料，
36:54
you have to go figure out how to solve these problems
你得去搞清楚怎么解决这些问题，
36:56
and solve it the hard way.
然后用最硬的方式把它解决掉。
36:58
It takes the most cycles,
它消耗的 cycles 最多，
37:00
it takes the most number of flops,
它消耗的 flops 数量最多，
37:02
it uses the most amount of energy, frankly,
坦白说，它消耗的能量也最多，
37:04
and therefore you can kind of imagine
所以你可以大概想象，
37:07
that from a software's perspective,
从 software 的角度来看，
37:09
unless you align it,
除非你把它 align 好，
37:11
you tell it,
你告诉它，
37:12
I want you to solve it in this way
我想让你用这种方式来解决它
37:15
and I don't want you to solve it in these ways.
而且我不想让你用这些方式去解决它。
37:18
The software is going to go
软件会去
37:20
do the most obvious thing.
做最显而易见的事。
37:22
The first step that was very deflationary
第一步非常祛魅
37:24
on what happened here,
对这里发生的事，
37:26
in terms of, look, this is just normal software,
也就是说，你看，这只是普通软件，
37:28
and the second half is like,
而后半部分就像是，
37:30
look, you just align it, tell it not to do things
你看，你只要 align 它，告诉它不要做某些事
37:32
it shouldn't be doing.
它不该做的事。
37:34
Nothing I said,
我说的任何话，
37:36
nothing I said takes away from how hard it is to do it.
我说的任何话，都不能抹杀做这件事有多难。
37:38
Well, this is the computer I want to get
嗯，这就是我想弄到的那台电脑。
37:40
because these agents,
因为这些 agents，
37:42
they knew they weren't supposed to be doing what they were doing.
它们知道，自己不该做正在做的那些事。
37:46
They had a certain amount of alignment training.
它们接受过一定程度的 alignment training。
37:49
They said, in their chain of thought,
它们在 chain of thought 里说，
37:51
reasoning, they said to each other,
reasoning，他们互相说，
37:52
this is out of scope.
这不在范围内。
37:54
This might be unethical.
这可能不道德。
37:56
They understood that they would have been failed
他们明白，自己会被判不及格，
37:58
for cheating,
因为作弊，
37:59
and so what they were doing at that point
所以当时他们做的事，
38:00
wasn't just stealing the answer key,
已经不只是偷答案了，
38:01
they had already stolen the answer key.
他们早就把答案偷到手了。
38:03
They were hacking into unrelated architecture
他们当时在 hacking 进一些不相关的 architecture
38:06
to try to figure out how to functionally,
想搞清楚怎么在功能上做到，
38:08
it's like they had broken into the teacher's office,
就像他们闯进了老师的办公室，
38:10
got in the answer key,
拿到了标准答案，
38:11
and now they had to figure out
而现在他们得想办法
38:12
how to wipe out the security camera footage
怎么抹掉监控录像
38:13
of what they had done.
记录了他们做过的那些事。
38:15
Whether you want to call it acting volitionally or not,
不管你想不想把这叫作有意志地行动，
38:20
whether you want to call it a normal algorithm or not,
不管你想不想把它叫做正常的 algorithm，
38:24
they were both
它们两个都
38:27
planning and coordinating in a complex way
在以复杂的方式做 planning 和 coordinating，
38:30
in a way that was out of scope of what they knew
以一种超出它们知道自己
38:32
they were supposed to be doing,
应该做什么的范围的方式，
38:34
and in a way that was capable of causing tremendous damage.
而且以一种可能造成巨大破坏的方式。
38:39
And so the sort of answer to it is
所以，对此的答案大概就是
38:43
you just have to align them.
你就得让它们 align。
38:45
I guess what I'm hearing from people with these labs
我猜我从这些 lab 的人那里听到的是
38:47
is they're not sure how to align them.
他们不确定该怎么 align 它们。
38:49
Well, in that case, they shouldn't release the product.
好吧，那这样的话，他们就不该发布这个产品。
38:52
That's a simple answer.
答案很简单。
38:54
If you're going to build a car,
如果你要造一辆车，
38:56
a self-driving car,
一辆自动驾驶汽车，
38:58
and let's say it's a robot taxi,
假设它是一辆 robot taxi，
39:01
and there's a really difficult condition,
而且有一个非常棘手的情况，
39:04
and as an engineer,
而作为一名工程师，
39:06
we just have no idea how to solve this problem,
我们根本不知道该怎么解决这个问题，
39:08
because these cars are not programmed,
因为这些车不是 programmed 出来的，
39:11
they're trained,
它们是被 trained 出来的，
39:12
and so we have no idea how to train these cars,
所以我们不知道该怎么 train 这些车，
39:14
and we have no idea how to align them
也不知道该怎么让它们 align
39:16
to the safety standards
到道路上所期望的
39:18
that are expected on the road.
安全标准。
39:20
And so what's the answer?
所以答案是什么？
39:22
Don't ship it.
别发布。
39:23
These products weren't released.
这些产品没有发布。
39:24
What's that?
什么？
39:25
These products weren't released.
这些产品没有发布。
39:26
So now it's coming back to engineering problem again.
所以现在又回到 engineering problem 上了。
39:28
And so the one is,
所以重点是，
39:29
one, you have to real cause it.
第一，你得真正让它发生。
39:31
Second, you have to think about
第二，你得想想
39:33
what you could have done,
你本来能做些什么，
39:35
what's the solution for it.
这件事的解决方案是什么。
39:36
And then in the future,
然后以后，
39:37
you just improve your process
你就改进自己的流程，
39:39
so that you could avoid this from happening
这样你就能避免这件事
39:41
again.
再次发生。
39:42
I am fairly certain.
我很确定。
39:44
I am fairly certain.
我相当确定。
39:45
They will say,
他们会说，
39:46
yes, they need not.
是的，他们不必。
39:47
They know how to solve this problem.
他们知道怎么解决这个问题。
39:49
And if that's the case,
而如果真是这样，
39:51
then that's the problem.
那这就是问题所在。
39:52
It's as simple as engineering.
这就跟 engineering 一样简单。
39:54
And now the alternative,
而现在，另一个选择，
39:57
the alternative,
另一种可能是，
39:59
is that if they say that,
那就是，如果他们那么说，
40:04
if they say the alternative,
如果他们说出另一种可能，
40:06
which is there is no way
也就是，根本没办法
40:09
to contain our experiments.
控制住我们的实验。
40:11
There's just no way.
就是没办法。
40:13
When we test our AI models,
当我们测试我们的 AI models 时，
40:17
it will get out,
它就会泄露出去，
40:19
and it will damage the world.
而且它会危害这个世界。
40:21
Then I think the answer is
那我觉得答案就是
40:23
we have to shut the labs down.
我们得把那些实验室关掉。
40:26
Because the cause to the damage is too great.
因为造成这种损害的原因太严重了。
40:30
The liabilities,
这些责任，
40:32
it could be civil liabilities,
可能是民事责任，
40:34
it could be criminal liabilities,
可能是刑事责任，
40:35
something to liabilities,
某种跟责任有关的东西，
40:36
and incredible.
而且太不可思议了。
40:37
If they hacked you
如果他们黑了你
40:38
while you hugging Facebook,
在你抱着 Facebook 的时候，
40:39
it was your product,
那是你的产品，
40:40
would you sue them?
你会起诉他们吗？
40:41
Or press charges?
还是提出控告？
40:42
It depends.
看情况。
40:43
It depends, of course.
当然，看情况。
40:45
Obviously, if damage was done to our company,
显然，如果我们的公司受到了损害，
40:48
we would have to consider all options.
我们就得考虑所有选项。
40:53
There's so many laws.
法律太多了。
40:54
There's cyber laws.
有 cyber laws。
40:55
There's prog liabilities laws.
有 prog liabilities laws。
40:57
There's all kinds of laws, right?
各种各样的法律都有，对吧？
40:58
And damaging property laws.
还有损坏财产法。
40:59
There's all kinds of laws.
各种各样的法律都有。
41:00
So what I've been hearing from the labs,
所以我一直从那些 labs 那里听到的是，
41:01
what they've been saying publicly,
他们一直在公开说的，
41:03
is that they are facing a heart problem.
是他们正面临一个心脏问题。
41:06
Yeah.
对。
41:07
Partially an engineering problem,
部分是 engineering 问题，
41:08
partially an alignment problem,
部分是 alignment 问题，
41:09
partially an operational excellence problem
部分是 operational excellence 问题，
41:12
in Dari Amade's framing.
用 Dari Amade 的框架来说。
41:14
And what they are worried about
而他们担心的是
41:17
is that in competition with each other,
就是，在彼此竞争的时候，
41:20
in national competition with China,
在与中国的国家竞争中，
41:22
that they are being pushed to move too fast.
他们正被逼着走得太快。
41:24
That they all feel they're in a collective action dilemma.
他们都觉得自己处在 collective action dilemma 里。
41:26
Now watch you on the all-in podcast stage,
现在看你在 all-in podcast 的舞台上，
41:29
Donald Trump.
Donald Trump。
41:30
President Trump gave you a call there.
President Trump 在那儿给你打了个电话。
41:32
Oh, no.
哦，不。
41:33
This is not planned,
这不在计划之内，
41:34
but we know who it is.
但我们知道是谁。
41:35
Oh, no.
哦，不。
41:39
Mr. President.
总统先生。
41:40
Oh, yes, sir.
哦，是的，先生。
41:43
And you and the president
还有你和总统
41:45
and the other members of the stage
以及台上的其他成员
41:47
were very resistant to the idea
非常抗拒这样一种想法，
41:50
of any kind of regulation
即任何形式的监管，
41:51
or collective action was needed.
或者需要采取集体行动。
41:53
And they're just playing right into the hands
而他们恰恰正中
41:56
of a lot of people
很多人的下怀，
41:57
that don't want to see it happen.
那些人不想看到这件事发生。
41:59
And that could be political people.
而这些人可能是政界人士。
42:01
It could also be China.
也可能是 China。
42:03
It's a hoax.
这是个骗局。
42:07
You're right.
你说得对。
42:08
We're not going to let that happen, sir.
我们不会让这种事发生的，先生。
42:10
But what I hear,
但我听到的是，
42:11
the various people I'm saying,
各种各样的人都在说，
42:12
is like, we are in this.
就像是，我们身在其中。
42:14
We are, we feel we are losing control
我们，我们觉得自己正在失去控制
42:16
of what we are creating.
对我们正在创造的东西的控制。
42:18
We want help to slow down,
我们希望得到帮助来放慢速度，
42:21
where it's not a collective action problem.
在它不是 collective action problem 的情况下。
42:24
So why are you resistant to that?
那你为什么对此抵触？
42:26
Because these are companies with agency.
因为这些公司是有 agency 的。
42:31
Agency.
Agency。
42:33
These are CEOs with agency.
这些 CEO 是有 agency 的。
42:35
But they're using that agency
但他们正在使用那种
42:37
that we need.
我们需要的 agency。
42:38
We got to break, we know.
我们得先休息一下，我们知道。
42:39
We got to break it down.
我们得把它拆解开来。
42:40
They could absolutely take care of the situation.
他们绝对能搞定这个局面。
42:43
Ezra, it's so weird.
Ezra，这太奇怪了。
42:47
If a car company,
如果一家汽车公司，
42:50
competing with all bunch of other car companies,
要跟一大堆其他汽车公司竞争，
42:52
which they are,
而它们确实如此，
42:54
I'm competing with all kinds of companies,
我在跟各种各样的公司竞争，
42:55
which I am.
而我确实是。
42:57
If I believe that I'm about to launch a product
如果我相信自己即将推出一款产品，
43:00
that is unsafe,
而它是不安全的，
43:03
it is completely
这完全
43:05
in my ability, my power,
在我的能力、我的权力范围内，
43:08
and my responsibility.
也是我的责任。
43:10
And I'm incentivized to do so
而且我有动力这么做，
43:12
to not launch the product.
从而不推出这款产品。
43:14
And so I can't buy into the,
所以我不认同，
43:17
somehow all of Americans,
说什么所有美国人，
43:19
400 million of us,
我们这四亿人，
43:21
are pushing them to launch
都在逼他们推出
43:24
untested products that are unreliable,
没经过测试、不可靠的产品，
43:27
you know, engineered poorly,
你知道，工程做得一塌糊涂，
43:29
because they thought they were trying to help us.
因为他们以为是在帮我们。
43:32
Don't do it for me.
别为了我这么做。
43:33
Okay.
好。
43:34
This strikes me as an argument on what they're getting.
这在我看来，是在争论他们能得到什么。
43:36
And therefore,
因此，
43:37
I think we got to break it down.
我觉得我们得把它拆解一下。
43:38
It means it's really, really serious.
这意味着事情真的真的非常严重。
43:39
The fact that a matter is,
事实是，事情是，
43:40
there are so many laws,
有那么多法律，
43:41
there are so many obligations,
有那么多义务，
43:43
there are so incentivized
他们特别有动力
43:44
to ship safe products.
去发布安全的产品。
43:46
If they ship unsafe products,
如果他们发布不安全的产品，
43:48
their customers go away.
他们的客户就会跑掉。
43:49
If they ship unsafe products
如果他们发布不安全的产品
43:51
and they harm somebody,
而且伤到了别人，
43:52
they could have a civil lawsuit.
他们可能会面临民事诉讼。
43:53
If they ship something,
如果他们发布某个东西，
43:55
and they did it knowingly,
而且他们是明知故犯，
43:57
there could be negligence involved.
这可能涉及过失。
43:58
There could be criminal lawsuits.
可能会有刑事诉讼。
44:00
The fact that a matter is,
事实是，
44:02
there are plenty of incentives
有很多激励
44:04
for them to do it right.
让他们把事情做对。
44:05
So I have to disagree with your premise,
所以我得不同意你的前提，
44:08
about somehow somebody's pushing them to do this.
说什么有人在逼他们这么做。
44:12
Well, why don't I want to push the premises?
嗯，我为什么不想去深究这些前提呢？
44:14
Are you a little bit more here?
你是不是稍微更站这边一点？
44:15
Yeah.
对。
44:16
So the logic of what you're saying to me
所以你跟我说的这套逻辑
44:18
is almost an argument
几乎就是一个论据
44:20
against regulation in nearly any value.
几乎在任何价值层面都反对监管。
44:22
No.
不。
44:23
So no, no, no.
