本集简介
双语字幕
仅展示文本字幕，不包含中文音频；想边听边看，请使用 Bayt 播客 App。
这是iHeart播客，保证真人发声。
This is an iHeart podcast, Guaranteed Human.
如果你的书架和你的推荐页对你来说同样重要，那你来对地方了。
If your bookshelf and your for you page are equally important to your personality, welcome home.
《Pro Society》是一档每周更新的播客，既有读书会的味道，又有群聊的感觉，适合那些认为《傲慢与偏见》和《爱情岛》值得同等讨论的人。
Pro Society is a weekly podcast that's part book club, part group chat for anyone who thinks Pride and Prejudice and Love Island deserve the same level of discourse.
每周，我们都会把书籍、网络和流行文化串联起来，邀请你喜欢的作家、读书博主，还有一堆想太多的观点。
Each week, we're connecting the dots between books, the Internet, and pop culture with your favorite writers, book talk creators, and plenty of overthought opinions.
对。
Yeah.
我迷上了。
I'm obsessed.
我迷上了。
I'm obsessed.
在iHeartRadio应用、Apple播客或任何你收听播客的地方，收听《Pro Society》。
Listen to pro society on the iHeartRadio app, Apple podcasts, or wherever you get your podcasts.
在新播客《Solita》里，我们分享女性独自旅行时那些真实又混乱的经历。
On the new podcast, Solita, we share the messy reality of traveling alone as a woman.
我可以等四个小时坐下一班公交车，或者让这个陌生男人骑摩托车载我一程。
I can wait four hours for the next bus, or this random dude is offering me a ride on his motorcycle.
我选了第二个选项。
I chose option b.
我是朱莉·皮涅罗，我独自旅行，因为那是一个我不用先问别人就能说“好”的难得空间。
I'm Julie Pinheiro, and I travel by myself because it's a rare space where I can say yes without asking anyone else first.
我的使命是重新定义“solita”这个词，把同情换成无限可能。
I'm on a mission to reclaim the word solita, trading the pity for possibility.
在iHeartRadio应用、Apple播客或任何你收听播客的地方，收听《Solita》。
Listen to solita on the iHeartRadio app, Apple Podcasts, or wherever you get your podcasts.
嘿。
Hey.
我是海斯·达文波特。
This is Hayes Davenport.
还有肖恩·克莱门茨。
And Sean Clements.
我们主持的播客叫《好莱坞手册》。
We host the podcast Hollywood Handbook.
每周我们都会和一位娱乐圈人士聊天，试着帮他们发展事业，也看看他们能给我们带来什么。
Each week, we talk to someone in show business and try to help them with their careers and see what they have to offer us.
大家都玩得很开心，而且没人因为经纪人让他们上我们的节目而生气。
Everyone has a good time, and no one gets mad at their publicists for letting them do our show.
我们请过很多大牌嘉宾，比如莎拉·谢尔曼、亚当·斯科特、丹尼·麦克布莱德、本·斯蒂勒，还有很多其他大人物，他们能有今天全靠我们。
We've had a lot of great guests like Sarah Sherman, Adam Scott, Danny McBride, Ben Stiller, and a lot of other big shots that wouldn't be where they are without us.
你可以在iHeartRadio应用、Apple Podcasts或者任何你收听播客的地方收听《好莱坞手册》。
Listen to Hollywood Handbook on the iHeartRadio app, Apple Podcasts, or wherever you get your podcasts.
我是霍达·科顿比，播客《霍达·科顿比的快乐101》的主持人。
I'm Hoda Cottonbee, host of the podcast Joy one zero one with Hoda Cottonbee.
你知道，我一直在寻找灵感和有用的工具来最大化快乐。
You know, I'm always searching for inspiration and useful tools to maximize joy.
比如我的朋友亨利·温克勒给那些想当演员的人，还有那些希望脱颖而出的人提建议的时候。
Like when my friend Henry Winkler offered advice to aspiring actors and really anybody hoping to stand out.
你被聘用，不只是为了填补时间和空间。
You're not hired just to fill time and space.
你被聘用，是为了用你自己来填满那个空间。
You're hired to fill the space with you.
在iHeartRadio应用、Apple Podcasts或任何你收听播客的地方，收听Hodakotbi的《Joy 101》。
Listen to Joy one zero one with Hodakotbi on the iHeartRadio app, Apple Podcasts, or wherever you get your podcasts.
Call Zone Media。
Call Zone Media.
你好，欢迎收听《Better Offline》。
Hello, welcome to Better Offline.
天哪。
Goddamn.
我是你的主持人，Ed Zitron。
It's your host, Ed Zitron.
我回来了。
I am back.
Better Offline。
Better Offline.
今天我们请回了节目的老朋友，GamersNexus的Steve Burke，来聊聊半导体公司在AI繁荣时期的表现如何。
And we are returning with a returning champion of the show, Steve Burke of GamersNexus, who's here to talk to us about how well the semiconductor companies are conducting themselves in this era of AI opulence.
我觉得这么说挺合适的。
I think it's a good way of putting it.
对，没错。
Oh, yeah.
对他们挺满意的。
Happy with them.
哦，他们啊，他们很棒。
Oh, well, they're oh, they're great.
他们做得太好了。
They're so good.
小丑行业从来没有这么多候选人。
The clown industry has never had more candidates.
这太奇怪了。
It's so strange.
所以你发布了那个精彩的视频，关于那个，嗯，
So you put out this wonderful video about the well, it's
就是，现在外面已经没有不是小丑的人了。
just we're out there are no more non clowns.
我觉得，这，这，这，就是，挺搞笑的。
I it it it's just it's it's funny.
不。
No.
现在小丑车里已经没有不是小丑的人了。
There are no non clowns in the clown car at this point.
我觉得这很准确。
I think it's accurate.
在我们录制这期节目的时候，他们最近宣布OpenAI和SB Energy要建一个数据中心，英伟达在后面兜底，但其实也不算真正的兜底。
As we record this, open they've recently announced that OpenAI and SB Energy are building a data center with NVIDIA backstopping it, but it's not really a backstop.
史蒂夫，这感觉不像是一个真实的行业。
This doesn't feel like a real industry, Steve.
这一切都是假的，都是装出来的。
It's all it's all pretend and make believe,
我觉得，到这个时候。
I think, this point.
我是说，是的，这是我最近听到的最喜欢的一句话。我今天确实看到了那条新闻，但我最喜欢的是吉姆·克莱默说的，我太喜欢数据了，太喜欢数据中心了。
I mean, yeah, it's my favorite quote I've heard recently, I did see that news today, but my favorite quote was Jim Kramer saying, I like data so much data center so much.
我真希望自己能住在一个数据中心里。
I wish I could live in one.
我应该让他住进去。
I should make him.
对。
Yeah.
你是不是上过CNBC，还是那是……
You should you've weren't you on CNBC, or was that
我最近上过几次CNBC，他们还没问过我关于吉姆·克莱默住数据中心的事。
I was on CNBC a few times recently, and they've not asked me about Jim Kramer living in a data center yet.
没有。
No.
他们还没问过？
They they haven't?
对。
Okay.
我以为他可能在找房产中介呢。
I thought maybe he'd looking for a realtor.
我希望他真的去住一个，这样他就能坐在里面被吵聋，我猜，因为据我所知，数据中心挺吵的。
I want him to go and live in one just so that he can sit around and just get deafened, I assume, because they're pretty loud from my experience.
是的。
Yeah.
我是说，现在事情太多了，因为本地有个数据中心市政会议，本来定在明天开的，我本来打算去参加的。
I mean, it's there was there's so much going on right now because, locally, we had a data center town council meeting that was supposed to happen actually tomorrow, and I was gonna go to that.
当地居民的反对太强烈了，结果申请方撤回了他们的数据中心申请。
And the the pushback from locals was so much that they the applicant withdrew their application for the data center.
这是在哪里？
Where was this?
在罗利。
This was in Raleigh.
具体是罗利东南部，他们原本看中的地方。
Southeast Raleigh specifically is where they were looking at.
因为那边已经有一些数据中心了。
Because there are few data centers over there.
对吧？
Right?
我知道Core Weave在那里有些设施。
I know Core Weave have some crap.
是的。
Yep.
Core Weave，他们在西部山区那边有不少设施，而且那边还有一些以前挖矿的老厂房，你知道的，有些已经改造成数据中心了，或者正在改造中。
Core Weave, they they have a good amount of stuff out near the mountains out west, and there's some of the one of the old, like, former crypto mining facilities, you know, has been converted or in the process of being converted into a data center as well out there.
所以，是的，我知道他们到处都在建。
So but, yeah, I know they're they're popping up everywhere.
看来终于，我想，公众对数据中心建起来意味着什么有了更广泛的认识。
It's just it seems like finally, I guess, there's a more widespread public understanding of what it means when a data center pops up.
而且，我和团队里的某个人聊过这个。
And it's I was talking with someone on the team about this.
数据中心早就存在了。
Data centers have existed forever.
对吧？
Right?
是的。
And it's Yeah.
有意思。
Interesting.
确实很多。
A lot of them.
对。
Yeah.
而且，就是这种专门用于AI处理的数据中心，确实名声不太好。
And they it's just like this particular type of data center for AI processing, it rightly gets a much worse wrap.
你知道，它耗电量大得多。
And it's, you know, it's far more power consumption.
通常耗水量也更大，占用的土地面积也可能更多。
It's more typically more water consumption and potentially more land area.
我觉得，它们把其他那些一直存在的数据中心的名声也给搞坏了。
And it's they've they've sort they've sort of ruined it for the rest of the data centers that have been around since the beginning of time, I feel like.
不过，你知道，它们都是大型建筑。
But, you know, they're all big structures.
就是这些新的，尤其是那些AI处理方案，通常更吵，也更耗电。
It's just these new ones, and particularly, the AI processing solutions tend to be louder and more power hungry.
所以，我还是很高兴看到大家终于对这些事情有所了解，并且开始关注了。
And so I I'm just happy to see, though, that people are, you know, are are finally informed on this stuff and showing up.
我挺喜欢那些政客们的反应，他们直接说，不。
I do like the amount of politicians who are like, no.
这对就业有好处。
It's good for jobs.
这对就业没好处。
It isn't good for jobs.
这对就业一点好处都没有。
It's not good for jobs at all.
哦，这对当地经济有好处。
Oh, it's good for the local economy.
