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。
This is an iHeart podcast.
百分百真人制作。
Guaranteed human.
经营企业却不考虑做播客？
Run a business and not thinking about podcasting?
再好好想想。
Think again.
听播客的美国人比听Spotify和Pandora广告支持的流媒体音乐的人还多。
More Americans listen to podcasts than ad supported streaming music from Spotify and Pandora.
作为排名第一的播客平台，iHeart的规模是第二名和第三名加起来的两倍。
And as the number one podcaster, iHeart's twice as large as the next two combined.
了解播客如何助力您的业务。
Learn how podcasting can help your business.
拨打844844
Call 844844
当种族隔离是法律时，一位神秘的黑人俱乐部老板查理·菲茨杰拉德有自己的规矩。
When segregation was a law, one mysterious black club owner, Charlie Fitzgerald, had his own rules.
白天隔离，夜晚融合。
Segregation in the day, integration at night.
就像踏入了另一个世界。
It was like stepping in another world.
他是个商人、罪犯，还是英雄？
Was he a businessman, a criminal, a hero?
查理是那种必须被摧毁的力量的典范。
Charlie was an example of power that had to crush him.
《查理的地盘》，来自Atlas Obscura和Myrtle Beach旅游局。
Charlie's Place from Atlas Obscura and Visit Myrtle Beach.
在iHeartRadio应用、Apple Podcasts或您收听播客的任何平台收听《查理的地盘》。
Listen to Charlie's Place on the iHeartRadio app, Apple Podcasts, or wherever you get your podcasts.
这是全新的我，也是旧日的他们。
It's the new me, and it's the old them.
今年妇女历史月，播客《如果你知道得更多》与安伯·格莱姆斯一起，聚焦那些将失误转化为动力、将教训转化为力量的女性。
This woman's history month, the podcast, If You Knew Better with Amber Grimes, spotlights women who turn missteps into momentum and lessons into power.
我当时那种一心只想达成目标的狭隘视角，其实源于我想为我们创造更好生活的愿望。
My, like, tunnel vision of, like, I gotta achieve this was off the strengths of, like, I wanna make a better life for us.
《如果你知道得更多》带来真实的声音，讲述那些亲历过这些故事的女性，深入探讨职业转型、感情教训以及改变一切的心态转变。
If you knew better brings real talk from women who've lived it, unpacking career pivots, relationship lessons, and the mindset shifts that changed everything.
请在 iHeartRadio 应用、Apple 播客或你收听播客的任何平台收听安伯·格莱姆斯的《如果你知道得更多》。
Listen to if you knew better with Amber Grimes on the iHeartRadio app, Apple Podcasts, or wherever you get your podcasts.
你越认真听孩子说话，你们的关系就会越亲密。
The more you listen to your kids, the closer you'll be.
所以我们问了孩子们：你希望父母听到什么？
So we asked kids, what do you want your parents to hear?
我有时候觉得没人认真听我说话。
I feel sometimes that I'm not listened to.
我只希望你们能更常听我说，和我一起分析情况，并引导我走向成功。
I would just want you to listen to me more often and evaluate situations with me and lead me towards success.
倾听是一种爱的表现。
Listening is a form of love.
请访问 sounditouttogether.org，获取帮助您支持孩子及其情绪健康的资源。
Find resources to help you support your kids and their emotional well-being at sounditouttogether.org.
网址是 sounditouttogether.org。
That's sounditouttogether.org.
本节目由广告委员会和 Pivotal 联合呈现。
Brought to you by the Ad Council and Pivotal.
Call Zone Media。
Call Zone Media.
你好，欢迎收听《更好的离线生活》。
Hello, and welcome to Better Offline.
我是你们的主持人，埃德·齐特隆。
I am, of course, your host, Ed Zitron.
现在
Now
我们现在处于喷子季。
we're in hater season.
喷子季就是我邀请一些人来，允许他们变得粗鲁，因为你知道，这个播客通常都非常克制和礼貌，尤其是在提及人物时，特别是科技高管，我们从未称他们为混蛋、蠢货、笨蛋、垃圾、烂人、屁精、fucknuts 或 knob ends。
Hater season is when I bring people on and allow them to get rude because, you know, this podcast is usually so reserved and nice in the way we refer to people, especially tech executives, which we've never referred to as fuckwits or morons or dumbasses or shitheads or shitheels or asswipes or fucknuts or knob ends.
我们从来都没这么做过。
We've never done any of that ever.
不过，今天加入我的是 Duckbill 公司的首席云经济学家科里·昆恩。
Nevertheless, joining me today is chief cloud economist at Duckbill, Corey Quinn.
科里，你最近怎么样？
Corey, how are doing?
我非常高兴能来这里，因为通常在其他所有地方，我都得保持克制。
I am delighted to be here because normally in all the other places I find myself, I have to be so reserved.
你得收着点劲儿。
You have to you have to pull your punches.
科里，今天就是我们的喷子季。
Corey, I I today's hater season.
谈谈你对亚马逊云服务的看法吧。
Tell me about your feelings on Amazon Web Services.
天哪。
Oh, dear lord.
这感觉很奇怪。
It's it's like it's weird.
我其实并不讨厌亚马逊。
I don't actually hate Amazon.
人们以为我讨厌，是因为我说过的话和做过的事。
People think I do because of the things I say and the things that I do.
但如果我真的讨厌它，那这个
But if I hate it, then this
你这个月的举动。
your month actions.
但十年过去了，我还在写关于它的通讯。
But ten years in, I write a newsletter about it.
如果我真的讨厌这家公司，只希望它遭遇不幸，首先，这本身就是一种心理问题，我该申请禁制令了。
If I actually hate the company and want nothing but ill things to happen to them, first off, that's a pathology, and I need a restraining order.
其次，每当有人认为我跟这家公司有私怨，我就特别不爽。
And secondly, it always bugs me when people think that I have an axe to grind against the company.
因为都十年了，你真的觉得，如果我真想重创他们，现在这已经是我能做的最好的方式了吗？
Because ten years in, are you seriously suggesting this is the best I would be able to do if I actually wanted to hit them where it hurts?
拜托。
Come on.
但说实话，最近他们确实烦得我够呛。
But, yeah, they've been annoying the piss out of me lately.
所以AWS挺奇怪的，也许你能给我解释一下。
So AWS is weird because and maybe you can explain this to me.
在我整个科技职业生涯中，我总听到一个说法：你的AWS账单就是会一直涨。
Over my entire tech career, I've heard this constant story that your AWS bill just it just goes up.
它就是会无缘无故地增加。
It just arbitrarily increases.
不过目前还不清楚他们是否在提价，虽然我知道提价确实存在。
It's not really clear whether they're doing price increases, though I know those exist too.
发生什么事了？
What happened?
但偶尔，最近这种情况越来越多了，是的。
But occasionally, increasingly these days, yes.
但到底是什么原因导致账单上涨呢？
But why what is it that makes the bill go up?
为什么？为什么这种情况会如此频繁地发生？
Why and why does this happen so often?
我曾经以为这是个阴谋，但后来我意识到，亚马逊在组织能力上根本不可能策划出这种事。
I used to think it was a conspiracy, and then I realized Amazon is in no way, shape, or form organizationally competent enough to pull something like that off.
这其实是按使用量收费的必然结果，再加上一些混蛋的做法——好吧，我们要推出一项服务，然后对你们每上传的一GB数据都收费。
It is the nature of charging per usage and honestly assholes where, okay, we're going to launch a service and we're going to charge you for every gigabyte you put in it.
这就是我们该做的。
That's all we're to do.
很好。
Great.
很简单。
Straightforward.
人类可以对此进行合理化解释。
Humans can rationalize around that.
轻而易举。
Easy peasy.
他们在2000年首次推出S3测试版时就这么做了。
They did that when they first launched s three in beta back in 2000
存储，对吧？
Storage, and right?
这是一个大型对象存储系统。
It's a big object store.
把它想象成一个无限大的硬盘，差不多就是这么回事。
Picture it as an infinite sized disk and you're pretty close.
但他们最初只对存储在那里的数据收费。
But they would they originally they only charged for the data you stored there.
很好。
Great.
人们存储了大量大小为零字节的对象。
People stored a bunch of objects that were zero byte in size.
于是他们只是利用这些对象的名称以及查询对象是否存在及其名称的能力，当作一个免费的数据库。
So they just used the names of those objects and the ability to query whether object existed or not and what its name was as a free database.
对。
Right.
当时，根据传说，这不仅仅是为了免费使用服务。
This at the time also, according to legend, didn't just it wasn't just about taking free services.
这还严重拖慢了 S3 当时的性能。
It was also just crippling the performance of s three at the time.
因此，他们开始对每次请求都收取一小笔费用。
So what they did is they started also charging every request you make would charge a small fee.
我记得大概是每千次请求收一分钱左右。
I think it's something like a penny per thousand requests if memory serves.
别。
Don't
别拿我说的话当真。
quote me on that.
等重要时再查一下。
Look it up when it matters.
这样就解决了这个问题。
And that solves that problem.
但随后你又会遇到这类情况，被修补、与各种东西挂钩，再与更多东西绑定。
But then you have things like that that are patched on and tied to things and tied to things and tied to more things.
在任何复杂的组织中，情况往往是突然间你有了一个每年花费一亿美元的混乱局面。
And in any complex organization, what happens is is now suddenly you have this morass of a $100,000,000 a year in spend.
很好。
Great.
里面有什么东西在花掉这么多钱？
What's in there that's costing all the money?
嗯，所有东西都在花。
Well, everything.