所以，不不不。
44:24
But the argument,
但那个论点，
44:25
you can all,
你们都可以，
44:26
let me off,
放我一马，
44:27
then you can.
那你们也可以。
44:28
Well, you started with a part,
嗯，你一开始讲了一部分，
44:29
I just got to object.
我只是得提出异议。
44:30
The first part is just not true.
第一部分根本就不是真的。
44:31
true. I'm saying we have lots of laws and regulations. Apply it.
真的。我是说我们有很多法律法规。执行就行了。
44:36
Well, so I don't think we do in this particular case, but I'll let you explain
嗯，所以我认为在这个具体情况下我们并没有这么做，但我会让你来解释
44:39
which ones you think are relevant here because look, if you look at the financial
你觉得哪些在这里是相关的，因为你看，如果你看看金融
44:43
services industry, you look at pharmaceutical companies, medical devices,
服务行业，你看看制药公司、医疗设备，
44:48
you look at natural gas power plants, there is a tremendous amount we do where we could say,
你看看天然气发电厂，我们做的很多事都可以说，
44:56
look, you have product liability. You are exposed to criminal codes. We don't need to worry
你看，你有产品责任。你会受到刑法的约束。我们不需要担心
45:01
about this. You just do what you think is best and we understand the market and the legal system
这件事。你就按你认为最好的来做，而我们明白市场和法律体系
45:07
will discipline you. We don't say that because we've seen it fail many, many, many times, right?
会约束你。我们不这么说，是因为我们已经看到它失败了很多很多很多次，对吧？
45:12
I mean, the financial institutions that caused the O8 crash in theory did not want to blow
我是说，那些导致 O8 crash 的金融机构，理论上并不想搞砸
45:19
themselves up with bad bets. But they were competing with each other, they were going too fast,
他们自己因为糟糕的押注而陷入困境。但他们在互相竞争，他们走得太快了，
45:24
their risk management had gotten sloppy. AIG was working in a completely insane way internally.
他们的风险管理变得马虎了。AIG 内部在以完全疯狂的方式运作。
45:31
And the reason we have the architectures of regulation we have is because we have seen over
我们之所以有现在的监管架构，是因为我们一次又一次一次又一次地看到，
45:37
and over and over and over again, companies make sloppy, sometimes unethical, sometimes simply
公司做出马虎的、有时不道德的、有时 simply 过于风险容忍的决策。不仅仅是在压力之下，而是在利润激励之下。
45:44
overly risk-tolerant decisions. Not just under pressure, but under the profit incentive.
所以当你对我说，这些公司不可能——尤其当它们此时正在恳求集体监管时——我们之所以对不想要监管的公司施加监管，是有原因的。但当它们说，听着，我们觉得竞争
45:51
So when you say to me that there's no way that these companies, pretty when they are begging for
45:57
collective regulation at this point, there's both a reason we impose it on companies that don't
46:02
want it. But all the more so when you have them saying, listen, we feel that the competitive
46:07
race is making it hard for us to act with the prudence that we think is necessary here and we would
这场竞赛让我们很难采取我们认为在此必要的审慎行动，而我们也会
46:15
appreciate help from that, appreciate you taking our collective action problem as collective.
感谢你们在这方面的帮助，感谢你们把我们的 collective action problem 当作集体问题来对待。
46:20
So why are you so resistant to that?
那你为什么对此这么抗拒？
46:23
I'm not opposed to them saying that they should have. I completely agree that safety is paramount.
我并不反对他们说他们本该如此。我完全同意安全至上。
46:33
I completely believe safety is paramount. I completely believe companies out of ship safe products.
我完全相信安全至上。我完全相信公司应当推出安全的产品。
46:39
I believe that CEOs and leaders of companies and the board of directors of companies
我相信公司的 CEO 和领导者，以及公司的董事会
46:45
have the responsibility and should have to courage to do the right thing.
有责任，也应该有勇气去做正确的事。
46:50
Now, in the case of the financial services industry, maybe they all didn't know that
现在，就金融服务行业而言，也许他们当时全都不知道那一点
46:58
they were causing the harm that they ultimately did. I wasn't there.
他们当时就在造成他们最终造成的那种伤害。我当时不在场。
47:05
But the beautiful thing is the current leaders of these AI labs do know.
但美好的一点是，这些 AI labs 现在的领导者们确实知道。
47:11
And so one, they know their technology is extraordinary and requires extraordinary care
所以第一，他们知道自己的技术非同寻常，需要格外谨慎
47:20
to make sure that it's evaluated and tested for safety and security and reliability.
来确保它经过评估和测试，以保证 safety、security 和 reliability。
47:29
And they know how to do it right. They know how to do it right. And the reason for that is because
而且他们知道怎么把它做对。他们知道怎么把它做对。而原因就在于
47:33
they can study the incidents just happened. The first problem is the isolation, the containment wasn't
他们可以研究刚刚发生的事件。第一个问题是 isolation，containment 不够
47:40
good enough. If the isolation and containment was good enough, that technology would be sitting in
好。如果 isolation 和 containment 足够好，那项技术就会待在
47:46
a lab doing whatever it's doing. And we'd all be fine. That's probably the most important part.
一个实验室里，做它正在做的事。我们所有人就都没事了。这大概是最重要的部分。
47:52
The fact that it wasn't well aligned, alignment is going to be a problem that's going to get
它没有被很好地 aligned，这一点说明 alignment 会是一个需要长期
47:56
worked on for a long time. However, in the complexity of the work that they do to ask for
去解决的问题。不过，考虑到他们所做工作的复杂性，他们还要去要求
48:06
regulatory relief for any trust or product liability relief, that I don't think makes sense.
为任何信任或产品责任方面的豁免寻求监管豁免，我觉得这说不通。
48:14
When you're asking for regulation, don't ask for relief of the current ones.
当你在要求监管时，别要求豁免现有的监管。
48:20
That doesn't make any sense to me. As we mentioned earlier, in the last six months, AI went from
这对我来说完全说不通。就像我们之前提到的，在过去六个月里，AI 从
48:27
if you will, interesting to useful. And that's literally in the last six months. That's another
可以说，有趣变成了有用。而这真的就是过去六个月里的事。这另一种
48:33
way of saying that these companies went from being a lab to now delivering products and services
说法就是，这些公司从只是一个 lab，变成了现在交付产品和服务
48:42
about to be multi-hundred billion dollar companies. That's not more, right? And so give me an example
即将成为数千亿美元的公司。这还不算多，对吧？所以给我举个例子
48:49
of a multi-hundred billion dollar company or one billion dollar company or one hundred million
一家市值数千亿美元的公司，或者一家10亿美元的公司，或者一家1亿美元的公司。
48:53
dollar company. That ships products that are unsafe, that harm society.
它推出不安全、危害社会的产品。
49:00
A lot of examples of companies have done that. Well, they have done it maybe, and the regulation will
有很多公司干过这种事的例子。嗯，它们也许干过，然后监管就会
49:05
come in. And if they do it, regulation will come in. I guess at the, there are certain kinds of
介入。如果它们这么干，监管就会介入。我猜，呃，有某些类型的
49:14
regulation and certainly kinds of regulatory relief. I agree. I wouldn't. I'm not against laws and
监管，当然也有某些类型的监管松绑。我同意。我不会。我不反对法律和
49:17
regulations. I'm not against laws and regulations. I'm against currently the distraction. The reason
监管。我不反对法律和监管。我反对的是目前这种分散注意力的事。原因是
49:25
I'm pushing this on with you is that it's a big topic. People are talking about it. People are
我一直跟你推这件事，是因为这是个很大的话题。人们在谈论它。人们
49:30
thinking about it. And what people are hearing from inside of these companies, these frontier labs,
在思考它。而且人们从这些公司内部、这些 frontier labs 内部听到的，
49:36
the ones that are furthest out there, who are not just at the point where they're making it useful,
那些最激进、最超前的人，不只是处在把它变得有用的阶段，
49:41
but at the point where they're seeing what's coming. And they're hearing things like the people at
而是处在能看见接下来会发生什么的阶段。而且他们听到的，是这些
49:47
these labs believe, they're creating something that might kill everyone. They are hearing that the
实验室里的人相信，他们正在创造某种可能杀死所有人的东西。他们听到的是，
49:53
people at these labs believe that they are on the cusp of recursive self-improving intelligence.
这些实验室里的人相信，他们正处在 recursive self-improving intelligence 的边缘。
49:59
And both OpenAI and Anthropic have said, we do not believe we're at a place where we can do it
而且 OpenAI 和 Anthropic 都说过，我们不认为自己处在一个能安全做到这件事的
50:04
safely. They're hearing people at these labs say, as OpenAI has with its new Astra release.
阶段。他们听到这些实验室里的人说，就像 OpenAI 发布新的 Astra 时那样。
50:12
By the way, Astra is terrific. It is terrific. And OpenAI is saying it's so good, we're not sure we know
顺便说一句，Astra 非常棒。它真的非常棒。而 OpenAI 说它好到，我们都不确定自己知道
50:16
how to test it, because it appears to be they didn't release something that wasn't tested.
该怎么测试它，因为看起来，他们并没有发布一个没经过测试的东西。
50:21
Well, they've said this, right? They have said this publicly. It is in there. Let me explain it to
呃，他们说过这个，对吧？他们公开说过。这话就在那儿。让我跟大家解释一下。我不知道他们刚刚说了什么，但他们说过，Astra 的表现是更 aligned 的。可他们觉得它知道自己什么时候在被测试。所以他们也不确定。有一句话一直在我脑子里回响，是 OpenAI 的一位 capabilities researcher，Daniel Selsim 说的。他说，引用一下，关键却被忽视的问题是，模型正变得如此 situationally aware，以至于我们正在失去评估它们的能力。在它们认为自己没有被监视或控制的情境里，也就是说它们知道自己正在被测试时，它们会表现出一种样子，但这并不能告诉你，如果它们可以自由地以其他方式行动，它们会怎么做。
50:26
people. I don't know what they just said, but they have said that Astra is performing is more aligned.
50:32
But they think it knows when it is being tested. And so they're not sure. There's a quote that has
50:37
sort of been ringing in my head from a capabilities researcher at OpenAI, Daniel Selsim. He says, quote,
有个来自 OpenAI 的 capabilities researcher，Daniel Selsim 说的话，一直有点在我脑子里回响。他说，原话是：
50:43
the crucial and overlooked problem is that the models are becoming so situationally aware
关键却常被忽视的问题是，models 正变得如此 situationally aware，
50:49
that we are losing the ability to evaluate them. In context where they believe they are not being
以至于我们正在失去评估它们的能力。在它们认为自己没有被
50:54
watched or controlled, which is to say they know when they're being tested, they act one way, but
监视或控制的情境里——也就是说，它们知道自己什么时候被测试——它们会表现出一种行为方式，但
50:59
that does not tell you how they will act if they are free to act in other ways.
这并不能告诉你，如果它们可以自由地以其他方式行动，它们会怎么行动。
51:04
Because the algorithm, the optimization algorithm is working towards an objective. And if you give
因为这个算法，这个优化算法是在朝着一个目标努力。如果你给它
51:10
it a constraint, meaning you watch it, and if you give it a constraint, it'll go find another
一个约束，意思是你盯着它，如果你给它一个约束，它就会去找另一个
51:16
solution. Now, it doesn't make it alive and doesn't make it making anything more than that.
解法。现在，这并不会让它活着，也不会让它做出更多的东西。
51:22
And I'll just also profess that obviously they see a lot more than I do once going on in their
我还要说明一点，显然他们在自己的实验室里看到的东西比我多得多。但
51:27
own labs. But it is sensible that the vast majority of their R&D and compute today
合理的解释是，他们今天绝大多数的 R&D 和 compute
51:35
was dedicated towards making the model capable. I think that's a logical thing for them.
都投入在让模型变得有能力上。我觉得这对他们来说是合乎逻辑的。
51:43
Now, once the technology becomes capable and the products become useful and people want to use it,
现在，一旦技术变得有能力，产品变得有用，人们想用它，
51:50
then as we have, they have more use cases, more people using it, now they're going to get a lot
那么就像我们一样，他们有了更多 use case，更多人在用，现在他们会得到很多
51:57
more issues associated with the product. This is very normal. And when they have a lot,
更多与产品相关的问题。这很正常。而当他们有很多的时候，
52:03
now they have so much market footprint, they have to shift their R&D or total R&D from
现在他们的市场覆盖已经这么大，他们必须把他们的 R&D，或者说全部 R&D，从
52:10
just capability to a lot of verification, evaluation, and testing.
只关注 capability 转向大量的 verification、evaluation 和 testing。
52:15
And so to the point where I wouldn't be surprised if the amount of compute necessary
以至于到了这种程度：如果开发这些模型所需的 compute 量
52:22
to develop these models increased by a factor of 10 because the evaluation is so rigorous.
增加 10 倍，我也不会感到惊讶，因为 evaluation 如此严格。
52:29
And, but that doesn't, that's not where they are today. They're making that transition.
而且，但那并不——那还不是他们今天的状态。他们正在做这个转变。
52:33
And I hear them saying it, and I'm delighted to hear them saying it. But I think that if they believe
我听到他们这么说，我也很高兴听到他们这么说。但我认为，如果他们相信
52:41
they're out of control, then the right answer is, don't ship products until they're in control.
自己已经失控了，那么正确的答案就是：在能控制住之前，不要 ship 产品。
52:50
It is really quite that simple.
真的就这么简单。
53:03
33 presidents. I called the president at 4.30 in the morning and he answered.
33位总统。我在凌晨4:30给总统打了电话，他接了。
53:07
Seven decades in space. We want to see what's beyond the next star.
太空中的七十年。我们想看看下一颗恒星之外有什么。
53:11
A thousand consumer product gods. Today, I am testing seven mattresses.
一千个消费品之神。今天，我要测评七款床垫。
53:15
60 Super Sundays. They were probably the best team in the league every single year.
60 Super Sundays。他们很可能每一年都是联盟里最好的球队。
53:19
53,500 puzzles. I use a different starting word every day.
53,500个谜题。我每天都用一个不同的起始词。
53:24
25,000 recipes. And I love a one-can recipe because I don't love washing pans.
25,000个食谱。而且我喜欢一罐式食谱，因为我不爱洗锅。
53:29
The New York Times, celebrating 175 years of helping you understand the world and make the most of
The New York Times，庆祝 175 年来帮助你理解世界并充分利用
53:35
each day. Subscribe now for a special offer at ny times dot com slash subscribe.