只要数据中心还在建，对当地经济就有好处，但一旦建完，好处就没了。
It's good for the local economy as long as the data center is being built, and then that stops.
而且即便如此，实际的经济到底有多好呢？
And even then, it's how much how good is the actual the economy?
我们其实不太清楚。
We don't really know.
对，就是这样。
And it's Yeah.
看着这一切发生真的很让人沮丧，因为感觉数据中心相关的事情好像把一切都带坏了。
It's frustrating to watch this happen because it feels like something about data centers is just corrupting everything.
比如，它带坏了——我记得今天早些时候看你视频，看Lisa Su谈论这个，就是AMD的Lisa Su，在这之前我真的会觉得，她是个好CEO。
Like, it corrupted I remember watching your video earlier today, watching Lisa Sue talk about this, like watching Lisa Su of AMD, who was before this, I would have genuinely been like, that's one of the good CEOs.
她很聪明。
She's a smart one.
她不会屈服于营销炒作。
She doesn't give in to marketing hype.
她是个很实在的人。
She's no bullshit person.
听她讲这些事，感觉她就像吸了油漆一样。
Listening to her talk about this stuff, it's it's like she's been huffing paint.
对。
Yeah.
她确实挺受欢迎的。
She was she was pretty well liked.
我的意思是，这些年来我和AMD的经历有好有坏。
I mean, I my experiences with AMD over the years have been mixed.
但总的来说，AMD至少保持了表面上，如果不是实际上的话，那种以工程为先的公司形象。
But broadly speaking, AMD has at least maintained the appearance of, if not the actual, you know, practice of being sort of an engineering first company.
而且Lisa Su确实有工程背景。
And Lisa Sue technically comes from an engineering background.
我敢肯定她现在不怎么亲自做工程了，但这仍然算数。
I'm sure she doesn't do a lot of it these days, but that still counts for something.
是的。
And Yeah.
黄仁勋也是工程师出身，这点得公平说。
And so does Jensen, to be fair to him.
不过，丽莎在过去一年左右，经常和迈克尔·克拉齐奥斯一起在白宫亮相，讨论他们的AI行动计划。
But, yeah, Lisa, over the last year or so, has shared the stage a lot with Michael Kratios from the White House to talk about their AI action plan.
那个计划里，他们提到要扫清数据中心建设中的清洁水和清洁空气法规障碍。
And in that plan, they describe things like getting clean water and clean air regulations out of the way of data center construction.
好。
Good.
哦，好。
Oh, good.
对。
Yeah.
对。
Yeah.
我是说，这些内容要是写进新闻稿，公关人员看了都得愣一下。
I mean, the the things to put in your press release that would make a PR person go, wait a minute.
那空气清洁这部分，我们到底该怎么包装呢？
You're so the the clean air part, how are we gonna how are we gonna spin that exactly?
比如说，AI会制造更多清洁空气。
Like, well, AI will make more clean air.
对。
Right.
也许吧。
Maybe.
是的。
Yeah.
而且这真的挺搞笑的，因为重新看那个视频，我见过几个这样的片段，那些高管们听起来就像末日邪教的一员似的。
And it it's so funny as well because rewatching that video, I've seen a few of these clips of these various executives sounding like they're part of a doomsday cult or what have you.
而且看着那些两年前还在谈论推理、高带宽内存，还有AI GPU的实际技术特性的人，现在却只谈财务上的泛泛之谈，真的很奇怪。
And it's so strange watching people that even two years ago were like, yeah, and here's the we're talk they're talking about things like inference and high bandwidth memory and, like, the actual financial sorry, the actual technical features of AI GPUs to just speaking in financial generalities.
就像，这会是一个300亿、1000亿、250万亿规模的TAM行业。
It's like, will be a $30,000,000,000, 100,000,000,000, 250 bazillion dollar TAM industry.
它会变得非常大。
It will be so big.
它会变得超级大。
It will be so huge.
我们很喜欢。
We love it.
而且你提到了这一点。
And you made the point.
现在更多是关于金融工程，而不是电气工程了。
It's it's become more about financial engineering than electrical engineering.
对。
Yeah.
对。
Yeah.
整个事情感觉我们完全偏离了技术的本质。
The whole thing feels like we've completely lost the plot of any of the technology.
比如说，除了少数例外，我现在几乎听不到这些高管们在他们的各种播客里讲什么。顺便说一句，他们现在都特别喜欢上播客、上电视节目，露面次数比我见过的很多CEO都要多得多。
Like, I I with some exceptions, I almost never hear these executives now when they're doing their various podcasts, which by the way, they've all gotten really onto the the, like, podcast and TV show and a lot more appearances than I've really ever seen from a lot of the CEOs.
以前他们一般只在大型主题演讲时才出来，然后就会消失半年。
Normally, they've, you know, they've historically kinda only wheeled them out for the big keynotes, and then they disappear for six months.
所以这也挺不一样的。
So that's been different too.
但是，是的，在黄仁勋、苏姿丰还有其他所有人的所有露面中，尤其是山姆·奥特曼，你现在几乎听不到任何真正技术性的内容。
But but, yeah, it's out of all the appearances from Jensen and Lisa and everyone else, Sam Altman especially, you almost never hear anything actually technical now.
全都是多少亿美元、多少吉瓦，对话就这么开始也这么结束。
It's all how many billions of dollars or how many gigawatts, and that's kind of the start and the end of the conversation.
偶尔他们会说这会创造很多管道工和电工的工作，但同时又会回避这些工作最多只能持续几年的事实。
Occasionally, they'll talk about how it's gonna create a bunch of plumbing and electrician jobs while kind of, you know, skirting past the fact that those will last for a couple years at most.
然后，你知道的，建筑建好了，这些工作也就没了。
And then it's you know, the structure is built so those jobs go away.
但你说得对，是的。
But to your point, I yeah.
我觉得工程方面已经完全淡出了，我有点好奇，我对互联网时代的技术细节了解得不是特别深，比如它是怎么一步步建起来的。
I think it's the engineering side has completely faded, which I I wonder if I'm I'm not that that well informed on sort of the .com era technicals, like how it actually built up.
当然，我知道发生了什么。
Obviously, I'm aware of what happened.
但我在想，如果你回头看那时候，你会不会看到类似的轨迹？
But I wonder if you look back then, you know, do you see a similar trajectory?
是不是一开始扎根于技术，然后技术慢慢淡出，最后就只剩下纯粹的金钱？
Is it like it starts rooted in technicals and then it that that falls away over time and it becomes just pure money?
还是说这次不一样，因为讨论的
Or is this different the way it's
方式不同？
being discussed?
回头看，互联网早期太原始了，几乎每篇文章都在重新解释互联网是什么。
Looking back, the early days of the Internet were so vestigial that almost every article was basically reexplaining what the Internet was.
他们谈论的方式都是些泛泛之谈，比如，好吧，有这个，那个，就大概像，对，互联网，你知道，它会变得挺大的。
And the way they talked about it was in these vast generalities of like, okay, well, so there's this, that, like, just vaguely like, yeah, the Internet, you know, it's pretty it's gonna be pretty big.
或者他们会展开一些很大的伦理讨论。
Or they'd get into very big ethical discussions.
而且最初对互联网的很多反对声音是，嘿，我们就要创造一个巨大的未经核实信息的沼泽。
And a lot of the original pushback on the Internet was, hey, we're just gonna create a giant swamp of unverified information.
这是问题吗？
Is this a problem?
而人们真的会说，闭嘴，你这个白痴。
And people were genuinely like, shut up, you fucking idiot.
那永远不会发生。
That'll never happen.
这太搞笑了。
It was it's so funny.
当时对互联网的反对声音其实挺聪明的。
The pushback on the Internet back then was actually quite smart.
但大多数情况下，人们会说，对，这可能会变成任何东西。
But for the most part, people were like, yeah, this could be anything.
但是当你读到像Windstar这样的公司的文章时，那是一家电信公司，基本上处于一种循环关系中，它从朗讯借钱来支付给朗讯，然后用那笔钱来偿还朗讯的贷款，而他们正是用那笔贷款来支付朗讯的。
But when you read the articles about like Windstar, which was a telecoms company that basically was in a circular relationship where it borrowed money from Lucent to pay Lucent and then used that money to pay Lucent back for the loan that they were using to pay Lucent.
当你读到他们的文章时，有个有趣的事实，我唯一一次能找到有人提到年化收入的地方，就是在泡沫时期。
And when you read their articles fun fact, the only time I've ever been able to find a bubble where someone mentioned annualized revenues.
哦，唯一另一次是
Oh, the only other
唯一另一次也是同样的胡扯。
only other time is the same bullshit.
但他们会说，是的，他们会用这些天线覆盖整个国家。
But it would be like, yeah, and they're gonna blanket the entire country with these dishes.
它们会无处不在。
They're gonna be everywhere.
这会是那种非常夸张、数字非常大，但技术细节非常少的东西。
It's gonna be the it was very highfalutin, very big numbers, but very very light on the technical details.
不过比现在还是要详细一些。
Still heavier than it is now.
有意思的是，如果我们从三万英尺的高空来看，摘下我的批评眼镜，戴上我用的那种乐观眼镜，现在我们看到的是黄色和蓝色。
Because the funny thing is is if we think about this 30,000 foot view taking away my critic glasses, putting on the the booster glasses I use, now we can see is yellow and blue.
我觉得，这不正是技术最重要的时候吗？
The thing that I think is, isn't this the time when the technology is the most important?
因为他们还没把这个做成盈利的。
Because they haven't made this profitable yet.
确实还没有。
They really haven't.
很明显，他们还没有。
Like, it's very obvious they haven't.
而且，每次我读到像维拉·鲁宾这样的项目，比如下一代GPS维拉·鲁宾，我都会想。
And yet, like, the I keep every time I read about, like, Vera Rubin, for example, next generation GPS Vera Rubin.
对吧？
Right?
每次读到这些，我都会看到同样的东西，就是，他们效率提高了10倍或者25倍。
Every time I read about them, I get the same shit, which is, yeah, they're 10 x more efficient or 25 x.
而且它们在DeepSeek r1上效率提高了25倍，那还是一年半前的模型。
And also they're 25 x more efficient on DeepSeek r one, a year and a half old model.
对。
Right.
而且，是的，你不能说，我们还没做过基准测试。
And, yeah, you can't like, we haven't benchmarked that.