但如果你看任何单个项目，觉得它没被使用，好吧。
But if you look at anything individual and you don't think it's being used, okay.
我要把它关掉。
Am I gonna turn I'm gonna turn it off.
如果我猜对了，就能省下一点钱。
If I'm right, it's going to save me a bit of money.
如果我猜错了，就会导致生产中断，现在我面临一个严重的问题，无法服务我的客户。
If I'm wrong, it's going to take down production, and now I have a serious problem trying to serve my customers.
所以，稳妥的做法是不要关掉任何东西。
So the safe bias is don't turn things off.
而且，你假设是个工程师。
Also, you're an engineer, hypothetically.
我意识到这可能会冒犯你所在地区的人。
I realize that can be an insult to your part of the world.
我暂且接受这一点。
I I Roll with it for a moment.
对。
Right.
你是个工程师。
You're an engineer.
你希望某个东西启动，但没人让你启动它。
You want something to spin up, and no one is letting you spin it up.
你会每隔十分钟就去烦他们，直到你获得所需的权限。
You will take up residence annoying the hell out of them every ten minutes until you get the access that you need.
对。
Right.
对。
Right.
你做好你的工作。
You you do your job.
事情都办妥了。
Things are done.
现在有两件事对你不利。
Two things now work against you.
第一，你不得不请求、恳求他们帮你启动它。
One, you had to ask and beg and plead them to spin that up.
你希望保留它，以防以后还需要，因为你不想再经历一遍那种过程。
You want to hang on to it in case you need it again because you don't want to go through that.
但还有一个组织层面的问题。
But also an organization.
没错。
Exactly.
没错。
Exactly.
是的。
Yeah.
这并不是亚马逊的错。
This is not Amazon's fault.
这是人性的错。
This is human nature's fault.
他们还没修复这个问题，不过我认为他们正在努力，他们在迭代 Alexa Plus。
They haven't patched that yet, though they're I think they're trying to They work iteration of Alexa Plus.
为什么不呢？
It why not?
我开玩笑的。
I'm kidding.
那东西就是用来卖广告的。
That thing just sells ads.
我们没问题。
We're good.
天啊。
Jesus.
等等。
But wait.
我是个工程师，我一直在为我的观点发声。
So I'm an engineer, and I've I've advocated for my thing.
是的。
Yes.
而现在你已经不再使用它了。
And now you're done using it.
没人会去，也没有任何行动号召让你站起来去
No no one is going there's no there's no call to action to get you off your ass to go
把它
and turn it
关掉，也不会有人来催你或烦你。
off, or no one will come and badger you about it.
它只是自然而然地存在着。
It just it tends to exist.
这就是为什么人们愤世嫉俗地说，你在云上付费的不是你使用的东西，而是你忘记关闭的东西。
It's it's why people cynically say you're not charged for the things you use in the cloud so much as you're charged for the things you forget to turn off.
我猜亚马逊很喜欢这一点。
And Amazon loves this, I'm guessing.
他们喜欢这种云锚定系统。
They love the fact that there's just this cloud anchor system.
我想说，在某种程度上，与普遍看法相反，他们并没有你想象的那么喜欢，为什么呢？
I I would say that on some level, contrary to popular belief, they don't love it as much as you'd think because Why no?
我试着在我的博客上运行，已经启动了八次这个博客，所以现在它的费用是应有费用的八倍。
I'm trying to run on my blog, and I've spun up eight attempts at this blog, so it now costs eight times what it should.
太好了。
Great.
而如果他们不谨慎对待这些毫无作用的浪费性使用，云服务就会贵得让人裤衩都吓飞了。
And the narrative becomes if they're not careful with all that wasted usage that's doing nothing is, The cloud is pants shittingly expensive.
我们应该回到数据中心。
We should go back to data centers.
这个问题的另一方根本没人。
There's really no one on the other side of this issue.
那为什么亚马逊不帮助人们关闭这些不必要的服务呢？
So why doesn't Amazon help people turn tune these things down and turn it off?
他们曾在组织的某个部分尝试过，但这可能会让你惊讶。
They made some abortive attempts in part of the organization, but this is gonna surprise you.
亚马逊并不是你见过的最有组织能力的公司。
Amazon is not the most organizationally competent company you're ever gonna meet.
但这在实际中意味着什么？
But what does that mean in practice?
只是因为他们分不清前后左右吗？
Is it just that they don't know their ass from their ear hole?
他们根本不懂事情是怎么运作的？
They don't really know how things work?
以上所有？
All of the above?
孤岛。
Silos.
全是孤岛。
It's all silos.
有超过200项服务，我们早就超过了那个阶段——我可以非常有说服力地谈论一些根本不存在的AWS服务，而亚马逊员工也不会指出我错了。
There are over 200 services, and we long ago crossed the point where I can speak incredibly convincingly about AWS services that do not exist to Amazon employees and not get called out on it.
因为这世上谁会知道亚马逊所做的一切呢？
Because who in the world knows everything that Amazon does?
我不认为有人知道。
And I don't think guy.
我现在说话听起来就像AI一样可信。
I sound as plausible as AI these days.
是的。
Yeah.
所以这就像一个迷宫，一个由各种SKU和产品类别组成的庞然大物，
I so it's just an a labyrinth, a leviathan of different SKUs and product categories that
如果要精确的话，超过三百万个SKU。
Over 3,000,000 SKUs, if you want to be precise.
是的。
Yes.
天啊。
Jesus.
这些都只是不同类型的虚拟机之类的东西吗？
And those are all just different kinds of virtual machines and the like?
好吧。
Okay.
我们就从这里开始。
We'll start there.
当然。
Sure.
虚拟机。
Virtual machines.
你想在弗吉尼亚州，美国东部1区启动一台虚拟机。
You wanna spin up a virtual machine in Virginia, US East 1.
好的。
Okay.
你会选择哪一种？
Which one are you going to pick?
因为大约有700种不同类型的虚拟机可供选择。
Because there are approximately 700 different types you can spin up.
你会选对吗？
Are you going to pick the right one?
当然，你根本选不对。
Of course, you're fucking not.
而且，你怎么才能选对呢？
Also, how do you pick the right one?
问题是，看起来那些决定在AWS上选择哪种虚拟机或启动什么实例的人都是工程师。
Like, this is the thing, because it seems that the people making these decisions of which virtual machines to get or whatever spin up on AWS are engineers.
对吧？
Right?
通常是这样。
Usually.
软件工程师。
Software engineers.
软件工程师不一定就是基础设施专家，对吧？
Software engineers aren't necessarily infrastructure experts, are they?
你注意到了。
You've noticed.
是的。
Yeah.
很多时候，公司会统一采用某种特定规格的实例。
Very often companies will standardize on one particular size of instance across the board.
为什么？
Why?
因为在开发阶段，这是最初创始工程师选定的，我们之后会再回头看看。
Well, because in development, it's the one that the original founding engineer picked and we'll come back and revisit this later.
十年前，没人重新审视过。
Ten years ago, no one has revisited.
就这么定了。
Take that.
他们确实有一个我挺喜欢的工具。
They do have a tool that I like.
而且还是免费的，我敢肯定有人为此睡不着觉，叫作计算优化器，它会分析你正在运行的工作负载，告诉你：这个应该调大。
It's even free, which I'm sure someone loses sleep over called Compute Optimizer that looks at your running workloads and says, hey, this one should be bigger.
这个应该调小。
This one should be smaller.
刚推出的时候，我非常怀疑。
I was extraordinarily skeptical when it came out.
它在判断这一点上比我自建的工具还要出色。
It is better at figuring that out than I am with the tooling that I built internally.
所以再次说明，我非常高兴能关掉它。
So once again, I'm thrilled to turn it off.
这是亚马逊最出色的功能之一。
One of the it's one of the Amazon at its best things.
我只能推测，负责这个功能的人正面临被裁员或降级的风险，因为
I can only assume that the people responsible for that are being eyed for layoffs or pips because
安迪·贾西和几名NBA射手现在正位于他们所在位置之外。
Andy Jassy and several NBA snipers are outside of their location right now.
这让人很沮丧。
It's frustrating.
我喜欢安迪·佩奈斯的安排方式。
I I like how Andy Penais has ordered.
AWS，确实如此。
AWS, truly.
有些事情发生了变化。
Something changed.
他得到了一份正常人根本不可能想要、但也无法拒绝的工作。
He got a job that no one in their right mind could possibly want, but also could not refuse.
他感觉自己出卖了灵魂。
And feel like he sold his soul.
是的。
Yeah.
当然。
Absolutely.
当他刚刚担任AWS首席执行官的时候。
When he just he was the CEO of AWS.
他确实是。
And He was.
当他发表主题演讲、做演示时，当我有幸偷偷溜进去旁听时，你能看到他身上无法抑制的人性在边缘流露出来。
He when he gave keynote talks, when he gave presentations, when I was fortunate enough to basically sneak my way into them, you could see the irrepressible humanity leaking out around the edges.
他曾经是一个真实的人。
He was a person.
现在他只是一个象征性的人物。
Now he is a figurehead.
他不能再直接与客户交流了。
He doesn't get to talk to customers
为了明确一下。
Just to be clear.
而是对国会。
To Congress.
这种转变发生在他成为亚马逊首席执行官还是AWS首席执行官的时候。
This shift was when he became CEO of Amazon or CEO of AWS.
亚马逊的首席执行官。
CEO of Amazon.
AWS是他自己的领地，据我所知，他可以在那个公司里为所欲为。
AWS was his own fiefdom, and he could do whatever he wanted, my to my understanding, within that company.