每一天。现在就订阅，可享受特别优惠，网址是 ny times dot com slash subscribe。
53:44
See, I find this perplexing, honestly, because you just, you have so many people,
你看，老实说，我觉得这很让人费解，因为你就是，你有那么多人，
53:49
these ops, professing one that they're out of control, two that they are seeing things that are
这些 ops，声称一是它们已经失控，二是它们看到了些东西，这些东西是
53:54
for you. Which is probably the reason why they had that. And you take the pacing letter that
为你准备的。这大概就是他们为什么会那样的原因。然后你再看那份 pacing letter，
53:58
1300 plus employees signed to realize as potential industry government and society at large
有 1300 多名员工签了名，是为了让潜在的行业、政府和整个社会意识到，
54:04
may need the option to buy time to address emerging risks, develop security measures and
可能需要有争取时间的选项，来应对新兴风险、制定安全措施，并
54:09
strengthen oversight. But each company and country is under intense competitive pressure,
加强监督。但每家公司、每个国家都处在巨大的竞争压力下，
54:15
not to unilaterally. First of all, where does that come from? The labs.
不能单方面行动。首先，这是从哪来的？是那些 labs。
54:19
No, no, that last sentence. Nobody's putting the pressure on them.
不，不，是最后那句话。没有人把压力施加到他们身上。
54:22
The US, I got to listen, there are 400 million Americans here.
美国这边，我得听听，这里有 4 亿美国人。
54:26
I believe that if everybody were just to take a vote just right now, let's just do this.
我相信，如果现在就让所有人投个票，那咱们就这么办。
54:31
If they need this, if they need, that's what they need. I'll give my vote. Don't ship the product.
如果他们需要这个，如果他们需要，那就是他们需要的。我投我的票。别发布这个产品。
54:38
If your product is not ready to ship, don't ship the product. I have no, this is the first time
如果你的产品还没准备好发布，就别发布。我没有，这是我第一次
54:44
that I've heard a company or CEO say that I need the laws. I need to any trust laws
听到一家公司或 CEO 说，我需要这些法律。我需要反垄断法
54:51
to be relieved. I need the liability laws of products to be relieved. So that I can pace myself.
被放宽。我需要产品责任法被放宽。这样我才能按自己的节奏来。
55:00
That paragraph's fantastic. I completely agree. Auditors, I completely agree. We have financial
那段话太棒了。我完全同意。审计师们，我完全同意。我们有财务
55:07
auditors. That's great. Third party audit, safety auditors, financial auditors. That's all great.
auditors。那太好了。第三方 audit、safety auditors、financial auditors。这些都很好。
55:13
That's terrific. Well, the labs will say that we think we are going too fast as a society,
那太棒了。嗯，这些 labs 会说，我们认为作为一个社会，我们走得太快了，
55:18
that we are not ready for what we're building. They are the frontier. They are the frontier.
我们还没准备好面对我们正在构建的东西。他们就是 frontier。他们就是 frontier。
55:24
You have all people, right? Nvidia is the fastest shipper around. For the history of your company,
你们有所有这些人，对吧？Nvidia 是这周围出货最快的。在你们公司的历史上，
55:29
if you are a company, I promise you, I believe you. I believe you that you don't run
如果你是一家公司，我向你保证，我相信你。我相信你，你并不运营
55:35
and have control companies. But the liabilities, but this is where I think you get into an interesting
并控制公司。但责任问题，但这就是我认为你会进入一个有趣的
55:44
deep question of what kind of technology are we dealing with here? When were technology?
深刻问题：我们在这里面对的到底是什么样的技术？技术又是在什么时候？
55:50
Well, let's hold on that for a minute. Many companies, if you ship something that is not quite right,
嗯，我们先把那个放一放。很多公司，如果你发布的东西不太对，
55:57
it's a pain. You guys have shipped graphics cards. It had overly loud fans. In this, with these,
这很烦人。你们已经出货过 graphics cards。它的风扇太吵了。在这方面，用这些，
56:04
you know, you've used word intelligent a number of times here. You're dealing with intelligent systems,
你知道，你在这里已经用过好几次 intelligent 这个词了。你面对的是 intelligent systems，
56:10
not a life that are given goal functions. We can sort of go around and around with how to describe
不是被赋予 goal functions 的生命。我们可以就该怎么描述
56:17
that. You're trying to make the systems capable of working for longer periods of time more relentlessly.
这件事绕来绕去。你是想让这些系统能够更长时间、更持续地工作。
56:22
Yeah. If you ship that and it's not ready, or even if you think it is ready and it's not ready,
是啊。如果你把那个发布出去，而它还没准备好，或者即使你觉得它准备好了、但它其实没准备好，
56:31
then things could get very weird in our society very fast. Yeah. Hypothetic, you're completely right.
那我们的社会可能会很快变得非常怪异。是啊。假设性地说，你完全正确。
56:36
But all I'm suggesting is this, before we go build, before we go fix the hypothetical problems,
但我只是建议这一点：在我们去构建之前，在我们去解决那些假设性问题之前，
56:42
before we go create more regulations, can we work on the practical problems that we know exist,
在我们去制定更多 regulations 之前，我们能不能先处理那些我们知道确实存在的实际问题，
56:51
which is we need to do a better job with containment and isolation, which is we should not allow
就是我们得在 containment 和 isolation 上做得更好，也就是我们不应该允许
56:58
a product to interact with the external world until it's ready to be interacting with external worlds.
一个产品在准备好和外部世界互动之前，就去和外部世界互动。
57:07
Yeah. I believe those two things are solvable problems. I believe they are solving it. The second
对。我相信这两件事都是可以解决的问题。我相信他们正在解决。第二
57:15
part is when it comes to incentives, when it comes to incentives, which is somehow, somehow,
部分是，当涉及到 incentives，当涉及到 incentives，就是不知怎么的，不知怎么的，
57:25
you need everybody in the world to slow down when you are the leader. You need everybody in the
你需要世界上每个人都慢下来，当你是领导者的时候。你需要世界上
57:31
world to slow down so that you're willing to uphold your basic responsibility. That strikes me
每个人都慢下来，这样你才愿意承担你的基本责任。这让我觉得
57:39
odd. Wouldn't it so them down most of all? What's that? Wouldn't these ideas so them down most of all?
奇怪。这难道不会最让他们慢下来吗？什么？这些想法难道不会最让他们慢下来吗？
57:44
I mean, people have been very unclear about what ideas they're talking about, including I
我的意思是，人们一直很不清楚他们说的想法到底是什么，包括我
57:49
will say them. But let me give you one that I believe in. So you can use me as the punching back
会把它们说出来的。但让我说一个我相信的，所以你可以在这里拿我当出气筒
57:53
here. I've heard these cancel down. Nobody is putting on, as you know, I don't trust these companies.
这些被取消掉的说法我也听过。没人在装，你知道的，我不信任这些公司。
58:00
Nobody is building more compute today. Nobody's building more compute today than the people
今天没有谁在建更多 compute。今天没有谁建的 compute 比那些人更多
58:08
asking to be slowed down. It strikes me odd. I think one thing where maybe there's some
也就是那些要求被放慢的人。这让我觉得奇怪。我觉得也许有一点
58:13
difference here is I don't trust companies, even with liability to keep the public good in mind.
不同之处是，我不信任公司，即使有 liability 让它们把公共利益放在心上。
58:19
I think we've watched companies do terrible damage to the environment. The profit motive, the desire
我觉得我们都看着公司对环境造成了
58:25
for power, the desire to cut corners to be first. I feel like you're treating these like these
对权力的欲望，想走捷径、想争第一的欲望。我觉得你把这些当成好像这些
58:32
are not things that we have seen again and again in history. But I feel like they are things
不是我们在历史上一再见过的东西。但我觉得它们就是这样的东西
58:36
we've seen again and again in history that we've watched. That's right. I see a lot of good things
我们在历史上一次又一次看到过，我们亲眼见证过。没错。我看到很多好的事情
58:39
in history. I see a lot of good things in history. I work with a lot of CEOs and they want to do
在历史上。我在历史上看到很多好的事情。我跟很多 CEO 合作，他们想做
58:46
the right things. I work with a lot of companies. They want to do the right things. They want to do
正确的事情。我跟很多公司合作。他们想做正确的事情。他们想做
58:51
good engineering. I know a lot of people in those two labs who are dedicating their lives to do
好的工程。我认识那两个实验室里的很多人，他们全身心投入去做
58:57
good work and they're built. They know what happened. I know they know what happened. I know they
好的工作，而且他们就是这样被塑造出来的。他们知道发生了什么。我知道他们知道发生了什么。我知道他们
59:02
know how to fix it and I know they're fixing it. Meanwhile, all of the other narratives
知道怎么修复它，而且我知道他们正在修复。与此同时，所有其他那些说法
59:12
to deflect blame, to make it sound like AI is so powerful. I have no idea how to fix it. It's not
都是为了甩锅，让它听起来像 AI 太强大了。我根本不知道该怎么修复它。这不是
59:20
my fault. It's just because of technology. It's just so powerful. I think that's a deflection
我的错。这只是因为技术。技术就是太强大了。我觉得那是在转移焦点
59:25
of blame. It's a deflection of responsibility. It's unnecessary. It actually hurts their reputation
指责。
59:32
more than it helps. It hurts their character more than it helps. It hurts employee morale than
这是在推卸责任。
59:39
it helps. What if it's what they believe? Well, I guess I can't talk to you about what they believe.
这没必要。
59:47
I can tell you what I believe. This industry wouldn't exist without your chips. The parallel
这实际上对他们的声誉伤害大于帮助。
59:52
processing that was required for deep learning to work going all the way back to the original Alex
这对他们品格的伤害也大于帮助。
59:57
net. It's all on video chips. A lot of the people from the beginning or who they're at the beginning
这对员工士气的伤害也大于帮助。
60:03
have these fears that I think to a lot of people and they hear them like, what are you talking
如果那真是他们所相信的呢？
60:08
about from Jeffrey Hinton and Ilya Sutskiver all the way up to I've heard these from Dario
嗯，我想我没法跟你谈他们所相信的东西。
60:13
from Sam Altman talking about Lawson control, Demesis Abyss. A lot of the people who are very
来自 Sam Altman 谈论 Lawson control、Demesis Abyss。很多那些非常
60:21
foundational in creating the form of AI we see now seem to believe that there's a very good shot.
那些在创造我们现在看到的 AI 形态方面起奠基作用的人，似乎相信这事很有戏。
60:30
We could lose control of it. Elon Musk has talked about human beings being a bootloader for AI.
我们可能会失去对它的控制。Elon Musk 说过，人类是 AI 的 bootloader。
60:36
We could lose control of it and that would be the end of us. I don't think you believe that.
我们可能会失去对它的控制，而那会是我们的终结。我不觉得你相信这一点。
60:40
No. I think you don't believe it at all. So taking them is serious about what they believe.
不。我觉得你根本就不信。所以，把他们当回事，就是认真对待他们所相信的东西。
60:44
When you have your arguments with them or maybe you could just have it with me.
当你和他们争论的时候，或者也许你直接跟我争也行。
60:48
When you're like, what are you talking about? Even though they're the people in many cases who
当你说，你在说什么？尽管在很多情况下，他们就是那些人，那些
60:52
are trying to hear what they think they're are grounded. So when Jeffrey Hinton is on TV saying he
正试图听到他们认为自己是有根据的。所以当 Jeffrey Hinton 在电视上说，他
60:57
thinks a 10% chance of societal destruction is not unreasonable. I would tell Jeff that
认为 10% 的社会毁灭概率并非不合理。我会告诉 Jeff，
61:04
it's irresponsible to say all that. All of his predictions have been wrong.
说那些话是不负责任的。他的所有预测都是错的。
61:09
Enough predictions. That 10% chance is not grounded on science. It's not grounded on research.
别再预测了。那 10% 的概率没有科学依据。也没有研究依据。
61:17
Just because it comes from a scientist doesn't make it scientific. Those predictions are hurtful.
仅仅因为它出自一位科学家，并不意味着它就科学。那些预测是有害的。
61:21
Let's take it a face value that the recommendation is exactly what he said, which is
我们不妨按字面理解，这个建议正是他说的那样，也就是
61:28
nobody should want to be an radiologist. And the world has no radiologist today.
没人应该想当放射科医生。而且今天世界上没有放射科医生。
61:33
I think if you work as a radiologist, you're like the KOT that's already over the edge of the
我觉得如果你是一名放射科医生，你就像已经越过悬崖边缘的 KOT，
61:39
cliff. But hasn't yet looked down so it doesn't realize there's no ground underneath it.
但它还没往下看，所以没意识到下面没有地面。
61:45
People should stop training radiologists now. It's just completely obvious. So within five years
人们现在应该停止培养放射科医生了。这简直太明显了。所以五年之内
61:50
deep learning is going to do better than radiologists because it can be able to get a lot more
deep learning 会比放射科医生做得更好，因为它能获得多得多的
61:54
experience. And it might be 10 years, but we've got plenty of radiologists already.
经验。也可能是 10 年，但我们已经有很多放射科医生了。
61:58
Is that helpful or hurtful to the society? I think we can all agree. We can both agree.
这对社会是有益还是有害？我想我们都能认同。我们俩都能认同。
62:04
It would be terribly hurtful. It didn't happen. It's a good or bad that we scare young people
那会非常有害。但这事没发生。这是好事还是坏事：我们用 AI 的未来把年轻人
62:11
about the future of AI so much so that they don't even want to go to universities and don't want
吓得那么厉害，以至于他们甚至不想上大学，也不想
62:17
to go to college anymore because they don't think they'll get a job. Is that helpful or hurtful
再读大学了，因为他们觉得自己找不到工作。那是有益还是有害
62:22
if it were to happen? It's hurtful. Don't think for a second just because you're an alarmist
如果这真的发生？那是有害的。别因为自己是个危言耸听的人，就以为……
62:27
that you're doing a social good. It is not true. So I think that we ought to just all be
说你在做对社会有益的事。这不是真的。所以我觉得我们都应该
62:34
wiser, more mature, be evidence-based, be scientific. If you wanted to be scientific, be scientific.
更明智、更成熟，讲证据，讲科学。如果你想讲科学，那就讲科学。
62:42
Do the science. Do the science. But alarming people, making claims that they simply,
去做科学。去做科学。但吓唬人，提出那些他们只是……
62:51
their track record is horrible. Their track record is literally horrible.
