Meta Shifts the Blame + Do Data Center Bans Work? + The Final HatGPT ·

00:00

Now, I thought you were going to bring up another story, which happened last night, which is

我还以为你要说另一个故事，就是昨晚发生的那件事。

00:05

so you and I last night are having drinks with, let's say, a very fancy technology person.

就你和我昨晚跟，怎么说呢，一位很厉害的技术圈人士喝酒。

00:11

Yeah.

对。

00:12

A legend, Ethan.

Ethan，一个传奇人物。

00:13

A legend.

传奇。

00:14

And we're sitting there at this very small table at the restaurant, very nice restaurant.

我们坐在那家餐厅里一张很小的桌子旁，餐厅非常高档。

00:19

And all of a sudden, you just do like a Mr. Bean, like spill an entire glass of wine

然后突然之间，你就跟Mr. Bean似的，把一整杯红酒全洒在自己身上了。

00:25

on yourself.

靠你自己。

00:27

And you revert, like, you are not a clumsy person, but you are like tall, which often

然后你又变回那样，你不是笨手笨脚的人，但你就是个子高，这常常

00:33

leads to you just like sort of flailing your limbs around.

导致你手脚乱舞。

00:36

So you are like trying to clean up the wine on yourself while also like making a joke

所以你就是在试着擦掉身上的酒，同时还在开玩笑

00:41

about how you just got so excited talking about the future of media that you like spilled

说你刚才因为聊媒体的未来太激动，以至于把酒

00:45

your wine on yourself.

洒在了自己身上。

00:46

Kevin, it was so horrible.

Kevin，那真的太惨了。

00:48

And you have to keep in mind, like Kevin and I don't even drink anymore, okay?

而且你得知道，我和Kevin现在居然连酒都不喝了，好吗？

00:53

But all of a sudden, there's some very good wine being poured in front of us and our guests

但突然之间，有人在我们和我们客人面前倒上了非常好的酒。

00:57

are drinking wine.

在喝葡萄酒。

00:58

And I just had the feeling of like, this is going to be one of those times where I have

然后我就有种感觉，就像，这将会是那种我喝一杯酒的时刻。

01:00

a glass of wine.

一杯葡萄酒。

01:01

Yeah.

对。

01:02

And how does my body repay me by almost immediately spilling half a glass of what I believe was

然后我的身体是怎么回报我的？几乎立刻就把半杯酒洒了，我觉得那还是

01:08

a very nice shardonnay all over my shirt?

一杯非常好的 shardonnay，全洒在我的衬衫上？

01:11

So I don't know how our guest received what I did, but I was very nervous.

所以我不知道我们的嘉宾怎么看我的举动，但我非常紧张。

01:17

I thought it was charming, because you're so composed, and it was just a moment where

我觉得那很迷人，因为你那么从容，而那一刻只是

01:23

you turned into Mr. Bean for 30 seconds, and I appreciated that.

你刚刚有30秒变成了 Mr. Bean，我还挺欣赏的。

01:26

I was flailing, bro.

我刚才在瞎扑腾，兄弟。

01:27

I'm Kevin Russo, tech columnist at The New York Times.

我是 Kevin Russo，The New York Times 的科技专栏作家。

01:35

I'm Casey Neumann from Platformer, and this is art board this week.

我是 Platformer 的 Casey Neumann，这是本周的 art board。

01:38

Metta agrees to a $17 billion settlement over child safety, why they caved, and what

Metta 同意就儿童安全问题达成170亿美元的和解，为什么他们让步了，以及这对 social media 意味着什么。

01:43

it means for social media.

接下来，Princeton University 的计算机科学教授 Arvinder Ryan 将重返节目，分享他关于为什么禁止 data centers 不会减缓 AI 进展的想法，最后，全是帽子。

01:45

Then, Princeton Computer Science Professor Arvinder Ryan in Returns to the Show to share his

然后，普林斯顿大学计算机科学教授Arvinder Ryan回归节目，分享他对于为什么禁止data centers不会减缓AI进展的看法，最后，这就是所有曾经达成过的一致。

01:50

thoughts on why banning data centers won't slow AI progress, and finally, it's all hats

这可能是科技史上最大的事件之一。

01:55

on deck for the final at GPT.

接下来是GPT的决赛。

02:04

Well, Casey, we got some big news just before we started recording today, which is that

嗯，Casey，我们今天开始录制前刚得到一个重大消息，那就是

02:09

Metta has settled its big multi-state lawsuit with all of the attorney's generals.

Metta已经和所有州的总检察长解决了那起大规模跨州诉讼。

02:17

Attorney's generals?

总检察长？

02:18

Attorney's generals.

总检察长。

02:19

I always mess that up.

我老是说错。

02:24

They have agreed to pay up to $17.1 billion and make a series of changes to their platforms.

他们同意支付高达171亿美元，并对其平台进行一系列改革。

02:33

This is a big deal.

这可是件大事。

02:34

It is a really big deal.

这真的很重大。

02:36

This is a case that has, that I have been following since it was filed in 2023.

这个案子我从2023年立案起就一直在关注。

02:42

Initially, you can go back and read the first thing I wrote about it, which is that I

最初，你可以回去读我写的第一篇相关文章，我当时觉得这案子立案时看起来并不那么有说服力。

02:45

didn't think that the case has filed looked all that compelling, but then Kevin, an unredacted

但后来，Kevin，一份未删减的诉讼版本出来了，我心想：哎呀，这家公司要麻烦了。

02:51

version of the suit, came out, and I thought, oh, this company is in trouble.

确实，快进到今天，这是 Metta 有史以来同意支付的最大和解金。

02:55

And indeed, fast forward to today, and this is the largest settlement that Metta has

而且很可能是科技史上最大的之一。

03:00

ever agreed to.

47个州，加上DC和美国领土。

03:01

And one of the biggest probably in the history of tech.

这两个案件是相关的，因为它们都涉及儿童安全。

03:03

Yeah, I think so.

对，我觉得是这样。

03:04

So let's talk about the actual suit because there have been so many of these kind of lawsuits

那我们来说说这个具体的诉讼吧，因为这类诉讼实在太多了，很难全都理清。

03:08

going around that it can be hard to keep all of them straight.

这不是三月份New Mexico的那个案子，那个案子里Metta因为多项儿童安全违规被勒令支付一大笔钱。

03:12

This is not the New Mexico case from back in March, where Metta was ordered to pay a bunch

这个有点像是那种大型的综合性多州诉讼和调查。

03:17

of money over various child safety violations.

47个州加上DC和US属地。

03:20

This is sort of the big omnibus multi-state lawsuit and investigation.

这些案件有关联，因为它们都涉及儿童安全。

03:25

47 states plus DC and US territories.

而且这里真正关键的有两件事，Kevin。

03:30

The cases are related in that they're both about child safety.

这些案件有关联，因为它们都涉及儿童安全。

03:34

And while that New Mexico case was really important, the journey there only ordered Metta

虽然新墨西哥州那个案子确实很关键，但整个过程到最后只判了 Metta 支付 3.75 亿美元的儿童安全违规罚款，虽然后来法官又追加了 5.67 亿美元。

03:38

to pay $375 million for child safety violations, although the judge later did tack on an additional

但这个案子的规模就是大了好几个数量级。

03:44

$567 million.

而且这里真正涉及两个主要问题，Kevin。

03:46

But this one is just orders of magnitude larger.

Kevin，这个案子就是直截了当地违反了一项法律，这项法律本应禁止像 Metta 这样的公司在没有父母允许的情况下收集 13 岁以下儿童的信息。

03:50

And there were really two major things at issue here, Kevin.

而且这里真正有两大问题，Kevin。

03:54

Kevin was just a straightforward violation of a law that is supposed to prohibit companies

Kevin 只是直接违反了一项法律，这项法律本应禁止像 Metta 这样的公司在未经父母允许的情况下收集13岁以下儿童的信息。

04:01

like Metta from collecting information about children younger than 13 without their parents

你是13岁。

04:06

permission.

但是在那份未删减的诉讼版本里，里面有大量、大量的证据表明 Metta 完全知道有数百万13岁以下的孩子在使用这个平台。

04:07

We talk all the time on the show about how we have very few privacy protections in this

我们在节目里经常聊到，在这个国家我们几乎没什么隐私保护。

04:10

country.

但这就是我们仅有的那点保护。

04:11

But this is what we do have.

你不能在没有父母同意的情况下收集一个12岁孩子的数据。

04:12

You cannot collect data about a 12 year old without getting their parents permission.

所以Metta长期以来做的就是，他们说，行，你必须年满13岁才能加入。

04:17

And so what Metta did for a long time was they just said, well, you can't join unless

但那份未删减的诉讼版本里，有大量大量的证据表明，Metta完全知道有数百万13岁以下的孩子在使用这个平台。

04:20

you're 13.

但在那发生之前，律师们算是插了进来，然后达成了协议。

04:21

But the thing that was in that unredacted version of the lawsuit was just tons and tons of evidence

此前，Metta 预计它会从与诈骗相关的广告中赚取大约100亿美元。

04:26

that Metta absolutely knew that millions of kids under 13 were using the platform.

Metta完全知道有数百万13岁以下的孩子在使用这个平台。

04:31

They were not asking for their parents permission.

他们没有征求父母的同意。

04:34

And so that's what I wrote in my column.

所以这就是我在专栏里写的。

04:35

I think the AGs have met a dead to rights here.

我觉得各州总检察长在这里已经是铁证如山了。

04:38

And I think it was a major reason that this case was settled.

而且我认为这是本案能和解的一个主要原因。

04:40

And this one was settled, like, I guess Adam Miserie, the head of Instagram had already

而这起案件的和解，好像，我猜Adam Miserie，就是Instagram的负责人，已经作过证了。

04:45

testified.

Mark Zuckerberg原本预计要出庭作证。

04:46

Mark Zuckerberg was expected to testify.

但就在那之前，律师们算是插手进去，达成了和解协议。

04:48

But before that happened, the lawyers kind of got in there and struck the deal.

但在那发生之前，律师们就介入并达成了协议。

04:52

That's right.

没错。

04:53

And that leads into the second set of issues that were under consideration here, which were

这就引出了这里在考虑的第二个问题，也就是——

04:58

just all of the things that Metta does to get you to pick up your phone and look at Facebook

就是 Metta 为了让你拿起手机看 Facebook

05:02

and Instagram as many times a day as it humanly can.

和 Instagram，一天尽可能多次，而做的所有事。

05:06

Right.

对。

05:07

The addictive features.

那些让人上瘾的功能。

05:08

Yeah.

是的。

05:09

So, you know, what do we mean by that?

所以，我们这么说是什么意思呢？

05:10

Well, the fact that they're sending push notifications at all hours of the day or night

嗯，他们没日没夜地发送 push notifications，

05:15

that they're using ranking algorithms to try to show you the absolute most enticing material.

用 ranking algorithms 试图给你展示最诱人的内容。

05:21

The fact that they're not enforcing any real screen time limit.

他们根本没有实施任何真正的 screen time limit。

05:24

So there was just a litany of things that the company was doing to try to get teens to

所以公司做了一连串的事情，试图让青少年

05:29

look at it as much as possible.

尽可能多地去刷它。

05:30

The AGs came in and they said, hey, this is actually just addiction that you're trying

总检察长们过来说，嘿，你们想在这里制造的其实是

05:34

to create here.

一种成瘾。

05:36

And Metta said, well, I guess we're not going to fight that one anymore.

然后 Metta 说，好吧，看来我们不再反驳这一点了。

05:39

Yeah.

05:40

Financial settlement here is significant.

是的。

05:44

Around $12 billion is what Metta will be required to pay initially.

Financial settlement here is significant.

05:48

And you know, for Metta, that's real money.

这里的财务和解金额相当可观。

05:51

It's not like all of their money.

Around $12 billion is what Metta will be required to pay initially.

05:52

They'll be, you know, they'll be fine.

Metta 初期需要支付大约 120 亿美元。

05:54

They can answer and amortize that over a multi-year period right away.

And you know, for Metta, that's real money.

05:58

Not only that.

而且你知道，对 Metta 来说，这是真金白银。

05:59

Not only that, Kevin, but there was reporting from Jeff Horowitz in Reuters several months

不仅如此，Kevin，而且几个月前路透社的Jeff Horowitz有报道说，Metta预计今年从诈骗相关广告里能赚大约100亿美元。

06:02

ago that Metta expected that it would make about $10 billion in ads related to scams

之前，Metta预期它会在与诈骗相关的广告上赚取大约100亿美元。

06:08

this year.

所以他们很不幸地不得不把所有的诈骗钱都拿出来付这笔和解金。

06:09

So they will unfortunately have to give up all of their scam money to pay for this settlement.

所以他们不幸地得交出所有骗来的钱来支付这笔和解金。

06:14

Yeah.

是啊。

06:15

Just move the scam budget over to the settlement budget and it basically nets out.

只要把骗钱的预算挪到和解预算里，基本上就抵消了。

06:19

Yeah.

就把诈骗预算挪到和解预算里，基本就抵消了。

06:20

But there's an additional wrinkle, which is that the settlement dollars will go up to roughly

但还有一个额外的变数，就是和解金额会涨到大约

06:25

$17 billion if TikTok and YouTube also settle and agree to penalties and changes to their

如果TikTok和YouTube也和解并同意接受处罚和修改产品，那就是170亿美元。

06:32

product.

所以这到底是怎么回事？

06:33

So what is going on here?

基本逻辑就是，嘿，别让我们在这场争夺青少年注意力的战争中单方面缴械投降。

06:34

The basic logic is like, hey, don't make us unilaterally disarm in the war for teenagers

对吧？

06:40

attention.

就是说，如果其他公司都不用负责，凭什么要我们负责呢？

06:41

Right?

所以我们要做的，Kevin，就是把美国的青少年当人质。

06:42

Like, why do we have to do the responsible thing if these other guys don't have to as well?

就像，如果那些家伙都不用负责，凭什么我们要做负责任的事？

06:46

So what we're going to do, Kevin, is we're going to hold America's teenagers hostage.

所以凯文，我们要做的事就是，把美国的青少年当人质。

06:51

And we're going to say we're not actually going to give them the safest experience possible

然后我们要说的是，我们实际上不会给他们提供最安全的体验，

06:55

unless these other guys do too.

除非其他那些人也这么做。

06:57

And they are of course presenting this as industry leadership and creating a new safety standard

而且他们当然把这标榜为行业领导力，还设立了新的安全标准

07:01

for teenagers.

针对青少年的。

07:02

Yes, I love the way they're spinning this.

是的，我喜欢他们这种包装方式。

07:03

Have you seen the full page ad?