当时，唯一在乎这件事的人只有安迪·贾西和上帝。
Was, the only people that mattered where they're who had an opinion on this were Andy Jassy and God.
而上帝基本上缺席了，所以很好。
And God was sort of absent, so great.
我们就把这事交给安迪吧。
We're just gonna leave it to Andy.
安迪深入钻研了。
Andy went deep.
安迪对这些内容的每一个方面都了如指掌。
Andy understood every aspect of this stuff.
是一年前吗？
That a year ago?
他去拉斯维加斯重新打造了这个项目，并在舞台上发表了演讲。
He went to reinvent in Las Vegas and gave a talk on stage.
我想是因为他错过了它。
I think because he missed it.
他错过了能上台和人们讨论计算机的机会，因为他现在得把内裤发往世界各地。
He missed being able to get up there and talk about computers with people because now he has to ship underpants all over the world instead.
对。
Right.
但他懂技术吗？
But is he technical?
因为他是个NBA，不是吗？
Because he's an NBA, is he not?
这是其中之一
This is one of
我发现人们常常低估亚马逊领导层的一个危险之处在于，永远不要假设他们不理解自己所从事事物的技术细节。
the dangerous things that I found people underestimate Amazon leadership on is it is never the smart bet to assume that they don't understand the technical nuances of the things that they are working with.
他们在许多方面都深入到令人讽刺的程度，因此断言他们不懂技术从来都不是明智之举。
They go so sarcastically deep in so many different ways that it is never a smart move to say that they don't understand.
这几乎不同于我所接触过的任何其他科技公司，除了那些由技术创始人创办的小型公司。
It is unlike virtually any other tech company I've ever spoken with other than small scale ones with technical founders.
他非常清楚地理解这项技术，并亲自用它来构建东西。
He very clearly understands the technology and used it to build things himself.
是的。
Right.
但这就是我质疑的原因。
But here's the reason I I question that.
请说。
Please.
他在谈论生成式人工智能。
He's talking about generative AI.
我刚查了他发布的公告。
I just pulled up his notice.
哦，没人真正理解其底层原理。
Oh, no one understands that that works under the hood.
我们必须明确这一点。
Let's be very clear on that.
而且，是的，他在生成式AI这个问题上彻底失控了。
And, yeah, he lost the fucking plot when it came to Gen AI.
这就是关键。
That's the the thing.
因为他觉得，我们也在整个内部运营和配送网络中广泛使用生成式AI。
Because he's like, we're also using generative AI broadly across our internal operations, in our fulfillment network.
我们用AI来优化库存布局。
We're using AI to improve inventory placement.
所以不是生成式AI。
So not generative AI.
需求预测，绝对不是生成式AI。
Demand forecasting, definitely not generative AI.
还有我们机器人的效率，也不是生成式AI。
And the efficiency of our robots, not generative AI.
我不确定。
I don't know.
也许他确实知道自己在说什么。
Maybe he does know what he's talking about.
他现在就是一个他妈的街头叫卖者。
He's just a fucking carnival barker now.
他也在做同样的事。
He's doing the same.
一切似乎都是这样。
That it all seems to be.
要驳斥你刚才转述给我们的关于他所说的每一件事，方法非常简单。
And the trick to disprove everything you just said about that about everything that he said that you relayed to us is very simple.
找个亚马逊员工出去喝几杯，然后问问他们对公司内部使用AI的体验。
Go out with an Amazon employee for several drinks and then ask them about their experience of using AI within the company.
我知道。
Oh, I know.
确实是
It is
两家不同的公司。
two different companies.
这两种观点之间没有任何一致性。
There there is no, there is no, congruity between those two perspectives.
所以，科里，也许你能帮我一下，因为你确实知道。
So, Corey, maybe you can help me with this because you Certainly.
不用具体说我们谈的是什么，你就知道我后来才知道的那份合同，而我提前几个月就知道了，是的。
Without saying exactly what we're talking about, you knew about a contract that I knew about way before I knew, and I knew about it months early Yes.
我们讨论过一家大公司。
With a with a large company we discussed.
嗯。
Mhmm.
所以你人脉很广。
So you're very well connected.
在生成式AI出现之前，AWS至少是一个还算合理的运营，我们至少关注规模和实用的云计算。
It seems up until generative AI, like AWS was at least a somewhat sensible operation, that it was something that we had, if not focus, at least a focus on scale and useful cloud computing.
是的。
Yes.
到底是什么关于GPU的东西让所有人都疯了？
What the fuck is it about GPUs that's driving everyone insane?
因为
Because
我想指出一点，这一点在行业中正被忽视，因为我跟AWS的人谈过，也跟离开AWS的人谈过，我曾经稍微涉足过数据中心，所以我知道得足够多，知道自己不懂什么。
it's I want to point something out that I think is being lost industry because I've talked to folks at AWS about this and folks who have left AWS who have I I used to dabble working in data centers, which means I know just enough to know what I don't know.
我跟这些人谈过。
And I've talked to these folks.
当亚马逊承诺能够上线特定数量的算力时，那是真实的算力。
When Amazon commits to being able to bring certain amounts of capacity online, it is real capacity.
他们每建设一个数据中心、每扩展一个区域，都是数十亿美元的投资，从签订合同到实际服务客户流量，需要数年时间。
It is the each data center they build out each region they build out is a multibillion dollar investment, and it takes years to go from signed contract to it serving customer traffic.
在这件事上，他们是认真的。
They are real when it comes to this stuff.
还有其他一些基础设施。
There are a number of other building.
是的。
Yeah.
没错。
Well, yes.
而且他们真的在运营这些设施。
And they they're really running it.
他们在如何更好地运行这些系统方面已经实现了高度专业化，远超你我所能达到的水平。
They have operationalized how to run these things better than you or I would ever be able to do.
这才是他们的核心优势。
That is their superpower.
其他公司，当他们新建数据中心时，并没有这样做。他们只是，嗯，我们租了个仓库。
Other companies, and they're spinning up a data center, do not do They are well, we've we rented a warehouse out.
我们找来一些发电机放在后面，然后联系康卡斯特，看看他们能怎么帮我们接入互联网。
We round up throwing some generators in the back, and we're gonna call Comcast and see what they can do to get it hooked up to the Internet next.
当你说到其他公司时
When you say other
准备就绪的公司。
companies ready to go.
是的。
Yeah.
但你说其他公司，不是指谷歌和微软吧。
But you say other companies, you don't mean Google and Microsoft here.
不是。
No.
我在说甲骨文，不过没错。
I'm talk well, I I am talking Oracle, but yes.
我的意思是微软那种。
My Microsoft kind of.
微软只是同意了。
Microsoft just agrees.
微软在建设数据中心方面很糟糕。
Microsoft is shitty at building data centers.
Azure 在各个方面都令人失望。
Azure has been a disappointment across the board.
在新冠疫情期间，他们长期面临容量不足的问题。
Capacity shortfalls plagued them for a long time during COVID.
他们通常把数据中心里的两个机架就当作一个区域，这很奇怪，也很混乱。
They, generally tend to view two racks in a data center somewhere as a region, which is weird and messed up.
谷歌是认真的，但谷歌在其他技术取舍上做了些有趣的选择，我能理解他们为什么这么做。
Google is real, but Google has made other interesting technical trade offs that I can see why they did it.
但就基础设施而言，根据我的经验，AWS 是第一名。
But I would say when it comes to infrastructure, in my experience, AWS is number one.
谷歌紧随其后，位居第二。
Google is a close second.
我听说过一些关于谷歌的疯狂传闻，说他们搞一些叫‘stalks’的东西，就像突然搭建起临时数据中心一样。
I have heard some crazy shit about Google that they're doing these things called stalks where they just, like like, create these pop up data centers.
嗯嗯。
Mhmm.
很长时间以来都是这样。
The for a long time.
每家公司都会为了某些目的这么做。
Every company does that for points of stuff.
是的。
Yes.
不过，回到主要问题，AWS 看起来是一个务实且做实事的生意。
Anyway, but back to the major question, which is AWS seems like a sensible business that does real things.
我并不否认这一点。
I'm not disputing that.
是的。
Yes.
但它能印钱。
But It prints money.
我们把这一点说清楚。
Let's be clear on this.
尽管大家都记得，我帮助公司谈判。
Despite all remember, I help companies negotiate Prints money.
云合同，并了解他们的支出和资金流向，这意味着公司可以提出很多要求。
Cloud contracts and understand what they're spending and where the money is going, which means that companies can say a lot.
人们总是说很多话。
People say a lot of things.
但我关于这类事情的首要原则之一是，客户大多是在欺骗自己。
But one of my prime rules about this stuff is that customers lie mostly to themselves.
当他们告诉我发生了什么时，他们并非有意误导我，但我发现账单才是最终的真相来源。
They they're not intentionally trying to mislead me when they tell me what's going on, but I find the bills are the ultimate source of truth.
因为如果你没有为之付费，它真的存在吗？
Because if you're not paying for it, does it really exist?
根据我的经验，除了少数例外，这些公司中大多数的人工智能支出，占其总基础设施支出的比例在5%到7%之间徘徊。
And the AI spend in most of these companies hovers, in my experience with a few outliers, between 57% of their total infrastructure spend.
对。
Right.
并不是三倍。
It is not three times it.
那些每年与AWS签订3亿美元合同的公司，并不会因为生成式AI就立刻说，把合同金额提高到4亿。
I companies that are doing a $300,000,000 a year contract with AWS are not turning around saying, better make it 400 because of all the Gen AI.
这种情况根本不会发生。
That is not happening.
经营企业却不考虑播客？
Run a business and not thinking about podcasting?