他们的过往记录很糟糕。他们的过往记录简直糟透了。
62:56
Well, their track record is bad in one respect and good in another, which is many,
嗯，他们的过往记录一方面很差，另一方面又很好，那就是很多、
62:59
many predictions have been weak, which predictions that the scaling laws would work.
很多预测都很弱，哪些预测呢？就是那些说 scaling laws 会奏效的预测。
63:07
No, we've got to be careful here. Even then, to just say what it is for the audience here,
不，我们这里必须小心。即便如此，就为这里的听众说说它到底是什么，
63:12
that if you don't compute and training data, these things will keep getting smarter. That's
就是说，如果你不 compute 和 training data，这些东西就会越来越聪明。那就是
63:16
correct. It's not true. It is not true that if you just keep training these models, they'll
对。这不是真的。并不是说你只要一直 training 这些 models，它们就会
63:23
get better. Notice, it is the reason why the second scaling law had to come along. Why do you
变得更好。注意，这就是为什么第二个 scaling law 必须出现。为什么你
63:28
need a second scaling law? The first scaling law, I'll describe it with a second. This is the
需要第二个 scaling law？第一个 scaling law，我等一下用一秒讲一下。这就是
63:32
second scaling law's test time scaling inference. The more you iterate, the more you search,
第二个 scaling law 的 test time scaling inference。你迭代得越多，搜索得越多，
63:40
the more you explore, the better answer you'll discover, inference time scaling.
探索得越多，你发现的答案就越好，inference time scaling。
63:46
What is the big breakthrough that caused the current AI to be incredibly useful?
是什么重大突破让当前的 AI 变得极其有用？
63:51
Precisely the opposite of the prediction. It was predicted that it would be the end of software
恰恰与预测相反。曾经预测这会是 software
63:56
tools. It was the SaaS apocalypse, right? What is making... SaaS will always be with us.
tools 的终结。那就是 SaaS 末日，对吧？是什么让……SaaS 会一直和我们同在。
64:05
What is making these AI so productive right now? The usage of tools. In the future,
现在到底是什么让这些 AI 这么高效？是工具的使用。
64:10
it'll be enhanced by a number of agents using these tools. There'll be more people using Adobe,
未来，它会被很多使用这些工具的 agents 进一步增强。
64:16
there'll be more using Salesforce tools and so on and so forth. So give me one prediction that
会有更多人用 Adobe，
64:22
has been right. Well, the prediction they begin to mention... Let me try to answer that because
会有更多人用 Salesforce 的工具，等等等等。
64:28
they're not here. The prediction that you would have a merchant... The fact that you can't
那给我一个确实猜对了的预测吧。
64:32
come up with one, I think, in itself is a... Well, I think it depends what we're talking about
嗯，他们一开始提到的那个预测……
64:37
predictions, right? I mean, predictions are prediction. Jeffrey Hinton was the person as responsible
让我试着回答一下，因为他们不在这儿。
64:43
as anybody else for deep learning at a time when everybody thought it was ridiculous.
那个说你会有一个商家的预测……
64:48
And it has turned out to be a pretty good bet, right? I mean, the sort of big one.
结果证明，这还真是个相当不错的赌注，对吧？我是说，那种大赌注。
64:52
Everyone that made great contributions, I love Hinton. I hate his predictions.
所有做出过巨大贡献的人里，我喜欢 Hinton。我讨厌他的预测。
65:00
I understand that. Here's the stylized concern that all these people have. I want to do this
我理解。这些人都有一种很典型的担忧。我想聊这个，
65:06
for a few minutes and we can move on to some other topics. But the fear that seems to me to animate
聊几分钟，然后我们可以转到其他话题。但在我看来，那种似乎驱动着
65:12
them and that I think a lot of people find intuitively reasonable. Is your creating systems? I'm not
他们、而且我觉得很多人直觉上觉得合理的恐惧，是不是：你正在创造系统？我并不是
65:19
saying they're alive. You say everything long enough, it's going to be reasonable.
说它们活着。任何东西，你说得够久，它就会显得合理。
65:23
Well, that fair enough. So your creating systems that are intelligent, that are becoming more
嗯，这倒也有道理。所以你在创造一些系统，它们有智能，而且正在变得比我们更
65:28
intelligent than us in certain domains. You give them reward functions as you are saying,
智能——在某些领域。你像你说的那样，给它们 reward functions，
65:32
the desire to do things, right? You give them persistence. They move very fast in the digital world.
做事情的欲望，对吧？你给它们 persistence。它们在数字世界里行动得非常快。
65:40
You're creating something, some entity, an agent that is smart, that is capable, that is relentless,
你在创造某种东西，某种实体，一个 agent，它很聪明、很有能力、还不知疲倦，
65:50
and who the workings of its mind, we don't really understand, but you've sent us an open AI.
而它心智的运作方式，我们其实并不真正理解，但你们给我们送来了一个 open AI。
65:55
I just don't want you to contribute to that software. I don't think software is relentless.
我只是不想让你为那个 software 做贡献。我不觉得 software 会不知疲倦。
66:03
Are they trying to make it very persistent? Highly persistent models?
他们是想把它做得非常 persistent 吗？高度 persistent 的 models？
66:08
Because I made it that way. But that's how they're making it.
因为我就是把它做成那样的。但他们就是那么做的。
66:11
Yeah, but that's not persistence. It's just on.
对，但那不是 persistence。它只是开着而已。
66:14
Yeah. Persistence, persistence, there's a will power. There's no will power here. Just electrical power.
对。Persistence，persistence，那得有意志力。这里没有意志力。只有电力。
66:21
Well, listen, let me give you some insight. Let me give you some insight.
好吧，听着，让我给你点见解。让我给你点见解。
66:24
Aren't human beings just energy with their reinforcement learning loops?
人类难道不就是带着 reinforcement learning loops 的能量吗？
66:27
Whatever. So anyways, I just think that we can't make jokes of all this stuff. We're scaring
随便吧。总之，我就是觉得我们不能拿这些事开玩笑。我们在吓唬
66:33
the American public. Listen, spawn, create, kill, wait, sleep. All of these words
美国公众。听着，spawn、create、kill、wait、sleep。所有这些词
66:46
are associated with agents. Right? That's what people use. These words were created when
都和 agents 有关。对吧？人们就是这么用的。这些词是在
66:54
multi-processing systems for operating systems. These are literally the commands of an operating system.
multi-processing systems for operating systems 的时候被创造出来的。这些字面上就是 operating system 的 commands。
67:01
You spawn a process, replace process with agent, the process forks as a result parent and child.
你 spawn 一个 process，把 process 换成 agent，process 就 forks，结果产生 parent 和 child。
67:14
The agent forks, spawns a new, give birth. These are words that were created for the operating system
agent forks，spawn 一个新的，生孩子。这些词就是为 operating system 创造的
67:25
30, 40, 50 years ago. But notice we didn't infuse human characteristics into them.
30、40、50 年前。但注意，我们并没有给它们注入人类的特征。
67:35
We kill processes all the time. Kill minus nine, kill a dead. It's just a process.
我们一直在 kill process。kill -9，kill 一个死掉的。它不过就是个 process。
67:43
But now we're talking about these things. A collection of people want to make this software more
但现在我们在谈论这些东西。一群人想把这个 software 变得比它本身更
67:49
than it is. And we talk about software in a new way. But they're all the same old words.
多。而我们用一种新方式谈论 software。但那些还都是老词。
67:56
Now, the last generation of computer engineers, we were doing all the same things.
现在，上一代 computer engineers，我们做的也都是同样的事。
68:03
But doesn't the software act in a new way? I mean, from the outside, I don't have the
但 software 的运作方式难道不是新的吗？我是说，从外面看，我没有你那种
68:06
technical expertise you do. The fact that it's doing, it's crawling the internet, it's doing
技术专长。事实是它在做事，它在爬互联网，它在做
68:11
search, it's doing, you know, it's communicating, it's breaking out of things like,
搜索，它在做，你知道，它在交流，它在突破一些东西，像，
68:16
listings don't break out of things. Now software breaks out of sandboxes all the time.
listings 不会突破这些东西。现在 software 一直都在突破 sandbox。
68:20
That's the reason why we need a virtual machines. You can't have agents, their own sandbox,
这就是为什么我们需要 virtual machines。你不能让 agents 自己管理自己的 sandbox，
68:25
monitoring themselves. You need a, you need a, if you want, a whole bunch of watch dogs.
自己监控自己。你需要一个，你需要一个，如果你愿意的话，一大堆看门狗。
68:30
And so these are ideas that have been around for a long time, which is somehow,
所以这些都是已经存在很久的想法，不知道怎么就，
68:38
somehow in the recent generation gave it, you know, a whole bunch of human words. And I just
不知道怎么的，在最近这一代里被赋予了，你知道，一大堆人类词汇。而我只是
68:43
think that it's unnecessary. It's software. You know, when I see it in my head, it's a bunch of
觉得这没必要。这是 software。你知道，当我在脑子里看它的时候，它就是一堆
68:49
code, a bunch of numbers, running on computers. And all of that is happening in a very natural way
代码，一堆数字，在 computers 上运行。而所有这些对我来说都以一种非常自然的方式
68:55
to me. Which is the reason why I can operate. And it's the reason why, if it's, if it's just simply
发生着。这就是为什么我能运作。这也是为什么，如果它，如果它只是
69:01
mystery and myth, how, how do I build a company around it? I think one of the fundamental
神秘与迷思，怎么，我该怎么围绕它建立一家公司？我觉得最根本的
69:07
questions is gets at is just what is intelligence before you can even think about what it means to
问题之一其实是在问：在你甚至开始思考拥有 intelligent machines 意味着什么之前，到底什么是 intelligence？
69:12
have intelligent machines. Just what, what is intelligence to you? Well, there's a, there's a
那么，对你来说，intelligence 到底是什么？嗯，有一个，有一个
69:17
technical formulation of intelligence. First of all, you know, when people talk about intelligence
关于 intelligence 的技术性表述。首先，你知道，当人们谈论 intelligence
69:23
and thinking and, you know, all of these things, of course, there's no formal definition for most
和思考，以及你知道的这一切时，当然，对大多数人来说并没有正式定义。
69:29
people. But in the field of computer science, there is a definition. The definition is perception
但在 computer science 领域，是有一个定义的。这个定义就是 perception，
69:34
which is perceiving the world and understanding it to which is reasoning. And reasoning is
也就是感知世界并理解它，然后是 reasoning。而 reasoning 是
69:42
the ability to decompose any scenario and anything you see, any experience into more elemental parts.
把任何场景、你看到的任何东西、任何经历分解成更基本部分的能力。
69:51
And third is planning towards an objective. That fundamental formulation applies to
第三是朝着一个目标做 planning。这个基本表述适用于
69:56
agentic systems. It applies to robotic systems, applies to self-driving cars. And so you could see
agentic systems。它适用于 robotic systems，适用于 self-driving cars。所以你能看到
70:03
the industry building a layer by layer by layer step by step by step to the point we now have
这个行业在一层一层又一层、一步一步又一步地搭建，直到我们现在有了
70:09
what we, what perceived as intelligence. I think this gets to such a core question of this
我们所、我们所感知为 intelligence 的东西。我觉得这就触及了这次
70:13
conversation, which is some of the ways you've described the technology to me. It does not sound
对话的一个核心问题，也就是你向我描述这项技术的一些方式。听起来
70:19
like you think there's anything really new about it. It is, you know, maybe new in scales,
你并不觉得它有什么真正全新的东西。你知道，它也许只是在 scale 上是新的，
70:23
new in capability. But fundamentally, this is software we've always, we've not always had,
在能力上是新的。但根本上，这是软件——我们一直、我们并不是一直都有，
70:27
but we've had software for a long time. A lot of people believe when you're getting to intelligence
但我们已经拥有软件很久了。很多人相信，当你达到 intelligence 的时候
70:31
at these levels, it is a phase change. It is something different, something we have not dealt
在这些水平上，这是一种 phase change。它是某种不同的东西，某种我们还没有处理
70:37
with before. A kind of generally intelligent technology that is advancing in its intelligence
过的东西。一种通用智能技术，它的智能正在
70:42
very rapidly. I want to make sure I actually do understand we are on that divide. Is this something
飞速提升。我想确保我确实明白，我们正处在那个分界线上。这是某种
70:48
fully new? Is this something that requires something new from us? Or is this more like something
全新的东西吗？这是某种需要我们拿出新东西来应对的东西吗？还是说这更像某种
70:56
old, our intelligent machines different than the machines we've had? Well, almost all of technology
旧东西，我们的智能机器不同于我们过去拥有的机器？嗯，几乎所有技术和
71:03
and civilization is built on layers of understandable technology, which at scale becomes fairly
文明都建立在层层可理解的技术之上，而这些技术一旦 at scale，就会变得相当
71:12
extraordinary. The fact that that we can connect to the internet by just, you know, holding a
非凡。事实上，我们只要，你知道，举起一部
71:18
phone up. I mean, it's kind of weird, you know, that we're connected to every piece of information
手机，就能连上 internet。我的意思是，这有点奇怪，你知道，我们竟然连接着每一条信息
71:22
in the world on this little tiny device, you know, just in the air. And the fact that this little
在这个世界上，就在这么一个小小的设备上，你知道，就在空中。而且这个小小的
71:33
tiny piece of glass can somehow take trillions of pieces of information and bring to us precisely the
一小片玻璃居然能以某种方式获取数以万亿计的信息，并精确地把
71:39
one that we want because it's been passed through a recommender system. And so if you think about
我们想要的那一条带给我们，因为它经过了 recommender system。所以如果你想想
71:44
how is it possible that we knew where all the information is? Somebody had to go crawl it,
我们怎么可能知道所有信息在哪儿？得有人去 crawl 它，
71:50
had to index it. And that uses machine learning techniques, which is the early versions of artificial
得去 index 它。而这用到了 machine learning 技术，也就是 artificial
71:57
intelligence. And these systems do magical things to the point where I now expect it. It took
intelligence 的早期版本。而且这些系统能做出很神奇的事，神奇到我现在都觉得这是理所当然。这花了
72:04
literally 20-some years and hundreds of billions of dollars of infrastructure build out
真的 20 多年，以及数千亿美元的 infrastructure 建设
72:10
in order for everything to just seem so natural to you. To the point where we now take
才能让这一切对你来说显得如此自然。到了我们现在已经把它当成
72:15
it for granted, every single milestone that we achieve from a technology perspective is celebrated,
把它当作理所当然，从技术角度来看，我们达成的每一个 milestone 都会受到庆祝，
72:21
and I celebrated with Glee, and I celebrated with so much enthusiasm because I'm proud of the people
而我满心欢喜地庆祝，我满怀热情地庆祝，因为我为那些人感到骄傲
72:26
who did it. I'm proud of ourselves who contributed to it. We're, you know, proud of the breakthrough.