你看到那则整版广告了吗？

07:05

No.

没有。

07:06

So meta is, according to my guys at my colleague at the times, meta is preparing to release

所以Meta，据我的线人，也就是我在《泰晤士报》的同事说，Meta正准备发布——

07:11

a full page ad, an open letter in the country's major newspapers this week calling for TikTok

这周在各大报纸上有一整版广告，是一封公开信，呼吁TikTok

07:20

and YouTube to quote, join us in supporting teens says basically, you know, we all platforms

和YouTube，原话是“加入我们支持青少年吧”，基本上就是说，你知道，我们所有平台

07:30

should empower parents and support teens in these ways because we know that when teens

都应该用这些方式赋能家长、支持青少年，因为我们知道，当青少年

07:34

are restricted on one app, they simply move to another.

在一个app上受到限制，他们就会直接转到另一个app。

07:37

So basically, they are worried that if they have to sort of unilaterally disarm and implement

所以基本上，他们担心的是，如果他们不得不单方面自废武功并实施

07:43

these new features, which we'll talk about in a second into their apps, teenagers will

这些新功能——我们待会儿会谈到——放到自己的app里，青少年就会

07:47

just go to TikTok and YouTube instead.

反而直接跑到TikTok和YouTube上去。

07:49

Yeah.

对。

07:50

And I mean, look, like is that narrowly true?

我是说，你看，这严格来说是真的吗？

07:53

Yes.

是的。

07:54

But please do not wait until truly the last possible second to do the bare minimum effectively

但拜托，不要真的等到最后一刻才去做最低限度的有效举措，

08:00

at gunpoint from 47 attorneys general and then say, we're excited to announce our new

在47位州总检察长的枪口下，然后说，我们很兴奋地宣布我们新的

08:06

leadership position in teen safety, like the depth of cynicism in this approach is breathtaking

在青少年安全方面的领导地位。这种做法的厚颜无耻程度让我瞠目结舌，

08:12

to me, even as a person who's covered this company for a really long time.

对我来说，哪怕我已经报道这家公司很久了，也是如此。

08:15

Yes.

是的。

08:16

So let's talk about the product changes that are being required as part of this settlement.

那我们谈谈这次和解协议要求的产品变更吧。

08:22

What is meta being forced to do or implement?

Meta 是被迫要做什么或者实施什么措施？

08:26

So it is a bunch of things and I will just hit some of the highlights.

所以有很多方面，我就挑几个重点说说。

08:30

A big one is that there's now a default two hour time limit that is cumulative across

一个重要变化是，现在默认有一个两小时的时限，而且是跨平台累计的。

08:35

both Facebook and Instagram.

也就是 Facebook 和 Instagram 都算在一起。

08:37

So your teen will be able to spend a mere two hours a day browsing the feed.

所以你的孩子每天最多只能刷两个小时的动态。

08:43

There will also be a default block on its apps between midnight and 6 a.m.

另外，在午夜到早上六点之间，Meta 的应用还会默认被屏蔽。

08:50

So your teen will not be able to like post to their feed or look at stories during that

所以你的孩子在这段时间里不能点赞、发动态或查看 Stories，

08:54

time when presumably they should be asleep.

也就是他们本该睡觉的时候。

08:57

And also for the first time, Kevin, they will now mute push notifications between 8 a.m.

而且这也是第一次，Kevin，他们现在会在上午8点到下午3点之间把 push notifications 设为静音，因为当然在过去十多年里，他们一直在学校不断打断孩子们，试图让他们看 Instagram。

09:04

and 3 p.m. because of course for the past decade plus they have been continuously interrupting

但他们还能这么做吗？我该怎么在高中化学课上和我那个 Nancy nasty Nancy AI chatbot 聊天？

09:09

children at school to try to get them to look at Instagram.

好消息是，Kevin，direct messages 不在此计划之内。

09:12

But can they do that anymore?

但他们还能那么干吗？

09:13

How am I supposed to chat with my Nancy nasty Nancy AI chatbot during high school chemistry

我高中化学课上还怎么跟我的 Nancy nasty Nancy AI chatbot 聊天啊？

09:18

class?

集体诉讼呢？

09:19

Here's the great news, Kevin.

好消息来了，Kevin。

09:20

Direct messages have been exempted from this plan.

私信不在此计划的范围内。

09:22

Oh, good.

哦，好。

09:23

And continue to get push notifications about all your Instagram DMs while you're trying

而且在你试图学微积分的时候，还会继续收到你所有 Instagram DMs 的 push notifications。

09:27

to learn calculus.

还有几个变化要提一下。

09:29

A few more changes to mention.

他们终于会默认隐藏 Instagram 帖子上的 like count 了。

09:32

They will finally hide the like count on Instagram post by default.

所以这里有一些研究似乎表明，这可能会对青少年有一些适度的心理健康益处。

09:38

So there has been research here that seems to show that this this may have some modest mental

他们还将禁用一些他们所谓的 extreme makeup filters。

09:44

health benefit to teenagers.

对青少年的健康益处。

09:46

They're also going to disable some what they call extreme makeup filters.

他们还会禁用一些他们所谓的极端美颜滤镜。

09:51

I would love to know how much makeup you have to be wearing to for it to be considered

我很想知道，你到底要化多浓的妆才能算得上“极端”。

09:55

extreme.

比如那个 juggalo 滤镜。

09:56

Like the juggalo filters.

Kevin，我们现在说的是哪种妆？就是一个青少年用了 juggalo 化妆滤镜那种。

09:57

Which make up are we talking here when a teenager puts on a juggalo makeup filter,

它给他们一种在现实生活中可能无法实现的美貌幻觉。

10:02

Kevin.

而这一点我看不出来。

10:03

It gives them a sort of illusions of beauty that they may not be able to realize in real

这会给他们一种在现实生活中可能无法实现的美貌幻觉。

10:09

life.

而我看不出这一点。

10:10

And I can't see that.

他们一直坚持说，我们不是社交媒体。

10:11

You can't unsee that.

你没法装作没看见。

10:12

So those are the big ones.

所以那些就是最重要的几个。

10:15

One other unbelievable change to mention, you know, I said before that the meta is holding

还有一个难以置信的变化要提，你知道，我之前说过，Meta正把

10:21

America's teenager hostage based on what TikTok and YouTube do here.

美国的青少年当人质，而这取决于TikTok和YouTube在这里怎么做。

10:25

So there's another provision, which is that if TikTok and YouTube will agree to meta's

所以还有另一项规定，就是如果TikTok和YouTube同意Meta的

10:33

terms that meta will reduce the daily limit for Facebook and Instagram to one hour and

条款，Meta就会把Facebook和Instagram的每日限制降到一小时，并且

10:39

to expand night mode hours to 10 p.m. to 7 a.m. which will be up from midnight to 6 a.m.

把night mode的时段从午夜到早上6点扩大到晚上10点到早上7点。

10:47

So meta is basically saying like, look, it would be a real shame if something happened

所以Meta基本上是在说，你看，如果真的发生什么事，那就太遗憾了。

10:51

to the children of this country.

致这个国家的孩子们。

10:53

But you have a way out simply agree to our terms and then we will expand these hours.

但你们有一条出路：只要同意我们的条件，我们就会扩大这些时长。

10:59

It's a really interesting bit of like game theory now.

现在这真的有点像博弈论，特别有意思。

11:02

Like if you're TikTok or YouTube and you're looking at this, I imagine that you are saying,

就比如如果你是TikTok或YouTube，看到这个，我猜你会说，

11:07

well, obviously we're not going to like join you in your like, you know, your court mandated

嗯，显然我们不会加入你们的那些，你知道的，法院强制或和解强制要求的改变，因为不像你们，我们对孩子更好。

11:12

or settlement mandated changes because unlike you, we are better for kids.

我尤其能想象YouTube会对此反驳。

11:16

I can imagine especially YouTube pushing back on that.

他们一直坚持说，我们不是社交媒体。

11:19

They have always insisted like, we are not social media.

他们一直坚持说，我们不是社交媒体。

11:22

We are more like TV.

我们更像电视。

11:23

We are like, we are not doing the kinds of like addictive and predatory things that some

我们就像，我们不像其他一些应用那样做那些让人上瘾和掠夺性的事情。YouTube有教育用途。

11:29

of these other apps are doing their educational uses of YouTube.

很多，你知道，很多学校在课堂上使用YouTube。

11:32

A lot of, you know, a lot of schools use YouTube in the classroom.

显然，这跟Instagram或Facebook是不同的东西。

11:35

Clearly, this is a different thing than like Instagram or Facebook.

我想TikTok对于自己为什么和这两个平台都不一样，也会有自己的说法。

11:39

I imagine that TikTok will have their own view on why they are different than either of

但就像，你觉得那些作为主要竞争对手的其他平台会在或应该在…方面做什么？

11:43

these platforms too.

这些平台也是。

11:44

But like, what do you think those other platforms met as main competitors will or should do in

但比如说，你觉得那些其他平台，也就是 meta 的主要竞争对手，会或应该怎么做，

11:49

response to this sort of weird hostage situation?

来应对这种奇怪的绑架局面？

11:52

I actually assume that they probably will adopt these standards because we now know that

我其实估计他们很可能会采纳这些标准，因为我们现在知道，

11:57

if they don't, they might have to go sign their own $17 billion settlement.

如果他们不这样做，他们可能就不得不去签自己那份 170 亿美元的和解协议。

12:02

So I suspect that meta believes that these things are actually coming and this will now

所以我怀疑 meta 认为这些标准真的会到来，而这件事现在会

12:06

be a talking point so that if and when TikTok and YouTube do commit to these things, meta

成为一个话题点，这样如果 TikTok 和 YouTube 真的承诺这些标准，meta

12:11

will go out and say, we led the industry and calling for new standards and we got our

就会站出来说，我们引领了行业，呼吁制定新标准，而且我们得到了我们的

12:15

peers to adopt these standards which we set.

让同行们采纳我们设定的这些标准。

12:17

So that is sort of the next wave of cynicism that I'm expecting here.

所以这就是我预计的下一波冷嘲热讽。

12:21

And we did it totally voluntarily with no court requiring us and no threat of, you know,

而且我们完全是自愿做的，没有法院要求，也没有那种，你懂的，

12:27

many billions of dollars in penalties if we didn't.

如果不做就会面临数十亿美元罚款的威胁。

12:30

You know, I was going back to read my first column about this when the case was originally

你知道，我正打算翻回我关于这件事的第一篇专栏，当时这个案子最初

12:33

filed and meta had a comment that said something to the effect of, you know, we're disappointed

立案的时候，Meta 发了一个评论，大意是，你知道，我们很失望，

12:36

that the state AGs didn't work with us on something more constructive and instead pursued

州检察长们没有和我们一起做更有建设性的事，而是推动了

12:41

this series of changes.

这一系列调整。

12:42

I'm looking at this.

我看着这些，

12:43

I'm like, this seems pretty constructive to me.

心想，这对我来说还挺有建设性的。

12:45

It seems like we got a lot further than we did just waiting for meta to adopt some new standards.

看起来我们比单纯等待 Meta 采纳一些新标准走得远多了。

12:49

I think that the product changes are a lot bigger deal than the financial piece of the settlement

我认为产品层面的变化比这里和解协议中的财务部分重要得多。

12:55

here.

真的，这些不是那种边缘上的表面调整。

12:56

It is really like that these are not sort of cosmetic tweaks around the edges.

这会要求……它会让 Instagram 或 Facebook 的青少年用户体验变得非常不同。

13:00

It will require like, it will make for a much different experience for a teenage user

这会要求，呃，它会给青少年用户带来非常不同的体验。

13:05

of Instagram or Facebook.

Instagram 或 Facebook 的。

13:07

It will.

会的。

13:08

But even, you know, even that said, and I do agree with you, Kevin, I think it's important

但是，你知道，即使如此，而且我确实同意你的看法，Kevin，我认为重要的是

13:12

to remember that more and more democracies around the world are just banning these apps

要记住，世界上越来越多的民主国家正在直接封禁这些 apps

13:16

for teenagers period.

对青少年，就是这样。

13:17

And so that is the real alternative that meta is staring down the barrel of here.

所以，这就是 meta 现在正眼睁睁面对的真正结局。

13:22

And it is willing to go, you know, pretty far by its limited standards to try to avoid

而且它愿意——你知道——以它有限的标准来看，已经愿意走得相当远，试图避免

13:27

the fate that it is now facing around the world.

它在世界各地如今正面临的这个命运。

13:29

Yeah.

对。

13:30

And so do you think this actually is going to change the sort of mental health picture

那你觉得这真的会改变心理健康方面的局面吗

13:35

for teens like if you had to guess are these the kinds of changes that actually do make

对青少年来说，如果你要猜的话，这些变化真的会带来

13:40

social media healthier for kids or is this just kind of a company doing the bare minimum

社交媒体对孩子更健康，还是这只是公司在做最低限度的事？

13:46

is it is required to do to at least stay out of bigger trouble?

这是为了避免更大的麻烦而必须做的吗？

13:50

Well, at the at the risk of overusing an analogy, Kevin, I do think that cigarettes are relevant

好吧，凯文，虽然我有点怕用太多类比，但我确实觉得香烟在这里很相关，

13:56

here in the sense that the way that we gradually got people to stop smoking was by just making

就是说我们让人们逐渐戒烟的方式，其实就是让抽烟这件事在很长一段时间里变得越来越难、体验越来越差，对吧？

14:03

it a little harder and a little worse experience over a long period of time, right?

香烟的价格一直在涨。

14:09

The price of those cigarettes kept going up.

能抽烟的地方越来越少。

14:12

The number of places that you could smoke kept going down and the amount of information

关于吸烟有害健康的信息，也无处不在地渗透进整个环境。

14:17

about the negative health benefits just kind of continue to permeate the environment.

关于负面的健康影响，就是会持续地渗透到环境中。

14:21

And eventually most people said, you know what, I don't think this is for me.

最终大多数人会说，你知道吗，我觉得这不适合我。

14:25

I think we're going to see a similar thing here where it's just going to get a little

我觉得我们会在这里看到类似的情况，就是会变得有点

14:28

bit harder all the time for kids to use social media.

更难，而且一直会这样，对于孩子们使用社交媒体来说。

14:31

They're going to continue to hear about all of the terrible outcomes that, you know, lots

他们会继续听到所有那些可怕的后果，你知道，很多

14:35

of teens are having on these apps and over time, they're just going to sort of move into

青少年在这些 App 上遭遇的，而且随着时间推移，他们就会有点转入

14:39

safer and different spaces.

更安全、更不同的空间。

14:40

Yeah.

是啊。

14:41

They'll start vaping instead or whatever.