再想想吧。
Think again.
听播客的美国人比使用Spotify和Pandora等广告支持的流媒体音乐服务的人还多。
More Americans listen to podcasts than ad supported streaming music from Spotify and Pandora.
作为排名第一的播客平台，iHeart的规模是第二名和第三名加起来的两倍。
And as the number one podcaster, iHeart's twice as large as the next two combined.
所以，无论你的客户听什么，他们都会听到你的信息。
So whatever your customers listen to, they'll hear your message.
此外，只有iHeart能将你的信息扩展到广播电台的受众。
Plus, only iHeart can extend your message to audiences across broadcast radio.
你觉得播客能帮助你的业务吗？
Think podcasting can help your business?
想到播客，就想到iHeart。
Think iHeart.
流媒体、广播和播客。
Streaming, radio, and podcasting.
拨打844844开始吧。
Call 844844 to get started.
那就是844844。一个富有野心、善意、强势且富有的母亲在黑人社区中是什么样子。
That's 844844 A ambitious, well intentioned, ferocious, and wealthy mother looks like in the black community.
在这个女性历史月，播客《保持积极，亲爱的》颂扬了女性在生活混乱时依然选择疗愈、目标和信仰的力量。
This woman's history month, the podcast, keep it positive, sweetie, celebrates the power of women choosing healing, purpose, and faith even when life gets messy.
展开剩余字幕（还有 480 条）
爱不是一个终点。
Love is not a destination.
你必须每天为之努力。
You have to work on it every day.
《Keep It Positive, Sweetie》为关于自我价值、爱、成长以及以优雅与坚韧应对生活的坦诚对话创造了空间，由那些激励他人、鼓舞人心并勇敢说出真相的女性主导。
Keep it Positive Sweetie creates space for honest conversations on self worth, love, growth, and navigating life with grace and grit led by women who uplift, inspire, and tell the truth out loud.
我与上帝进行了多次对话，我知道为什么花了二十年。
I have several conversations with God, and I know why it took twenty years.
要收听这些内容及其他更多节目，请在iHeartRadio应用、Apple播客或您收听播客的任何平台收听《Keep It Positive, Sweetie》。
To hear this and more, listen to keep it positive, sweetie, on the iHeartRadio app, Apple podcast, or wherever you get your podcast.
这是全新的我，也是旧日的他们。
It's the new me, and it's the old them.
每个人都在自己的
Everybody's on their
旅程中，而你的旅程与他们的不同。
journey, and your journey is different to theirs.
今年妇女历史月，播客《如果你知道得更多》与安伯·格莱姆斯合作，聚焦那些将失误转化为动力、将教训转化为力量的女性。
This woman's history month, the podcast, if you knew better with Amber Grimes, spotlights women who turn missteps into momentum and lessons into power.
我认为，从我出身的环境来看，我来自布朗克斯。
I think coming out of where I came from, I'm from The Bronx.
我觉得我小时候非常贫穷。
I think I grew up really poor.
但那时我并不知道，因为我一直用我的创造力来美化生活。
I didn't know that then because I very much use my creativity to romanticize life.
我觉得我妈妈做得非常好，你退后一步看，就会惊叹：天啊。
And I'm like, my mom did a really good job of, like, you step back and you're like, woah.
我不知道我们是怎么熬过来的。
We I don't know how we made it.
所以我的一生很大程度上是靠着生存的本能，只为到达下一个阶段。
So a lot of my life was, like, built out of, like, survival to get to the next place.
我的动力，我那种一心一意要变得更好的执着。
Like, my drive, my, like, tunnel vision of, like, I gotta be better.
我必须实现这个目标，源于我想为我们创造更好生活的愿望。
I gotta achieve this was off the strengths of, like, I wanna make a better life for us.
《If You Knew Better》邀请亲身经历过的女性分享真实对话，深入探讨职业转型、感情教训以及改变一切的心态转变。
If You Knew Better brings real talk from women who've lived it, unpacking career pivots, relationship lessons, and the mindset shifts that changed everything.
请在 iHeartRadio 应用、Apple 播客或您收听播客的任何平台收听 Amber Grimes 主持的《If You Knew Better》。
Listen to if you knew better with Amber Grimes on the iHeartRadio app, Apple Podcasts, or wherever you get your podcast.
你好。
Hi.
我是《Bookmarked》节目的主持人丹妮尔·罗贝特，这个播客由蕾西的读书俱乐部出品。
I'm Danielle Robet, host of Bookmarked, the podcast by Reese's Book Club.
本周在《Bookmarked》中，我们几乎就是在举办一场终极女孩之夜。
And this week on Bookmarked, we're basically hosting the ultimate girls night.
蕾西·威瑟斯彭、詹妮弗·加纳、朱迪·格雷尔、丽塔·威尔逊、加里·莱斯以及作者劳拉·戴维。
Reese Witherspoon, Jennifer Garner, Judy Greer, Rita Wilson, and Gary Rice, and author Laura Dave.
她们正是 Apple TV 系列剧《他告诉我的最后一件事》第二季背后的女性角色。
These are the women behind season two of the Apple TV series, The Last Thing He Told Me.
我们正在讨论如何将一本书打造成热门剧集，以及真正实现这一目标需要什么。
We're talking about turning a book into a hit show and what it really takes to
将一个故事生动地呈现出来。
bring a story to life.
对我来说，最重要的衡量标准是我是否愿意把这本书推荐给别人？
The most important metric for me is do I want to share this book with somebody?
这才能建立起社群，这也是我们读书会的核心理念，我们创办它的初衷就是为了连接人们。
That's what creates community, and that's the main thesis of our book club and why we started it, was just to connect people together.
在 iHeartRadio 应用、Apple 播客或你收听播客的任何平台收听由 Reese 的读书会出品的《Bookmarked》。
Listen to the Bookmarked by Reese's Book Club podcast on the iHeartRadio app, Apple Podcasts, or wherever you get your podcasts.
但那正是我想表达的重点。
That's kind of my point, though.
就像亚马逊云服务看起来并不像是一个追逐潮流的公司，即使它追逐潮流，也是以一种更可持续的方式。
It's like Amazon Web Services doesn't seem like a business that chases fads, or if it does, it chase them in a more sustainable way.
值得一提的是，他们几乎完全放弃了区块链，我当时还担心他们会深陷其中。
They're now To their credit, they docked they docked blockchain almost entirely, and I I was worried they were gonna fall down that rabbit hole.
但我的意思是，他们今年的资本支出高达两千亿美元。
But, I mean, they're doing $200,000,000,000 in Capex this year.
嗯。
Mhmm.
他们到底在人工智能身上看到了什么？
What is it they see in AI?
因为按照你刚才的说法，一个三亿美元合同的百分之七到八，那简直是垃圾。
Because to your point just now, what, like seven to 8% of a 300,000,000 contract, that's dog shit.
这根本不可能收回那笔资本支出，桑迪。
Like, that's that's not gonna pay back that CapEx, Sandy.
我相信，
I believe,
这是我的看法。
this is my belief here.
我只能基于我所看到的亚马逊的所作所为来判断。
I can't base it on anything other than what I've seen Amazon do.
根据我所看到的客户情况，亚马逊正在迅速部署并出售他们正在快速搭建的算力。
I and what I've seen from customers, the Amazon is selling the capacity that they are spinning up they are spinning up as fast as they can spin it.
我有两个问题。
The question that I have is twofold.
第一，假设你们投资了这么多AI数据中心，但AI泡沫破裂、消退了，怎么办？
One, okay, you you invest in all these AI data centers and the AI bubble pops, deflates, whatever.
这些设施和资本支出还能被重新利用吗？
How repurposable are those facilities and that capital expenditure?
如果是数据中心和网络设备，是的，非常容易重新利用。
If it's data centers and networking, yes, very repurposable.
是的。
Yeah.
GPU呢？
GPUs?
嗯，再过几年，我们都会大量投入在线游戏。
Well, we're all gonna get really into online gaming in a few years.
我不知道。
I don't know.
但它们没有视频输出。
Except they don't have video out.
它们没有DisplayPort接口。
They don't have DisplayPort.
是的。
Yes.
但我们仍然称它们为图形处理器，这我很喜欢。
Yet we still call them graphics, which I I love.
我觉得我喜欢那些人
I think it's I like the people who
我也听过有人称它们为通用处理器，但这恰恰与GPU的性质相反。
I like I've also heard people say general purpose units, which is the opposite of what a GPU is.
因为它们根本不是通用的。
Like, it is very much not general purpose.
这通过一份新闻稿公之于众，他们与OpenAI签署了一份380亿美元的合同。
It was it became public through a press release that they have signed a $38,000,000,000 contract with OpenAI.
与
With
疯狂。
insane.
我毫不怀疑，因为我对此做了一些调查。
There is zero doubt in my mind that because I did some digging on this.
AWS有没有可能提供380亿美元的算力给OpenAI？
Is there line of sight for AWS to bring online $38,000,000,000 of capacity to provide to OpenAI?
有。
Yes.
有。
Yes.
有的。
There is.
我的疑虑在于，OpenAI值不值这380亿美元？
Where my doubts come in, is OpenAI good for the money?
不值。
No.
我这里有一些问题，主要是基于阅读你的作品。
I have some questions here, mostly from reading your work.
这正是关键所在。
And that's the thing.
看起来，GPU和大语言模型让所有人都崩溃了。
It seems like something about GPUs and large language models have broken everyone.
这正是我想表达的，因为我读了所有关于AWS如何实现盈利的报道——因为我是个怪人。
That's kind of the point I'm getting to because like everything I've read about AWS and I read because I'm a dickhead.