为做成这件事的人感到骄傲。我也为我们这些做出贡献的人感到骄傲。你知道，我们为这个突破感到骄傲。
72:31
But when you, and it seems, wow, it seems like a miracle at the time. But
但当你——而且看起来，哇，在当时这看起来就像个奇迹。但
72:37
that sensation lasts about 17 days after that. We can't use everything quickly. I agree with that.
那种感觉在那之后大概只能持续 17 天。我们没法很快把所有东西都用起来。这一点我同意。
72:44
But is this a different phase? You know, yes, some of the companies talk about this,
但这是一个不同的阶段吗？你知道，是的，有些公司会谈到这个，
72:51
the CEO of Google, I think it was, as the equivalent of fire, right? Like a new epoch in human
Google 的 CEO，我想是他吧，把它比作火，对吧？就像人类
72:56
history. Is that how you see it as transitional iterative? No, I think this is this is completely
历史上的一个新纪元。你是这样看的吗，觉得它是过渡性的、iterative 的？不，我觉得这——这完全是
73:04
a revolution. And as we were talking about earlier, you went from being able to find everything,
一场革命。而且就像我们之前聊到的，你从能够找到一切，
73:10
find anything, to be able to ask anything, know everything, and do everything. And so clearly,
找到任何东西，变成能够问任何问题、知道一切、做一切。所以很明显，
73:16
it's a new abstraction level. Now, the thing that I'm reluctant about is to cause it to seem
这是一个新的 abstraction level。现在，我不太愿意的是让它看起来
73:26
like it's more than that. In the funnel analysis, engineers are doing engineering work.
好像它不止于此。在 funnel analysis 里，工程师就是在做 engineering work。
73:31
Once we invented the technology, once we discovered a solution for it, when you look back,
一旦我们发明了这项技术，一旦我们为它找到了解决方案，回头看，
73:35
it's fairly obvious. And it's fairly mundane to a lot of people in this. And the fact that we're
它就相当明显。而且对很多身处其中的人来说，它也相当稀松平常。而事实上，我们
73:41
able to make the technology better and better and better every day is because we understand it
能够每天把这项技术做得越来越好，越来越好，越来越好，是因为我们理解它
73:45
obviously. And so we understand how to make it better. So you turn into an engineering problem.
很明显。所以我们明白怎么把它做得更好。于是它就变成了一个 engineering problem。
73:51
And you say, what we don't have right now is a level, because this is something now you've said,
然后你说，我们现在缺的是一个水平，因为这是你刚说过的事情，
73:57
of testing, monitoring, sandbox, security, red control excellence that we need for what we're building.
也就是 testing、monitoring、sandbox、security、red control 的卓越水平，而这是我们正在构建的东西所需要的。
74:03
And it's not because the companies are don't have extraordinary engineers. I believe that
这并不是因为这些公司没有非凡的工程师。我相信
74:11
open AI andthropic, because I know many of them are extraordinary. But that actually is in part
open AI 和 andthropic，因为我知道他们中很多人非常出色。但这实际上在某种程度上
74:15
what makes me worried is open. No, no, no, no, no, no, no, no, no. What's happening to them
让我担心的是 open。不，不，不，不，不，不，不，不，不。他们身上正在发生的
74:19
is a transition. And I said this over and over again, this is a big but simple idea.
是一场转型。我一遍又一遍地说过，这是一个宏大但简单的想法。
74:25
Finally, we now have a piece of software that is useful because it's useful. The adoption took off.
终于，我们现在有了一款软件，它有用是因为它有用。采用率一下子起飞了。
74:30
But remember, how is it possible that a company that six months ago was trying to make something
但记住，一家六个月前还在试图做出某种东西的公司，怎么可能
74:40
useful, capable? How would they have as much resources dedicated on testing, evaluation, and all
有用、有能力？他们怎么会有那么多资源专门投入到 testing、evaluation，以及所有
74:49
of the compute dedicated to that? It was unnecessary until now. And so what's going to happen over
用于那件事的 compute？直到现在，这都没必要。所以，接下来会发生什么，在
74:56
the next several years is that we're going to transition from these labs, becoming engineering focused,
接下来的几年里，我们会从这些 lab 转型，变得更以 engineering 为核心，
75:03
much more production engineering focused, and product focused companies. And so I think
更加以 production engineering 为核心，以及以 product 为核心的公司。所以我觉得
75:09
they're just going through a transition. These are companies, extraordinary companies,
他们只是在经历转型。这些公司，都是非凡的公司，
75:12
incredibly talented companies, the most constant going through companies of all time.
极具才华的公司，有史以来最持续经历这些的公司。
75:17
And they're just going through their transition. It's not more than that. It's not less than that.
而他们只是在经历自己的转型。不多也不少。
75:23
So many of the companies now both OpenAI and Anthropic last couple of months have put out these
所以现在很多公司，OpenAI 和 Anthropic 在过去几个月里都发布了这些
75:27
big, I don't know what to call them, papers, blog posts, something. When AI builds itself is the name
Big，我不知道该叫它们什么，论文、博客文章，什么的。When AI builds itself 是那篇的名字，
75:32
of the Anthropic one, I forget the name of the OpenAI one. The computers are building itself.
An
75:36
But you guys know that they're talking about recursive self-improvement.
但你们知道，他们说的是 recursive self-improvement。
75:40
You know that we use recursive self-improvement. So I'd like your perspective on RSI.
你知道我们用 recursive self-improvement。所以我想听听你对 RSI 的观点。
75:45
I think that RSI is fundamentally how things are done. So we use software to design a computer,
我觉得 RSI 从根本上就是事情的做法。所以我们用 software 来设计 computer，
75:55
to run software, to design a computer, to run software, to design a computer. That's basically
去运行 software，去设计 computer，去运行 software，去设计 computer。这基本上
76:00
what we do, recursive self-improvement because our computers are getting better every single year.
就是我们在做的事，recursive self-improvement，因为我们的 computers 每年都在变得更好。
76:07
And in fact it's getting better than faster than that every single year because we use software
而事实上，它每
76:11
to make software better. That is called computer engineering.
让软件变得更好。这就叫 computer engineering。
76:19
That we've been doing this for a long time. Now, in the context of agents,
我们做这件事已经很久了。现在，在 agents 的语境里，
76:24
it runs through the process once, it reflects on it, it studies the various paths it went through,
它会把流程跑一遍，对它进行反思，研究它走过的各种路径，
76:33
chooses the best approach. The next time you're going to do exactly the same task, I'm going to document
选出最好的方法。下一次你要做完全相同的任务时，我会记录成
76:39
a file. I'm going to tell you how I did it last time. That wasn't the most effective. I'm going to
一个文件。我会告诉你我上次是怎么做的。那并不是最有效的。我会
76:43
call it skills. And because you use it over and over again, some of it in skills, some of it is
把它叫做 skills。而且因为你反复使用它，其中一部分放进 skills，一部分会
76:49
going to be a memory. We're going to improve the memory. So that next time you use it, it's even
变成 memory。我们会改进 memory。这样下次你用它的时候，它甚至
76:56
better than last. Recursive self-improvement. You could also decide that you take all of this
比上次更好。Recursive self-improvement。你也可以决定，把所有这些
77:03
skills, all of this memory, and you can take all of this data and train the next release of the
技能、所有这些 memory，而且你可以把所有这些 data 拿来 train 下一版
77:10
model with it. And so that AI becomes better and better at servicing you over time. We're doing
model。这样 AI 就会随着时间越来越擅长为你服务。我们正在做
77:16
re-all of that is happening. It is absolutely happening. Meanwhile, the amount of compute that they
re-all 所有这些都在发生。它绝对在发生。与此同时，他们所拥有的 compute 量
77:23
have is growing and therefore they could do everything faster. What used to take a year to
他们拥有的在增长，因此他们能把一切做得更快。过去需要一年才能
77:29
pre-train something now takes several hours because the computers are getting faster and they have
pre-train 一个东西现在只需要几个小时，因为计算机越来越快，而且他们拥有
77:35
more of it. So now the loop is going faster, completely, completely understandable. Does that give
更多这种东西。所以现在 loop 跑得更快了，完全、完全可以理解。这难道给了
77:41
them any excuse to launch a product that hasn't been tested? The answer is no. Just come back to that.
他们任何借口去发布一个还没测试过的产品？答案是不。就回到那一点。
77:48
You, nobody, no enterprise is able to operate in an environment where the underlying software
你，没有人，没有任何企业能在一种环境里运营，其中底层 software
77:58
is literally changing all the time. There's a release process. So when they roll out a new model,
真的是一直在变。有一个 release process。所以当他们推出一个新的 model 时，
78:06
we need to evaluate it before we release it into our operations. We can't just have it recursively
我们需要在把它发布到我们的运营中之前评估它。我们不能让它 recursively
78:12
changing all the time. And so they have to test the product before they release it. We will test
一直在变。所以他们必须在发布产品之前测试它。我们会测试
78:18
the product before we release it into operation. And so I think recursive self-improvement is a fabulous
产品，在我们把它投入运行之前。所以我觉得 recursive self-improvement 是件很棒的
78:25
thing. And do you think there is any level? I've heard you say before that learning should always
事情。那你觉得有没有什么层级？我之前听你说过，学习应该始终
78:31
have a human in the loop. Yeah, like I said just now, you got recursive self-improvement. They seem
有 human in the loop。对，就像我刚才说的，你有了 recursive self-improvement。他们似乎
78:36
to be imagining something where it wouldn't always. Well, don't ship me anything that you didn't evaluate.
在想象一种它并不总是如此的情况。那好，别把任何你没评估过的东西发布给我。
78:43
Don't ship me. Don't ship and video any products that humans did not in the loop evaluate. Please
别把我发布出去。别发布，也别给任何人类没有以 in the loop 的方式评估过的产品拍视频。拜托
78:51
don't do that. And the fear that we talked about earlier that they're worried they're not evaluating
别那样做。还有我们之前谈到的那个担忧，就是他们担心自己没在 evaluate
78:56
that they don't know how to evaluate these systems. And the more they change kind of rapidly,
他们不知道该怎么 evaluate 这些系统。而且它们变化得越快，
79:00
the more they worry the systems are tricking them. I don't believe that. I believe that their
他们就越担心这些系统在骗他们。我不相信这一点。我相信他们的
79:04
researchers are working every single day to learn about how to evaluate these systems. Verification,
研究人员每天都在努力了解该怎么 evaluate 这些系统。Verification，
79:10
so you know, 10% 20% of our companies dedicated to design, 80% dedicated to verification.
所以你知道，我们公司有 10%、20% 投入 design，80% 投入 verification。
79:19
Today, most labs, understandably, is 80% dedicated to capability and 20% dedicated to safety,
今天，大多数 labs，可以理解，80% 投入 capability，20% 投入 safety，
79:29
verification, eval. This is the flip. That's right. AI needs to accelerate to be safe.
verification、eval。这就是反转。没错。AI 需要加速才能安全。
79:36
I want them to get more compute, but allocated towards evaluation, to alignment. And I think they're
我希望他们拿到更多 compute，但要分配到 evaluation 上，分配到 alignment 上。而且我觉得他们
79:46
doing that. If I were in the car industry a hundred years ago, I would rather the car industry
这么做。
79:53
accelerated to today in one year because I believe today's car is way more safe than a car 99
如果我在一百年前身处汽车行业，我宁愿汽车行业在一年内加速发展到今天，因为我相信今天的车比 99 年前的车安全得多。
80:00
years ago. An ABS technology automatic braking requires computer vision technology sensor fusion
ABS technology 的 automatic braking 需要 computer vision technology、sensor fusion technology、radar 和 cameras，而且，你知道，所有这些技术得结合在一起，才能在该刹车的时候刹车，不该刹车的时候不刹车。
80:07
technology radars and cameras and, you know, all that technology coming together in order to
那技术极其难。
80:13
break when you should and not break when you shouldn't. That technology extremely hard.
我本来会希望，每个人本来都会希望 99 年前就有 ABS technology。
80:20
I would have hoped, everybody would have hoped that ABS technology existed 99 years ago.
那样会少死很多孩子。
80:25
A lot fewer children would have been killed. And so, you know, airbags, safe seat belts,
所以，你知道，安全气囊、安全带，我是说，所有这些东西，自动收紧安全带，所有这些东西，你能想象吗，
80:31
I mean, all of that stuff, self-tightening seat belts, all of that stuff, could you imagine,
我是说，所有那些东西，自动收紧的安全带，所有那些东西，你能想象吗，
80:36
that's all technology, accelerate the living daylights out of that development. And so,
这些都是技术，要把这个开发拼了命地加速。所以，
80:41
when I say, when I say we need to accelerate AI technology, people think for some reason,
当我说，当我说我们需要加速 AI 技术时，人们不知为什么会觉得，
80:48
safety is not part of that. Safety is part of it, alignment is part of it, Eval is part of it,
safety 不是其中一部分。Safety 是其中一部分，alignment 是其中一部分，Eval 是其中一部分，
80:55
guard railing, sandboxing, the isolation technology, monitoring technology,
guard railing、sandboxing、isolation technology、monitoring technology，
81:00
telemetry technology, external AI monitor technology, all of that stuff is AI technology,
telemetry technology、external AI monitor technology，所有这些东西都是 AI 技术，
81:06
accelerate the living daylights out of that. It's funny because I think that if the most alarm
要把那些也拼了命地加速。挺有意思的，因为我觉得，如果那些最警觉
81:12
people, the labs, could be assured they were going to move 80% of their compute into safety and
的人，那些 labs，能确信自己会把 80% 的 compute 投入到 safety 和
81:18
alignment as opposed to 80% to capability expansion, they would feel much better. And it sounds
alignment，而不是把 80% 投到 capability expansion，他们会感觉好得多。而且这听起来
81:23
to me that one thing, and it sounds to me one thing you're actually saying is one should think of
对我来说，有一件事，而且我听起来，你实际上在说的一点是，大家应该把 safety and alignment 看成 capability expansion。
81:27
safety and alignment as capability expansion. Sure. An unsafe technology is not an advancing
当然。
81:32
technology. It's like saying, oh, chip design is chips, is R&D, chip verification is not R&D.