他们会改抽电子烟或者什么的。

14:42

Whatever the social media.

不管什么社交媒体。

14:43

They'll be over.

它们就会过时。

14:44

I'm calcium turning their allowance money into less allowance money.

我看啊，就是把他们的零花钱变成更少的零花钱。

14:49

I realize that sounds very cynical, but I do feel like there is some parallel here with

我知道这听起来很愤世嫉俗，但我确实觉得这里跟

14:53

the cigarette industry because they had this massive multi-state settlement back in 1998.

烟草行业有相似之处，因为他们在1998年达成了一项巨额跨州和解协议。

15:00

And out of that came this sort of regime of like you have to put, you know, you have

而从中就产生了一种规则，就像你得，你懂的，你得

15:05

to, you can't advertise cigarettes in certain places.

要，你不能在某些地方做香烟广告。

15:08

And I think what really moved the needle on smoking was not any of those things.

我觉得真正让吸烟率下降的并不是这些。

15:12

It was like, as you said, it was sort of the cultural milieu and the like, the fact

而是，像你说的，有点像是文化氛围之类的，事实是

15:16

that there just weren't smoking sections and restaurants anymore.

就是餐厅里已经不再有吸烟区了。

15:20

And like the sort of vibe shift of smoking in culture.

还有吸烟在文化里的那种氛围转变。

15:24

And so I wonder if what you're saying is like this is, this is an important symbolic step

所以我在想，你说的是不是，这是一个重要的象征性步骤

15:29

and, you know, making the apps a little less addictive will mean that sort of teens have

而且，你知道，让这些 apps 不那么容易上瘾，就意味着那些青少年会……

15:33

some distance to say, like, wait a minute, do I like what this is doing to me and my friends?

稍微拉开点距离来说，比如，等一下，我喜欢它对我和我朋友造成的影响吗？

15:37

But that it might actually be some of these cultural factors that end up making more

但实际上，可能正是这些文化因素最终会产生更大的

15:41

of a difference when it comes to use of social media.

在社交媒体使用方面的差异。

15:43

That's what I think.

这就是我的想法。

15:44

And look, I mean, there are limits here, you know, I think that people are going to continue

而且你看，我的意思是，这也有局限，你知道，我认为人们会继续

15:48

to watch a lot of short form video, you know, I don't yet see an off ramp from that

看很多短视频，你知道，我还没看到从那个

15:54

particular phenomenon.

特定现象中退出的途径。

15:56

But some of the, you know, other mechanisms here, I do think will just make social media

但其中一些，你知道，其他的机制，我确实认为会让社交媒体

16:02

a little bit less fun to use and, you know, maybe that will actually result in teens

用起来没那么好玩了，而且，你懂的，可能这真的会让青少年少用一些。

16:06

using it less.

嗯。

16:07

Yeah.

说到Meta这家公司，他们现在正在达成和解，并且会做出这些改变。

16:08

Say about meta as a company that they are making the settlement now and that they are

我觉得他们意识到州检察长们在某些方面掌握了对他们非常不利的证据。

16:12

going to make these changes.

这个案子有意思的地方在于，它将由一个人来裁决，就是这位法官，Ivan Gonzalez Rogers。

16:13

I think they realized that the state AG's had them dead to rights on some of this stuff.

我觉得他们意识到，州总检察长们在一些事情上已经让他们无可抵赖了。

16:19

This was an interesting case in that it was going to be decided by one person, which

这个案例很有意思，因为它将由一个人来裁决，而这个人

16:22

is this judge, Ivan Gonzalez Rogers.

就是法官 Ivan Gonzalez Rogers。

16:26

And she has a reputation for being really tough.

而且她向来以非常强硬出名。

16:29

And I think they listened to the first few days of testimony.

而且我觉得他们听了前几天的证词。

16:32

They were looking at all of the lawsuits that they had already lost recently on similar subjects.

他们在看他们最近在类似问题上已经输掉的所有诉讼。

16:38

And they estimated that the fallout from this could be over a trillion dollars, like getting

而他们估计这事的后果可能超过一万亿美元，就像接近

16:42

close to their total market capitalization as a company.

他们整个公司的市值。

16:46

So this case truly was too dangerous for a meta to pursue all the way to the end.

所以这个案子对Meta来说确实太危险了，没法一路追到底。

16:51

Yeah.

是啊。

16:52

I mean, I'm also curious if you think this will have sort of knock on effects for them

我的意思是，我也很好奇你觉得这是否会对他们产生某种连锁反应。

16:57

as a company.

作为一家公司。

16:58

I'm thinking about like the Microsoft antitrust investigations of several decades ago.

我在想比如几十年前微软的反垄断调查。

17:04

And one of the sort of conventional narratives that came out of that was that all of the antitrust

当时有一种比较常见的说法是，所有这些反垄断调查、司法部、审判等等，都对微软产生了影响。

17:11

investigations and the justice department and the trial and all of this had an effect on

不是因为拆分令在上诉中被推翻，而是因为这件事太让人分心了。

17:16

Microsoft.

而且当时形势正在发生变化，却分散了太多高管的注意力。

17:17

Not because the breakup was reversed on appeal, but it was so distracting.

不是因为拆分在上诉中被推翻，而是因为它太让人分心了。

17:22

And it took away so many executives attention at a time when things were shifting under their

而且它占用了那么多高管的注意力，当时事情正在他们的

17:26

feet.

脚下发生转变。

17:27

And everything just became very cautious and there were lawyers in every meeting.

然后一切都变得非常谨慎，每个会议里都有律师。

17:31

Like they missed the platform shift to mobile, they missed this, they missed search because

比如他们错过了向移动端的平台转型，错过了这个，错过了搜索，因为

17:37

they were just so tied up in all of this expensive and distracting litigation.

他们就是被这些既费钱又分心的诉讼给缠住了。

17:40

So I'm curious if you think there's any parallel to meta here.

所以我很好奇，你觉得这里跟 Meta 有相似之处吗？

17:44

And just the amount of time and energy that this company is now having to spend defending

还有光是这家公司现在不得不花在辩护上的时间和精力，

17:49

itself in these lawsuits.

来应对这些诉讼。

17:50

It's a good question.

这是个好问题。

17:51

That is a story that the people at Metta know very well.

Metta 的人对这个故事再熟悉不过了。

17:55

And this has always been a very paranoid company.

而且这家公司一直都非常偏执。

17:57

They have always assumed and have lectured their employees about how likely it is that

他们一直假设，也一直告诫员工，他们有多大的可能会

18:01

they will get out competed in the marketplace.

在市场竞争中被淘汰出局。

18:04

And so truly they think about this constantly.

所以他们确实一直在想着这件事。

18:07

And I think you have seen them react by shifting all of the resources that they did to AI.

而且我觉得你们已经看到他们的反应了，就是把所有资源都转向了AI。

18:13

Look at how this company talks about itself.

看看这家公司是怎么谈论自己的。

18:15

They do not talk about a bright bold future for Facebook and Instagram and connecting

他们谈的不是Facebook和Instagram的光明未来，也不是连接

18:19

the world.

整个世界。

18:20

I mean, they'll make noises about creators here and there where it's useful.

我是说，他们会在有用的时候，时不时拿创作者说事。

18:24

And when you look at what this company says it's doing, it's building super intelligence.

而且你看看这家公司说它在做什么，它是在打造 superintelligence。

18:28

It's releasing a Mac app that like plugs into your calendar to do work for you.

它要发布一个 Mac app，接入你的日历，替你干活。

18:32

Like this company's public posture is that it is running away from its own products to

就像这家公司的公开姿态是，它正在逃离自己的现有产品，去

18:36

build something completely different.

构建完全不同的东西。

18:37

That tells you a lot about what it thinks about the stuff it's already built.

这很能说明它对自己已经做出的东西是什么看法。

18:40

Yeah.

是啊。

18:41

I wanted to bring up this other meta story that came out today, which I thought was fascinating

我想提一下今天出来的另一则 meta 故事，我觉得特别有意思。

18:44

on a very different topic.

换一个完全不同的话题。

18:46

Katie Paul, the great reporter over at Reuters, has a story about Metta's attempts to

路透社的杰出记者 Katie Paul 有一篇关于 Metta 试图

18:52

overhaul its workforce.

彻底改革其员工队伍的报道。

18:54

Basically, Mark Zuckerberg and his lieutenants have been plotting for months now to sort of

基本上，Mark Zuckerberg 和他的副手们几个月来一直在策划

18:59

tear up the org chart at Metta and replace these sort of like big overstaff teams with

拆掉 Metta 的组织架构，并把那些大而臃肿的团队换成

19:03

these small nimble AI native pods, which could involve slashing the size of many teams

这些小而灵活的 AI native pods，这可能会导致许多团队的规模

19:09

across the company by as much as 60%.

在公司范围内削减多达 60%。

19:13

She reports that also this did not happen in the way that they had wanted to basically

她还报道说，这件事基本上并没有按他们想要的方式发生。

19:18

implies that Mark Zuckerberg sort of got cold feet sort of midway through this slashing

这暗示着 Mark Zuckerberg 在削减成本和重组进行到一半时有点打退堂鼓，取消了未来进一步削减的计划。

19:21

burn, reorganization, and called off planning for future cuts.

但在我看来，这家公司似乎意识到自己只是为上一个时代而建，还没有真正跨入新时代。

19:27

But to me, it seems like this is a company that realizes that it is just, it is built for

我想知道你是否认为社交媒体成瘾和伤害诉讼也是其中的一个因素。

19:31

the last era and it is not quite, it has not made the jump to the new era.

很难说，因为当然，社交媒体业务对他们来说仍然极其赚钱。

19:37

And I wonder if you think the social media addiction and harm lawsuits are factor in there.

Facebook 和 Instagram 就是在为这家公司印钱。

19:42

It's hard to say because of course, the social media business is still so incredibly profitable

这很难说，因为当然，社交媒体业务仍然极其赚钱。

19:47

for them.

对他们来说。

19:48

Facebook and Instagram just print money for this company.

Facebook 和 Instagram 对这家公司来说就是印钞机。

19:51

They remain hugely popular despite everything that we have talked about.

尽管我们聊了这么多，他们依然非常受欢迎。

19:55

In fact, you could argue that these lawsuits are essentially targeting the fact that they

事实上，你可以说这些诉讼针对的正是他们太受欢迎了。

20:00

are too popular.

同时，我觉得这家公司，这些年我跟他们的不少高管聊过，他们对自己手下的很多员工其实很不舒服。

20:02

At the same time, I think this company, having talked to many of their executives over

他们觉得员工想法太多。

20:06

the years, they got really uncomfortable with a lot of the employees they had.

员工们太把自己当回事了。

20:10

They thought their employees had too many ideas.

他们想要的东西太多。

20:12

They were too entitled.

他们太自以为是了。

20:13

They wanted too many things.

他们想要的东西太多了。

20:14

And I think they have frankly relished getting rid of lots of them.

而且我觉得他们坦白说很享受裁掉其中很多人。

20:18

And so I expect that to continue.

所以我预计这还会继续。

20:19

Like here's a prediction, those 60% cuts that they got cold feet about this year, I bet

比如有个预测：今年他们临阵退缩没敢砍的那60%的削减，我打赌

20:24

they try again next year when the AI systems are better.

明年AI系统更好的时候，他们会再试一次。

20:27

Yeah.

对。

20:28

I find this totally fascinating.

我觉得这特别有意思。

20:29

I think it'll be very interesting to see whether teens do actually go to sleep earlier

我觉得很有意思的是，看看青少年是不是真的会早点睡

20:33

and pay more attention in their classes or whether they just switched to some other more

然后在课堂上更专注，还是他们只是转到了其他更...

20:38

horrible app.

这 app 太糟糕了。

20:39

Here's the thing.

问题是这样的。

20:40

You cannot actually solve the teen mental health crisis at the level of app design.

你实际上无法在 app 设计的层面解决青少年心理健康危机。

20:44

Like you need more people involved and you need different solutions.

比如你需要更多人参与，也需要不同的解决方案。

20:47

But harm reduction is a very effective strategy to have in your toolkit.

但 harm reduction 是你工具包里非常有效的策略。

20:52

And I do view the changes that meta is making today as harm reduction.

而且我确实认为 Meta 今天所做的这些改变就是 harm reduction。

20:56

I do too.

我也是。

20:57

I mean, I feel better raising a kid in a world with these restrictions on these platforms

我的意思是，在一个对这些平台有这些限制的世界里抚养孩子，我会感觉更好。

21:01

than I would without them.

比没有他们时更好。

21:03

And I think there's a lot of, you know, we give a lot of guff to sort of regulators and

而且我觉得有很多，你知道，我们这节目里经常吐槽那些监管者和立法者在科技方面的事。

21:09

law makers when it comes to technology on this show.

但我觉得这一次他们很专注，也很坚持，而且

21:11

But I think this is one instance in which they were focused and they were persistent and

但我觉得这就是一个他们专注而且坚持不懈的例子，并且

21:16

they got the goods on one of the most important companies in the world.

他们拿到了全球最重要公司之一的把柄。

21:20

And they got what they wanted in the form of these stronger protections for kids.

而且他们如愿以偿，得到了对孩子们更强的保护。

21:24

So I think this is a case of democracy working.

所以我觉得这就是民主在起作用。

21:27

This was a case of state attorneys general doing what Congress tried and failed to do.

这是州检察长们做到了国会想做却做不到的事。

21:33

Congress considered legislation that would have mandated some of these changes.

国会曾考虑通过立法来强制要求其中一些改变。

21:37

But the state attorneys general got this over the finish line and I am grateful that they

但各州的 attorneys general 最终把这件事办成了，我很感激他们。

21:41

did.

是啊。

21:42

Yeah.

而且另一个原因是，我就不用再磕磕绊绊地说 attorneys general 了。

21:43

And the other reasons, it means that I will not have to fumble my way through saying

想象一下，为了你的宪法权利而斗争，就为了在凌晨3点给一个13岁的孩子发 push notification。

21:46

attorneys general.

州检察长们。

21:48

Imagine fighting for your constitutional right to send a 13 year old to push notification

想象一下，为了在凌晨3点给13岁孩子发 push notification 而争取你的宪法权利。

21:53

at 3 a.m.

时下的热门话题，也就是 data centers。

21:54

Saying that there was a new real that they might be interested in that at some fundamental

说有一种新的 'real'，他们可能在某个根本层面上对它感兴趣。

21:57

level.

这就是这个案例的关键所在，等我们回来，会听到普林斯顿计算机科学家 Arvina Ryan 对 data centers 的一个惊人看法。

21:58

That is what this case has been about when we come back, a surprising take on data centers

好吧 Casey，我们不能让这节目一个礼拜都不聊当天的热门话题，也就是 data centers。

22:05

from Princeton computer scientist, Arvina Ryan.