我们没法跟Anthropic或OpenAI做基准测试，因为如果做了，你可能会发现他们不赚钱。
It's not like we could benchmark with Anthropic or OpenAI because if we did that, you might find out they're unprofitable.
每次你都是这样，就像在一段糟糕的关系里，你不想问那些你知道会结束关系的问题。
And it's just every time you it's like a it's like being in a bad relationship where you don't wanna ask the questions that you know would end it.
你有没有注意到最近这种双重说法，因为你刚才在说，你知道，性能会是那个的10倍。
Do you know have you noticed the sort of dual narrative lately where because you were just talking about how you know, it'll be it'll be 10 x the performance of whatever.
所以那里有一个性能数字。
So there's a performance number there.
但与此同时，我不记得是不是Core Weaver，就在上周。
But then at the same time, I don't remember if it was Core Weaver who it was just last week.
我看到一个采访，有个高管在说，以前每GPU小时的成本是两美元，现在那些设备要四美元了。
I saw a an interview with some executive who was talking about the cost per GPU hour being two dollars previously, and now those devices are $4.
不知道你有没有印象。
I don't know if this rings a bell to you.
但是
But
对。
Yeah.
对。
Yeah.
对。
Yeah.
不。
No.
我记得这个。
I I remember this.
对。
Yeah.
所以那个采访构建的叙事就是，你看，即使是旧的、相对古老的技术，也就是三到六年前的技术，现在反而更贵了，或者现在更有效了，而且仍然有用。
And so the the narrative that was being constructed by that interview was one of look, even the old, the comparatively ancient technology from somewhere between, you know, three and six years ago is more expensive now or is more effective now and is still useful.
然后，显然，另一方面，你又有那种错失恐惧症，觉得你必须拥有最新的Vera Rubin，不然你还在干什么？
And then, obviously, on the other side of it, you have the FOMO of you need the absolute newest Vera Rubin or what the fuck are you even doing?
你知道吗？
You know?
对。
Right.
所以这就像是GPU领域里的鱼和熊掌兼得的情况。
So it's it's just it's the have your cake and eat it too set up on the GPU side.
但在消费者这边，也就是最终用户这边，我看到那个“我们的旧GPU比以往任何时候都值钱”的讨论，感觉就是，哦，这很糟糕。
But on the consumer side, like the end user side, all I saw with that, our old GPUs are worth more than ever conversation was like, oh, this is bad.
这就像，因为现在你，你，我觉得，是消费者定价的最后希望了，好吧。
This is like because now now you're just you're the last bastion of hope, I think, for consumer pricing was, okay.
好吧，也许我们能捡到上一代的边角料，现在他们连这个都要抢。
Well, maybe we'll get the scraps from last generation, and now they're saying they even need that.
其实呢，这里面有几个很有意思的点。
So Well, the thing is, there's a few wonderful parts of this.
第一，我看到上面说，你每小时不是拿两个多Black Quell，而是四个多。
Number one is I saw that it was saying there was like you're getting instead of getting two something per Black Quell, you're getting 4 something an hour.
对吧？
Right?
对。
Right.
这根本不够覆盖成本。
That is not enough to cover the costs.
差得远呢，远远不够。
That's not even fucking close enough.
我有个Oracle的熟人，他跟我说，这个人是有实际经验和信息的。
I have an Oracle contact who told me that and this is someone with actual info experience.
他们说，光运行一个GB 300，每小时运营成本就要6美元左右。
They were saying that it costs like 6 something dollars an hour in OPEX just to run a GB 300.
哇。
Wow.
所以我觉得每小时4美元根本不够。
So I don't think $4 an hour is gonna do it.
这就是问题所在。
And this is the story.
这个话题也没人愿意多谈。
No one really wants to talk about this one as well.
就是说，当你把现货价格卖给那些急需的人时，你能赚更多钱。
It's like, yeah, when you're selling spot pricing to the desperate, you can make more money.
但说到长期合同，说到更长的合同期限，成本就会下降。
But when it comes to the longer ones, when it comes to the longer duration contracts, that cost that that the cost drops.
而且降得非常厉害。
It drops dramatically.
甲骨文给OpenAI的合同里，每小时收2美元多一点的H100和H200 GPU费用。
Oracle charging 2 something an hour per h 100 GPU and h 200 GPU for OpenAI with their contract.
再说一次，没人真的想谈这个，因为如果你每小时只收2到4美元，那好吧。
And not again, no one really wants to talk about this because it's like, if you're getting 2 to $4 an hour for this shit, okay.
所以如果你一直保持满负荷使用，也许十年才能回本。
So maybe you'll break even in ten years if you have constant utilization.
对。
Right.
如果那能行的话。
If that works.
就算那样，天哪，你的毛利率才30%，兄弟。
And even then it's like, wow, you have a gross margin of 30%, mate.
你想开航空公司吗？
You you want to start an airline?
这他妈的怎么回事？
Like like, what the fuck?
这真算不上什么正经生意。
This isn't really much of a business.
但接着你又听到苏姿丰、黄仁勋或者山姆·奥特曼那些人说的话。
But then you go and you hear these things that Lisa Su or Jensen Huang or Sam Altman are saying.
就好像，没错，下一次工业革命，规模会非常庞大。
It's like, yes, the next industrial revolution, it's gonna be huge.
你会赚很多钱。
You're gonna make so much money.
每个人都会变得富有。
Everyone's gonna be rich.
每个人都会超级有钱。
Everyone's super rich.
每个人都去买GPU，买七个，能买多少买多少。
Everyone's buy a GPU, buy seven of them, buy as many as you can.
我们得把它们卖出去。
We need to sell them.
有时候感觉就像生活在另一个世界一样。
And it feels like living in an alternate reality sometimes.
因为
Because
对我来说，这个数字并没有那么夸张，你知道吗？
the math isn't that crazy to me, you know?
嗯，我觉得这也是政府在这方面的投资，或者说政府投资在不断增加，或者政府里的个人投资，不是在一个……
Well, I I think it's also the government investment in all of this or the increasing government investment or the individuals in the government invest in not in a
哦，对。
Oh, yeah.
产能。
Capacity.
我是说，特朗普对英伟达的投资。
I mean, well, the Trump's investment in Nvidia.
对吧？
Right?
对。
Yeah.
数百万美元的交易。
Millions of dollars transacted.
对。
Yeah.
如果你的书架和你的“为你推荐”页面对你的人格同样重要，欢迎回家。
If your bookshelf and your For You page are equally important to your personality, welcome home.
这里是《散文社》，每周一期的播客，既是读书俱乐部，又是思想女儿、流行文化迷以及那些认为《傲慢与偏见》和《爱情岛》值得同等讨论度的人的群聊。
This is Prose Society, the weekly podcast that's part book club, part group chat for thought daughters, pop culture obsessives, and anyone who thinks Pride and Prejudice and Love Island deserve the same level of discourse.
我是Eli Rallow，每周我们都会把书籍、互联网和每个人都在热议的话题串联起来。
I'm Eli Rallow, and every week, we're connecting the dots between books, the Internet, and the conversations everyone can't stop having.
我得去查一下
I'm gonna have to look up
这个故事。
this story.
我完全着迷了。
I'm obsessed.
我完全着迷了。
I'm obsessed.
从畅销书作家和你最喜欢的书籍讨论创作者，到最新的流行文化热点，什么都可以聊。
From bestselling authors and your favorite book talk creators to the latest pop culture moments, nothing is off the table.
就像是
It's like
如果你能把真实的信息藏在引人入胜的角色里，那这就是特洛伊木马。
if you can hide some real messages inside compelling characters, and that is a Trojan horse.
无论你是在寻找深度的文学探讨、聪明的流行文化对话，还是一个喜欢想太多的读者社区，你都来对地方了。
Whether you're looking for literary deep dives, smart pop culture conversations, or a community of readers who love to think a little too much, you're in the right place.
在iHeartRadio应用、Apple Podcasts或任何你收听播客的地方，收听Prose Society。
Listen to pro society on the iHeartRadio app, Apple Podcasts, or wherever you get your podcasts.
我们书中见。
See you between the pages.
在新播客《Solita》里，我们分享女性独自旅行的真实混乱。
On the new podcast, Solita, we share the messy reality of traveling alone as a woman.
我可以等四个小时坐下一班公交，或者这个陌生男人让我搭他的摩托车。
I can wait four hours for the next bus, or this random dude is offering me a ride on his motorcycle.
我选了第二个选项。
I chose option b.
我是朱莉·皮涅罗，我独自旅行，因为那是一个我可以不用先问别人就说是的稀有空间。
I'm Julie Pinheiro, and I travel by myself because it's a rare space where I can say yes without asking anyone else first.
我的使命是重新定义“solita”这个词，把同情变成可能性。
I'm on a mission to reclaim the word solita, trading the pity for possibility.
每次我试着
Every time I tried to
独处，却总是遇到人，而且
be alone, I kept meeting people, and
他们就说，你对我们笑了。
they were like, you smiled at us.
这里很少有人会微笑。
Not a lot of people smile around here.
当你独自一人时，你最容易感受到真实的世界，而不是那些被灌输的谎言。
It's when you're alone that you're most receptive to the world as it is and not the lies you're sold about it.
这段时间你可以挑战自己的极限，改变想法，或者发现一个全新的自己。
It can be a time where you push your limits, change your mind, or wake up to a new version of yourself.
所以无论你是独自旅行的老手，还是紧张得不敢预订第一次旅行，我都希望你能在iHeartRadio应用、Apple Podcasts或任何你收听播客的地方收听《Solita》。
So whether you're a solo travel veteran or you're too nervous to book your first trip, I hope you listen to Solita on the iHeartRadio app, Apple Podcasts, or wherever you get your podcasts.
嘿。
Hey.
我是海斯·达文波特。
This is Hayes Davenport.
我是肖恩·克莱门茨。
And Sean Clements.
我们主持Big Money Players网络上的播客《好莱坞手册》。
We host the podcast Hollywood Handbook on the Big Money Players Network.
我们的节目对第一次听的听众来说非常友好。
Our show is extremely accessible to first time listeners.
每周我们都会和一位演艺圈人士，也就是所谓的“圈内人”聊天，试着帮他们发展事业，看看他们能给我们带来什么。
Each week, we talk to someone in show business, aka The Biz, and try to help them with their careers and see what they have to offer us.