我专门去读了AWS实现盈利之前的所有相关报道。
I went and I read all of the previous coverage of the lead up to AWS becoming profitable.
顺便说一句，凯文·罗斯就在他们盈利前一个月，还写了一篇文章说还需要一段时间。
Kevin Roose, by the way, literally a month before they did so, wrote an article saying it would be a while.
那个混蛋的三个屁点。
What fucking three points in that asswipe.
总之，我读到的所有内容都是说，亚马逊花得太狠了。
Anyway, everything I read was everyone saying, oh, Amazon's being really spendy.
他们每个季度的资本支出高达60亿到70亿美元。
They're spending 6 to 7,000,000,000 in CapEx a quarter.
哦，天哪。
Oh, no.
而总的资本支出达到了690亿美元。
And then the total CapEx of 69,000,000,000.
但亚马逊方面的说法都是：我们在打造一个可持续的东西。
But everything from Amazon was like, we're building something sustainable.
我们看到了它背后的收入潜力。
We see the revenue potential behind it.
一切都非常无聊。
It was all very boring.
现在说到AWS，他们谈论大语言模型时，听起来就像都吸了死藤水一样。
Now with AWS, everything they talk about with large language models, it sounds like they've all been huffing Ayahuasca or something.
他们听起来疯了。
They sound insane.
确实如此。
They do.
这告诉我，只有两种可能。
Which tells me one of two things is true.
或者老实说，这就是现实世界。
Or honestly, this is the this is the real world.
不同于推特，在现实中，两个复杂的对立事物完全可以同时为真。
Unlike on Twitter, two complex competing things can in fact be true at the same time.
其中之一是，他们仍在做大量那些无聊的事情，但所有的舆论关注都集中在那些离谱的项目上。
One of them is they're still doing a lot of that boring stuff, but the the publicity all accrues to the things that are far flung.
这还只是亚马逊整个公司2000亿美元的开支。
This is also $200,000,000,000 across the entirety of Amazon.
这还包括配送中心。
That includes distribution centers.
这还包括低地球轨道，基本上就是亚马逊的Starlink。
That includes low Earth orbit, Basically, Amazon basic Starlink.
我想他们现在称之为LEO。
I think they're calling it LEO now.
这还包括他们的AI胡扯、工厂浪费、机器人项目、办公室装修等等。
It includes, yes, their AI nonsense, their factory boondoggles, their robotic stuff, office build outs, etcetera, etcetera.
他们从不明确说明其中有多少是AI服务，而且即使他们说明了，我也不信。
They don't give clear breakdowns on how much of that is AI services, and frankly, wouldn't trust them if they did.
自从几年前我在amazon.jobs上看到一则招聘广告，要求有其他AI服务如S3的经验以来，我就不再相信了。
Not since I saw a job ad a few years back on amazon.jobs saying experience with other AI services like s three.
S3存储服务。
S three storage service.
那根本不是AI。
That's not AI.
可以用来为人工智能提供数据，但我听说
Can use it to feed AI, but I hear
Anthropic 是我听说过的最大的 S3 客户之一。
Anthropic is Anthropic is one of the largest s three customers I've heard.
我听说
I've heard
到目前为止，我从七家不同的公司那里听说，他们是 S3 最大的客户。
that I've heard from seven different companies so far that they are the largest customer of s three.
是的。
Yeah.
这结果还挺有意思的。
It's funny how that tends to work out.
天啊。
God.
我好怀念以前做云软件的时候，那时候每个人都自称是独家合作伙伴。
I miss I miss working in cloud software so much where everyone is the everyone's the exclusive partner.
每个人都是第一名的合作伙伴，每个人都是最大的客户。
Everyone's the number one partner and everyone's the largest.
这太
It's so
但 qualifiers 就这样出现了。
But And the qualifiers get there.
比如，我们是澳大利亚最大的 S3 客户。
Like, we're the largest s three customer in Australia.
很好。
Great.
在东北地区，早上6点到午夜之间很好。
Good for In the Northeast Region between the hours of 6AM and 12AM.
没错。
Exactly.
好的。
Okay.
关于资本支出，情况是这样的。
So here's the thing with the CapEx.
我确信其中很大一部分是用于人工智能的，因为如果你看看他们过去几年的资本支出，大概在400亿、500亿、600亿左右。
I'm sure a lot of it is for is for AI just because if you look at their previous years of CapEx, was like 4050, '60.
是的。
Yeah.
406亿、63亿、52亿、83亿、1310亿、2000亿美元。
4061, '63, '52, '83, a hundred and thirty one, two hundred billion.
长期以来，平台基因组学的杰拉尔德·菲茨杰拉德一直在追踪超大规模厂商的资本支出。
Gerald Fitzgerald over at Platform Endomics has been tracking CapEx spend to the hyperscalers for a long time now.
他是最早指出所谓‘云伪玩家’——特别是IBM和甲骨文——的人之一，因为他们虽然大谈云计算，却根本没有在资本支出上投入，而要实现这些，你必须建设数据中心。
And he was one of the first people to call out the, cloud pretenders, IBM and Oracle specifically, because they're making all cloud noises, but they're not spending anything on CapEx and you kind of have to build the data centers for this to work.
别担心，甲骨文正在处理这件事。
Not to worry, Oracle's taking care of that.
我想知道的是，当你开始说，好吧，我们打算投入，即使假设全部用于人工智能，也没关系，没问题。
What I want to know, my question is when you start, okay, we're going to spend, even assume it's all AI, whatever, Fine.
2000亿美元用于人工智能支出。
$200,000,000,000 on AI spent.
其中有多少是以支票形式支付给英伟达的？
How what how much of that is just in the form of a check written to NVIDIA?
这正是问题所在。
And that's the question.
因为据我了解，我应该展示一下图表。
Because from my understanding my understanding, the I should bring up the chart.
我现在就现场演示一下。
I'm just gonna do something live on air.
我正要调出一张图片，当然，现在找不到。
I'm just bringing up an image and, of course, can't find it.
不过，我有一张图，显示亚马逊并不完全依赖外部供应商，它自己研发训练用GPU，它们现在还在用Inferentia吗，还是
Nevertheless, there is an image I have that Amazon doesn't well, Amazon builds its own the training GPUs and do they still use Inferentia, or is it
只用Tranium？
just Tranium?
听起来像是我爷爷被诊断出的病。
Sounds like something my grandpa got diagnosed with.
是的。
Yeah.
但他们真的用吗？
But do they?
比如，Inferentia，原本以为它存在于
Like, Inferentia, thought was there in
哦，他们经常提到它。
Oh, they talk about it a fair bit.
他们谈论Tranium。
They talk about Tranium.
问题是，你没看到有人在实际中使用它。
And the thing is is what you don't see is people using it in the wild.
我研究了一些o lama的东西，那是你自己运行的那部分。
I looked at a bunch of stuff with o lama, that's the run your own stuff there.
我大约一年前搜索过“Inferentia”这个术语。
I searched about a year ago for the term Inferentia.
它只在一条被关闭的拉取请求中出现过一次，该请求因六十天无人回复而被关闭。
It showed up once in a pull request that had been closed after sixty days because no one responded to it.
没有人真正使用这个技术。
No one is using this in the real world.
一年前，也就是现在这个月份，他们推出了Tranium 2，当时Anthropic的发言人登上了主题演讲舞台谈论它，是的。
A year ago, you're in the month now, they had with Tranium two, I think it was, they had speakers from Anthropic on keynote stage talking about it, which yeah.
好吧。
Okay.
他们投资了数十亿美元在Anthropic上？
They invested how many tens of billions into ropic?
是的。
Yeah.
我敢肯定，让高管上台演讲是有合同要求的。
I'm sure there's a contract requirement around sending an exec to talk on stage.
另一个是苹果。
And the other one was Apple.
苹果从不向自己的员工、彼此之间，更不用说公众，承认他们正在做什么。
Apple doesn't admit what they're doing to their own employees, to each other, let alone publicly.
所以，是的，这看起来就像一次口腔排便一样不自然。
So, yeah, that looked about as natural as an oral bowel movement.
不错。
Nice.
但那些
But those
是我见过的唯一两家在做这件事的公司。
are the only two companies I've seen doing anything with it.
根据我这张包含各种星标和公司名称的图表，亚马逊的GPU是通过富士康，也就是鸿海精密工业股份有限公司获得的。
So based on this chart I've got, which involves various stars and company names, it looks like Amazon gets their GPUs through Foxconn, also known as Hon Hai Precision Corporation Limited.
我喜欢中国和台湾的名字，由
I love Chinese and Taiwanese names, by
就是这样。
the way.
台积电几乎肯定在其中。
TSMC is almost certainly in there.
不。
No.
台积电不在其中，因为台积电制造芯片。
TSMC isn't because TSMC builds the chips.
不。
No.
这些是服务器制造商。
These are the server manufacturers.
他们自己生产服务器，因为他们拥有Anapurna。
Ships themselves because they have anapurna.
他们进行实验室收购和设计，但不进行晶圆制造。
Do the labs that they bought and do the design, but they don't do the fab.
所以他们发货，我假设亚马逊内部制造芯片的Anapurna部门会把芯片发给富士康、量子计算公司和捷普。
So they ship the I assume that anapurna, which is the internal place that Amazon makes its chips, ships them, it looks like, to Foxconn, quantum computing, and Jabil.
然后他们把这些芯片装进服务器里。
And then they put them in the servers.
所以，实际上是他们在向这些公司付款。
So, really, it's them cutting the checks to them.