是啊。

22:28

Well Casey, we couldn't let a week go by on this show without talking about the hot topic

我会说这是年度热门话题，因为好像从一月份开始，就几乎……

22:33

du jour, which is data centers.

当下最热门的话题，也就是data centers。

22:35

Yeah.

嗯。

22:36

I would say it's the hot topic due year because it seems like since January, it's almost

我会说这是今年的热门话题，因为看起来从一月份开始，几乎

22:42

all we hear about.

我们听到的全是这个。

22:43

Yes.

对。

22:44

And there's been so many, we've done so many segments and so many shows about the data

而且已经有很多很多了，我们做过那么多专题，那么多节目，都在讲这个 data

22:47

center backlash and what's driving it and what's, you know, where it's going and what

center 的 backlash，是什么在推动它，你知道的，它会往哪走，又

22:52

it means for the AI companies and for AI progress as a whole.

对 AI 公司和整个 AI 的进展意味着什么。

22:56

But today I thought we should drill into one specific post and one specific angle that

但今天我想深入探讨一篇具体的文章，一个具体的角度，

23:01

I thought was really interesting on data centers and these data center, moratorium and

我觉得关于 data center 以及这些 data center 的暂停和

23:05

bands.

禁令。

23:07

This came from Arvina Ryan in a former hard fork guest professor of computer science at

这是来自 Arvina Ryan 的观点，她曾是 Hard Fork 的嘉宾，普林斯顿大学计算机科学教授，也是被广泛认为是严肃批评 AI 炒作的人之一。她是《AI as normal technology》这篇文章的合著者。我们大概一年半前请她来聊过。她一直和我们一样在关注 data center 的反弹，并且得出了一个让人意外的结论：停止建设新的 data center 并不会显著减缓 AI progress。

23:11

Princeton and someone who is widely considered a serious critic of AI hype.

Princeton和某个被广泛认为是AI炒作的严肃批评者的人。

23:17

He's the co author of AI as normal technology, the essay.

他是那篇文章《AI as normal technology》的合著者。

23:21

We brought him on to talk about about a year and a half ago.

我们大约一年半前请他上节目聊过。

23:24

And he has been following this data center backlash as we have and has come to a surprising

他也和我们一样一直在关注这个data center引发的反弹，并且得出了一个令人惊讶的

23:29

conclusion.

结论。

23:30

And that conclusion is that stopping the construction of new data centers will not

而这个结论是，停止新建data centers将不会

23:36

meaningfully slow AI progress.

AI 进展明显放缓。

23:39

Yeah.

嗯。

23:40

And in particular, he gives some real numbers to this argument.

而且他特别为这个论点给出了一些具体数字。

23:44

He says that if a typical US state and acts a one year moratorium on new data center

他说，如果一个典型的美国州对新的 data center 建设实施为期一年的暂停令

23:49

construction, it would only slow AI efficiency progress by five to ten hours.

那只会让 AI efficiency 的进展减慢五到十个小时。

23:55

So basically, if your concern is all this stuff seems to be moving too fast, opposing data

所以基本上，如果你担心的是这一切似乎进展太快，反对在你所在的州或城市建设 data centers

24:00

centers in your state or your city is not an effective way to slow it down.

并不是减缓它的有效方式。

24:04

Yeah.

对。

24:05

And there are some questions about how much opposition to data centers is really about

而且有一些疑问，关于对 data centers 的反对到底有多少是真正关于

24:10

AI progress.

AI progress。

24:11

I think there are a lot of other factors will get into it.

我觉得还有很多其他因素会牵扯进来。

24:14

But we did think that Arvina's point was worth dwelling on a bit because I suspect that

但我们确实觉得Arvina这个观点值得多聊一下，因为我怀疑

24:18

as we move forward and more and more data centers get proposed, there are going to be

随着我们往前推进，越来越多的data center被提出来，会有

24:23

a number of people who see this as a meaningful way of getting some control and agency back

不少人把这看作一种在AI未来里重新拿回一些掌控感和主动权的方式。

24:28

in the AI future.

对。

24:29

Yeah.

而且我甚至听到一些整体上对AI持乐观态度的人也在这么说。

24:30

And I've even heard it expressed by people who are generally supportive of an optimistic

而且我甚至听到那些通常持乐观态度的人也这么说过。

24:36

about AI, but are worried about some of the risks of this moving too quickly, which

关于AI，但又担心这东西发展太快会有一些风险，这

24:40

is like, well, maybe the data center backlash is sort of based on these like, you know,

就像是，嗯，也许对data center的反弹有点基于那些，像是，你知道，

24:45

these sort of local fears about water use and electricity.

这种地方性的对用水和电力的担忧。

24:49

Maybe they're not quite fact-based when they're out there protesting data centers.

也许当他们出去抗议data centers的时候，这些担忧并不完全基于事实。

24:54

But if the net effect is slowing down AI progress, that might be a good thing.

但如果最终效果是拖慢AI进展，那可能是件好事。

24:58

And so I really was challenged by and provoked by Arvina's work on the subject because he's

所以我确实被Arvina在这个问题上的研究挑战和启发了，因为他

25:04

someone who's been very skeptical of the claims being made by some of the AI lab leaders.

一直对某些AI lab领导人提出的说法非常怀疑。

25:10

And yet here he is sort of saying, well, stopping data centers is not the way to stop AI progress.

然而他在这里却有点说，嗯，阻止data centers并不是阻止AI进展的方式。

25:14

All right.

好的。

25:15

Well, before we bring him in, Kevin, should we quickly do our AI disclosures?

嗯，在把他请进来之前，Kevin，我们要不要快速做一下 AI 披露？

25:18

Sure.

当然。

25:19

I work for the New York Times, which is doing open AI, Microsoft and Proplexity.

我在《纽约时报》工作，它正在和 OpenAI、Microsoft 和 Proplexity 打交道。

25:22

And my fiance works at Anthropic.

而且我未婚夫在 Anthropic 工作。

25:30

Arvina Ryan, and welcome back to HardFork.

Arvina Ryan，欢迎回到 HardFork。

25:32

Thank you, Kevin.

谢谢你，Kevin。

25:33

Thank you, Casey.

谢谢你，Casey。

25:34

Great to be here.

很高兴来到这里。

25:35

So it has been nearly a year and a half since we last had you on the show to talk about

距离我们上次请你上节目谈论……已经快一年半了。

25:40

your AI as normal technology paper, which argued that essentially there are all these

你的那篇《AI as normal technology》论文，它认为本质上存在所有这些

25:45

bottlenecks that are going to prevent AI from being universally adopted throughout society

bottlenecks，它们会阻止 AI 在整个社会中被普遍采用

25:50

and getting the rapid takeoff that some people out here in San Francisco believe is imminent.

并阻止其获得快速 takeoff，而旧金山这里的一些人认为这种 takeoff 即将来临。

25:56

And so it was a little surprising to me that we are having you back to talk about one

所以让我有点意外的是，我们又请你回来谈论一个……

26:01

sort of potential bottleneck that you don't think is going to be a big issue to continue

算是那种你觉得继续下去不会成为大问题的潜在 bottleneck 吧。

26:06

to AI progress, which is all of this backlash to an opposition to the construction of data

说到 AI 进展，这一切都是对反对 data 构建的反弹。

26:12

centers.

中心。

26:13

So tell us about your thesis on data centers and whether they will or won't slow down

所以请和我们说说你对 data centers 的论点，以及它们会不会减缓

26:20

AI progress.

AI 的进展。

26:21

Absolutely.

确实。

26:22

I think the backlash can be a real constraint to AI progress, but maybe not if it's directed

我认为反弹确实可能制约 AI 进展，但也许如果它是针对

26:28

to worry data centers.

data centers 的担忧，就不会。

26:30

It's a different kind of backlash if it's directed at, for instance, governments or other

这是一种不同的反弹，如果它针对的是，比如说，政府或其他

26:35

decision makers using AI for consequential decisions.

使用 AI 做出重大决策的决策者。

26:38

I think that would have much more of an impact and we're maybe seeing some of that, but

我觉得那会带来大得多的影响，我们也许已经看到了一些，但

26:42

most of it is directed towards data centers.

大部分都集中在 data centers 上。

26:46

And the reason I think it could be a good way to push back if you're concerned about

而且我认为，如果你担心的是

26:53

noise or water or other local environmental concerns.

噪音、用水或其他当地环境问题，这可能是一个很好的抵制方式。

26:56

But if you're actually concerned about AI progress as a whole, what is this going to mean

但如果你真正关心的是整个 AI 进展，这会意味着什么

27:01

for society, for jobs, what is this going to mean for safety, and you generally want

对社会、对工作来说，对安全又意味着什么，而你通常想要

27:06

to push back on the pace at which this technology is developing data centers to me are not

去抵制这项技术发展的速度——data centers 在我看来并不是

27:11

the way to go.

正确的方向。

27:12

Well, so let me make the naive case that not building data centers would slow down AI

好吧，那么让我先提出一个比较天真的观点：不建设 data centers 会拖慢 AI 的进展。

27:18

progress.

然后你可以给我算一笔账，Arvin，为了这个环节，我会假装自己懂数学。

27:19

And then you can give me your math, Arvin, and I'll pretend like I understand math

所以我觉得这个天真的观点大概是这样的：嗯，我们知道，要把 AI 推广到整个社会，我们需要训练越来越大的 models。

27:23

for the purposes of this segment.

这意味着要建更多的 data centers 来训练这些 models。

27:24

So I think that the naive case might go something like, well, we know that to sort of spread

而且我们还需要这些 data centers 来 serve 这些 models。

27:32

AI throughout society, we need to train ever bigger models.

随着 AI 遍布整个社会，我们需要训练越来越大的 model。

27:36

That means building more data centers that can train those models.

这意味着要建更多能够训练那些模型的 data centers。

27:38

And then we also need those data centers to serve those models.

然后我们也需要那些 data centers 来 serve 那些 models。

27:42

And so the fewer data centers we have, the fewer places there are where we could train

所以data center越少，我们能training这些models、再serve给其他人的地方也就越少。

27:46

these models and serve them to other people.

所以基于这些理由，我能想象有人会想，那好吧，我就阻止这个data center在我家后院建起来，也算为阻止AI尽一份力。

27:48

And so for those reasons, I could imagine somebody thinking, well, maybe I'll just stop

这种说法有什么问题呢？

27:51

this data center from being built in my backyard and I will be doing my part to stop AI.

我觉得这是个很好的思考起点，但接下来我们得深入细节。

27:57

What's wrong with that argument?

所以你很巧妙地把training和inference，或者说serving models区分开了。

27:59

I think that's a good starting point to think about it, but then we have to get into

我觉得那是个很好的思考起点，但接下来我们得进入……

28:02

the details.

细节。

28:03

So you helpfully separated training and inference or serving the models.

所以你很明确地把 training 和 inference 或者说 serving 模型区分开了。

28:07

When it comes to training, most of the data centers, as I understand it, are not used

说到 training，据我所知，大多数 data centers 并不是用来做 training 的。

28:11

for training.

这需要一定程度的专门化，那些是在特定的 clusters 里完成的。

28:12

That requires some degree of specialization and those are done in particular clusters.

所以零零散散停掉几个 data centers，根本不会让 training 变慢。

28:17

And so stopping a few data centers here and there is not going to slow down training

要做到那一步，你需要某种国家层面的 moratorium，甚至是全球性的。

28:21

at all.

所以真正的问题在于 inference。

28:22

For that, you need some kind of national moratorium or even global.

那么，停掉几个 data centers——无论是在某个本地社区，还是搞一个州级的 moratorium——

28:26

So the real question is about inference.

所以真正的问题在于 inference。

28:28

So is stopping a few data centers, either in one's local community or a statewide moratorium

那么，是停止本地社区的少数几个 data center，还是实施全州范围的暂停？

28:35

or something like that, which is where the action really is?

或者类似的东西，那才是真正的重点所在？

28:39

Is that playing a meaningful part in decreasing the total amount of AI capacity available in

这在减少全球可用的 AI capacity 总量上是不是起着重要作用？

28:45

the world?

所以要想把这个变得更具体一点、能做点计算，我们就得看看特定 capability level 下可用的总 capacity，对吧？

28:46

And so to turn that into something a little bit more concrete where we can do some math,

而这之所以重要，是因为 AI 的进展方式中有一个人们通常不太想到的，并不是靠新建 data centers 和部署新的 GPUs。

28:51

we have to look at the total amount of capacity available at a given capability level.

我们必须看看在特定能力水平下，总共有多少 capacity 可用。

28:56

Right?

对吧？

28:57

And so the reason that is important is that one way in which AI is advancing that people

而这之所以重要，是因为 AI 发展的一条常被忽略的路径，并不是靠新建 data center 和部署新 GPU。

29:03

who don't often think about is not through building new data centers and putting new GPUs

所以真正起作用的其实是前两个因素——你能用现有 data center 做些什么。

29:08

into them, but by making more out of the ones that we already have.

投入到它们里面，而是通过把我们已经有的那些用得更好。

29:12

The industry is getting more efficient at building AI to do a certain kind of task at a certain

这个行业在构建 AI 完成特定类型、特定能力水平的任务时，正变得越来越高效，用更少的电力和更少的机器。

29:17

capability level using less power, less machines.

然后第二个因素是，GPU，更新的 GPU 能效更高。

29:22

And then the second factor you have is that GPUs, newer GPUs are more power efficient.

而这些 GPU 并不全都是投入到新的 data centers 里。

29:27

And those are not all going into new data centers.

它们也会投入到现有的 data centers 里。

29:29

Those are also going into existing data centers.

所以真正主导了来自新的 data centers 容量增长的，其实是前两个因素，也就是你能用现有的 data centers 做些什么。

29:32

So it's really those first two factors, what you can do with the existing data centers

这才是真正主导新增 capacity 的因素，远超新建 data center 带来的部分。

29:36

that really dominates the increased capacity from new data centers.

这确实主导了新建数据中心带来的 capacity 增长。

29:39

So you're saying that data centers are becoming efficient at a rate faster than demand is

所以你的意思是，data centers 的效率提升速度比需求增长还快，或者至少能跟上需求增长，那么这里停一个 data center，那里停一个 data center，根本不足以抵消你刚才说的那些 efficiency gains 带来的影响。

29:47

growing or at least to the extent that demand is growing, stopping a data center here,

没错。

29:53

a data center there isn't going to be enough to counteract the effects of these efficiency

software 和 hardware 带来的 efficiency gains，比起纯粹新建 physical buildings 带来的 capacity gains，大约高一个量级。

29:58

gains that you just described.