每个人都玩得很开心，没有人会因为经纪人让他们上我们的节目而生气。
Everyone has a good time, and no one gets mad at their publicist for letting them do our show.
我们有很多
We've had a lot of
很棒的嘉宾，比如莎拉·谢尔曼。
great guests like Sarah Sherman.
也没那么棒。
Not that great.
其实，你能帮我
Actually, can you help
选一下吗？你能帮我弄一下这个造型吗？
me pick can can you help me shape this?
对。
Yeah.
要不我们反过来
How about we flip
把它倒过来？
it on his head?
不。
No.
行吗？
K?
要不我们换个角度？
How about we flip this around?
行吗？
K?
那如果那个你以为很正常的惊喜女嘉宾，其实才是表现古怪的那个人呢？
How what if the surprise woman who actually you think is the straight man is actually the one who's acting weird?
亚当·斯科特。
Adam Scott.
这段时间你一直以为亚当·桑德勒会来上节目。
This entire time, you've been expecting Adam Sandler to come to visit the show.
我绝不会接受。
I I would never accept.
放弃那个想法吧。
Go of that.
我有，是的。
I have yeah.
我绝不会对那种事抱有任何期待。
I would never set expectations for something like that to happen.
如果宇宙允许那样的事情发生，是的。
If the universe allows for something like that to happen Yeah.
我愿意接受
I'm open to
对。
it.
我永远都会接受的。
I'm always going to accept.
你知道吗？
You know?
丹尼·麦克布莱德，可以。
Danny McBride, fine.
是的。
Yeah.
看到了这种脆弱的男子气概。
Saw this fragile masculinity.
嗯。
Mhmm.
散发出来。
Exuding.
说实话，我走在走廊上就能闻到那股味儿。
I could smell it walking down the hallway, to be honest with
你。
you.
嗯。
Mhmm.
我当时就想，那味儿真冲。
I was like, that Smelled.
我还以为是大麻呢。
I was like, is that weed?
嗯。
Mhmm.
不是。
Nope.
是脆弱的男子气概。
It's fragile masculinity.
对。
Mhmm.
是的。
Yep.
本·斯蒂勒，不是。
Ben Stiller, no.
因为很多时候，在
Because a lot of times in out
脱离上下文的情况下，人们不理解，甚至不知道圈子是什么。
of context, people don't get it or even know what a circle is.
你明白我的意思吗？
You know what I mean?
就像，就像，
Like, like,
哦，那是什么状态？
oh, what's the state?
展开剩余字幕（还有 480 条）
大概就是这样。
It's around
除了你，其他人都很蠢。
Everyone's so dumb except for
你。
you.
假的。
Fake.
对。
Right.
假的，而且也不感兴趣，或者就像，哦，我有我的烟斗。
Fake and and also not interested or just like, oh, I've got my pipe.
我不在乎你的生活。
I don't care about your life.
还有很多其他大人物，没有我们他们也不会有今天的地位。
And a lot of other big shots that wouldn't be where they are without us.
哦，顺便说一句，威尔·法瑞尔亲自告诉我们他超爱这个节目，还想每一集都来当嘉宾，但他就是
Oh, and by the way, Will Ferrell told us personally he loves the show and he wants to be a guest on every episode, but he's
太忙了。
just so busy.
在iHeartRadio应用、苹果播客或者任何你收听播客的地方，收听《好莱坞手册》。
Listen to Hollywood Handbook on the iHeartRadio app, Apple Podcasts, or wherever you get your podcasts.
我是霍达·科顿比，《霍达·科顿比的快乐101》播客的主持人。
I'm Hoda Cottonbee, host of the podcast Joy one zero one with Hoda Cottonbee.
了解我的人都知道，我一直在寻找灵感和有用的工具来最大化快乐。
If you know me, you know I'm always searching for inspiration and useful tools to maximize joy.
这个播客让我们通过和有趣的人进行有意义的对话，一起发现这些。
This podcast lets us uncover those things together through meaningful conversations with fascinating people.
比如我的朋友亨利·温克勒给那些想当演员的人，还有任何想脱颖而出的人的建议。
Like when my friend Henry Winkler offered advice to aspiring actors and really anybody hoping to stand out.
你被聘用不是为了填补时间和空间，而是用你自己来填满那个空间。你肚子里有货，就跟着感觉走。
You're not hired just to fill time into space you're hired to fill the space with you you have something in your guts you go with it.
肖恩·约翰逊回忆了她意识到现在的丈夫安德鲁就是对的那个人那一刻。
And Shawn Johnson recalled the moment she realized her now husband, Andrew, was the one.
他完全做自己，毫不掩饰，对好莱坞一无所知。
He was just so unapologetically himself and knew nothing of Hollywood.
他之前从没单独坐过飞机。
He hadn't been on a plane before without his family.
那太纯粹了。
It was so pure.
在iHeartRadio应用、Apple Podcasts或任何你收听播客的地方收听《Joy 101 with Hoda Kotb》。
Listen to Joy one zero one with Hodakotbi on the iHeartRadio app, Apple Podcasts, or wherever you get your podcasts.
今天早些时候我和团队里的一个人聊天，他提到，他问我，你知道，现在有这么多数据中心，存储需求也这么大，如果普通人用ChatGPT或Claude最多就是做个表格或者当个聊天机器人，至少，那并不需要那么多处理能力和存储。
I was talking with someone on the team earlier today, and he made the comment he he was asking me, you know, if there's all these data centers and this massive demand for storage now as well, If if the average person is using ChatGPT or Claude to make a spreadsheet at most or be a chatbot, you know, at least, like, that doesn't require all that processing power and storage.
那到底是为了什么呢？
And so what is it?
然后对我来说，我不知道这是不是太阴谋论了，但我把这些点连起来，觉得政府真的在投资，我想，数据中心和AI公司的成功，或者至少是它们的普及。
And and then to me, it's I don't know if this is too conspiracy brand or not, but the dots I'm connecting are the government's really invested in the, I guess, the success or at least the prevalence of data centers and AI companies.
其中一部分可能很简单，就是没有哪届政府愿意在自己任期内看着经济垮掉。
Some of that could just be as simple as no administration wants the economy to fail on their watch.
所以，你知道，他们当然会投入其中。
So, you know, they they're invested in it.
但还有，你会看到像Flock摄像头到处出现，还有Axon摄像头和无人机。
But also, you see stuff like flock cameras popping up everywhere and Axon cameras and drones.
而所有这些，才是真正需要数据中心的地方。
And all of that, that's what requires data centers.
就是说，那……
Like, that's
真的需要吗？
Does it, though?
问题是这样的。
Here's the thing.
为了处理……
To process
不过，这需要GPU吗？
it require the GPUs, though?
我觉得是的，这有两方面原因。
I I think so it's two things.
存储肯定是需要的。
The storage definitely is
存储是需要的，但不需要AI数据中心来做这个。
The the storage, but you don't need an AI data center for that.
我觉得可能唯一需要GPU的原因——这个想法我还在琢磨，还没完全确定。
I I think the only reason you might, and this is still something I'm kind of playing with the idea of, I'm not sure on yet.
但我觉得GPU有用的原因可能是，开始把关于人或物体的大量复杂信息矩阵相互关联起来。
But I think the reason you might find the GPUs useful is to start connecting large complex matrices of information you have on people or objects to each other.
比如，这应该是GPU擅长的。
Like, this is something that You know, this is this is something that GPUs should be good at.
理论上，但问题是，我要反驳一下，因为我确实收到不少邮件，人们说这是我最担心的，政府监控。
In theory, the the but this is the thing, to push back on this because I do get a fair amount of emails people saying this is this is, like like, this is the thing I'm afraid of, the government surveillance.
对。
Uh-huh.
我觉得有很多不同的政府机构已经直接和甲骨文合作，而且已经拥有那些东西了。
The Department of I think there's multiple different government agencies that work directly with Oracle and already have and already have that.
他们已经有了一堆GPU。
Like, they already have they already have, like, bunches of GPUs.
所以如果他们打算这么做，他们早就做了。
So if they were gonna do this, they were gonna do this.
他们可能已经做完了。
They've probably already done it.
对吧？
Right?
嗯，我觉得区别在于，很多这些新东西是由私营公司大规模推出的，你知道，就像那些摄像头一样。
Well, I think the difference is a lot of this new stuff is being rolled out by private companies in large volumes, you know, just like the cameras alone.
我不知道。
I don't know.
大概六个月前，有八万个Flock摄像头，现在我不确定具体数字了。
Something like six months ago, it was 80,000 flock cameras, and now I'm not sure what it is.
肯定超过十万了，而且这是一家私人公司。
It's definitely well over a 100 And it's a private company.
数据量非常大。
There's a lot of data.
我不确定他们目前处理这些数据的程度，但处理图像或视觉识别数据，对比视频帧，确实是非常耗费计算资源的任务，需要数据中心。
I'm not sure to what extent they currently process this stuff, but definitely to process image or visual recognition data against video frames is a very computationally intensive task, that would require data centers.
对。
Right.
但我想说的是，所有这些数据集，你不需要那么多数据中心来处理。
But I'm just saying that all of these data set you don't need all of these data centers for that.
不需要。
No.
确实不需要。
Definitely not.
你确实不需要。
And you definitely don't.
而且他们也不是在做CDN那种需要把内容分发到全球各地的事情。
And it's not like they're doing CDN shit where they need to have it delivered to all points of the world.
我不是说这永远不会发生。
I'm not saying that this could never happen.
只是我觉得他们建这么多GPU数据中心，并不是因为这个原因。
It's just the I don't think the reason they're building all these GPU GPU data centers is because of that.
我觉得他们这么做，是因为他们根本不知道自己在干什么。
I think they're doing all of that because they don't know what they're fucking doing.
我觉得可能是……
Like, I think they I think it may
其实很简单。
actually just be really simple.
说清楚一点，我觉得他们建这些数据中心，并不是为了那个目的。
To be clear, I don't think they're trying to they're building all of them for that.
我觉得这确实是它们的一个用途，也是政府可能感兴趣投资的一个原因。
I I think that is a use case of them and is a reason that, you know, government may be interested in being invested as well.
所以，总的来说，这不是主要的用途，你看，这就像之前讨论的把软件和硬件所有权转移到云端一样，他们并不是为了把游戏从本地搬到云端才建所有这些数据中心的。
So, yeah, by and large, no, that's not like the main use case and you see things like it's kind of the same discussion as the taking ownership of software and hardware into the cloud where, no, they're not building all of the data centers to take gaming away and put in the cloud.