我会密切关注这些台湾公司的月度财报，因为这里有个关键点。
I'm gonna be watching the monthly earnings of those companies in Taiwan quite viciously because here's the thing.
如果这2000亿美元的资本支出都是真的，如果他们真觉得自己能拿到这么多钱的话。
If it's all $200,000,000,000 of CapEx, if they really think they're gonna get paid that much.
但会发生什么呢？
But what happens?
亚马逊有没有遇到过过度建设的情况？
Has Amazon ever faced the situation where they did an overbuild?
这种情况以前发生过吗？
Has that ever happened to them?
他们做了什么？
What did they do?
是的。
Yes.
在疫情期间，他们对此大肆宣传。
They made a bunch of noises about it during the during the pandemic.
他们过度建设了他们的配送中心。
They overbuilt their distribution centers.
啊，但那是人们在买东西。
Ah, but that's people buy stuff.
没错。
Exactly.
这正是我的观点。
That was my position on it.
感觉就像是，好吧。
It feels like, okay.
他们放慢了步伐，但并没有关闭这些设施，而是等待人们逐渐适应。
Like, they slowed it down, and they didn't shut them down, but they waited for people to grow into it.
问题是，你用一个仓库来发货内裤。
The thing is is a warehouse that you use to ship out underpants.
太好了。
Great.
五年后，这个仓库的用途和今天差不多。
That has the same utility more or less five years from now as it does today.
很多这类芯片都是贬值资产。
A lot of these chips are depreciating assets.
嗯。
Mhmm.
天啊。
Like, oh, man.
没人会说：‘我真想赶紧把那台七年前的电脑搬进来，好干点正事。’
I can't wait to get that seven year old computer in here so I can do some real work, says no one ever.
另外，Trainium 尤其是，他们现在已经是第二代、第三代了？
Well also, Trainium especially, they've got what they're on generation two, three now?
他们正在推出第三代产品，同时提前预告第四代产品，
They're shipping three right in time to pre announce four,
然后试图效仿奥斯本电脑公司自己搞一套。
and then a desperate attempt to Osborn Computer themselves.
是的，解释一下这个典故。
Yeah, explain that reference.
哦，奥斯本电脑公司是八十年代的一家公司。
Oh, Osborn Computer was a company back in the eighties.
他们推出了奥斯本一号，他们的首席执行官出去搞了一大堆媒体宣传，说：是的，这很棒，但奥斯本二号将会彻底碾压它，
There they launched the Osborn one and their their CEO went out and did a whole press junk and he's like, yeah, this is great, but Osborn two is gonna blow the crap out
了它。
of it.
哦，就像英伟达那样。
Oh, like NVIDIA.
大家都说，太好了。
And everyone's like, great.
那我们就不买Osborn One了。
We're not gonna buy the Osborn one then.
这家公司随后倒闭了，Osborn Two也从未上市。
The company went under and never shipped the Osborn two.
好吧。
Well, okay.
那就跟NVIDIA不一样了。
Not like Nvidia then.
问题是这样。
This is the thing.
这肯定会让亚马逊在Trainium芯片的成功上陷入大麻烦。
Surely that sets Amazon up for a big problem with those Tranium chips success.
即使在某种奇怪的情况下，Trainium四代或五代非常出色——我觉得不太可能，但只是说，这难道不意味着他们会囤积一堆过时的Trainium芯片吗？
Like, even if in some weird world, Tranium four or five was amazing, I don't think it will be, but just saying, doesn't that mean they're gonna be sitting on a bunch of obsolete Tranium?
这感觉非常可疑。
It just feels very questionable.
你知道，你可以说OpenAI很多事，而且你也确实说了。
You know, you can say a lot about OpenAI, and you do.
但他们发布那份380亿美元合同的新闻稿时，根本没有提到Trainium。
But that press release where they announced that $38,000,000,000 contract, they didn't mention Trainium.
他们没提，对吧？
They didn't, did they?
这让我觉得，如果Trainium真像他们说的那样在模型训练上有一半那么厉害，OpenAI肯定会大力推广它。
Which tells me that if it were half as good at this at the model training as they say it is, OpenAI would be all about it.
是的。
Yeah.
我不认为这是真的。
I am not seeing that to be true.
嗯，那个有点紧张的萨姆·阿尔特曼，他确实和Cerebras签了个协议。
Well, clammy Sam Altman is he signed a thing with a Cerebras.
你是Cerebras的吗？
You you're of Cerebras?
萨姆·阿尔特曼有没有没签过协议的人？
Is there anyone that Sam Altman has not signed deals with?
我们来想想。
Let's think.
Grok。
Grock.
他还没和Grok签过协议。
Haven't signed one with Grok yet.
好的。
Okay.
很好。
Good.
很好。
Good.
我的意思是，其实，抱歉。
I mean, actually, sorry.
为了澄清一下，他还没和任何一个签署协议。
Just to be clear, hasn't signed with either.
嗯，我想他不会和Grok签协议，因为从技术上讲，Grok是个竞争对手。
Well, I guess it would he wouldn't sign one with Grok because technically Grok is a competitor.
我的意思是，g r o q，那个让人头疼的快速推理。
I mean, g r o q, the annoying fast inference.
我不确定。
I don't know.
到目前为止，我对Duhm和那些人已经很反感了，他们肯定会说这一集太技术化了。
At this point, my hate at Duhm and people are gonna say this episode was too technical.
而我的回答是：别激动。
And the answer is calm down.
这是我的播客。
This is my podcast.
我只是在想，他们在什么时候才会认输，好吧。
I'm just I'm wondering at what point they they say uncle because okay.
两千亿美元，其中大约一千二百五十亿美元是用于人工智能的。
$200,000,000,000 and let's say a 125,000,000,000 of that is AI.
对吧？
Right?
人工智能芯片。
AI chips.
大概需要四到五年才能安装完所有这些设备。
Takes four or five years probably to install all of that.
目的是什么？
To what end?
四五年后会发生什么？
What happens in four or five?
他们会制造出什么？
Are they gonna make?
我的意思是，可能会是
Well, I mean, would be
这正是没人能向我解释清楚的地方。
That's what no one is able to explain to me.
我们今天有哪些使用场景，如果我们的推理能力增加十倍，就能实现x？
What use cases today do are we looking at where, wow, if we had 10 times as much inference as we do today, then x would be possible.
求解x。
Solve for x.
我没听到过。
I'm not hearing it.
没有。
No.
我也没听到过。
I'm not either.
我甚至都没像往常一样当个喷子，尽管现在正是喷子季。
I'm not even being my usual hates self, even though this is hater season.
我真的从任何人那里都得不到这个答案，即使是那些支持者也不例外。
It's I genuinely can't get that answer out of anyone, even the boosters.
就好像对算力有着永不满足的需求一样。
It's like there's this insatiable demand for compute.
我不
I don't
不知道我们是不是要通过JSON来召唤神明作为第一步。
know if We're gonna summon god through JSON is step one.
第二步是，我们要问神该怎么做。
And step two is we're gonna ask god what to do.
我的意思是，这正是关键所在。
I mean, that's the thing.
我认为他们——你觉得有看过这些公司的收入预测吗？
I do think that they I think that, have you seen the revenue projections for these companies?
你有没有看过甲骨文的现金流？
Have you ever seen, like, Oracle's cash flow?
所以如果你看到这个
So if you see this
这件事也破了，意思是每次甲骨文的联合创始人拉里·埃里森去发表主题演讲或讲话时，都会先被提醒：你即将听到的一切，都是律师版本的虚构想象。
thing breaks also known as there's a reason that every time Larry Ellison, cofounder of Oracle, goes and gives a keynote or talks somewhere, they preface it with a disclosure that is the lawyer version of everything you are about to hear is a fanciful imagining of reality.
别听他的。
Don't listen to him.
他今天累坏了。
He's he's he's he's had a long day.
他还没午睡呢。
He hasn't taken his nap.
该死。
Fuck.
如果HR guy变成杰里·斯蒂勒那样。
If HR guy goes Jerry Stiller.
关键是，你看看Anthropic、OpenAI和甲骨文的现金流图，它们都遵循同一个模式：连续多年现金流惨淡，长期为负。
He the thing is, you look at these cash flow positive these cash flow diagrams for Anthropic, OpenAI and Oracle, they all follow the same thing, which is terrible cash flow, negative cash flow for years and years and years.
到了2029年，一切都会改变，数字开始上升。
And then in 2029, everything changes, number go up.
而到这个时候，我真心觉得他们认为自己会在2029年发明出AGI，这几乎是他们唯一的计划，因为他们并没有
And I I, at this point, genuinely think that they think they're going to invent AGI in 2029, that that's the only plan they have because They're not
他们并没有做那些如果真相信这一点就会去做的事，他们不会做这些事情。
they're not doing the things that They're not would if they genuinely believe that, they would not be doing these things.
我们认为，我们距离推出能改变一切的AGI只有三年了。
We think that we are three years away from launching AGI that will change everything.
那我们该怎么办？
So what are we gonna do?
我们会想办法在ChatGPT里插入广告。
We're gonna find a way to put ads into chat gippity.
当你觉得自己正站在奇点边缘时，这根本不是该有的做法。
That that is not the play when you are when you think you're the verge of the singularity.
我的意思是，没人懂，我也不知道。
I mean, nobody Demes habib I don't know.
我不喜欢他到那种程度。
I don't like him to such an extent.
我拒绝学习他的名字。
I refuse to learn his name.
那个深脑的人。
The deep mind bloke.
嗯。
Mhmm.
PeekinAWS。
PeekinAWS.