我正试着理解这件事，因为我知道你说的是对的。

30:00

That's right.

没错。

30:01

The efficiency gains both from software and from hardware are about an order of magnitude

软件和硬件方面的效率提升大约有一个数量级。

30:05

more than the capacity gains from literally new physical buildings.

比实实在在新建物理建筑带来的 capacity 增长还要多。

30:09

I'm trying to wrap my head around this because I know that what you're saying is true.

我努力想搞清楚这件事，因为我知道你说的是真的。

30:13

The models get more efficient to serve over time, right?

模型随时间推移，serve 起来越来越高效，对吧？

30:16

The labs keep discovering new ways of running their models more efficiently so that the

实验室们不断发现更高效运行模型的新方法，使得

30:21

query that might have eaten up a certain amount of compute last year, this year only needs

去年可能消耗一定量 compute 的 query，今年只需要

30:27

a tenth of that amount.

那个量的十分之一。

30:30

But my understanding is that those efficiency gains are the product of increased compute

但我的理解是，这些效率提升是更多 compute 的产物

30:35

and having bigger models that are able to help researchers discover the efficiency tricks

以及拥有更大的模型，这些模型能够帮助研究人员发现效率技巧

30:41

in the new algorithms that allow them to serve the models much more cheaply.

在新算法中，这些算法让他们能以低得多的成本 serve 模型。

30:45

So is it the overall size and availability of compute important in keeping that efficiency

那么，compute 的整体规模和可用性在保持这种效率方面重要吗？

30:52

curve going?

曲线怎么样了？

30:54

Absolutely.

没错。

30:55

But again, I think we're looking at really, really marginal changes to the total capacity.

但话说回来，我觉得我们看到的对总容量造成的变化真的是微乎其微。

31:03

We're looking at 0.1% of the national or world's capacity of data centers that can be affected

我们说的是，可能受到影响的只有全国或全球 data centers 容量的0.1%，

31:08

by the actions that one community or even one state can take.

而这是某个社区甚至某个州所能采取的行动造成的。

31:13

So the way to look at it is, yes, it's true that compute is an input even for future development

所以看待这个问题的方式是，没错，compute 也是 AI 未来发展的一个 input，

31:20

of AI, but whatever decrease you bring about in total capacity through stopping data centers,

但无论你通过叫停 data centers 使得总容量下降多少，

31:32

whether that compute is going to be used for serving models or for further AI research,

无论这部分 compute 是用来 serving models 还是用于进一步的 AI research，

31:37

that same amount of progress can be made through these gradually continuing efficiency

同样的进展可以通过这些持续渐进的 efficiency improvements 来实现，对此我算过数字。

31:42

improvements at a rate that I did the numbers on this.

如果某个州停止新建 data centers，那就相当于大概 10 小时。

31:46

If one state stops new data center construction, that translates to something like 10 hours.

那就是 AI 行业整体上通过 efficiency improvements 追上被放弃的 compute 所需的时间。

31:51

That's how much it takes aggregated for the AI industry to catch up through efficiency

这太反直觉了，因为所有公司都在拼命建新的 data centers，你知道，今年大部分时间里，大多数前沿实验室都处于这种 capacity crunch 中，基本上在尽可能多地卖 AI。

31:56

improvements to the compute that has been foregone.

对已经放弃的 compute 的改进。

31:59

It's just so counterintuitive because of how hard the companies are all working to build

这真的太反直觉了，因为所有公司都在拼命努力建造

32:03

new data centers, you know, like throughout most of this year, most of the big frontier labs

新的 data centers，你知道，像是今年大部分时间里，大多数 big frontier labs

32:09

have been in this capacity crunch where they're basically selling as much AI as they can

都处于这种 capacity crunch 状态，基本上能卖多少 AI 就卖多少

32:14

make.

做到。

32:15

So it certainly seems like there is something very important to them about being able

所以很明显，对他们来说，能够建造这些 data centers 是非常重要的事情。

32:20

to build these data centers.

那么，Arben，你怎么看这件事？

32:21

So how do you think about that, Arben?

是的。

32:23

Yes.

所以这些都是真的。

32:24

So that is all true.

而且这和我之前说的并不矛盾。

32:25

And that doesn't contradict what I said.

而这就是把这两件事协调起来的方法。

32:26

And here is the way to square both of those things.

1. 而这就是调和这两件事的方法。

32:29

So this progress that's happening through improved GPUs or improved software efficiency

所以这种通过改进 GPUs 或改进 software efficiency 带来的进步

32:35

for the most part, that is shared by all AI companies.

在很大程度上是所有 AI 公司共有的。

32:39

And so that factor kind of cancels out between companies and the competitive advantage that

所以这个因素在公司之间基本就抵消了，而竞争优势

32:44

one company is going to have over another really comes down to how much compute they can

一家公司相对于另一家能有多大，真的取决于它能有多少 compute 可以

32:49

control.

控制。

32:50

And so while the total amount of compute is not a big factor in the aggregate rate of

所以虽然 compute 的总量在 AI 进步的总体速率方面不是一个重要因素，

32:56

AI progress, the relative amount of compute is a surprisingly big factor in the relative

但 compute 的相对量却出人意料地是个大因素，影响着相对的

33:02

competitive positions of companies, both of those things are true at the same time.

公司竞争地位，而且这两件事同时成立。

33:07

Interesting.

有意思。

33:08

So regardless of what moratoriums might do for overall AI progress, if you are an individual

所以，不管暂停措施对整体AI进展会有什么影响，如果你是一个独立的实验室，尽可能多地拿到compute仍然是特别重要的事。

33:14

lab, it's still really important to you that you get as much compute as you can get your

没错。

33:17

hands on.

哪怕只是为了不让竞争对手拿到。

33:18

Exactly.

对。

33:19

If only to stop your competitors from getting their hands on.

我觉得这个问题让我联想到很多1970年代反对核能的情况。

33:21

Yeah.

2. 是啊。

33:22

I mean, I think a lot about this in terms of the nuclear power opposition in the 1970s

3. 我的意思是，我经常想到1970年代反核能的那股势力。

33:27

and 1980s, where you had this, these like, you know, series of accidents, including

到了80年代，就有那么一系列的事故，你知道的，包括 Three Mile Island。

33:31

Three Mile Island.

在那之后，很大一部分美国民粹主义者，或者说至少是政治上活跃的那批美国民粹主义者，认定核能是又坏又危险的，我们不应该再建。

33:32

And after that, like a huge chunk of the American populist, or at least the politically

所以基本上这场运动是成功的，我是说，美国差不多有30年没再建核电站。

33:37

active American populist, decided that nuclear power was bad and dangerous and we shouldn't

而这并没有让技术停止，只是某种程度上改变了它的地理分布——

33:41

build it.

4. 建起来。

33:42

And so there was basically a successful campaign, I mean, we didn't build nuclear power

5. 所以基本上有一场成功的运动，我是说，我们根本没有建造核电站

33:47

in America for like 30 years.

6. 在美国差不多有30年。

33:49

And it didn't stop the technology, but it just sort of changed the geography of where

7. 而它并没有阻止这项技术，只是有点改变了地理上的位置

33:54

it could be built.

它是可以建成的。

33:55

So instead of happening in America, it was happening in France and other places around

所以，这件事没有发生在美国，而是发生在法国和世界上其他确实建了 nuclear power 的地方。

33:59

the world that did build nuclear power.

那你觉得，美国对 data center 的反对，可能的结果是不是这样：它不会真正阻止这项技术，但可能会改变它建在哪里的分布？

34:01

Is that what you think is the likely outcome of data center opposition in America that it

你说得对，我不认为它会阻止这项技术。

34:05

won't sort of stop the technology, but that it could, it could change the distribution

我也非常怀疑，它是否会显著改变这个分布。

34:09

of where it's being built?

8. 也就是它被建造在哪里？

34:10

You're right that I don't think it'll stop the technology.

你说得对，我不认为这会阻止这项技术。

34:12

I'm also very skeptical of whether it's going to meaningfully change the distribution

我也非常怀疑它能否真正改变分配。

34:17

of where it's built.

关于它建在哪里。

34:18

State versus state maybe, but if I understand correctly, one reason stopping a lot of

州与州之间也许吧，但如果我没理解错的话，阻止很多核电站计划

34:24

nuclear plans from being built in the US was so successful was by raising the regulatory

在美国建设之所以如此成功的一个原因，是通过提高监管

34:30

cost.

成本。

34:32

And that's just a very different model from how the AI data center backlash is proceeding.

而这和AI data center抵制活动的进展方式是非常不同的模式。

34:37

It's not, you know, it's not nationwide regulation that forces AI developers wherever

不是那种，你知道的，不是全国性法规强制AI开发者无论在这个国家的哪里

34:42

they are in the nation to go through a whole lot of extra review or a whole lot of extra

都要经过一大堆额外的审查或者一大堆额外的

34:48

safety technology that is going to raise their cost 10 fold and change their timelines, deployment

安全技术，这会让他们的成本提高十倍，并改变他们的时间线和deployment。

34:54

timelines 10 fold.

timelines 拉长十倍。

34:56

It's rather, you know, moving it from one location to another, which locally might seem

这更像是，你知道，把它从一个地方搬到另一个地方，这在局部可能看起来

35:01

like a big win, but at a national level, I don't see it doing much.

像是很大的胜利，但在国家层面，我不觉得它能起多大作用。

35:05

Yeah.

是的。

35:06

That's interesting.

这很有意思。

35:07

And I'm curious, Arvind, if stopping AI progress is not a realistic goal of the data center

而且我很好奇，Arvind，如果阻止AI进展并不是data center

35:14

opposition that we're seeing right now, like, are there other things that it could accomplish

我们现在看到的反对运动的一个现实目标，那么有没有其他它能实现的事情

35:18

that may be constructive rather than futile?

可能是有建设性的，而不是徒劳的？

35:21

Absolutely.

没错。

35:22

I've looked a little bit at what is driving the anti-data center movement.

我稍微研究过是什么在推动 anti-data center 运动。

35:27

And I'm certainly not claiming that this kind of, you know, slowing AI as the main goal

而且我当然不是说，这种，你知道的，把减缓 AI 作为主要目标

35:32

of the movement.

这场运动的。

35:33

A lot of it is driven by local concerns.

其中很多是地方性担忧。

35:35

A lot of it is driven by procedural concerns, the lack of transparency, the way that local

很多是程序性担忧，缺乏透明度，以及当地

35:40

politicians enter into these deals without giving citizens a voice.

政客们签订这些协议时不给公民发言权。

35:45

But, you know, whatever the ultimate reasons are, I would say that what is happening right

但是，你知道的，无论最终原因是什么，我想说的是，现在正在发生的事情

35:49

now is arguably a pretty rational way to go about it because you have to have the credible

现在这可以说是相当理性的做法，因为你必须要有可信的禁令和暂停威胁，才能迫使公司回到谈判桌前。

35:54

threat of bands and moratoria in order to force companies to come to the negotiating table.

以禁令和暂停的威胁来迫使公司回到谈判桌前。

36:02

So companies are making a lot of money, I think, maybe communities can channel their

所以公司在赚很多钱，我觉得，也许社区可以把反对意见转化成一种诉求，要求AI公司直接付款、投资到当地社区，而且力度要比我们今天看到的AI公司大得多，这样当地社区才能真正受益。

36:06

opposition in a way that they can ask for direct payments and investment into local

反对的一种方式，让他们可以要求直接支付和对当地的投资。

36:10

communities much more than we're seeing today from AI companies so that local communities

你觉得是什么在推动AI进步，如果不是compute的可用性的话？

36:16

can benefit.

可以受益。

36:17

What do you think does drive AI progress, if it's not availability of compute?

你之前说过algorithmic efficiency，就是那种model变得更便宜的概念。

36:23

You've said algorithmic efficiency, the sort of notion that these models become cheaper

你说过 algorithmic efficiency，也就是这些模型变得更便宜的概念。

36:27

to serve over time.

长期地服务下去。

36:29

But what else should people be if they are interested in slowing down or stopping AI progress?

但如果有人想减缓或阻止 AI 的进展，他们还能做什么？

36:33

Like what is the better fight to be having?

比如说，什么样的仗更值得打？

36:36

Yeah, I mean, there are many different dimensions of progress, right?

对，我是说，进步有很多不同的维度，对吧？

36:39

So one can think about what is the most powerful, capable model out there and that might

所以你可以想想，外面最强大、最有能力的 model 是什么，而这可能

36:44

matter to you a lot if what you're concerned about is some of the incidents we've seen

对你来说就很重要，如果你担心的是我们看到过的一些事件

36:49

like the hugging face incident and some of the safety risks.

比如 Hugging Face 事件和一些安全风险。

36:52

There's another sense in which you might want to stop, you know, the rapid march of AI,

还有一种意义上，你可能想要阻止，你知道的，AI 的快速前进。

36:58

which is people doing things with AI that are not really suitable for AI, delegating

也就是人们用 AI 做那些并不真正适合 AI 的事情，把太多的决策委托给

37:04

too much decision making to AI.

AI。

37:06

So when I hear things like the colloquially named AI psychosis among CEOs, right?

所以当我听到那种俗称的 CEO 群体中的 AI 精神病时，对吧？

37:12

CEOs, you know, using cloud code to try to replace what an employee does and of course

CEO 们，你知道的，用 cloud code 来尝试取代员工的工作，当然

37:17

it does a first cut version which superficially seems like it does the job firing a bunch

它做出了一个初版，表面上看似乎能胜任工作，解雇了一大批

37:22

of people and then recognizing, oh shit, now when something goes wrong, there's nobody

人，然后发现，哦糟了，现在出了问题，没人

37:26

to fix the mess hiring people back.

收拾烂摊子，于是又把人们招回来。

37:29

We've heard this kind of story many times over and of course, you know, in some cases,

这类故事我们已经听过很多次了，当然，你知道，在某些情况下，

37:33

these models are good enough to in fact, replace certain tasks that people do, but the level

这些模型已经足够好，实际上可以替代人们做的某些任务，但那种

37:39

of premature decision making that we've seen from CEOs has been a cause for concern.

过早决策的程度，我们从CEO们身上看到的，一直令人担忧。

37:43

I would have hoped that these people who frankly have a lot of money on the line would

我本来希望，这些坦白说押了很多钱的人，会

37:48

just for rational self interested reasons make better decisions that take into account

仅仅出于理性的自利原因做出更好的决策，考虑到

37:53

not just AI's potential, but also its limitations, you know, here and now.