可能也就是为了GeForce NOW建那么一两个。
That's like maybe a couple here and there for GeForce NOW.
对吧？
Right?
但这只是另一个用途，怎么说呢，我觉得归根结底就是，不管你是私营企业还是政府，不管你遇到什么问题或者觉得自己有问题，数据中心都能帮你解决。
But it's just another use case where I don't know, it's it's kinda like I I guess what it comes down to is whatever the problem is that you have or think you have as a private business or a government, the data center is gonna fix it for you.
对。
Right.
你知道，这就像是一个——
You know, there's It's a
包治百病的万能药，尽管你去查一下企业实际的回报率就会发现并非如此。
panacea for your business even though when you go and look up the actual returns for businesses.
有一项OpenAI的研究出来了，结果特别搞笑。
There was an OpenAI study that came out that said, it said it's hilarious.
它说，每个员工的收入和使用的token数量之间没有任何关系。
It was like, there is no connection between revenue per employee and the amount of tokens used.
好吧。
Okay.
四年过去了。
Four years in.
四年过去了。
Four years in.
四年了。
Four years.
我们现在在哪儿呢？
And where are we?
我们就在这里，但什么进展都没有。
We're here, and we are nowhere.
我们真的是一点进展都没有，除了黄仁勋得搞出一些奇怪的财务工程，有点像安然，但不是安然，是那种安然风格、安然边缘的东西，就为了尽可能多地卖GPU。
We are absolutely fucking nowhere other than the fact that Jensen Huang has to create, like, some sort of weird financial engineering, not quite Enron sort of Enron, Enron esque, Enron adjacent thing to sell as many GPUs as possible.
我能问什么呢？
I What can I ask?
作为我认识的最了解我最喜欢的公司——安然的人，你觉得，什么会打破这个局面？
So as as the person I know who is the most knowledgeable on my favorite company, Enron, What to you, what what's the what would break the dam?
你知道，怎么从像安然变成真的就是，不，这根本就是安然。
You know, what bridges it from being Enron esque to being more actually just, no, this is literally Enron.
就像，他们就是这么干的。
Like, this is what they did.
你知道，我们缺少的那个决定性特征或者独特之处是什么？
You know, what's what's the defining or what's the distinct feature that we're missing?
如果我是黄仁勋，我跟你说，艾德，你会说我是安然风格的。
If I'm Jensen Huang and I say to you, Ed, you could you say I'm Enron esque.
我真的就是想成为安然。
I really just wanna be Enron.
你说，我得怎么做才能达到那个地步？
You know, what do I need to do to get there?
但问题是，问题是，安然那完全是赤裸裸的违法。
And the thing is the thing is, though, like, Enron was just nakedly illegal.
这就是最大的区别。
Like, that's that's the big comparison thing.
就像安然会把这些特殊目的实体拿来，把它们的债务
It's like, Enron would take these SPVs and take their debt
嗯。
Uh-huh.
或者把公司里那些烂的部分丢进去，然后说，看啊。
Or their bet bad parts of the company and just dump them there and be like, well, look up.
这他妈的太聪明了。
This this is fucking brilliant.
这没什么问题。
There's nothing wrong with this.
我觉得这只是一个观点，完全没有财务信息，我只是看了OpenAI的财务报表。
I think that this is an this is an opinion with no with absolutely no financial information I found just looking at the financials from OpenAI.
OpenAI的财务报表里有个奇怪的地方，我之前提过，就是他们把亏损转移到不同的实体上。
OpenAI has a weird thing in their financials that I've mentioned, which talks about moving losses onto different entities.
对。
Mhmm.
这确实很奇怪。
That's that's weird.
我不能说这就是安然，但闻起来很像。
That's I can't say it's Enron, but it smells like it.
另一方面，英伟达一直很聪明，他们从来不用自己的信用。
NVIDIA, on the other hand, has been smart in the they never NVIDIA never uses its credit.
英伟达总是会投资，比如这个SB Energy的交易。
NVIDIA always gets some they invested for this so for example, this SP Energy deal.
英伟达保证投资15亿美元给SB Energy，他们正在建数据中心，而且如果数据中心建成，英伟达会保证其价值高达1060亿美元。
NVIDIA is guaranteeing they're investing $1,500,000,000 in SB Energy who's building the data center, and they are guaranteeing the value of the data center up to a 106,000,000,000 if it gets completed.
现在，因为金融市场和信贷市场是由那些没有物体恒存性的人运作的，他们看到这个，就会说，好吧，NVIDIA已经保证这个数据中心会做到这一点。
Now, because we have financial markets and credit markets run by people with no object permanence, they see that and they go, well, NVIDIA has guaranteed that this data center will do this.
他们不会联想到，如果数据中心没建成会怎么样。
They don't make the link that what happens if the data center doesn't get built.
好的。
Okay.
而且NVIDIA在那里做的另一件事是，他们说，好吧。
And the other thing that NVIDIA did there is they said, okay.
如果OpenAI租不了这个，你可以试着找别人。
Well, if if OpenAI can't rent this, you can try and find someone else.
如果你找不到别人，我们会付你差价，无论你能卖多少钱，和那个1080亿之间的差额。
And if you can't find someone else, we will pay you the difference between whatever you can sell it for and what it sells and what what is a 108,000,000,000.
再说一次，永远不会完全NRO。
Again, moving never quite NRO ing.
永远不会NRO起来。
Never NRO ing up.
他们非常小心，不会完全走到NRO那一步。
We're they're very careful to not go full NRO.
那么，完全NRO是不是真的就跨入了明显的违法范畴？
So is is the full NRO really that that leap into just distinct illegality?
是的。
Yes.
好的。
Okay.
这就是为什么，比如安然公司，他们就是在明目张胆地欺诈。
And that's the that's why so for example, with Enron, they were just committing fraud.
你无法理解。
You can't get it.
安然公司从未发布过资产负债表，这简直太疯狂了。
Enron never published a balance sheet or, like which is fucking crazy.
比如，任何看到那家公司的人都会说，哦，他们没发布资产负债表。
Like, anyone who saw that company was like, oh, they didn't publish a balance sheet.
那没问题。
That's fine.
其实早就该预料到了。
Kinda kinda should have seen that coming.
对。
Right.
这个确实该想到的。
Should have seen that one.
不过，我最近在付费内容里也提过这个。
That being said, I've said this in the premium recently.
还有很多像通用电气这样的公司，会直接投资客户，借钱给他们。
There are a lot more like General Electric who would just invest in their clients, lend them money.
再说，英伟达并没有借钱出去。
Again, NVIDIA hasn't lent money.
这就是他们真正聪明的地方，没有真的变成债权人。
That's where they're real they've been quite smart to not actually become a creditor.
嗯哼。
Uh-huh.
所以他们会为CoreWeave的租赁提供担保，也就是说，如果CoreWeave付不起，他们同意代为支付。
So they'll back they backstop a CoreWeave lease, as in they agree to pay if CoreWeave can't pay.
他们投资CoreWeave。
They invest in CoreWeave.
他们不是提供自己的信用，而是说，我们同意从CoreWeave购买一定量的计算资源，这样你就可以说CoreWeave有收入了。
They are instead of offering up their credit, they're like, we agree to buy a certain amount of compute from CoreWeave so you can say CoreWeave has revenue.
他们利用金融手段。
They exploit the financial
或者那个基金会，当然。
Or the one foundation, of course.
也可以。
Can also
哦，天哪。
Oh, god.
我忘了那件事。
I forgot about that.
等等。
So wait.
他们是通过Coreweave租的，对吧？
What is the they rent it through Corewave.
对吧？
Right?
就是那个奇怪的非营利组织？
That's this weird nonprofit?
对。
Yeah.
我觉得这就是联系所在。
I think that's the what the connection was.
我之前研究过这个。
I had researched it a while ago
为了做一个视频。
for a video.
已经有一段时间了。
It's been a while.
对，没错。
But yeah.
所以胡安基金会是个非营利组织。
So the Juan Foundation is some kind of nonprofit.
他们也从CoreWeave租了一些东西。
They also rent some from CoreWeave.
是的。
Yeah.
而且他们基本上就是说，你知道，我们租这个GPU容量，我记得是为了学生。
And it's and they basically say, you know, we're renting this GPU capacity for I think it was for students.
它
It
是的。
was Yeah.
他们捐赠了价值一亿零八百万美元的CoreWeave AI计算资源用于研究。
They donated a $108,000,000 worth of CoreWave AI computing resources for research.
真是太棒了。
Is just like Beautiful.
我觉得这太不可思议了，因为真的太美好了。
So crazy to me because it's so beautiful.
对。
Yeah.
一方面，你经营并拥有一家销售GPU的公司的一部分。
You have this on one hand, you you run and own part of a company that sells the GPUs.
另一方面，你又有这个基金会，应该是你自己的事业。
On the other, you've got this foundation that's supposed to be, you know, your own thing.
然后它还跟第一个实体的客户扯上了关系，嗯。
And then it's it's getting involved with the customer of the first entity Mhmm.
为了给客户创造需求。
To create demand for the customer.
但这一切都是合法的。
But it's all legal.
公司。
Company.
都是合法的。
It's all legal.
对。
Yeah.
没错。
Right.
合法的，宝贝。
It's legal, baby.
别担心。
Don't worry about it.
都是合法的。
It's all legal.
完全没问题。
It's totally fine.
一切都好。
It's all good.
别担心。
Don't worry about it.
没事的。
It's fine.
但问题是这样的。
But this is the thing.
当你向财经记者、商业记者追问这件事时，他们就会这么说。
When you push back on financial journalists, business journalists about this, that's what they say.
都是符合GAAP会计准则的。
It's all GAAP legal.
这完全合法。
It's totally legal.
这里面没有任何违法的地方。
There's nothing illegal about this.
而且就好像，如果某件事不是字面上违法的话——
And it's as if if something isn't if something isn't literally illegal
那就没问题。
Then it's okay.
那就完全没问题，而且实际上大家应该多做这种事。
Then it's totally fine, and actually people should do it more.
而且，如果CoreWeave的收入中有12%来自其最大供应商NVIDIA，这也不算是什么坏信号。
And it's not a bad sign if, what, I think, like, 12% of CoreWeave's revenue comes from NVIDIA, its largest supplier.