我试着。
I try
我不去试图将AI或其创造者拟人化。
not to I try not to anthropomorphize either AI or its creators.
是的。
Yeah.
迪皮斯·门比皮斯，就是那个每隔几周就做一次采访的人，他说上帝即将现身。
Dipis Menbipis, he's the one that every few weeks does an interview and he's like, and then God is going to come out.
我真的很害怕计算机将会做什么。
I'm really scared of what the computer will do.
他听起来不像那样。
He doesn't sound like that.
他倒是有这愿望。
He wishes.
而且这仅仅是
And it's just
你把它说得太吸引人了。
You make it sound exciting.
是的。
Yeah.
没错。
Exactly.
当这些人在电视上看起来有趣、吸引人、滑稽的时候，他们其实只是些无聊透顶、乏味的混蛋。
It sounds fun and interesting and comical when these people are just boring, boring people, boring motherfuckers.
但这些都是巨大的过度承诺和一些奇怪的东西，比如，让我们来看看这个。
But it's just these massive overpromises and these weird fucking, like, these weird things where I like, looking let's look at this.
今年，仅谷歌、亚马逊、Meta和微软这四家公司就在资本支出上投入了6500亿美元。
There are $650,000,000,000 in CapEx being spent this year just by Alphabet, Amazon, Meta, and Microsoft.
这些钱都花到哪儿去了？
Where is it going?
说真的，这是个简单的问题。
Genuinely actually, simple question.
我真的不知道这些钱现在都流向哪里了。
I actually don't know where it's going anymore.
迟早会让他们资产负债表上出现一个巨大的窟窿。
It's leading to a giant hole in their balance sheet at some point.
没错。
Right.
你会发现，所有成功的超大规模云服务商，包括那些冒牌的，都有其他业务来为这项投入融资。
You'll notice that all of the hyperscalers, all of them, that have succeeded, even the pretenders, have other businesses that finance this.
是的。
Yes.
从来没有一家超大规模云服务商是刚起步就直接以云服务商身份成功的。
There has never been a successful hyperscaler that launched as a hyperscaler.
最接近这个情况的例子是Cloudflare，但我甚至质疑你是否该称他们为超大规模云服务商。
The closest story you've got to that, and I I question whether you'd call them a hyperscaler, is Cloudflare.
因为他们最初是做CDN的，之后才逐步增加了其他云服务。
Because they start as a CDN and then just added the rest of the cloud stuff.
我们来看看他们的市值有多大。
And let's see how big their stock is.
他们的市值是590亿美元。
I got a $59,000,000,000 market cap.
嗯，他们
Well, they're
一家真正的公司。
a real company.
是的。
Yeah.
你不会想投资一家不做出夸张承诺的公司，但这真是太离谱了。
You wouldn't wanna I wouldn't wanna invest in a company that wasn't making egregious promises, but it's just Jesus.
这太奇怪了，因为即使他们认为人工智能会很重要，但他们今天花的每一分钱都不太可能实现这一点。
It's just so strange because even if they think AI is gonna be big, surely all the money they're spending today is not gonna do that.
这些芯片根本不可能做到。
Like, all of these chips aren't gonna do it.
他们今天训练的模型也做不到。
The models they train today aren't gonna do it.
这就像他们正在
The It's like they're
试图从一件事跳到另一件事，不断自我启动。
trying to bootstrap from thing to thing to thing.
让我担心的是这样一个故事：好吧。
The story that concerns me is I okay.
你去看看这些公司的订阅计划。
You take a look at these companies and their subscription plans.
最高档的是200美元。
Top end of it is $200.
很好。
Great.
我自己花钱订阅，只是为了派一只长翅膀的猴子给我建个垃圾网站，这太棒了。
I I pay it myself just so I can dispatch a winged monkey to build me a shitpost website, which is awesome.
有时候它能成功，有时候我得提示它做不同的事情。
And sometimes it works, and sometimes I have to prompt it to do different things.
很好。
Great.
它在做糟糕的前端方面，比我做得好。
It's better at shitty front end than I am at shitty front end.
太棒了。
Terrific.
我不会花五美元买这个。
I'm not gonna spend $5 on that.
是的。
Yeah.
但Zoom的模式并没有这样，我们说的不只是我一个人。
But the model for Zooms did not and we're not just talking me here.
我们实际上说的是所有人。
We are talking effectively everyone.
这些人的潜台词就像是：你的老板会把你解雇，然后拿走你的薪水，和AI公司平分。
It's like the unspoken message of a lot of these folks is that your boss is gonna fire you and then take your salary and split it with the AI company.
对。
Right.
而且你还能用AI编写所有代码，而不是像人们常说的那样。
And also the you will be able to code everything with AI versus what people love that I quote the same thing.
卡尔·布朗说，让简单的事情更简单，让困难的事情更困难。
Carl Brown saying, makes the easy things easier and the hard things harder.
就像我们在Duckbill开发软件一样。
Like We're building software at Duckbill.
我们正在招聘工程师。
We are hiring engineers.
这是因为我们很蠢吗？
Is this because we're stupid?
我不认为这是真的。
I don't believe that to be true.
经营一家企业却不考虑播客？
Run a business and not thinking about podcasting?
再想想吧。
Think again.
听播客的美国人比使用Spotify和Pandora等广告支持的流媒体音乐服务的人还多。
More Americans listen to podcasts than ad supported streaming music from Spotify and Pandora.
作为排名第一的播客平台，iHeart的规模是第二名和第三名总和的两倍。
And as the number one podcaster, iHeart's twice as large as the next two combined.
因此，无论你的客户听什么，他们都会听到你的信息。
So whatever your customers listen to, they'll hear your message.
此外，只有iHeart能将你的信息扩展到广播电台的受众群体。
Plus, only iHeart can extend your message to audiences across broadcast radio.
你觉得播客能帮助你的业务吗？
Think podcasting can help your business?
那就选iHeart。
Think iHeart.
流媒体、广播和播客。
Streaming, radio, and podcasting.
让我们在iheartadvertising.com为你展示。
Let us show you at iheartadvertising.com.
那就是iheartadvertising.com。
That's iheartadvertising.com.
在黑人社区中，一位雄心勃勃、善意满满、充满斗志且富有的母亲是什么样子。
A ambitious, well intentioned, ferocious, and wealthy mother looks like in the black community.
这位女士的黑人历史月播客《保持积极，亲爱的》颂扬了女性在生活混乱时依然选择疗愈、目标与信仰的力量。
This woman's history month, the podcast, keep it positive, sweetie, celebrates the power of women choosing healing, purpose, and faith even when life gets messy.
爱不是一个终点。
Love is not a destination.
你必须每天为之努力。
You have to work on it every day.
《保持积极，亲爱的》为关于自我价值、爱、成长以及以优雅与坚韧应对生活的坦诚对话创造了空间，由那些激励他人、鼓舞人心并勇敢说出真相的女性主导。
Keep it Positive Sweetie creates space for honest conversations on self worth, love, growth, and navigating life with grace and grit led by women who uplift, inspire, and tell the truth out loud.
我多次与上帝对话，我知道为什么花了二十年时间。
I have several conversations with God, and I know why it took twenty years.
要收听这些内容及其他更多节目，请在 iHeartRadio 应用、Apple 播客或您常用的播客平台收听《保持积极，亲爱的》。
To hear this and more, listen to keep it positive, sweetie, on the iHeartRadio app, Apple Podcasts, or wherever you get your podcasts.
这是全新的我，也是旧日的他们。
It's the new me, and it's the old them.
每个人都在自己的旅程中。
Everybody's on their journey.
你的旅程和他们的不同。
And your journey is different to theirs.
这个女性历史月的播客《如果你知道得更多》由安伯·格莱姆斯主持，聚焦那些将失误转化为动力、将教训转化为力量的女性。
This woman's history month, the podcast, If You Knew Better With Amber Grimes, spotlights women who turn missteps into momentum and lessons into power.
我觉得从我出身的地方走出来，我来自布朗克斯。
I think coming out of where I came from, I'm from The Bronx.
我觉得我小时候非常贫穷。
I think I grew up really poor.
那时我并不知道，因为我用我的创造力来美化生活。
I didn't know that then because I very much use my creativity to romanticize life.
我觉得我妈妈做得真的很棒，你退后一步看，就会惊叹。
And I'm like, my mom did a really good job of like, you step back and you're like, woah.
我不知道我们是怎么挺过来的。
We I don't know how we made it.
所以我的很多人生都是为了生存，只为到达下一个阶段。
So a lot of my life was, like, built out of, like, survival to get to the next place.
我的动力，我那种专注的目标，就是要变得更好。
Like, my drive, my, like, tunnel vision of, like, I gotta be better.
我想要达成这些目标的动力，来自于我想为我们创造更好的生活。
I gotta achieve this was off the strengths of, like, I wanna make a better life for us.
《If You Knew Better》邀请亲身经历过的女性分享真实对话，剖析职业转型、感情教训，以及改变一切的心态转变。
If You Knew Better brings real talk from women who've lived it, unpacking career pivots, relationship lessons, and the mindset shifts that changed everything.
请在 iHeartRadio 应用、Apple 播客或你收听播客的任何平台收听 Amber Grimes 主持的《If You Knew Better》。
Listen to If You Knew Better with Amber Grimes on the iHeartRadio app, Apple Podcasts, or wherever you get your podcast.
你好。
Hi.
我是丹妮尔·罗贝，Reese 书友会播客《Bookmarked》的主持人。
I'm Danielle Robet, host of Bookmarked, the podcast by Reese's Book Club.
本周在《Bookmarked》中，我们基本上是在举办一场终极女孩之夜。
And this week on Bookmarked, we're basically hosting the ultimate girls night.