不仅仅是AI的潜力，还有它的局限性，你知道，此时此刻。

37:57

So this in my view would be a much better target for opposition and that's, you know,

所以在我看来，这会是一个更好的反对目标，而且，你知道，

38:04

less at the community level and more at the workplace level, better aligning incentives

更少在社区层面，更多在工作场所层面，更好地对齐激励机制

38:09

and decision making and information between the individual workers who in many ways have

以及个体工人之间的决策和信息，这些工人在很多方面拥有...

38:16

a much more grounded understanding of what AI can and can do in a particular company and

对公司和管理层来说，对AI能做什么、不能做什么会有更接地气的理解，这你知道，它在很多方面是很有前瞻性的，这可能是好事，但也可能带来一些错误的决策。

38:21

management, which, you know, has has its in many ways is forward looking and that can

是啊，与此同时，我觉得现在可能有很多人确实想在AI快速发展的世界里拥有更多的控制权和主动权，而且我猜他们不会满意那种“你可以靠组织的惯性来拖慢这一切”的说法。所以你觉得，在接下来的方向上，还有没有别的东西能给人更多……

38:27

be a good thing, but can also make some misguided decisions.

是件好事，但也可能做出一些错误的决定。

38:31

Yeah, at the same time, I think there are probably a lot of people out there who do want

是的，与此同时，我觉得可能有很多人确实想要。

38:37

to sort of have more control and agency in a world where it seems like AI is advancing

在一个AI快速发展的世界里，想要拥有更多的掌控力和自主性，而且我怀疑他们不会满足于那种“你可以靠组织的惯性来减缓这一进程”的说法。那么，你觉得未来还有没有其他东西能给人带来更多的……

38:43

quickly and I suspect they're not going to be satisfied with the answer of like, well

而且我怀疑他们不会满足于像这样的答案，嗯

38:48

you can just rely on the inertia of your organization to slow the progress of this.

工业革命时期，他们基本上是通过暴动来进行集体谈判的。

38:54

So are there other things out there on the horizon that you think could give people more

那么，你觉得未来还有其他东西能给人们带来更多吗？

38:59

of that feeling of control?

那种掌控感吗？

39:02

Absolutely.

绝对。

39:03

I think we have a lot of control in how we ourselves use AI.

我觉得我们在自己使用AI的方式上有很多掌控力。

39:07

So this is something I constantly confront in my own use of these tools.

所以这是我在自己使用这些工具时经常面对的问题。

39:12

I mean, I am a pretty heavy user of AI tools and AI agents and I find that by default,

我的意思是，我是AI工具和AI agent的重度用户，我发现默认情况下，

39:17

they do things that I'm not very happy with in ways that I feel like I'm losing control.

它们会做一些我不太满意的事情，让我感觉自己正在失去控制。

39:23

So if I ask an agent to, you know, find papers, materials, reports on a given topic so

所以如果我让一个agent去，你懂的，找某个主题的论文、材料、报告，以便

39:30

that I can start to analyze it, it will without my asking for it, turn that into its own

我可以开始分析它，它就会在我没有要求的情况下，把它变成自己的

39:36

analysis, its own opinion.

分析，它自己的观点。

39:38

And so I have to put specific things in my prompts so that I can get it to do the grunt

所以我必须在 prompts 里放具体的东西，才能让它做那些苦力

39:42

work for me not to do my thinking for me.

活，而不是替我做思考。

39:45

So that's, you know, just one simple example.

所以，你知道，这只是一个简单的例子。

39:47

But the broader point is that these tools are not only powerful, they're very, very flexible

但更宏观的点是，这些工具不仅强大，而且非常非常灵活

39:52

and how we can use them.

也在于我们如何使用它们。

39:53

And I think each of us should not just accept the way that the developer has created the

而且我认为我们每个人都不应该只是接受开发者创造工具的方式，

39:57

tool, but configure them, personalize them in ways that meet our appropriate comfort

而应该以符合我们适当舒适度的方式去配置它们、个性化它们。

40:03

level for what should be delegated to AI and what should be in the domain of people.

关于什么应该交给AI、什么应该由人来负责的层面。

40:08

That's not an answer to your broader question about AI as a kind of, you know, powerful global

这并不能回答你那个更广泛的问题，就是关于AI作为一种强大的全球力量，以及我们如何能减缓它。

40:14

force and how we can slow that down.

但我觉得在个人层面，我确实想强调，我们其实有很大的主动权。

40:16

But I think at an individual level, I do want to emphasize that we have a lot of agency.

马克思主义历史学家Eric Hobbesbaum有一句关于工业革命时期Luddites的名言，说他们基本上是在用暴动来进行集体谈判。

40:21

The Marxist historian, Eric Hobbesbaum has this famous phrase about the Luddites of the

也就是说，所有那种对机器的反对，都是一种集体谈判的形式，因为那些行业没有工会，而那是他们表达不满的唯一方式。

40:28

Industrial Revolution, that they were conducting collective bargaining by riot basically

所有那种对机器的反对都是一种集体谈判，因为那些行业还没有工会化，表达不满的唯一方式就是这个。

40:34

that all of the sort of opposition to machines was a form of collective bargaining because

所有那些对机器的反对其实是一种集体谈判的形式，因为

40:40

those industries had not been unionized and the only means of expressing their discontent

那些行业还没有工会化，表达不满的唯一方式就是

40:46

that workers had was to break the machines that were threatening their jobs or threatening

工人当时能做的就是砸掉那些威胁他们工作的机器，或者威胁

40:51

to concentrate wealth to the owners of the factories rather than to the workers.

要集中财富到工厂主手里，而不是给工人。

40:57

And I've been thinking a lot about this, you know, obviously, and thankfully, we're not

我一直在想这个问题，你知道，显然，而且幸好，我们还没有

41:01

at the point of like violent riots over data centers yet, but I think there is a possibility

到那种因为data centers而发生暴力骚乱的地步，但我觉得有可能

41:05

that we end up there.

我们会落到那一步。

41:07

But I think your point, Arvind, about this being a form of kind of collective bargaining

但我觉得，Arvind，你说的这点，就是这有点像是一种集体谈判

41:11

without any formal mechanism, like this is, to me, when I see people protesting data centers,

没有任何正式机制，对我来说，当我看到人们抗议data centers时，

41:17

I think this is about what's in it for these communities, you know, if the data center

我觉得这关乎这些社区能从中得到什么，你知道，如果data center……

41:22

of companies came in and said, we're going to build this data center, but we're also going

有些公司过来说，我们要建这个 data center，但我们也打算

41:25

to build you a high school and a really nice public park and improve your roads and make

给你们建一所高中和一个很棒的公共公园，改善你们的道路，让

41:30

your everyday life better.

你们的日常生活变得更好。

41:32

To me, that feels like a much better deal than they're getting now.

在我看来，这比他们现在得到的好处划算多了。

41:36

And so I think of this sort of data center backlash as a form of kind of collective

所以我觉得这种对 data center 的抵制，某种程度上是一种集体

41:41

bargaining by another name.

谈判的另一种形式。

41:43

Well, and, you know, the sort of traditional form of collective bargaining has been pretty

嗯，而且你知道，传统的集体谈判形式一直相当

41:47

effective here, right?

有效，对吧？

41:48

We've seen workers in Korea rise up and threaten to go on strike.

我们在韩国看到工人们起来抗议，威胁要罢工。

41:50

If they didn't get a greater share of the profits, the record profits that their companies

如果他们拿不到更大份额的利润，那些他们公司因为AI泡沫而实现的创纪录利润的一部分。

41:54

were realizing in part due to the AI bubble.

所以我觉得，传统老派的工会化在这里能起很大作用。

41:57

So I think good old fashioned unionization can do a lot here.

我们在好莱坞也看到了，对吧，你知道，就像上次那些大工会联合起来，他们谈判了很多针对AI的保护措施。

42:00

We also saw it in Hollywood, right, where, you know, like the last time the big unions

是的。

42:04

got together, they negotiated for a lot of protections against AI.

百分之百同意集体谈判这个观点。

42:07

Yeah.

是的。

42:08

A hundred percent agree with the collective bargaining perspective.

百分之百同意集体谈判这个视角。

42:11

All right.

好的。

42:12

Well, Arvin, thank you so much for coming on.

嗯，Arvin，非常感谢你来上节目。

42:14

We really appreciate your expertise and wisdom as always.

我们真的很感谢你的专业和智慧，一如既往。

42:17

And yeah, we'll, we, I guess we won't see you out there on the picket lines for the

而且，是啊，我们——呃，我猜我们不会在data center外面的纠察线上看到你。

42:22

data center.

我们会在电子表格里看到你。

42:23

We'll see you in the spreadsheets.

我对这个角色很满意。

42:24

I'm happy with that role.

谢谢。

42:28

Thank you.

谢谢。

42:29

This has been really fun.

这真的很有意思。

42:30

Thank you.

谢谢。

42:34

When we come back, hats all folks.

等我们回来，帽子就全剧终了。

42:37

It's time for the last installment of Hat GPT.

是时候做 Hat GPT 的最后一期了。

42:51

Well, Casey, we have a bitter sweet final segment of our episode today.

好吧，Casey，今天我们有一个苦乐参半的最终环节。

42:56

We are doing the last ever Hat GPT.

我们要做史上最后一次 Hat GPT。

43:06

That's right, Kevin, after almost four years of slips of paper being lovingly handcrafted

没错，Kevin，将近四年来，一张张纸条被用心手工制作，然后塞进各种时尚帽子里，而这是我们最后一次取纸条的机会。

43:12

and fed into a variety of stylish hats, this will be our final opportunity to take a slip

然后放进各种时尚的帽子里，这将是我们最后一次机会去拿一张纸条

43:19

from the hat, discuss its contents, and then when one of us gets bored, we'll say to

从帽子里拿出来，讨论里面的内容，然后当我们中有人无聊时，我们会说

43:23

the other stop generating.

让对方停止生成。

43:25

We've been through so many hats, at least three by my count, and so many slips, and we'll

我们经历了这么多顶帽子，至少三顶，我记得，还有这么多纸条，我们

43:31

never forget them.

永远不会忘记它们。

43:33

Hats are now out of fashion, and so we are retiring it as a vehicle for podcast content

帽子现在已经过时了，所以我们不再把它作为播客内容的载体了。

43:40

creation.

创造。

43:41

All right.

好的。

43:42

Let's do one final trip through the hat.

让我们最后再来一轮吧。

43:48

Wow.

哇。

43:49

Sometimes I forget that we record in front of a live studio audience.

有时候我都忘了我们是在现场观众面前录的。

43:54

That's beautiful.

太美了。

43:55

Our producers just hit us with the applause, sound effect, and Casey.

我们的制作人刚刚给我们放了掌声、音效，还有Casey。

44:00

This is, I think, our longest, was this our first ever segment that had its own title

这，我觉得，是我们最长的——这是我们第一个有自己标题的环节吗

44:05

in gimmick?

带噱头的？

44:06

I think it was.

我觉得是。

44:07

Our first ever gimmick segment, and it's only fitting that we go out in a bang.

我们的第一个噱头环节，就该轰轰烈烈地收尾。

44:12

My only lament is that because we're doing this segment today, we won't be able to continue

我唯一的遗憾是，因为今天要做这个环节，我们就没法继续我们那个“每期节目都推出一个新环节”的梗了，直到节目结束。

44:18

with our bit of launching a new segment for every show until the end of the show.

好的，Casey。

44:22

All right, Casey.

让我们最后一次打开帽子。

44:25

Let's open the hat one last time.

好的，Kevin，我们今天的第一项，Apple Vision Pro 被用在手术里了——一项来自 UC San Diego 的研究，把 Vision Pro 当作泪道手术的主要显示器。这个手术嘛，我当然更熟悉它的另一个名字——endoscopic decryosist，哦，我的天，这也太难念了，decryosist，decryosist，decryosist，rhinostomy。

44:28

All right, Kevin, our first item today, the Apple Vision Pro is being used in surgery,

好的，Kevin，我们今天的第一条，Apple Vision Pro 正在被用于手术，

44:36

a study from UC San Diego in which the Vision Pro was used as a primary display for a tear

来自UC San Diego的一项研究，该研究将Vision Pro用作主显示屏，用于泪

44:42

duct procedure, which, of course, I know better as an endoscopic decryosist, oh, my God,

道手术，当然，我更熟悉的是 endoscopic decryosist，哦，我的天，

44:48

that's really hard, decryosist, decryosist, decryosist, rhinostomy.

这真的好难说，decryosist，decryosist，decryosist，rhinostomy。

44:58

Let's go.

开始吧。

45:00

Anyway, there were 32 total procedures and key findings included that by using the Vision

总之，总共有32个流程，关键发现包括通过使用Vision...

45:07

Pro, the operating times were 19% shorter, they had 100% functional success with no post-operative

Pro，手术时间缩短了19%，他们实现了100%的功能性成功，且无术后并发症。

45:14

complications and a significantly lower surgeon-reported workload.

并发症更少，而且外科医生报告的workload也明显更低。

45:19

Kevin, have we finally found an ideal use for the Vision Pro?

Kevin，我们到底有没有找到Vision Pro的理想用途啊？

45:23

Yes, it turns out that the people most in need of the Vision Pro were the Vision Pro's.

是的，结果发现，最需要 Vision Pro 的其实就是 Vision Pro 自己人。

45:29

And by that, I mean, the surgeons operating on people's eyeballs.

我是说，那些在人的眼球上动手术的外科医生。

45:34

Subgeneraating.

子生成。

45:35

Wait, no, I have more to say about this.

等等，不，关于这个我还有话要说。

45:39

Okay, what more do you have to say?

好，你还有什么要说的？

45:40

As a Vision Pro owner, I would like to formally offer my Vision Pro, which is sat on my shelf

作为Vision Pro用户，我想正式把我那个在架子上吃灰了一年多的Vision Pro提供给任何想在手术中使用它的外科医生。

45:48

collecting dust for the last year plus to any surgeon who would like to use it in an

哦，我以为你是说要给我的泪腺动手术呢。

45:53

operation.

哦，那个我也会做。

45:54

Oh, I thought you were going to offer an operate on my tear ducts.

你觉得这是不是那种，就是有个app，然后有点像……

45:56

Oh, I will also do that.

哦，我也会这么做的。

45:59

Do you think it's a sort of thing where there's just like an app and it's kind of like

你觉得这是不是那种情况，就是有个 app，然后有点像……

46:02

paint by numbers, you know, and it's just like poke here, prod here, a snip there,

你知道的，就像按数字填色一样，这儿戳一下，那儿捅一下，再剪一下，

46:05

and congratulations.

然后恭喜你。

46:06

Your tear ducts work again.