嗯，Jensen确实说过，我们在这个视频里又放了那段片段。
Well, Jensen did say we played that clip again in this video.
哦，对。
Oh, yeah.
对。
Yeah.
他确实说过类似的话，没有英伟达他们就不存在。
He he did say something to the effect of without NVIDIA, they don't exist.
你知道吗？
You know?
有些这些，耶。
Some of these Yay.
就像，我就是喜欢，我喜欢那玩意儿。
Like, it's just like I love the I love that shit.
我喜欢它，因为就像，哇。
I loved it because it's just like, wow.
好。
Okay.
你这么说，是因为你知道这种事应该让美国证券交易委员会去管，然后他们就会说，哇。
You're just saying it because you know should something like that should have the SEC go, and like, woah.
哇。
Woah.
你这么做是什么意思？
What do you mean you're doing that?
因为我敢肯定，关于来回交易是有规定的。
Because I I swear there is regulation around round tripping.
对。
Yeah.
我想应该是有的。
Meant to I believe.
也许吧。
Maybe.
不过，谁在乎呢？
I mean, who gives a shit?
对吧？
Right?
很多时候就是这样，你去跟那些AI吹捧者聊，你会说，对，这是个问题。
It's it's so much of it is when you go to, like, when you go to the AI boosters, you're like, yeah, this is a problem.
他们就会说，啊，其实这挺好的。
They go, ah, it's actually great.
我前几天看到我们最喜欢的分析师之一丹尼尔·费勒发帖说，人们会批评这是循环融资，但我看到的是历史上最成功的风险投资家之一。
I saw one that one of our favorite analysts that Daniel Feller the other day posting, like, people will criticize this as a circular financing, but what I see is one of the most successful venture capitalists in history.
嗯哼。
Uh-huh.
然后你就会觉得，是啊。
And it's like, yeah.
确实。
True.
小丑吹的气球里全是空气，只要小丑一直吹，它们就一直鼓着。
The the balloons that the clown inflates are so full of air as long as the clown keeps inflating them.
而且这
And it's
Axios 刚问了他，你知道，就是问这是不是泡沫之类的问题。
just asked by Axios, you know, about is this a bubbler or something like that?
然后黄仁勋说，泡沫总有一天会来的。
And Jensen made his comment of, well, the bubble will come someday.
但不是今天。
Not today.
不是今天，但会在五到十年后。
It's not today, but it's in five to ten years.
对。
Yeah.
他还说我们现在正处于这个阶段的初期。
And and he describes how we're in the beginning of it.
他下一个词是以 B 开头的，我当时想，拜托，一定要是泡沫。
The next word started with a b, and I was like, please be bubble.
可惜他说的是建设扩展。
He said build out, unfortunately.
然后，我也不知道。
And the I don't know.
你能看得出来。
You could see it.
首先，我不想真的去分析他的肢体语言，但他确实有一种非常谨慎的态度，就是那种我得仔细想想再回答的感觉。
First of all, I don't wanna, like, analyze the the body language really, but he did definitely have kind of a it it was it was a very measured I need to carefully think about this answer split
想象一下科莱特·克雷斯拿着枪。
Imagine collet cress with a gun.
这就像那个CFO。
This is like the CFO.
就像在说，别说，Jensen。
It's like, don't say it, Jensen.
嗯，他前几天说计算应该成为一种可投资的资产类别。
Well, he said the other day that compute should become an investable asset class.
然后呢？
And it's Then what?
就是这样，没错。
It's just the exactly.
嗯，就是那些说算力应该像商品一样交易的人。
Well, it's the people who were like, compute should be traded like a commodity.
然后就像，拿枪指着你，告诉我你觉得算力是什么意思。
And it's like, gun to your head, tell me what you think compute means.
对。
Yeah.
那到底是什么意思？
What does that even
什么是算力？
What is compute?
作为一种资产类别？
As an asset class?
嗯，GPU应该作为资产类别来交易，但算力也应该作为商品来出售。
Well, asset class GPU should be traded as that, but also compute should be sold as a commodity.
不过，算力这东西。
Compute, though.
这些说法其实都讲不通。
None of these really make any sense.
只要稍微想一想，
If you think about it for even one second,
就好像——那我是不是得把我那间堆满测试过GPU的储藏室，在报税时单独列一项，因为那些GPU理论上能产生的算力？
it's like Does that mean I need to declare my storage room full of GPUs we've tested as as some form of a some for some separate line item on the taxes going forward because of the theoretical compute they could produce?
你现在是商品交易员了。
You're a commodities trader now.
哦，太好了。
Oh, great.
如果你的书架和你的“为你推荐”页面对你来说同样重要，那欢迎回家。
If your bookshelf and your For You page are equally important to your personality, welcome home.
这里是“散文社”，一档每周更新的播客，既是读书俱乐部，也是思想女儿们、流行文化狂热者，以及那些认为《傲慢与偏见》和《爱情岛》值得同等讨论度的人的群聊。
This is Prose Society, the weekly podcast that's part book club, part group chat for thought daughters, pop culture obsessives, and anyone who thinks Pride and Prejudice and Love Island deserve the same level of discourse.
我是Eli Rallow，每周我们都会把书籍、互联网和大家都停不下来的话题串联起来。
I'm Eli Rallow, and every week, we're connecting the dots between books, the Internet, and the conversations everyone can't stop having.
我得去查一下
I'm gonna have to look up
这个故事。
this story.
我迷上它了。
I'm obsessed.
我迷上它了。
I'm obsessed.
从畅销书作家和你最喜欢的书籍讨论创作者，到最新的流行文化热点，什么都可以聊。
From bestselling authors and your favorite book talk creators to the latest pop culture moments, nothing is off the table.
就像如果你能把一些真实的信息藏在引人入胜的角色里，那就是特洛伊木马。
It's like if you can hide some real messages inside compelling characters, and that is a Trojan horse.
无论你是在寻找文学深度探讨、聪明的流行文化对话，还是一个喜欢想太多的读者社区，你都来对地方了。
Whether you're looking for literary deep dives, smart pop culture conversations, or a community of readers who love to think a little too much, you're in the right place.
请在iHeartRadio应用、Apple Podcasts或任何你收听播客的地方收听Pro Society。
Listen to Pro Society on the iHeartRadio app, Apple Podcasts, or wherever you get your podcasts.
我们书页间见。
See you between the pages.
在新播客《Solita》中，我们分享女性独自旅行时那些混乱的真实经历。
On the new podcast, Solita, we share the messy reality of traveling alone as a woman.
我可以等四个小时坐下一班公交车，或者让这个陌生男人用摩托车载我一程。
I can wait four hours for the next bus, or this random dude is offering me a ride on his motorcycle.
我选了第二个选项。
I chose option b.
我是Julie Pinheiro，我独自旅行，因为那是一个我可以不用先问别人就能说“好”的难得空间。
I'm Julie Pinheiro, and I travel by myself because it's a rare space where I can say yes without asking anyone else first.
我致力于重新定义“solita”这个词，把同情变成可能性。
I'm on a mission to reclaim the word solita, trading the pity for possibility.
每次我试图
Every time I tried to
想独处的时候，我却总是遇到人，然后
be alone, I kept meeting people, and
他们就说，你对我们笑了。
they were like, you smiled at us.
这附近很少有人笑。
Not a lot of people smile around here.
正是当你独自一人时，你最能真实地感受这个世界，而不是那些被灌输的谎言。
It's when you're alone that you're most receptive to the world as it is and not the lies you're sold about it.
这段时间可以让你突破极限、改变想法，或者唤醒一个全新的自己。
It can be a time where you push your limits, change your mind, or wake up to a new version of yourself.
所以，无论你是独自旅行的老手，还是紧张得不敢预订第一次旅行，我都希望你能在iHeartRadio应用、Apple Podcasts或任何你收听播客的地方收听《Solita》。
So whether you're a solo travel veteran or you're too nervous to book your first trip, I hope you listen to Solita on the iHeartRadio app, Apple Podcasts, or wherever you get your podcasts.
嘿。
Hey.
我是海斯·达文波特。
This is Hayes Davenport.
还有肖恩·克莱门茨。
And Sean Clements.
我们在Big Money Players网络主持播客《好莱坞手册》。
We host the podcast Hollywood Handbook on the Big Money Players Network.
我们的节目对新听众非常友好。
Our show is extremely accessible to first time listeners.
每周我们都会和演艺圈的人聊天，也就是所谓的“娱乐圈”，试着帮他们发展事业，看看他们能给我们带来什么。
Each week, we talk to someone in show business, aka The Biz, and try to help them with their careers and see what they have to offer us.
每个人都玩得很开心，没有人会因为经纪人让他们上我们的节目而生气。
Everyone has a good time, and no one gets mad at their publicists for letting them do our show.
我们有很多
We've had a lot of
很棒的嘉宾，比如莎拉·谢尔曼。
great guests like Sarah Sherman.
也没那么棒。
Not that great.
其实，你能帮我
Actually, can you help
选一下吗？你能帮我弄一下这个造型吗？
me pick can can can you help me shape this?
好的。
Yeah.
要不我们把它倒过来？
How about we flip it on his head?
不行。
No.
行吗？
K?
要不我们把它反过来？
How about we flip this around?
行吗？
K?
不。
No.
那如果那个你以为很正常的女人，其实才是那个行为古怪的人呢？
How what if the surprise woman who actually you think is the straight man is actually the one who's acting weird?
亚当·斯科特。
Adam Scott.
这整段时间，你一直在期待亚当·桑德勒来上节目。
This entire time, you've been expecting Adam Sandler to come to visit the show.
我绝不会接受。
I I would never accept.
对。
Have yeah.
我绝不会对那种事情抱有任何期待。
I would never set expectations for something like that to happen.
如果宇宙允许那样的事情发生，是的。
If the universe allows for something like that to happen Yeah.
我对此持开放态度。
I'm open to it.
我总是会接受的。
I'm always going to accept.
你
You
知道吗？
know?
丹尼·麦克布莱德，行吧。
Danny McBride, fine.
对。
Yeah.
看到了这种脆弱的男子气概。
Saw this fragile masculinity.
嗯。
Mhmm.
散发出来的。
Exuding.
说实话，我走在走廊上都能闻到那股味儿。
I could smell it walking down the hallway, to be honest with you.
对。
Mhmm.
我当时还想，这是不是大麻的味道？
Was like, is that I like, is that weed?