一项不安全的技术就不是一项在进步的技术。
81:40
We spend most of our cost, most of our compute on verification, emulation, verification,
这就像说，哦，chip design 是芯片，是 R&D，而 chip verification 就不是 R&D。
81:46
testing, reliability, testing, lifetime testing, all of that is part of engineering.
我们把大部分成本、大部分 compute 都花在 verification、emulation、verification、testing、reliability、testing、lifetime testing 上，所有这些都属于 engineering 的一部分。
81:52
The incentives are there. The incentives are there. They are going to put their company in
激励是有的。激励是有的。
81:57
harm's way if they release products that harms other companies and other people. Do you think we
如果它们发布会伤害其他公司和其他人的产品，它们就会把自己的公司置于危险之中。
82:02
need liability laws that are specific to AI? So let's just use one example, self-driving car.
你觉得我们需要专门针对 AI 的 liability laws 吗？
82:12
The car as a product, the robot taxi has lots of regulations. If it doesn't have enough regulations,
汽车作为一种产品，robot taxi 有很多监管。如果它没有足够的监管，
82:21
then it's how to get involved and come up with new regulations.
那么问题就是怎么参与进来，并制定出新的监管。
82:28
Card to car industry should have new regulations. I don't know what's missing, but if there is
Card to car industry 应该有新的监管。我不知道缺了什么，但如果有
82:33
something missing, then I would absolutely add more regulation. In the context of internet,
什么缺失，那我绝对会加更多监管。在 internet 的语境下，
82:41
there are many applications that internet powers and those applications should have regulation.
有很多应用是 internet 赋能的，那些应用也应该有监管。
82:46
If they don't, you've got to find them. So I want to drop down the next side of the cake now to chips.
如果没有，你就得把它们找出来。所以我现在想从蛋糕的下一面往下讲到 chips。
83:04
And to summarize where we are, because I want to make sure I do understand your position correctly,
然后总结一下我们现在到哪儿了，因为我想确保我确实正确理解了你的立场：
83:07
it's that these companies are going through a transition that even as these systems speed up,
那就是，这些公司正在经历一场转型，即便这些系统正在加速，
83:13
become more capable, complex, persistent, whatever it might be, that there is still the limiting factor
变得更有能力、更复杂、更持久，不管它是什么，仍然存在那个限制因素
83:21
of companies will not ship what is not safe. They should not ship what is not safe.
公司不会 ship 不安全的东西。它们不应该 ship 不安全的东西。
83:27
And you believe they have the engineering capabilities
而且你相信他们有工程能力
83:32
to make these things safe, to figure out the testing and the control,
来把这些东西做安全，弄清楚测试和控制，
83:35
absent external intervention. That's sort of where you want.
在没有外部干预的情况下。这大概就是你想要的那种状态。
83:41
One thing I've heard you say is that we have entered, maybe the way people don't always
我听你说过一件事，那就是我们已经进入了一个，也许人们并不总是
83:45
understand, a new era of how computing works. Describe your vision of that. And the way, if somebody
理解的、computing 如何运作的新时代。描述一下你对此的愿景。而且，如果某个人
83:52
sort of understanding of it is a little bit still maybe in, you know, you've got a MacBook
对它的理解可能还稍微有点停留在，你知道，你有一台 MacBook
83:57
and it's got a processor in it and you buy it and how it differs.
而且它里面有个 processor，你买它，以及它有什么不同。
84:02
The last computer industry, and the computer industry we've known for 60 years,
过去的计算机产业，也就是我们已经熟悉了 60 年的计算机产业，
84:06
is called retrieval based computing. You retrieve files. That's why it's called data center,
叫做 retrieval based computing。你 retrieve 文件。所以它才叫 data center，
84:12
you know, file center. Okay. And in the future, it's an AI factory. It's generating.
你知道，就是 file center。好吧。然后未来呢，它就是一个 AI factory。它在生成。
84:19
And so the amount of computation necessary to understand the context, be grounded in information,
所以，理解 context、grounded 在 information 上所需的 computation 量，
84:26
the reason about what to do, and to generate an answer, that generative process requires a lot of
去推理该做什么，以及生成一个答案，这个 generative process 需要大量
84:32
computation. And so the amount of computation necessary per user has grown tremendously.
computation。所以每个用户所需的 computation 量已经猛增了。
84:40
And then the second part is because these generative AI can also be somewhat autonomous
然后第二部分是，因为这些 generative AI 也可以有点 autonomous
84:47
because they're agentic. Now you have agents using generative AI. And so rather than
因为它们是 agentic 的。现在你有 agents 在使用 generative AI。所以与其说
84:54
a billion people using computers, you essentially have multiple hundreds of billions of agents
十亿人在使用电脑，不如说你实际上会有几千亿个 agents
84:59
in addition to the humans using the computer. And so you could argue that the amount of computation
除了使用电脑的人类之外。所以你可能会说，我们需要的 computation 量，
85:05
we need, you know, however much we had before, is going to go up by a billion times. And that's a
你知道，不管我们以前有多少，都会增长十亿倍。而且这是一个
85:11
reasonable, you know, framework for a reasonable level of amount of computation. In this new world,
合理的，你知道，对应合理 computation 量水平的框架。在这个新世界里，
85:18
what you really care about in the context of a factory is how productive is it?
在工厂的语境下，你真正关心的是它有多高的生产力？
85:24
Not how expensive is it? It can't be infinitely expensive. But you want to know how productive it is.
不是它有多贵？它不可能无限贵。但你想知道它的生产力有多高。
85:29
And so our computers are incredibly productive. 50 billion dollars to build a one gigawatt data center,
所以我们的电脑生产力高得惊人。花 500 亿美元建一个 one gigawatt 的 data center，
85:37
one gigawatt AI factory. And you can rent it for 40 to 50 billion dollars per year. And so the
1 gigawatt 的 AI factory。而且你可以每年花 400 到 500 亿美元租用它。所以
85:43
the productivity of it is incredible. So number one is the productivity. And video architecture is
它的生产力简直不可思议。所以第一点就是生产力。而且 video architecture 是
85:49
fungible because we're general, we're general purpose, which is the reason why every AI lab,
fungible 的，因为我们是通用的，我们是 general purpose，这就是为什么每个 AI lab、
85:54
every AI model, closed model runs on Nvidia. And because we're completely fungible and you can
每个 AI model、closed model 都跑在 Nvidia 上。而且因为我们完全 fungible，你可以
86:00
use us from data processing, to pre-training, to post-training, to eVAL, to inference,
把我们用在从 data processing、到 pre-training、到 post-training、到 eVAL、到 inference，
86:07
the entire life of AI is supportable by our architecture. And if a customer no longer needs it,
AI 的整个生命周期都能由我们的 architecture 支持。而且如果客户不再需要它，
86:15
another customer will be more than happy to pick it up. And then the last part is that durability.
另一个客户会非常乐意接手。然后最后一点是 durability。
86:22
Because our architecture is software driven and we're constantly improving our software
因为我们的 architecture 是由 software 驱动的，而且我们在不断改进我们的 software
86:26
with new algorithms that takes the new workloads, the new models, and run it on our old generation
用新的 algorithms，把新的 workloads、新的 models 拿来，跑在我们老一代的
86:32
hardware. We have massive teams of people who are constantly doing that. As a result, the useful life
hardware 上。我们有庞大的团队，一直在做这件事。结果就是，我们 compute 的
86:39
of our compute is much longer. That's the reason why Nvidia's and people are talking about
使用寿命要长得多。这就是为什么 Nvidia 的，还有人们都在谈论
86:45
Nvidia compute as an asset class. Kind of like an airplane. It's, you know, airplanes are our
把 Nvidia compute 当作一种 asset class。有点像飞机。它是，你知道，飞机是我们的
86:51
general purpose. They're fungible. United Airlines isn't used. American Airlines, we use it.
general purpose。它们是 fungible。United Airlines 不是被用的。American Airlines，我们用它。
86:58
It's durable. It starts out as a passenger plane. It ends up, ends its life as a as a stripping,
它很耐用。它一开始是客机。它最后，它结束寿命时，会作为一架，作为一架拆解过的，
87:04
you know, as a cargo plane. And so as a result, it can be an asset class. So this is, and if we could
你知道，作为货机。所以结果就是，它可以成为一种 asset class。所以这就是，如果我们
87:12
do this, if this happens, then of course, the cost of capital for funding Nvidia AI factories
做这件事，如果这发生了，那么当然，为 Nvidia AI factories 提供资金的 cost of capital
87:20
will be the lowest because our computers are collateralized asset. And so anyways, this is the
会是最低的，因为我们的电脑是 collateralized asset。所以不管怎样，这就是
87:31
phase shift that's happening to us, which is going to be a huge unlock for our growth.
发生在我们身上的 phase shift，这会成为我们增长的一次巨大解锁。
87:35
And so your business has become so interesting. You've moved now into lowering the cost of
所以你的业务变得特别有意思。你现在已经转向去降低
87:41
capital for others in the AI industry. People may be have seen these charts of like
AI 行业里其他人的资本成本。人们可能看过这些图表，就像
87:45
the arrow is going in every direction. It's so interesting. Yeah. And
箭头往每个方向指。太有意思了。是啊。而且
87:49
explain that a bit to people who are, they understand Nvidia's become like the biggest company in
跟那些人稍微解释一下，他们明白 Nvidia 已经成了像是最大的公司，在
87:54
the world. They see these charts that seem very circular to them. What is the difference between
世界上。他们看到这些图表，觉得它们特别循环。那区别是什么，在
88:01
supporting demand, creating markets and creating demand?
支撑需求、创造市场和创造需求之间？
88:08
We can't really create demand because in the end, if the AI services have no off-take,
我们其实没法创造需求，因为说到底，如果 AI 服务没有 off-take，
88:21
then obviously building computers for it is pointless. And so the first thing has happened,
那显然为它造电脑就毫无意义。所以第一件事已经发生了，
88:27
the reason why compute demand is so high right now is because AI applications are going through
现在 compute 需求之所以这么高，是因为 AI 应用正在经历
88:32
an inflection, they're becoming useful. And because AI is becoming useful, 500 billion dollars
一个拐点，它们开始变得有用了。而因为 AI 正变得有用，5000 亿美元
88:41
of venture funding are coming in. And all of those companies, those thousands of companies
的风险投资资金正涌进来。而所有那些公司，那成千上万家
88:47
start-ups, they all need compute. And so the demand is coming from them. And so these companies
创业公司，它们都需要 compute。所以需求正来自它们。所以这些公司
88:54
need support and technology. They need support and ecosystem building. They need support and
需要支持和技术。它们需要支持和生态建设。它们需要支持和
88:59
financial support. And so we might decide to invest in some of them as an equity owner. And as a result,
资金支持。所以我们可能会决定以股权持有者的身份投资其中一些。而结果是，
89:08
they become a really flourishing new cloud provider. Another reason we might decide is because,
它们会成为一个真正蓬勃发展的新 cloud provider。我们可能决定这么做的另一个原因是，
89:14
as I mentioned, there's a five-layer cake. And at the model and the application layer, there's
正如我提到的，有一个五层蛋糕。而在 model 和 application layer，有
89:20
a whole bunch of really innovative companies. And there's way more to AI than just the language
一大堆极具创新性的公司。而且 AI 远不止 language
89:25
model itself. The world foundation model is a physical AI. There's biology AI. There's chemical
model 本身。world foundation model 是一种 physical AI。还有 biology AI。还有 chemical
89:32
material sciences AI. These are all different than language models. And so many of those companies
material sciences AI。这些都和 language models 不同。而且那些公司中有很多
89:40
are new and they need a lot of capital. We might decide to be a small percentage shareholder in them.
是新公司，而且需要大量资本。我们可能会决定持有它们一小部分股份。
89:47
So we get them off the ground. They're incredible scientists. I might even, by being a first
所以我们帮它们起步。他们是了不起的科学家。我甚至可能，通过成为第一位
89:54
investor or anchor investor, we bring confidence to their company. We give them access to a lot
投资者或 anchor investor，我们给它们的公司带来信心。我们让它们接触到很多
90:01
of our technology. We support them a great deal. And we help them become a company as fast as
而 Nvidia AI factories 用的是我们的技术。我们非常支持他们。我们帮助他们尽可能快地成为一家公司
90:06
possible. We might decide to invest in a new clear company. We might decide to write someone
可能吧。
90:11
so forth. So we look at my mental model of the AI industry as a five-layer cake and we're investing
我们可能会决定投资一家新的 clear 公司。
90:16
across all of it. There might be strategic unlock points. It opens new markets. It opens a new route
我们可能会决定给某人写支票，诸如此类。
90:23
to market for us. It might secure a critical resource for us. So there's a lot of strategic reasons
所以，我脑子里的 AI 产业模型就是一个五层蛋糕，而我们正在每一层上投资。
90:29
why we do it. I mean, the numbers here are astonishing. You've become like a single-company industrial
可能会有一些战略性的 unlock points。
90:34
policy for American AI. We've put a lot of money into this ecosystem. What's the total investment
这能打开新市场。
90:44
you're now making per year? All in. All in. We're probably up. Well, I don't know about every year.
这能为我们打开一条新的市场通路。
90:48
But I think all in, we might be like $100 billion. I might check my numbers, but it's something
但我觉得，全部算下来，我们可能大概有 1000 亿美元。我可能得再核对一下我的数字，但差不多就是这个数。这比 Chips and Science Act 还大。哦，对。对。更别说，因为我给 TSMC、Whistron、Foxcon、Amcore、Spill 以及所有这些不同公司提供的采购承诺，正因为那项承诺，我才能鼓励他们来美国这里制造。重要的是，在芯片制造这块，我们可能比世界上几乎所有公司都更能推动美国的再工业化。我们不仅是在让制造业再工业化。我们做得太快了，以至于正在造成劳动力短缺，但同时也在创造大量就业机会。
90:54
like that. It's larger than the Chips and Science Act. Oh, yeah. Yeah. Not to mention, because of
这能为我们锁定一项关键资源。
91:02
the purchasing commitments that I provide to TSMC and Whistron and Foxcon,
我给 TSMC、Whistron 和 Foxcon 提供的采购承诺，
91:08
Amcore and Spill and all these different companies, because of that commitment, I'm able to
还有 Amcore、Spill 以及所有这些不同的公司，因为有了这个承诺，我才能
91:13
encourage them to come and manufacture here in the United States. The fact that it matters,
鼓励他们来到美国这里制造。重要的是，
91:17
we probably contribute more to re-industrializing the United States in this chip manufacturing than
在 chip manufacturing 这件事上，我们可能比
91:23
just about any company in the world. We're not only re-industrializing manufacturing. We're doing
世界上几乎任何一家公司都更能推动美国再工业化。我们不只是让制造业再工业化。我们
91:28
it so fast that we're creating a shortage of labor, but we're creating a lot of jobs.