你的泪管就又能用了。

46:07

I think about it more like you could just keep the surgeons as like a higher energy level

我更愿意这样想：你可以让外科医生保持在更高的能量水平上。

46:11

because they can be doing their surgery, but also like checking their email and also

因为他们可以一边做手术，一边查邮件，还有

46:14

playing Fruit Ninja.

玩 Fruit Ninja。

46:15

That makes sense.

有道理。

46:16

You know, I actually have a cheaper alternative, which is if your tear ducts are busted,

你知道吗，我其实有个更便宜的办法，那就是如果你的泪管坏了，

46:20

just look at the price tag for your Vision Pro.

看看你那Vision Pro的标价吧。

46:23

You'll be crying in no time.

你马上就得哭。

46:27

Stop generating.

别生成了。

46:28

All right.

好了。

46:29

This one comes to us from Kotaku.

这条是Kotaku那边来的。

46:31

Oh, boy, have you been following the GTA 6 chaos?

哎呀，你一直在关注GTA 6那摊子乱事吗？

46:35

I absolutely have.

我当然有。

46:37

It has been the industry's most anticipated video game for well over a year now, and yet

它成为游戏业界最期待的作品已经一年多了，然而……

46:43

Grand Theft Auto 6 is still not out Kevin, except in some ways it kind of is.

Kevin，Grand Theft Auto 6 到现在还是没发售，不过从某种意义上来说，它也算已经出来了。

46:48

So yes, so there was a leak of some footage, actually multiple leaks of footage from the

所以没错，确实有一段画面被泄露了，实际上是好几段新的 Grand Theft Auto 游戏画面，在 Rockstar——就是做 GTA 的那家游戏公司——计划好的盛大发布之前，就出现在了互联网上。

46:53

new Grand Theft Auto game that have appeared on the internet ahead of a big planned reveal

Rockstar 已经在网上公开向粉丝们道歉，为这些不幸的泄露事件。

47:00

by Rockstar, the gaming company that makes GTA Rockstar has taken to the internet to apologize

他们说，虽然很遗憾原本预期的游戏体验现在可能会被一些剧透影响，但我们希望大家能再多等一段时间，等到 11 月 19 日亲自体验这款游戏。

47:07

to fans for these unfortunate leaks.

向粉丝们，因为这些不幸的泄露。

47:11

They say that while it is unfortunate that the intended game experience may now be impacted

他们说，虽然很遗憾，原本预期的游戏体验现在可能会受到

47:16

by some spoilers, we hope that everyone will wait a bit longer to experience the game

一些剧透的影响，我们希望每个人都能再等一等去体验游戏

47:20

for themselves on November 19th.

在11月19日亲自体验。

47:22

Yeah.

是啊。

47:23

People are not going to wait.

人们不会等的。

47:24

They're going to look at every single leak as it comes out.

每一个泄露出来他们都会去看。

47:27

As you noted, there have been a series of them, and I mean, look, you know, I don't know

就像你提到的，已经有一系列了。我是说，你看，你知道，我不知道

47:31

how these leaks were obtained presumably through some sort of crime, so I don't want to advocate

这些泄露是怎么搞到的，大概是通过某种犯罪手段吧，所以我不想提倡

47:36

for that.

那样做。

47:37

But people are really, really frustrated, and in particular, they're frustrated, Kevin,

但人们真的真的非常沮丧，尤其是，Kevin，他们沮丧的是，

47:41

that to watch a trailer for Grand Theft Auto 6, the sort of thing that used to be released

为了看《Grand Theft Auto 6》的预告片——这种东西以前可是直接发布的。

47:47

completely for free since after all, it is just a marketing item.

完全免费，毕竟它只是个营销噱头。

47:51

This time around, Rockstar said, yeah, you can see the new trailer for Grand Theft Auto

这次 Rockstar 说，对，你们可以先看 Grand Theft Auto 6 的新预告片。

47:56

6.

如果你有 Netflix 订阅，因为我们会先在那里放出。

47:57

If you have a Netflix subscription, because we're going to put it there first.

不过，我觉得他们说了，你懂的，之后会放到 YouTube 和其他地方。

48:00

Now, I think they've said they're going to, you know, put it out on YouTube and everywhere

但这只是人们非常不满 Rockstar 的又一个原因，也是为什么我觉得看到这些泄露出来，他们其实没那么难过。

48:03

else later.

其他稍后再说。

48:04

But this is just one more reason why people have been very frustrated with Rockstar and

但这只是人们对 Rockstar 感到非常沮丧的又一个原因，而且

48:08

why I think they were not too sad to see some of these leaks come out.

以及为什么我觉得他们看到这些泄露出来也没那么难过。

48:11

Now, if you had to guess, what is the probability that these leaks were caused by a rogue open

那么，如果让你猜的话，这些泄露是由一个 rogue open AI agent 入侵 Rockstar Games 的电脑、窃取视频素材并上传到互联网上的概率有多大？

48:17

AI agent hacking into the computers at Rockstar Games, stealing the footage and uploading

我永远不会排除这种可能性。

48:22

it to the internet?

而且我觉得我们必须考虑这一点。

48:23

I will never rule that out.

从这些泄露中我知道的主要一点是，Grand Theft Auto 6 的一部分显然发生在一个天体营里，Kevin。

48:25

And I think we have to consider it.

真的吗？

48:26

The main thing I know from these leaks is that part of Grand Theft Auto 6 apparently does

我从这些泄露里知道的主要是，Grand Theft Auto 6 的一部分显然确实发生在裸体营地，Kevin。

48:31

take place in a nudist colony, Kevin.

真的？

48:33

Really?

直接在游泳池里。

48:34

And there is full frontal male nudity.

48:36

So let's just say I'm pre-ordering.

48:39

Stop generating.

48:42

Next up, Amazon Drone delivers package directly into a women's pool.

48:46

I'm going to guess if you're listening, you've already seen this.

48:48

This was a truly inescapable clip and we loved it so much.

48:51

A Texas woman named Lindsay Austin ran outside to film her first drone delivery.

48:56

She used one of these Amazon prime drones that we've featured in a past episode.

49:02

And the drone hovered over her pool before dropping the package straight into the water.

然后那个drone在她的泳池上方悬停，直接把包裹扔进了水里。

49:08

Do we know what was in the package?

我们知道包裹里是什么吗？

49:10

No.

不知道。

49:11

Like maybe it was a pool float or something thematically appropriate and the drone just

比如说可能是个泳池浮垫之类符合主题的东西，然后drone就自己做了决定。

49:15

made the decision.

它就想“我要把这个扔下去，嗯，我要帮他们省事，直接扔进泳池里。”

49:16

I'm going to drop this like, well, I'm going to save them the trouble and just drop it

也许是个往水里加氯的那种小装置，你懂的。

49:18

directly in the pool.

也许那是那种往水里加氯的小装置，你知道的。

49:19

Maybe it was one of those little systems for putting a chlorine into the water, you know,

而且还有0.2%的几率你的消息永远送不到，因为鸽子在途中迷路或者死了。

49:25

in which case why well done, Amazon Drone.

那样的话，为什么要说干得好呢，Amazon Drone？

49:27

Yeah, well done.

是啊，干得好。

49:28

Keep going.

继续。

49:29

Let's give the benefit of the doubt to the drone.

我们就姑且相信这个drone吧。

49:31

Okay.

好吧。

49:32

Stop generating.

停止生成。

49:34

Next up.

下一个。

49:35

And here's a question I truly never asked before.

还有一个我从来没真正问过的问题。

49:38

What if your text traveled as slowly as a carrier pigeon?

如果你的文字像信鸽一样慢悠悠地飞，会是什么感觉？

49:41

There are two new messaging apps doing the rounds online, Kevin, carrier pigeon and

最近有两款新的消息应用在网上挺火，Kevin，叫 carrier pigeon 和 roost，它们故意把消息速度放慢到和真实动物一样。

49:45

roost, and they deliberately slow messages to the speed of real animals.

也就是说，Kevin，如果你用 carrier pigeon，从洛杉矶发到纽约的短信要花22小时。

49:52

So this means Kevin that if you use carrier pigeon, a text traveling from Los Angeles to

而且还有0.2%的几率你的消息永远到不了，因为鸽子在路上迷路了或者死了。

49:57

New York City would take 22 hours.

你得换一只新鸽子，买一只要99美分。

50:00

And also there's a 0.2% chance your message will never arrive because the pigeon got

你得换一只鸽子，买只新的要花99美分。

50:07

lost or died in transit.

50:09

You will have to replace your pigeon and it costs 99 cents for a new one.

在运输途中丢了或者死了。

50:14

I really admire this because it's rare to see an app that is made this bad on purpose.

我真的很佩服这个，因为很少能看到一个 app 故意做得这么烂。

50:23

But this seems like something that might appeal to you as somebody who has done various

不过作为过去试过各种招数来戒掉手机的人，这似乎对你很有吸引力。

50:28

tricks to stop using their phone in the past.

不，这对我来说是个好消息，我跟你说为什么我并不是一个好的群聊参与者。

50:30

No, this is great news for me and I'll tell you why I am not a good group chat participant.

我是那种，你知道，让消息一直堆着不回复的人。

50:36

I am someone who, you know, lets the messages pile up.

我不太擅长及时回复。

50:40

I'm not very good about responding in a timely way.

你是个潜水者。

50:43

You're a lurker.

我更像是个潜水者。

50:44

I'm more of a lurker.

我更多是个潜水党。

50:45

I'm just, I'm very bursty like I'll respond to like 18 texts at once.

我就是这样，我回消息很一阵一阵的，比如我会一次性回18条消息。

50:49

But then I'll like put my phone down for six hours.

但接着我会把手机放下六个小时。

50:52

This is great for me because now whenever anyone's like, why isn't Kevin trying me and I can

这对我挺好的，因为现在每当有人问“为什么Kevin不联系我”的时候，我就可以

50:56

just say I did, but I sent my messages via carrier pigeon and they just died on the way

直接说“我发了啊，但我的消息是用信鸽寄的，它们在半路上就死了”

51:03

to your phone.

还没到你手机。

51:04

Yeah.

对啊。

51:05

The messages are dead.

消息都死了。

51:06

Sorry.

抱歉。

51:07

Yeah.

嗯。

51:08

You know, for all of the concerns I have about like using my phone too much, like sending

你知道，尽管我对自己用手机太多有不少担忧，但给别人发消息并不是我真正担心的。

51:13

messages to people is not really one of the concerns that I have.

你知道，更多的是发行商、创作者和社交应用发送 push notification。

51:17

You know, it's more of like publishers and creators and like social apps sending push

那才是我希望少一点的，而不是朋友的消息变少。

51:22

notification.

是的，不过如果你不擅长回短信，这倒是很好的借口。

51:23

Like that's the stuff that I want less of, not like fewer messages from my friends.

所以为此，我要称赞那些鸽子应用。

51:26

Yeah, it is great plausible to die ability though if you're bad at texting.

是啊，不过如果你不擅长回短信，这倒是提供了很好的推脱借口。

51:30

So for that, I want to commend the pigeon apps.

所以呢，我要表扬一下那些信鸽应用。

51:32

Here's what I would say.

我会这么说。

51:33

You can find yourself using this app to get less messages from your friends.

你可能会发现自己在用这个app，结果朋友给你发的消息变少了。

51:36

The problem is not your phone.

问题不在你的手机。

51:37

The problem is your friends.

问题在于你的朋友。

51:39

What is more annoying, someone in the group chat with with carrier pigeon installed on

更烦人的是，群聊里有人手机上装了信鸽，还是有人发绿气泡？你会把谁踢出去？

51:44

their phone or someone with a green bubble, who are you kicking out?

用绿气泡的人一点问题都没有。

51:48

There's nothing wrong with people that have green bubbles.

这句话真是太有阶级歧视了。

51:50

That is such a classist sentence.

这可真是句充满阶级歧视的话。

51:54

Kevin.

Kevin。

51:55

Okay.

好吧。

51:56

Man of the people.

人民之友。

51:57

How many group chats do you have with people with green bubbles?

你有多少个群聊是和有绿气泡的人一起的？

52:00

I people.

我，人们。

52:01

Let me see your phone.

让我看看你的手机。

52:02

Let me see your guy.

让我看看你的家伙。

52:03

I don't want to see your phone.

我不想看你的手机。

52:04

There are some federal crimes.

有一些联邦罪行。

52:05

Stop generating.

停止生成。

52:08

All right.

好吧。

52:09

AI agents are turning founders into insomniacs.

AI agents正在把创始人变成失眠者。

52:13

This is from the Wall Street Journal, which reported on AI startup employees working around

这是来自《华尔街日报》的报道，讲的是AI初创公司的员工们全天候照看一群AI agents。

52:17

the clock to babysit fleets of AI agents.

员工们会在奇怪的时间点醒来，看看某个agent在夜里做了什么，或者我猜是在半夜。

52:21

Workers are waking up at odd hours to check what an agent did overnight, or I guess in

工人们会在奇怪的时间醒来，检查agent在夜间做了什么，或者我猜在

52:25

the middle of the night.

半夜。

52:26

And there are some great quotes in here.

这里面有一些很棒的话。

52:28

For example, one founder says, it's like a drug.

比如，一位创始人说，这就像毒品一样。

52:30

I've never actually done drugs, but I imagine it's what it feels like.

我从来没吸过毒，但我猜就是这种感觉。

52:34

My advice to this man?

我对这位先生的建议呢？

52:35

Do drugs.

吸毒。

52:36

Do drugs.

吸毒。

52:37

Do drugs.

吸毒。

52:38

Do drugs.

吸毒。

52:39

I promise it is better than doing agents in the middle of the night.

我保证这比半夜搞 agents 好多了。

52:42

Another person quoted in the story says, I don't know if it's healthy for me to be out on

故事里引用的另一个人说，我不知道自己一边跑步一边盯着手表给我的 agent 授权做某件事，这算不算健康。

52:45

a run looking at my watch to give my agent permission to do something.

不，我会说这非常健康。

52:49

And no, I would say that's very healthy.

请时刻盯着你的 agent。

52:51

Please look at your agent at all times.

过马路的时候千万别看路。

52:52

Never look at the street as you cross it.

当然，也不要左右看。

52:55

Certainly, don't look both ways.

就看看会发生什么。

52:56

Just see what happens.

就看看会发生什么。

52:57

I think it'll probably be fine.

我觉得应该没什么问题。

52:59

I understand why people in the rest of the country want to wipe San Francisco off the face

我理解为什么这个国家其他地方的人想把旧金山从地图上抹去。这里的人的生活方式会让任何一个有血性的美国人脊背发凉。

53:04

of the map. People here are living in ways that would send shivers down the spine of any

那么，Kevin，你现在算是正处于转向 startup 创始人的半途中。你会半夜醒来去检查那些我猜你让它们帮我们搭建公司的 agents 吗？

53:09

red-blooded American. Now, Kevin, you're sort of mid-pivot into being a startup founder.