对。
Mhmm.
不是。
Nope.
那是脆弱的男子气概。
It's fragile masculinity.
对。
Mhmm.
对。
Yep.
本·斯蒂勒，不是。
Ben Stiller, no.
因为很多时候，在脱离
Because a lot of times in out of
上下文的情况下，人们不明白，甚至不知道圆圈是什么。
context, people don't get it or even know what a circle is.
你懂我的意思吗？
You know what I mean?
就像，他们会说，
Like, they're like,
这太蠢了。
what's the so stupid.
糟糕。
Crap.
除了Fake，其他人都是笨蛋。
Everyone's so dumb except Fake.
因为
For
对。
Right.
Fake也是，而且其他人要么不感兴趣，要么就是觉得，哦，我有
Fake and and also not interested or just like, oh, I've got
我的烟斗。
my pipe.
我才不在乎你的生活呢。
I don't care about your life.
还有很多其他大人物，没有我们他们也不会有今天。
And a lot of other big shots that wouldn't be where they are without us.
哦，顺便说一句，威尔·法瑞尔亲自告诉我们他喜欢这个节目，还想每集都来当嘉宾，但他实在太忙了。
Oh, and by the way, Will Ferrell told us personally he loves the show and he wants to be a guest on every episode, but he's just so busy.
请在 iHeartRadio 应用、Apple Podcasts 或任何你收听播客的平台上收听《好莱坞手册》。
Listen to Hollywood Handbook on the iHeartRadio app, Apple Podcasts, or wherever you get your podcasts.
我是霍达·科顿比，播客《霍达·科顿比的快乐101》的主持人。
I'm Hoda Cottonbee, host of the podcast Joy one zero one with Hoda Cottonbee.
如果你了解我，你就会知道我一直在寻找灵感和有用的工具来最大化快乐。
If you know me, you know I'm always searching for inspiration and useful tools to maximize joy.
所以这个播客让我们通过和有趣的人进行有意义的对话，一起发现这些东西。
So this podcast lets us uncover those things together through meaningful conversations with fascinating people.
比如我的朋友亨利·温克勒给有抱负的演员以及任何希望脱颖而出的人提供建议的时候。
Like when my friend Henry Winkler offered advice to aspiring actors and really anybody hoping to stand out.
你被雇佣不仅仅是为了填补时间和空间。
You're not hired just to fill time into space.
你被雇佣是为了用你自己来填满这个空间。
You're hired to fill the space with you.
你内心有某种东西。
You have something in your guts.
你就顺着它来。
You go with it.
肖恩·约翰逊回忆起她意识到现在的丈夫安德鲁就是对的那个人那一刻。
And Shawn Johnson recalled the moment she realized her now husband, Andrew, was the one.
他就是那么毫不掩饰地做自己，对好莱坞一无所知。
He was just so unapologetically himself and knew nothing of Hollywood.
他之前从没在没有家人的情况下坐过飞机。
He hadn't been on a plane before without his family.
那太纯粹了。
It was so pure.
在iHeartRadio应用、Apple Podcasts或任何你收听播客的地方收听《Joy 101》与霍达·科特布。
Listen to Joy 101 with Hoda Kotb on the iHeartRadio app, Apple Podcasts, or wherever you get your podcasts.
说到安然公司末期那个著名的事情，不过要说明一下，英伟达并不是在交易算力。
Well, the famous thing towards the end with Enron, though and just to be clear, NVIDIA is not trading compute.
但在安然末期，他们打算交易带宽吗？
But towards the end with Enron, they were gonna trade bandwidth?
我是说，如果黄仁勋能搞定这个，我们已经有人在交易电力了。
I mean, if Jensen gets it I mean, we've already got some people trading power.
Meta在做，苹果也在做。
You got Meta doing it, Apple doing it.
所以在这方面我们已经有点盼头了。
So that's we've already got a little bit to look forward to there.
但另一件事是，我个人希望看到英伟达在金融工程上做更多。
But the other thing is if I I wanna see in I personally wanna see NVIDIA do more financial engineering.
我想要比这多得多的东西。
I want way more than this.
我希望看到英伟达现在干脆从自己那里买算力。
I wanna I wanna see NVIDIA just buying compute from itself at this point.
其实他们离那也不远了。
I mean, they're not far from it.
不过，是的，能源这块也挺疯狂的，因为政府现在介入得很深。
Like but, yeah, I I the energy stuff too has been kinda wild because the government is getting really involved there.
你知道吗，几个月前，特朗普提到过那个自带发电厂计划的早期概念。
You know, like, months ago, Trump talked about the sort of early concept of the BYO power plant plan.
那是什么？
Which is what?
抱歉。
Sorry.
我肯定错过了那个。
I must have missed that one.
其实，那是我给它起的名字。
So, basically, that's my name for it.
但那个计划是让公司自己提供能源。
But he the plan of company is to provide their own energy.
所以如果他们的数据中心需要能源，那么这些公司就会以某种方式自己提供。
So if they need energy for their data centers, then the companies are either through one form or another going to provide that energy.
而且公关上的说法是，这有助于防止居民和当地人的能源成本上涨。
And the sort of PR play was that this would help prevent rising energy costs on residents and locals.
我说公关手段的意思是，特朗普总统在会议上真的说了，数据中心现在需要一些好的公关。
And when I say PR play, I mean, president Trump literally in the meeting said, data centers need some good PR right now.
特朗普先生，他确实可以在这方面做出公正的判断。
I mean, he could call in balls and strikes there, mister Trump.
所以这个想法就是，数据中心要更负责地自己建设和提供所需的电力，我确实能理解，如果这是起点的话。
And and so the concept of, you know, the the data centers will be more responsible for bringing online and provisioning their power that they need, which I I could certainly see how if that's the start.
对。
Yeah.
嗯，这确实是第一个问题。
Well, I mean, that's definitely the first question.
但从公关的角度来看，如果去掉所有细节，你可以看到这可能会赢得一些人的支持，如果他们不深入思考的话。
But from a PR standpoint, I guess, you strip all detail out of it, you could see how that that might win some people over if they don't really think too too deeply about it.
因为，就像，哦，好吧。
Because, like, oh, okay.
对。
Yeah.
我的意思是，他们自己掏钱，我就不用替他们付电费了。
I mean, they so they pay for their own shit, and I don't have to pay for their energy.
对。
Right.
但是能源这东西，如果不完全监管，真的很吓人。
But energy is, like, a terrifying thing to not be fully regulated.
能源搞错了可是很危险的。
Like, energy is dangerous to do wrong.
非常危险。
Like, dangerous.
你知道，
You know,
它可能
it can
会害死人。
kill people.
而且这东西的量也是有限的。
But also just there's only so much of it.
我觉得搞AI的人，尤其是他们，真的低估了电力供应的难度。
Like, it's I think people AI people especially really underestimate how difficult power is.
对。
Yeah.
但他们就是觉得，哦，是啊。
But they're just like, oh, yeah.
你知道，他就是随便搞个创业公司，然后一下子就搞定了。
You know, he's just fucking start up and he just boom.
对。
Right.
就随便搞搞，你需要电什么的？
Just fucking do you want you need some power or something?
这有什么好炫耀的？
What what's waving a dick about?
她他妈的，你知道，就是
She's fucking, you know, just
嗯，那都是燃气轮机。
Well, that's all the gas turbines.
对吧？
Right?
如果你直接把燃气轮机放在数据中心屋顶上，
If you just drop gas turbines on the roof of a data center,
真是疯狂的东西。
crazy they things.
我前几天读到一个故事，说燃气轮机快不行了，因为它们本来就不该一直运行。
I read the story the other day where it was saying the the gas turbines are dying because they're not meant to be They're not meant to be run all the time.
对。
Yeah.
它们应该是应急用的。
They're supposed to be for emergencies.
就像联邦应急管理署会部署的那种东西。
It's like stuff that FEMA will, you know, will deploy.
老兄，而且它们到处都是。
Dude, and they're everywhere.
我知道在星际之门，阿比林，它们简直无处不在。
I know at Stargate, Abilene, they're fucking they're everywhere.
他们到处都放满了。
They got them all over the shop.
真是太奇怪了。
It's just so strange.
我是说，你是个彻头彻尾的硬件专家。
I mean, you are a dyed in the wall hardware guy.
你搞硬件这么多年了，我想你已经清楚地意识到，这已经不再是关于硬件或者GPU或者其他什么的了。
You've been dealing with hardware for years, and I think you've successfully caught on to the fact that this is no longer about hardware or GPUs or anything.
这关乎的是钱。
It's about It's about money.
钱。
Money.
对。
Yeah.
对。
Yeah.
技术已经不重要了。
The technology stopped mattering.
我的意思是，现在连主题演讲都不怎么讲技术了，虽然我还是会认真看。但黄仁勋以前讲得特别技术化，即使在早期机器学习和深度学习那会儿，还没改叫AI的时候。
I mean, they don't even really even in the keynotes now, which I still watch really closely, but Jensen used to be so technical even with the early machine learning and deep learning stuff before they rebranded to AI.
他参加过一些最早的会议。
The guy was at some of those earliest conferences he did.
我去过那个特别尴尬的GTC，他和马斯克面对面坐在台上，我当时离他们只有50英尺，都能感受到那种替人尴尬的感觉。
I was at the really awkward GTC where he and Elon Musk sat across from each other on stage, and that was a I I could I could feel the second hand embarrassment from just 50 feet away from them.
而且那时候，虽然总是有讨好投资者的成分，但也有很多技术细节。
And all of that, there was there was always pandering to investors, but there was also a lot of technical detail.
比如说，我听完一场主题演讲，就能写出一篇非常详细的技术文章，介绍他们刚发布的新东西的架构。
Like, I could walk away from a keynote and write a really dense write up about the architecture of some brand new thing they just announced.
但现在呢，我听完主题演讲，到最后，说实话，我都不知道这一代和上一代有什么区别，除非我去查白皮书。
And now there's like, I'll watch Keynote, and at the end of it, I'm like, I I honestly I don't even know what the difference is between this generation and the last generation without going and looking up a white paper.
你懂吧？
You know?
对，就是这样。
Like, it's Right.
他们把所有技术细节都删掉了。
They've stripped it all.
虽然技术还在，但不知怎么的，就是……对。
And and there's still technology there, but somewhere, it just yeah.