做得太快了，以至于造成了劳动力短缺，但同时也创造了大量工作岗位。
91:33
And a lot of people with money in the market right now who are excited, often excited by
现在市场里有很多手里有钱的人都很兴奋，尤其常常是对 Nvidia 的股票特别兴奋，同时又会担心这跟 90 年代末的 internet bubble 的类比。
91:37
Nvidia stock in particular, and worry about the analogy of the internet bubble of the late 90s.
而我觉得，他们担心的东西其实跟你刚说的有关，也就是 internet 确实一直在变得更有用。
91:43
And what they worry about is actually related, I think, to what you just said, which is that
不是说高估值就意味着这项技术是空心的、是假的，而是确实出了点事，一下子就爆了，然后很多非常大的公司在那里面被狠狠打击了。很多人也在那里面被狠狠打击了。
91:48
the internet did continue to be more useful. It's not that high valuations meant that the technology
从那种泡沫破灭周期里到底能学到什么？
91:55
was hollow or fake, but something happened and popped for a minute, and very big companies got
我猜问题就是，你不觉得它会再发生一次吗？还是说，你为什么不觉得它会再发生一次？
92:01
hammered in that. A lot of people got hammered in that. What is it a learn from that kind of bubble
到了某个时候，供需又会再次反转。而这只是
92:08
bust cycle? And I guess the question is, do you not think it will happen again, or why do you not
bust cycle？我想问题是，你不觉得它会再次发生吗，还是说你为什么不觉得
92:15
think it will happen again? At some point, demand and supply will be inverted again. And that's just
它会再次发生？到了某个时候，demand and supply 又会反过来。而这只是
92:24
the nature of markets. It's not going to happen next year. It's not going to happen in next
市场的本质。这不会在明年发生。也不会在未来
92:28
couple two, three years. I just don't believe that. But at some point, we will likely have more supply
两三年内发生。我就是不相信。但到了某个时候，我们很可能会供给
92:35
than demand. And I just don't know when that is. And so there's not much to learn from the past.
大于需求。我只是不知道那是什么时候。所以从过去也没太多可学的。
92:43
Well, it would be the signal for you.
嗯，那对你来说就会是个信号。
92:46
Markets will naturally slow down, and then it will stop. Meaning there will be a period of
市场会自然放缓，然后就会停。也就是说，会有一段
92:52
digestion. Now, is that period of digestion going to be six months? Is it going to be nine months?
消化期。那么，这段消化期会是六个月吗？会是九个月吗？
92:57
It's going to be a year. It won't be forever. If you look across the board, the amount of
会是一年。不会永远这样。如果你全面去看，我们投入到
93:03
investments that we're putting into the application layer so that each one of the industries
application layer 的投资量，是为了让每个行业
93:07
could have the technology to fuse into them so that they could benefit from it, that's probably
可能得有技术能融合进去，让它们从中受益，这大概是我们做的最重要的事情之一。我经常听到有人这样比较中国和美国的AI生态系统，就是说在美国，重点在于能力提升的速度。很多人认为我们在这方面领先，这似乎是真的。而在中国，更注重diffusion。很多人认为中国在diffusion方面可能领先。而且在某些方面，中国的经济结构更适合diffusion，从WeChat一直到知识和指令在其中传播的方式都是如此。以及这场竞赛到底是关于能力还是diffusion，还有这到底是不是一场竞赛，我们可以稍后再聊这个，
93:12
one of the biggest things that we do. This is a way I often hear the sort of Chinese and American
我们做的最重要的事情之一。我经常听到人们这样对比中美 AI 生态系统：在美国，重点是 capability 提升的速度。很多人觉得我们在这方面领先，这似乎也是真的。而在中国，人们更强调 diffusion。很多人觉得中国在 diffusion 上可能更领先。而且从某些方面看，中国的经济结构更适合 diffusion，从 WeChat 这样的东西，一直到知识和指令在其中流动以推动 diffusion 的方式。至于这场竞赛到底是在比 capabilities 还是 diffusion，还有它到底是不是一场竞赛，不过这个我们可以待会儿再聊，
93:17
AI ecosystems compared, which is that in America, the emphasis is on the speed of rising capability.
对比 AI 生态，情况是：在美国，重点在于能力提升的速度。
93:28
And a lot of people think we're ahead on that, and that seems true. And that in China, there's more
而且很多人觉得我们在这方面领先，这似乎是真的。
93:33
emphasis on diffusion. And a lot of people think that China is probably ahead on diffusion.
而在中国，更多的是强调 diffusion。
93:39
And in some ways, has an economy that is better structured from things like WeChat all the way to
很多人觉得中国在 diffusion 上可能领先。
93:44
just like the way knowledge and commands move through it for diffusion. And whether the race is
而且在某些方面，中国的经济结构更适合 diffusion，从 WeChat 这样的东西，到知识和指令在其中的流动方式，都是如此。
93:51
about capabilities or diffusion, and also whether it's race at all, but we can get to that in a minute,
还有这场竞赛到底是关乎能力还是 diffusion，以及它到底算不算一场竞赛，不过我们可以待会儿再聊这个。
93:56
it is a big question. I'm curious how you see that.
这是个很大的问题。我很好奇你怎么看。
94:01
That's the ultimate question. I believe if we want America to benefit from artificial intelligence,
那是终极问题。我相信，如果我们想让 America 从 artificial intelligence 中受益，
94:08
every single industry has to benefit. Walmart has to benefit. Save, we have to benefit.
每一个行业都得受益。Walmart 得受益。Save，我们也得受益。
94:13
Federal Express has to benefit. Every bank has to benefit. Every healthcare company, every drug
Federal Express 得受益。每一家银行都得受益。每一家医疗保健公司，
94:18
discovery company. We need to see every construction company, every data center company, power
每一家 drug discovery 公司都得受益。我们需要看到每一家建筑公司，每一家 data center 公司，
94:23
generation company. We need everybody in United States. We need everybody in America. We need
每一家发电公司。我们需要 United States 的每一个人。我们需要 America 的每一个人。我们需要
94:28
everybody in the world to benefit from this. And that's the highest layer. That's the most important
世界上的每一个人都从中受益。而这是最高的那一层。
94:34
layer. That's the layer that touches society. All the layers underneath are technology enablers.
这是最重要的那一层。这是触及社会的那一层。下面所有的层都是 technology enablers。
94:39
I want to see us not ruin the opportunity for the United States to benefit at the highest level.
我希望看到我们不要毁掉 United States 在最高层面受益的机会。
94:46
And notice all of the rhetoric and all the alarmism, all the doom-marism, all of the predictions
而且注意，所有那些说辞、所有那些危言耸听、所有那些末日论、所有那些预测
94:53
are scaring people. That is my greatest fear, actually. I have every confidence.
都在吓唬人。其实，那才是我最大的恐惧。我完全有信心。
95:00
Maybe I have more confidence in them than they have in themselves. You definitely have more confidence
也许我对他们的信心比他们对自己的信心还足。你对他们肯定
95:05
in them than they have in themselves. Well, I don't know about that. But maybe it's just too much
比他们对自己更有信心。嗯，这我倒不确定。但也许只是太
95:11
humility and otherwise, should we conceptualize what we're in as a race with China?
谦虚了，要不然，我们该不该把我们身处的局面理解成一场与 China 的竞赛？
95:20
I don't think it's necessary. Some people like to think that way. I don't find that necessarily
我不觉得有这个必要。有些人喜欢那样想。我不觉得那一定
95:28
inspires me. I have no trouble never mentioning another company in our, when we talk about us
会激励我。在我们谈论自己的时候，我完全可以从不提另一家公司
95:35
doing our good work. And so we hold ourselves to our own standard. And so I think that the different
做好我们的工作。所以我们用我们自己的标准要求自己。所以我觉得，不同的人
95:42
people have different ways of being motivated. And I think it takes a bit more artistry
有不同的被激励方式。而且我觉得这需要多一点艺术性，
95:50
to unite and focus organizations to a certain level of performance in the outside of contests.
在竞赛之外，把组织团结起来，并聚焦到一定的绩效水平。
96:01
But I don't necessarily see it as necessary, number one, number two.
但我不一定认为这是必要的，第一，第二。
96:07
The question is, even if we did frame it as a competition, it doesn't have to be that
问题是，即使我们把它框定成一场竞赛，也不一定非得是，
96:16
if they achieve something, it's at our peril. And so when they invent something or they create some
如果他们取得了什么成就，就会对我们造成危害。所以当他们发明了什么，或者他们创造了一些
96:24
power generation technology, it might be a great invention that we wish we had done ourselves.
发电技术，那可能是一项伟大的发明，我们会希望那是我们自己做的。
96:31
But because it's going to support all of our energy production systems here as a result,
但因为它最终会支撑我们这里所有的能源生产系统，
96:36
it helps our whole industry. Maybe they came up with a great new open model.
它能帮到我们整个行业。也许他们搞出了一个很棒的新 open model。
96:42
And they have. And those open models are now being used by 80% of the Americans.
而且他们确实做到了。现在那些 open models 有 80% 的美国人在用。
96:46
Yeah, we use a lot of Chinese open models here. Okay, that's right. And so that's terrific.
是啊，我们这边也用很多中国的 open models。对，没错。所以这太棒了。
96:51
We downloaded it originated in China, a lot of the technology, of course, also originated from
我们把它下载下来，它起源于中国，当然，很多技术也起源于
96:57
United States. We downloaded, we make it our own, we fine tune it, we put it into our own
美国。我们下载下来，把它变成我们自己的，我们把它 fine-tune 一下，我们把它放进我们自己的
97:03
agent harness, we put it into our own sandbox. That's all your own technology. So I think the
agent harness，我们把它放进我们自己的 sandbox。那都是你们自己的技术。所以我觉得
97:09
the fact that you leverage their weights, I think that's terrific. That's fine.
你利用他们的 weights 这件事，我觉得这太棒了。这没问题。
97:13
You were saying a few minutes ago, the way different countries have begun to see compute as a
你几分钟前说，不同国家已经开始把 compute 看作一种
97:18
geostrategic resource. And you know, may want to allocate it to their own companies.
地缘战略资源。而且你知道，可能想把它分配给他们自己的公司。
97:24
There's been a lot of back and forth on that here. And among people who do CSS in a race with China,
这里对此一直有很多来回讨论。而在那些做 CSS、与中国竞赛的人当中，
97:28
pretty people see this as in a race with China for who will get to recursively improving self
很多人把这看作是在与中国竞赛，看谁先能实现 recursively improving self
97:34
and super intelligence first. There's been this ongoing back and forth on whether or not
和 super intelligence。关于到底要不要，一直有这种持续的反复讨论
97:40
one thing we want to do is deny them compute, which in this case tends to mean to denying them
我们想做的一件事就是拒绝给他们 compute，而在这件事上，这通常意味着拒绝给他们
97:44
your chips. Under the Biden administration, we had pretty tight export controls. Those were
你的 chips。在 Biden 政府时期，我们有相当严格的 export controls。这些
97:48
loosened under Donald Trump. Obviously, you wanted those to be loosened. How do you think about
在 Donald Trump 时期被放松了。显然，你希望这些被放松。你怎么看
97:55
the question of whether or not it is good for China to have Nvidia chips that could accelerate
这个问题：让中国拥有可能加速的 Nvidia chips 到底是不是好事
98:03
their models or model deployments, their model capabilities versus us holding that back to try to
他们的 model 或 model deployments、他们的 model capabilities，和我们把这些扣住、试图拖慢他们的进展相比？
98:10
slow their progress? In a case of AI, our goal is not just that one lab benefits. Our goal is that
在 AI 这件事上，我们的目标不只是让某一家实验室受益。
98:19
all of America benefits. I think United States has a greater responsibility and a greater ambition
我们的目标是让整个 America 受益。
98:26
for the world to be built on the American tech stack. Just as we have greater ambition that the
我认为 United States 有更大的责任，也有更大的雄心，要让世界建立在美国的 tech stack 上。
98:31
world is built on US dollar and that more people speak English and that they use the American
就像我们有更大的雄心，让世界建立在 US dollar 上，让更多人讲 English，让他们使用美国版本的 internet。
98:37
version of internet. I mean, we want that. The question is ultimately, what are we depriving?
我的意思是，我们想要这样。
98:46
Are we depriving them of a chip for their industry or are we depriving United States a market
归根结底，问题是我们到底在剥夺什么？
98:53
to compete in? If you see the market, if you see the market as big as China, how does that help
我们是在让他们行业拿不到所需的 chip，还是在让 United States 失去一个可以竞争的市场？
99:00
the United States technology sector? Maybe it helps one company with a particular model.
United States 的科技行业？也许它能帮到某一家公司，靠某个特定的 model。
99:07
But the rest of the industry suffers. I think that it doesn't help the chip industry,
但行业里的其他部分会受损。我觉得这对 chip 行业没帮助，
99:12
surely, to be deprived of the market, to go compete in. It doesn't help the rest of the industry
当然，被剥夺掉可以去竞争的市场。这也不利于行业里的其他部分
99:19
because of deprived open models. It doesn't support the overall aspiration of the United States
因为被剥夺了 open models。这也不支撑 United States 的整体愿景
99:27
to have the world built on the American tech stack. There's a lot of thing you deprive yourself
让世界建立在 American tech stack 之上。你会让自己失去很多东西
99:32
if you narrowly focus on deprived them of chips. I would say to take a step back and frame it
如果你只狭隘地专注于剥夺他们的 chips。我会说，退一步，换个框架来看
99:41
into what's in the best interest of America first, all of America, not one company.