不，我半夜醒来是因为焦虑，老派的那种。

53:13

Are you waking up in the middle of the night to inspect the agents that I assume you have

你提到这个真有意思。我也一直在这样。

53:18

building our company? No, I'm waking up in the middle of the night with anxiety, the old-fashioned

Stop generating.

53:22

way. It's funny you've mentioned that. I've been doing that too. Stop generating.

好的。哦，下一条。我一直很期待看到这个。

53:28

All right. Oh, this next one. I've been looking forward to seeing this.

好的。哦，下一个。我一直期待看到这个。

53:31

You did the last one. It's my turn. Okay, fine.

你做了上一个。轮到我了。好吧，行。

53:33

Don't take this from me. This is our last chance. I got greedy. All right, next out of the hat.

别抢我的。这是我们最后的机会了。我太贪心了。好啦，接下来从帽子里抽一个。

53:40

Oh, this one's a good one. This one comes to us from Business Insider. There's a new term for

哦，这个不错。这个来自 Business Insider。现在有个新词来形容

53:44

co-workers who blindly share AI output. Meet proxy. Coined by Nicholas Groen, the term Meet proxy

那些盲目分享 AI 输出的同事。叫 Meet proxy。由 Nicholas Groen 创造，Meet proxy 这个说法

53:51

describes people who blindly copy and paste the output of AI systems to their peers. Groen writes,

描述的是那些把 AI 系统的输出盲目复制粘贴给同事的人。Groen 写道，

53:57

by all means, prompt AI, but don't just relay the output. Read it, understand it, validate it,

尽管用 AI 生成内容，但不要只转发输出。要读它、理解它、验证它，

54:03

and then write a response in your own words. Casey, what do you make of Meet proxy?

然后用你自己的话写回复。Casey，你怎么看 Meet proxy？

54:08

And are you a Meet proxy? I really strive not to be a Meet proxy. I don't like the phrase and

那你是 Meet proxy 吗？我真的很努力不做 Meet proxy。我不喜欢这个说法，而且

54:16

I won't be using it. But do I know software engineers now whose job just consists of

我自己不会用它。但我确实认识一些软件工程师，现在他们的工作就只是

54:23

occasionally checking in on a coding agent and saying like, oh, yeah, do that or the agent will say

偶尔去看看 coding agent，然后说，哦，好，就这么做，或者 agent 会

54:30

to them something like, hey, I can't access this file and they'll just kind of go quickly open it.

对他们说类似这样的话：嘿，我访问不了这个文件，然后他们就会很快把它打开。

54:34

But it's kind of a funny phrase, but there is a kind of creeping human disempowerment in here that

但这说法挺有意思的，不过这里面有一种逐渐蔓延的人类失去主导权的趋势，

54:41

I think is actually bad. Yeah, don't be a Meet proxy. I think the stigma around Meet

我觉得那其实很糟。对，别当 Meet proxy。我觉得围绕 Meet proxy 的污名

54:45

proxies is positive. And if someone in your life, well, I also just want to say you have sent me

是正面的。而且如果你身边有人——嗯，我还想说，你确实给我发过一些

54:51

some, uh, some Claude Slop that struck me as the actions of a Meet proxy. Yes. Now I always try

一些，呃，一些 Claude Slop，我觉得那些行为就像是一个 Meet proxy。是的。现在我总是试着

54:58

to identify this Claude Slop when I do that. That's true. Yeah. Stop generating. All right. You're up.

在那么做的时候识别出这种 Claude Slop。确实。嗯，停止生成。好了，轮到你了。

55:05

Okay. Well, I've been looking forward to watching this one, Kevin, the Chinese robot Tian Gong

好的。

55:12

clocked a sub nine second hundred meter run in Beijing. As you probably know, the world

嗯，Kevin，我一直很期待看这一场，中国的robot Tian Gong在北京跑出了不到9秒的100米成绩。

55:19

humanoid robot games were in Beijing this week, more than 2000 robots from 16 countries competed

你可能知道，这周world humanoid robot games在北京举行，来自16个国家的2000多个robot参赛。

55:28

and in the 100 meter sprint prelims, two robots, both the Tian Gong Ultra and honors lightning robot

在100米短跑预赛中，Tian Gong Ultra和honors lightning robot这两个robot都击败了Usain Bolt在100米中9.58秒的世界纪录。

55:36

beat Usain Bolt's world record of 9.58 seconds in the 100 meter spread three days later, Tian

三天后，Tian Gong Ultra以8.86秒跑完，打破了自己开幕日创下的纪录，这是好消息。

55:46

Gong Ultra ran it in 8.86 seconds, breaking its own opening day record, which that's the good

坏消息是，这个robot刹不住车，径直撞上了缓冲墙，然后起火了。

55:54

news. The bad news is that the robot had trouble stopping and ran straight into a padded wall

但我要告诉你，如果我100米能跑8.86秒，

56:00

and burst into flames. But I'm telling you, if I ever ran a hundred meters and 8.86 seconds,

然后烧起来了。但我跟你说，如果我一百米跑8.86秒，

56:06

I too would burst into flames at the end of it just out of the sheer joy of having one. I have been

我也会在最后因为纯粹拥有它的喜悦而自燃。

56:11

totally obsessed with these humanoid robot games. This is my Olympics. Yeah.

我一直对这些 humanoid robot 比赛超级痴迷。

56:18

Less because of the achievements. I don't actually think it is that cool that a robot can run

这就是我的奥运会。

56:23

faster than Usain Bolt. Why don't you think that's cool? A car can also go faster than Usain Bolt.

对。

56:27

It's not, it's a robot. It doesn't, who cares? But the way that these robots run and what happens

倒不是因为那些成就。

56:34

to them after they finish is the most transfixing thing that I've seen all year. I want to show you

我其实不觉得一个 robot 能跑得比 Usain Bolt 快有什么酷的。

56:40

a clip from the humanoid robot games because it is astounding. Let's look. Okay, so they're running

你为什么觉得那不酷？

56:48

down the track. They're sprinting. They're sprinting boom. They all run into the wall. Now can they

车也能跑得比 Usain Bolt 快啊。

56:56

not program them to either stop or I don't know just not go not run into the wall?

不是编程让它们要么停下来，要么……我不知道，就是别往前走，别撞到墙上？

57:03

This one goes boom into the wall, falls back, falls down, sparks fly.

这个直接 boom 一声撞墙上，弹回来，倒下，火花四溅。

57:11

He had to be carrying it out of the stretcher. It is disturbing because they just go so completely

他得用担架把它抬出去。这让人不安，因为它们完全瘫软，而且因为是 humanoid，你会本能地同情它们。哦，我不会。

57:16

limp and because they're humanoid, you do instinctively sympathize with them. Oh, I don't.

哦，好吧。我觉得这是它们的报应，是它们试图在跑步上打败我们的傲慢。

57:22

Oh, okay. I feel like this is their come-up, it's for their hubris of trying to beat us at

当你看到机器人跑那么快，你会不会想，总有一天那种东西会在黑暗小巷里追我，而我毫无希望？我觉得我们为什么要造比 Usain Bolt 跑得还快的机器？是啊。干得好，所有 humanoid 参赛者。

57:31

running. When you see a robot running that fast, do you think like one of these days something

跑步。当你看到一个机器人跑得那么快，你会不会想，总有一天会不会有东西

57:36

like that is going to chase me down a dark alley and I've got no hope? I think why are we building

像那样在黑暗的小巷里追我，而我毫无希望？我觉得我们为什么要建造

57:40

the machines that run faster than Usain Bolt? Yeah. Well done to all the competitors in the humanoid

跑得比Usain Bolt还快的机器？是啊。做得好，所有参加humanoid的选手们

57:44

games, at least the ones who survived. It's not generating. All right, Casey, here we are. Final item.

游戏，至少是那些活下来的。

57:50

The bottom of the barrel, the bottom of the hat. I'm getting weirdly emotional. We've done so many of

它不是 generating 的。

57:57

these. This one comes to us from Alex Heath who wrote the cover story in this week's edition of

好了，Casey，我们到了。

58:04

Time magazine called Inside Open AI's reboot. He talked to a bunch of different executives and

最后一项。

58:11

leaders at Open AI who gave an update on their path to AGI. According to Heath, Chief Research

桶底的东西，帽底的东西。

58:18

Officer Mark Chen estimated that Open AI is 80% of the way to AGI. Sam Altman told him that Open AI

我莫名其妙地有点动情了。

58:26

was not quite yet at AGI, but that by the end of the year, the company would have an internal

我们已经做了好多期了。

58:33

system that he would call AGI. Casey, what do you make of this? I mean, my instinct is to be

这一个来自 Alex Heath，他为本期《Time》杂志撰写了封面故事，叫《Inside Open AI's reboot》。

58:41

skeptical, but when they say we're about 80% of the way there, that basically sounds right to me.

我持怀疑态度，但当他们说我们大概走了80%的路程时，我觉得基本上没错。

58:49

I don't know. It sound off in the comments if you disagree, but we are at a level where

我不知道。如果你不同意，就在评论里说出来吧，但我们现在已经达到了一个水平，

58:53

the computers use themselves and you can just type what you want into a box and more often than

电脑自己会使用自己，你只要把想要的东西打进一个框里，而且现在大多数时候

58:58

not these days. You'll get it. So I'm basically saying, yeah, that sounds about right to me. What do you

都能得到结果。所以我基本上是说，是的，这听起来差不多对。你觉得呢？

59:03

think? I mean, I've thought a lot about this question of like what AGI even is and how you know

我的意思是，我思考了很多关于AGI到底是什么、你如何知道

59:10

when you've reached it and what it means to you and what it means to me. I mean, this is literally

你已经达到了它、它对你意味着什么、对我意味着什么。我是说，这实际上就是

59:14

the sort of title question of my book that I spent the last year researching and where I came down

我这本书标题里的那个问题，我过去一年都在研究它，而我的结论

59:21

on this is basically that by any pre-2022 definition of AGI, AGI is here. Like if you took

基本上是，按照任何2022年之前的AGI定义，AGI已经在这里了。就像如果你拿

59:31

any of today's models, if you took Fable 5 or GPT 5.6 back in a time machine to 2017 and showed

现在的任何模型，如果你用时间机器把 Fable 5 或 GPT 5.6 带回到 2017 年，给当时搞 AI 的人看，他们肯定会说，嗯，没错，这显然就是 AGI。

59:40

them to the people who are building AI at the time, they would have said, well, yes, this is obviously

我觉得随着它越来越接近，系统越来越强，我们的目标线一直在移动，期望也变高了。

59:46

AGI. And I think as it's gotten closer, as the systems have gotten better, our goal posts have

现在嘛，你知道，每个人对 AGI 是什么、什么时候算真正达到了，都有自己的标准。

59:52

shifted and our expectations have been raised. And now, you know, everyone sort of has their own

但基本上，我们早就跨过了那个门槛——哪怕是十年前的计算机科学家也会觉得这符合 AGI 的定义。

59:57

different bar for what AGI is or when we should consider that we've achieved it. But basically,

嗯，我的标准是——一个系统是否足够先进，能造出时间机器，让你把 Fable 5 带回 2021 年给一位科学家看。

60:03

we have already long ago reached the threshold that computer scientists from even a decade ago would

我们早就达到了这样一个门槛，即使是十年前的计算机科学家也会

60:08

have said meets the definition of AGI. Well, my bar is can is a system sufficiently advanced that it

说这符合AGI的定义。嗯，我的标准是：一个系统是否足够先进，以至于它

60:15

can build a time machine so that you could take Fable 5 back to 2021 and show it to a scientist

能够建造一台时间机器，这样你可以把Fable 5带回2021年，给一位科学家看

60:21

and ask them is this AGI. So that's my bar. Okay, we'll get there. Maybe not by the end of the year,

然后问他们，这是不是 AGI。所以这就是我的标准。好吧，我们会达到的。也许不是今年年底之前，

60:26

but we'll get there 80% of the way there. Well, I think that they are sincerely convinced that

但我们会走完 80% 的路程。嗯，我认为他们确实由衷地相信

60:32

AGI is 80% of the way there. There are other companies that may take longer to announce that they

AGI 已经走完了 80% 的路程。还有其他公司可能需要更长时间才能宣布他们

60:37

have created AGI, but I think this is, I think it is a big deal that they are now talking about AGI

已经创造了 AGI，但我认为这——我觉得他们现在谈论 AGI 这件事很重要，

60:43

in the present tense as something that is more here than not. And I think it'll just be sort of a

用现在时态，把它描述成一种“已经到来”而非“还没到来”的东西。而且我认为这会是某种

60:48

marketing decision when they want to officially claim that they've gotten there. This is such an

营销决策，看他们什么时候想正式宣布已经达到了。这是一个非常

60:52

arbitrary threshold impossible to test for. The other thing that I'm certain about is that

主观武断、无法检验的门槛。我还确定的一件事是，

60:59

if and when Sam Alden does come out and say we have built AGI, there will be many, many people who

如果 Sam Alden 真的站出来说我们已经构建了 AGI，那将会有很多很多人

61:05

argue with him and say, no, you didn't. It still can't do X, Y and Z. It hasn't proved the

跟他争辩说，不，你没有。它还是做不到X、Y和Z。它没有证明那个假设。

61:10

hypothesis. It hasn't, you know, it still makes mistakes. I don't think there will ever be a sort

它没有，你知道，它还是会犯错。我觉得关于AGI是什么或者我们什么时候能实现AGI，永远不会有共识。

61:15

of consensus on what AGI is or when we have gotten there. But I think the people at OpenAI are

但我认为OpenAI的人对我们取得了多少进步这件事，对多于错。

61:21

more right than wrong about how much progress we've been making. Well, Kevin, with that, I'm going

好了，Kevin，既然如此，我请你停止生成。向你致敬。

61:27

to ask you to stop generating hats off to you. That's off to me and hats off to chat GPT.

那是向我和chat GPT致敬。

61:33

That was at GPT. That was at GPT. Wrong sound effect.

那是at GPT。那是at GPT。音效放错了。

62:19

Wewing Tam, Brooke Mentors, and Talia Hadad. You can email us at HartFork at nytime.com

Wewing Tam、Brooke Mentors和Talia Hadad。你可以给我们发邮件到HartFork at nytime.com。

62:26

with your time in the humanoid robotic games.

分享你在humanoid robotic games中的经历。

Meta Shifts the Blame + Do Data Center Bans Work? + The Final HatGPT

Play Queue