它已经不重要了，这真的很让人沮丧。因为即使我想报道纯技术层面的架构，它也会被包装成企业级产品，然后对观众说，我们报道这些是因为希望有一天它能以某种形式进入消费市场，过去通常都是这样做的。
It just doesn't matter anymore, which is real depressing because it's like, even if I wanna cover the pure technology side of the architecture, it goes into enterprise products and then say to the audience, you know, we we cover this stuff because hopefully, this comes to consumer at some point in some capacity, which is typically how it's been done in the past.
即使我现在想这么做，他们也不再像以前那样努力让这些资料广泛传播了，因为已经不重要了。
Even if I wanna do that now, it's like they're they're not even really putting the effort into making those materials as widely available as they used to be because it doesn't matter.
只有钱才是最重要的。
Only the money matters.
我就是看不出他们怎么解决这一切。
I just don't see how they solve all this.
比如，我最近一直在研究博通，基本上每次你读到关于半导体公司，特别是AMD、英伟达和博通的文章，都会觉得，对，就是这样。
Like, I've been digging into Broadcom recently and all like, pretty much every time you read about a semiconductor company or specifically AMD, NVIDIA, and Broadcom, it's like, yeah.
而且我的意思是，对于未来，基本上我们需要，我不知道，未来三年一切都要顺利，因为我们所有的收入增长都只来自这一块。
And I mean, for the future, pretty much we need, I don't know, everything to go perfectly for the next three years because all of our revenue growth is coming from only this.
比如博通，我觉得，未来几年他们超过60%的收入都来自人工智能，而其中大部分来自谷歌的TPU。
Like, Broadcom, I think, is I think, like, over 60% of their revenue in the next few years is coming just from AI, which is mostly coming from Google's TPUs.
我的意思是，这，我不知道。
I mean, it it I don't know.
你刚才说未来三年一切都要顺利，这让我想起小时候玩游戏，比如桌游。
The thing that comes to mind, like, when you said that everything needs to go perfectly for three years, it reminds me of just when I played games as a kid, like board games.
像国际象棋或者万智牌之类的。
So chess or magic the gathering or something.
特别是下棋，你知道，这是个回合制游戏。
But chess in particular where, you know, you're you're a turn based game.
就像你可以制定一个完美的计划，只要一切顺利，对手在接下来的三十分钟里每一步都走错，你就能赢。
It's like you could come up with this master plan of how you're gonna win as long as everything goes perfectly and they move their pieces in exactly the wrong way for the next, you know, thirty minutes.
但现实可不是这样的。
But, like, this is not how reality works.
而且这些公司现在看起来，我不知道，有时候听黄仁勋或者OpenAI的萨姆·奥特曼讲话，感觉他们现在也承认事情可能不会那么顺利，但就是得把面子撑住，直到某个重大突破出现，好让更多钱流进这个系统。
And it does seem like these companies right now are I I don't know if they're sometimes it feels like watching Jensen, some of the other, like, OpenAI's, you know, Sam Altman speak, it feels like there's a little bit of acknowledgment now of things might not go perfectly, but we just kinda need to keep the facade up long enough until there's some major breakthrough where we can get more money pumped into the system.
我最近看的感觉就是，他们在推销说我们需要再挖出下一个一千亿或五千亿美元。
And that's that's kinda what it feels like I'm watching lately is the sales pitch for we need to to dig up the next 100 or $500,000,000,000.
基本上就是，我们必须让这一切完美运转，不能出任何差错。
We need to make this all work perfectly without fail is pretty much it.
而且就算他们被问到，通常如果萨姆或者黄仁勋被问到，他们就会说，对，你知道，有些人会没钱了。
And even when they ask about, they they usually if Sammy's done this, Jensen's done, it's like, yeah, you know, some people will run out of money.
你觉得萨姆关于什么的新评论怎么样？
How'd you like Sam's new comments about What?
那个精灵？
The the genie?
哦，那个精灵，那确实挺有意思的。
Oh, the genie was that was a pretty good one.
但还有那些关于ChatGPT后代会看到你在电脑屏幕上做的一切的评论。
But also the comments about how some descendant of chat GBT will be able to see everything you ever do on your computer screen.
他在这方面真是厉害。
He's so good at this.
他在这方面真是厉害，因为，我肯定你还记得微软宣布Microsoft Recall的时候。
He's so good at this because it's like, I'm sure you remember when Microsoft did they announced Microsoft Recall.
对。
Yeah.
然后，大家的反应就是，我要杀了你。
And Oh, everyone's reaction was, I'm going to kill you.
我要烧了你的房子。
I'm going to burn down your house.
别碰我的电脑。
Do not touch my computer.
我会杀了你。
I will murder you.
而微软，本来几乎什么都能硬推的，结果也怂了。
And e Microsoft, who could push through just about anything, were just like, nah.
不行。
No.
我们干脆跳过这个。
We're just gonna skip this.
然后OpenAI大约一年后出现了，说：靠，兄弟，来吧。
And then OpenAI turns up about a year later and goes, fuck yeah, bro.
搞死我吧。
Fuck me up.
我们要看你做的一切。
We're gonna look at everything you do.
奥特曼在那儿说，是的，我们能看到你想做的一切，你做过的一切，还能建议你去做点别的。
And Altman is there being like, yeah, we can see everything you want you do, everything you you've done, and, suggest you could do something else.
这很棒，因为，再说一次，这又回到了AI的老话题上。
Which is great because, again, it comes back to the classic thing of AI.
当我们把AI当作社会和文化的一部分来思考时，我们会想，好吧，AI就是那个不用我们告诉它就知道一切的东西，不用我们真正操心就能做出明智的决定。
It's when we think about AI as a society, as a cultural thing, we think, okay, AI, it's this thing that just knows everything without us telling it, makes smart decisions without us really having to.
但这个AI是尽可能多地喂数据给它，也许有一天它会起作用。
But this AI is feed as much data as humanely possible to this and maybe it will work one day.
对。
Yeah.
嗯，而且你也会在类似的事情中看到这一点，比如这个视频我们还没发布，我不知道你们什么时候播出这期节目，所以这个可能会比我们早一点上线。
Well and and you also see this in things like this video is not up yet for us, and I don't know when you're airing this episode, so this might go up a little before us.
但我们现在有一些关于特定智能电视和其他设备的深度探讨。
But we have a deep dive into some specific smart TVs right now and other devices.
而且，你知道，如果你真的去仔细阅读这些公司的隐私政策，他们告诉你他们在做的事情真是疯狂。
And, you know, if you really go through the privacy policies of these companies, the the stuff they tell you they're doing is crazy.
而且，这还是他们明说会做的事情。
Like, and that's the stuff they tell you they're doing.
全都是数据收集和数据挖掘。
And it's all data collection and data harvesting.
现在呢，当然，他们又都加上了AI。
And now, of course, they've all got AI tacked onto it.
你知道，其中一家智能电视公司，他们专门为AI功能设了一个展示区，比如展示A，AI连接功能，你从头到尾读一遍。
You know, there's in one of these smart TV companies, they have a special exhibit for their AI features like exhibit a, AI connected functionality, and you read through it all.
结果他们就直接说，我们会卖你的数据。
And it's just it's they straight up say, we sell your data.
就是，直接
Like, just
那倒是挺好的。
That's tell good.
你知道，是的。
You know, it's yes.
就在那里。
Right there.
再加上像ACR这种自动内容识别技术，根据部署方式的不同，它有时能截取你智能电视屏幕上的内容，然后传回去分析，或者把它转换成数字指纹再传回去分析。
And this coupled with things like ACR, automated content recognition, which depending on the the deployment has the capability of screenshotting what you show, what you have on your screen, on your smart TV in some cases, and transmitting it back to be analyzed or converting it technically into a fingerprint, transmitting that back to be analyzed.
所有这些加起来就是海量的数据。
All of this is mountains of data.
他们都非常想用AI来处理这些数据。
They all really want to deploy AI to process it.
我们收到一些观众的爆料，说目前他们还在雇人手动查看截图数据之类的，来解析电视的广告画像。
We've received some tips from viewers who say that for now they're hiring staff to manually parse advertising profiles for TVs by looking at the screenshot data and stuff like that.
但最终目标是把它全部交给某种大语言模型，让它自动搞定这一切。
But the goal is to move it all into some kind of LLM that just magically does all this for you.
我觉得这大概就是OpenAI或Anthropic这类公司梦寐以求的客户吧。
And I guess that's kind of the dream client for a company like OpenAI or Anthropic.
但即使是这种理想客户，也只是一个B2B的用例。
But even in that dream client scenario, it's just one, it's a b two b use.
而且，这纯粹是一种邪恶的用途。
And two, it's like a strictly evil use.
对。
Yeah.
而且我也不太确定。
And it's also I don't know.
他们还没这么做。
They've yet to do it.
比如说，我觉得如果他们有一个超级秘密的监控数据源，效果特别好，而且有人愿意出比运营成本更高的价钱来买，但现实似乎恰恰相反——当你和普通企业客户聊天时，他们会说，AI的成本已经太高了。
Like, feel like if they had a super secret source that was great for surveillance that worked really well and that someone would be buying it and paying more for it than the the cost of operating it, which when the direct opposite seems to be happening, it seems the people that you when when you talk to when you talk to, like, an average enterprise customer, it's like, yeah, AI costs are already too high.
为什么还要用呢？
Why do you do it?
嗯，我所有的朋友都告诉我必须这么做。
Well, all my friends are telling me I have to do this.
如果我不这么做，就会面临那些吹嘘AI有多好的人带来的竞争压力。
If I don't do this, I'll face competitive pressure from people that are yelling about how good AI is.
关于这个，我有个有趣的故事。
I have an interesting one on that.
哦，说来听听。
Oh, go on.
对。
Yeah.
我有个朋友在一家游戏开发公司工作，而且是一家挺大的公司。
I have I have a friend who works at a game development company and a pretty large one.
据我了解，这家公司内部现在的指令是，使用AI是他们工作的一部分，是必须的。
And so the the directive, as I understand it from internal now at this company, is that AI use is an expected part of their job.
也就是说，如果他们不找办法去消耗token进行查询，那他们就没在好好干活。
Meaning, if they don't find ways to burn tokens on queries, then they're not doing their jobs.
太好了。
Great.
对。
Yeah.
关于 Bayt 播客
Bayt 提供中文+原文双语音频和字幕，帮助你打破语言障碍，轻松听懂全球优质播客。