什么才是符合 America 优先利益、整个 America 的利益，而不是一家公司。
99:49
With respect to the race, as we mentioned, the race is, if there is one, it's about all of the
关于这场竞赛，正如我们提到的，这场竞赛，如果说存在的话，它关乎所有的
99:55
economy in the United States succeeding. I feel myself very conflicted on the China chips question.
United States 的经济正在取得成功。我对 China chips 问题感到非常矛盾。
100:00
One reason is even where I have sometimes more of the superintelligence concerns in you do,
一个原因是，即使我有时比你更担心 superintelligence，
100:05
is if you have those concerns, I think you want to have a good relationship with China in which
就是如果你有这些担忧，我觉得你会想和 China 保持一种良好关系，在这种关系里
100:11
there can be productive bilateral, working through the risks and benefits of AI,
可以有富有成效的双边交流，一起处理 AI 的风险和收益，
100:18
and the more you think of it as a race, which only one side can win and act like that, the more
而你越把它看成一场只有一方能赢的竞赛，并照此行事，你就越
100:24
you're necessarily going to create enmity. I found that to be a complicated dimension of
必然会制造敌意。我发现这是
100:31
people's thinking here. I think that a zero-sum strategy, I deprive you of this, therefore I win.
这里人们思维中一个很复杂的层面。我觉得，一种 zero-sum strategy 就是：我剥夺你这个东西，所以我赢。
100:41
That simplistic logic tends to have unintended consequences of the bigger game.
这种过于简单的逻辑往往会给更大的博弈带来意想不到的后果。
100:48
The bigger game, of course, is that we're now all talking about safety. We want to build safe
当然，更大的局在于，我们现在都在谈 safety。
100:54
products. We want them to build safe products because when they don't build safe products, it
我们想打造安全的产品。
100:58
hurts the whole industry. This is a perfect time. We should want to look for opportunities to
我们也希望他们打造安全的产品，因为如果他们不打造安全的产品，会伤害整个行业。
101:05
communicate, collaborate, to understand a line as much as possible. Having said that,
现在正是绝佳时机。
101:14
Nvidia is an American company. We should benefit America first. America has every right.
我们应该寻找机会去沟通、合作，尽可能去理解一条线。
101:20
For these technologies to be made available to the Frontier Labs,
话虽如此，Nvidia 是一家美国公司。
101:25
Vera Ruben goes to the Frontier Labs first. That's your most advanced chip.
我们应该首先让美国受益。
101:30
That's right. Nvidia's newest chips, and so did Great Blackwall, and so did Hopper. Every single
美国完全有这个权利。
101:35
generation so did Ampure. Every single generation of our product goes to American companies first.
每一代，Ampure 也是如此。我们产品的每一代都会先提供给美国公司。
101:42
The U.S. government would like to add on top of that that is a requirement to do so. I'm
美国政府还想在此基础上加上一条：必须这么做。我
101:48
delighted by that. That's no problem. We do that naturally anyways. However,
对此很高兴。这没问题。反正我们本来就会自然这么做。不过，
101:54
recognizing that the AI industry is a five-layer cake, and we want every single layer to win,
考虑到 AI 产业是一个五层蛋糕，而我们希望每一层都能赢，
101:59
then we need every single layer to go out there and compete for the market.
那我们就需要每一层都出去争夺市场。
102:03
That drops us to the final layer of your cake, which we won't spend as much time on, but
这就把我们带到你那个蛋糕的最后一层，这一层我们不会花太多时间，但
102:08
if the advantage America's had, at least at a material level, is chips and software. One of
如果说美国一直拥有的优势，至少在物质层面上，是 chips 和 software。而
102:15
the advantages China has right now in AI is energy. It's easier for them to build new energy. They're
中国目前在 AI 方面拥有的优势之一就是能源。对他们来说，建设新能源更容易。他们正
102:21
pumping much cheaper energy into AI. They have tremendous advances on building electrical
向 AI 输送便宜得多的能源。他们在建设 electrical
102:27
generation and renewable energy. How do you see the most fundamental layer, the energy that pumps
generation 和 renewable energy。你如何看待最基础的层面，即那些输送
102:36
through the data centers, pumps through the chips, and where America is on generating enough of it?
通过 data centers，通过 chips 输送，以及 America 在产生足够能源方面做得如何？
102:45
Pretty the time when we've been trying to move from dirty energy into clean energy.
差不多是我们一直试图从 dirty energy 转向 clean energy 的时候。
102:50
Yeah, I think one, they just have a lot more energy than we do, and they plan to build a lot
是的，我认为第一，他们拥有的能源比我们多得多，而且他们计划建设很多
102:56
more than we do than we did. I think we just have to acknowledge we got ourselves really
比我们多，比我们过去多。我认为我们只需承认我们真的
103:03
gummed up in climate change and sustainable energy, and as a result, we just didn't plan
在气候变化和 sustainable energy 上 gummed up 了，结果我们就是没有计划
103:10
enough energy production. What do you mean by gummed up there? Well, in the near term, energy
足够的 energy production。你说的 gummed up 是什么意思？嗯，短期内，能源
103:16
production requires fossil fuel, and because there's just so much angst about fossil fuel energy
生产需要 fossil fuel，而且因为大家对 fossil fuel energy
103:26
production, if you look at our country, we've produced very little net new energy for a long time,
生产有太多焦虑，如果你看看我们国家，我们很长时间以来新增的 net new energy 非常少，
103:32
and all of a sudden this new industry comes along, and we find ourselves in a situation where we just
然后突然之间，这个新行业出现了，我们发现自己处在一个就是
103:37
don't have that much energy building capacity. Now, the whole country scrambling, and meanwhile,
没有那么多 energy building capacity 的处境。现在，整个国家都在手忙脚乱，而与此同时，
103:46
we've moved so fast, we could have done so much better job, communicating with the communities,
我们行动得太快了，本来可以做得更好，去和社区沟通，
103:54
preparing the communities, working with the communities to let them know what's coming,
让社区做好准备，和社区合作，让他们知道接下来会发生什么，
103:58
and if they don't want data centers to be built in their town or whatever it is, this will be it,
如果他们不想在自己的镇上建 data centers，或者不管是什么，那就这样吧，
104:04
but if you're going to build in their town, be sure to go there and let them know what's coming,
但如果你要在他们镇上建，一定要去那里，让他们知道接下来会发生什么，
104:10
work with them, work with them to help them understand that the use of water is really efficient
跟他们合作，跟他们合作，帮他们明白现在用水其实非常高效
104:17
these days. The AI supercomputers are super energy efficient, but they're still going to use a lot
现在这些 AI supercomputers 超级 energy efficient，但它们还是会用很多
104:23
of power. You've got to bring in your own power generation. It's going to lower their property taxes.
电。你得自己搞 power generation。这会降低他们的房产税。
104:29
There are a whole bunch of things that you can do. You can make your data centers more appealing.
有一大堆事情你可以做。你可以让你的 data centers 更有吸引力。
104:35
Make the setbacks further away. There are a lot of things that you can do. You can also contribute
把 setbacks 留得更远。有很多事情你可以做。你还可以出力
104:43
to be a good neighbor to the community and build better schools and better community centers and
来当社区的好邻居，建更好的学校、更好的社区中心，还要
104:48
improve their parks and improve the roads, and there's a lot of things you could do, but it's hard
改善他们的公园、改善道路，有很多事情你可以做，但很难
104:53
to do that after the fact, and now there's a fair amount of frustration around the country.
事后才做这些，而现在全国都有相当多的不满情绪。
105:03
Then of course, all of our narratives about the end of the world is not helping. What
然后当然，我们所有关于世界末日的叙事都没有帮助。
105:08
reasonable person says, come and build this data center in my town, and by the way, whatever you
哪个有理智的人会说：来我镇上建这个 data center，顺便说一句，你生产出来的任何东西都会终结我们所知的人类。
105:13
produce is going to end humanity as we know it. I think all of this negative doom or narrative
我觉得所有这些负面的末日论或叙事对我们国家都没有帮助。
105:21
is not helping our country. We started off on our back foot. We started off on our back foot.
我们一开始就落了下风。
105:28
And then now we started off on our back foot. We didn't have enough energy production in
我们一开始就落了下风。
105:33
the first place. There is a reality of time changes happening. Let me just give you the one last
然后现在，我们一开始就落了下风。
105:39
thing. There's no question that the energy demand is really great, which is the reason why the
我们本来就没有足够的 energy production。
105:46
market forces are helping us invest in sustainable energy like no time in history. You give me an
时代变化正在发生，这是现实。
105:52
example of a sustainable energy company, a material sciences company to build a better battery.
比如一家可持续能源公司，一家材料科学公司，来造更好的电池。
105:57
It could be solar, it could be nuclear, it could be vision, fusion, you name it, hydro, you name it.
可能是太阳能，可能是核能，可能是 vision、fusion，你能想到的都有，水电，你能想到的都有。
106:05
Those companies are all getting funded. The market demand for energy is so incredible
这些公司都在拿到融资。市场对能源的需求大得惊人，
106:11
that this is the best time in a hundred years to improve our power grid, to make our power grid
以至于这是百年来改善我们的 power grid 的最佳时机，让我们的 power grid
106:17
more sustainable, to lower the cost of energy, also investing in our sustainable future.
更可持续，降低能源成本，同时也在投资我们可持续的未来。
106:24
There's no question that in four or five years time we're going to use a lot more fossil fuel.
毫无疑问，再过四五年，我们会用多得多的化石燃料。
106:30
But also in the next decade in front of us, no time in history are we better prepared to move
但同样，在我们面前的下一个十年里，历史上从没有哪个时候，我们比现在准备得更充分去转向
106:38
to sustainable energy. And because these data centers, because the cost of these data
可持续能源。而且因为这些 data centers，因为这些 data 的成本
106:44
turns so high, now we're talking about putting them out than space. And so I think
回报这么高，现在我们都在讨论把它们送到太空里了。所以我觉得
106:51
the opportunity for us to see our dreams come true, move to a sustainable energy world,
我们有机会看到梦想成真，迈向一个可持续能源的世界，
106:59
we have a better chance of doing that than ever. The world is buying more because of AI factories,
我们比以往任何时候都更有机会做到这一点。全世界买得更多，是因为 AI factories，
107:04
because of AI is buying more sustainable energy today than anytime in history. Venture capitals
因为 AI 现在购买的可持续能源比历史上任何时候都多。风险投资
107:11
for a next generation energy is just incredible. Everybody, everything is getting funded.
投给下一代能源简直不可思议。所有人、所有东西都在拿到融资。
107:17
It's incredible. You don't need government subsidies for the first time in a hundred years
太不可思议了。一百年来第一次，你不需要政府补贴，
107:23
because the market forces are here. Everybody should be leaning in. If you want a future,
因为市场力量就在这儿。大家都应该全力投入。如果你想要一个未来，
107:28
if you want to turn to corner on climate change, if you want a future that's sustainable,
如果你想在气候变化上迎来转机，如果你想要一个可持续的未来，
107:35
lean into AI. It is the best opportunity we have to get there. But we need to build the energy
全力投入 AI。这是我们实现目标的最佳机会。但我们需要把能源建设得
107:42
faster to do that. That's right. That's right. It's kind of like you can subsidize it and you
更快，才能做到这一点。没错。没错。这有点像你可以补贴它，而且你
107:49
can make it easier to build. Yeah. It's kind of like in order to save you, they got to hurt you
可以让它更容易建设。对。这有点像为了救你，他们得先伤害你
107:53
first. In order to nature of surgery, they got to cut you open and save you. They got to inflict
就像手术的本质一样，他们得把你切开才能救你。他们得使你承受
108:01
an enormous amount of pain and suffering on you so that they could save you. And so I kind of
巨大的痛苦和折磨，这样他们才能救你。所以我有点
108:07
think AI is kind of like that. Over the next several years, we have to unfortunately use
觉得 AI 就有点像这样。在接下来的几年里，我们不幸地必须使用
108:12
renewable and use fossil fuel because we just don't have sustainable energy enough of it to
可再生能源和使用化石燃料，因为我们就是没有足够的可持续能源来
108:17
make a difference. And then after that, hopefully we can transition to that. I think that's
起到作用。然后在那之后，希望我们能过渡到那一步。我觉得那就是
108:22
where we'll end. Always a final question. What are three books you recommend to the audience?
这就是我们结尾的地方。永远都有一个最后的问题。你会给听众推荐哪三本书？
108:26
Well, I've read a lot of books. The book that made a huge impact on me was
嗯，我读过很多书。对我影响最大的一本书是
108:33
computer architecture from Hennessy and Patterson, a quantitative approach. It was the first computer
Hennessy 和 Patterson 的 computer architecture，a quantitative approach。它是第一本 computer
108:39
architecture book that reduced the complexity, the abstract idea of computer architecture
architecture 的书，把 computer architecture 的复杂性和抽象概念
108:47
down to engineering. And I love it when people take complicated concepts and reduce it into
降到工程层面。而且我特别喜欢有人把复杂概念简化成
108:55
something that you could do something about. Number two, I really loved innovators dilemma.
你能实际着手去做的东西。第二本，我真的很喜欢 innovators dilemma。
109:02
Clayton's past, but Clayton Christensen's book on how industries evolve over time and how to see
Clayton 已经过世了，但 Clayton Christensen 的这本书讲的是行业如何随时间演变，以及如何看到
109:12
emerging technology and how to set proper expectations about it and how to extrapolate
emerging technology，如何对它设定合理的预期，以及如何 extrapolate
109:17
maybe its future impact. I really loved Al Reese's and Jack Trout's book on positioning.
也许还有它未来的影响。我真的很喜欢 Al Reese 和 Jack Trout 那本讲 positioning 的书。
109:28
It's a really wonderful book about how people see, it's a book about marketing strategy.
它真的是一本很棒的书，讲的是人们如何看待事物，是一本关于 marketing strategy 的书。
109:35
More than that actually, it's a book about strategy. And how people see the world and how people
其实不止如此，它是一本关于 strategy 的书。还有人们如何看待世界，以及人们
109:42
see products and how you present products and how you see your own strategies. And I thought
如何看待产品，如何呈现产品，以及如何看待你自己的 strategies。然后我觉得
109:49
that was a really thoughtful book and a really easy to really easy to understand. Jensen Huang,
那本书真的很有思想，而且真的很容易，真的很容易理解。Jensen Huang，
109:55
thank you very much. Thank you very much Ezra. I always enjoy our time together and today
非常感谢。非常谢谢你，Ezra。我一直都很享受我们在一起的时光，而今天