蕾西·威瑟斯彭、詹妮弗·加纳、朱迪·格雷尔、丽塔·威尔逊、加里·莱斯以及作家劳拉·戴维。
Reese Witherspoon, Jennifer Garner, Judy Greer, Rita Wilson, and Gary Rice, and author Laura Dave.
这些女性是Apple TV剧集《他告诉我的最后一件事》第二季背后的主创。
These are the women behind season two of the Apple TV series, The Last Thing He Told Me.
我们将讨论如何将一本小说打造成热门剧集，以及真正让一个故事鲜活起来需要什么。
We're talking about turning a book into a hit show and what it really takes to bring a story to life.
对我来说，最重要的标准是我是否想把这本书推荐给别人？
The most important metric for me is do I wanna share this book with somebody?
这正是构建社群的方式，也是我们创办读书会的核心理念——让人们彼此连接。
That's what creates community, and that's the main thesis of our book club and why we started it was just to connect people together.
在iHeartRadio应用、Apple播客或您收听播客的任何平台收听Reese的读书会播客《Bookmarked》。
Listen to the bookmarked by Reese's Book Club podcast on the iHeartRadio app, Apple Podcasts, or wherever you get your podcasts.
是的。
Yeah.
而这正是关键所在。
And that's kind of the thing.
就是说，我不知道。
It's like, I don't know.
人们似乎还在招聘软件工程师。
People still seem to be hiring software engineers.
工程师被解雇，被人工智能取代。
The engineers are getting fired and replaced with AI.
这个说法似乎站不住脚。
Story doesn't seem to work.
而且每次我去读那些声称Claude代码改变他们整个人生的人的故事时，归根结底不过是，是的，把原本几个小时的工作缩短到了一小时。
And also, every time I go and read someone who claims that Claude code changed their entire life, it mostly comes down to, yeah, it took something from a few hours to an hour.
这挺酷的。
And it's like, cool.
很好。
Great.
好的。
Okay.
但它改变了我的生活。
But it changed my life.
它做了什么？
What did it do?
它教会我清晰地表达我的需求。
It taught me to clearly explain what I wanted.
是的。
Yeah.
但我也还是得去修正问题。
But, also, I still had to fix things.
如果所有这些钱只是用来制造稍微更高效的软件工程师，而即便如此，数据表明他们反而效率更低，我不确定。
It's like if if all of this money just went into creating slightly more efficient software engineers and even then the data suggests they're less efficient, I don't know.
这值得吗？
Was this worth it?
亚马逊当然不是亚马逊，我想他们从Anthropic那里捞到了一大笔钱。
And Amazon certainly isn't Amazon certainly isn't I guess they're getting the chunk out of Anthropic from like this.
他们可以宣称Anthropic的成功。
Like, they're they can claim Anthropic success.
他们可以让对方为Project Radiant中的Trainium付费。
They can have them pay for Trainium in Project Radiant.
作为平台公司，当人们在你之上构建有趣的东西时，你该如何在不显得自大情况下获得认可？
Platform company where when people build interesting things on top of you, how do you how do you take credit for that without being obnoxious?
我的意思是，这对亚马逊来说曾经是个问题吗？
I mean, has that ever been a problem for Amazon?
嗯，
Well,
并不是说他们担心显得自大。
not that that's not that they're worried about being obnoxious.
只是他们不擅长传达信息。
It's just they're bad at messaging.
我的意思是，我之前提到过他。
Well, I mean, they I mentioned him earlier.
帕诺斯·佩内特
Panos Penet
是的。
Mhmm.
帕诺斯·佩内特从微软的Surface团队转投亚马逊，负责领导Alexa Plus项目。
Is Panos Penet came over from Microsoft from the Surface team to, lead Amazon's Alexa Plus.
我的意思是，这确实有点过分了。
I mean, now there's something obnoxious.
现在这简直是生成式AI的胡扯。
Now there's some real generative AI bullshit.
他们现在在Alexa上也亏了数十亿美元。
They've what lost, like, billions on Alexa now as well.
天啊，亚马逊变得真奇怪。
God, what a strange company Amazon's become.
他们一直都很古怪。
They've always been peculiar.
问题是，到了一定规模时，我们有点古怪这一说法就不再成立了。
The problem is is at some point of scale, like, we're a little weird, no longer carries water.
这很好。
It's great.
你现在已成为全球经济的关键支柱之一。
You're now one of the linchpins of the global economy.
你得更有效地解释你在做什么。
You have to explain what you're doing a bit more effectively.
是的。
Yeah.
你觉得呢？
I I What do you think?
你认识亚马逊已经一百年了。
You've known you've known Amazon for one hundred years now.
你觉得如果AI泡沫破裂了会发生什么？
Do you think happens if the AI bubble bursts?
你实际上觉得他们是怎么做的？
What do you actually think they do?
比如，你认为他们会真的把这些东西搁置起来吗？
Like, what do you think are they the types to actually mothball this stuff?
他们会假装这件事从来没发生过吗？
Would they just pretend like this never happened?
其中一些方面。
Aspects of it.
但同样，他们在云计算的核心业务上赚得盆满钵满。
But there are again, they are making money hand over fist on the actual nuts and bolts of cloud computing.
这部分不会消失。
That's not going away.
他们的股价可能会下跌，因为没有公布惊人的增长数据，但他们提供全球计算机服务的业务非常强劲。
They they will their stock will tumble because they're not posting massive growth numbers, but they have a a banging business of providing computers to the world.
问题是，当他们的资产负债表出现巨大缺口时会发生什么？因为事实证明，所有人签署的数千亿美元合同，突然间他们没钱了。
The question is is what happens when there's a giant hole in their balance sheet because, well, it turns out that hundreds of billions of dollars in contracts everyone signed, suddenly they don't got the money.
我们该怎么办？
What do we do?
这将是一个系统性问题。
That's gonna be a systemic problem.
不会仅仅影响亚马逊。
It's not gonna just hit Amazon.
会波及很多人。
It's gonna hit a lot of folks.
我认为，在这个问题上，大家所期待的答案是政府救助。
And I think that the answer everyone's sort of hoping for in that space is government bailout.
不行。
No.
但亚马逊呢？
But of Amazon though?
将会是
Would be
太烦人了。
so irritated.
而且，他们到底在救助什么？
Also, what are they bailing out?
亚马逊不会破产的。
Amazon's not gonna go bankrupt.
即使亚马逊不得不对老年训练芯片进行巨额减值，比如高达300亿美元。
Even if Amazon had to take a massive impairment on elder training chips, and it was, like, 30,000,000,000.
但股东们呢？
But the shareholders at?
股东们。
The shareholders.
他们会生气的。
They'll be mad.
别误会。
Don't get me wrong.
但这是
But it's
不会毁掉亚马逊的。
not gonna destroy Amazon.
我的意思是，这可能毁了安迪·贾西的屁股。
Like, it's not I mean, it might destroy Andy Jassy's asshole.
可能会把安迪·贾西装进铁球扔进太阳，但感觉不会真把公司搞垮。
Like, it might be might send Andy Jassy in an iron ball into the sun, but it doesn't feel like it will kill them.
我只是不知道所谓的救助会是什么样子。
I just don't know what a bailout would be.
我真的很烦，当初安迪接任时我抱有很大期望，但公司似乎一直在沿着越来越糟糕的方向下滑。
I am so annoyed that I I had great hopes when Andy took over, but it seems like the company has just continued down the insuredification curve.
他们没在做任何令人惊讶的事。
They're not doing surprising things.
我2017年创办《LastWeekinAWS》通讯的原因是，每周他们都在解决大量面向客户的问题，当时就很难跟上，因为现在的问题和那时一样糟糕。
The reason I started the LastWeekinAWS newsletter in 2017 was that every week, they were fixing massive customer facing problems, and it was hard to keep track because they were just as bad then as they are now
他们到底在卖什么？
messaging What are selling them?
这些
These
2017年。
Twenty seventeen.
对。
Right.
现在，他们不再这么做了。
Now, they don't do that anymore.
这是惯性。
It's inertia.
我今天不可能像当年那样重新创办这个通讯简报并建立起读者群了，因为人们现在对这个平台的关注度远不如从前。
I I would not be able to start the newsletter and build in our readership today the way that I did back then, just because it people do not care nearly so much about the platform as they once did.
那你是什么时候注意到这种转变发生的？
So when did you notice the shift happened?
这是一下子全部涌现出来的，或许这就是描述它的最佳方式。
It was first graduate all at once, think is probably the way to frame it.
当我意识到自己从每周都面临一个问题——内容太多，哪些该选中——转变为觉得内容全是填充物，究竟有什么值得写的呢？
When I realized that I went from every week having the problem of there's so much stuff here, which which ones make the cut to there's so much filler here, where's anything actually worth writing about?
当你说到‘填充物’时，你指的是什么？
And when you say filler, what do you mean?
他们过去会谈论像 AWS Lambda 这样的东西，那是计算运行方式的一次变革性转变。
They used to say the they used to talk about things like AWS Lambda, a transformative shift in how a compute could be run.
他们用同样那种 corporate 的语气发布这些内容，比如‘达拉斯现在有了第三个 CloudFront 边缘节点’，连达拉斯本地人都根本不关心。
They put that out with the same enthusiasm corporate voice as there's now a third CloudFront edge location in Dallas, which even people in Dallas do not give a toss about.
对。
Right.
Lambda 是做什么的？
What does Lambda do?
本质上，你给它一段代码，当某些触发事件发生时——比如网页请求、定时触发，或者其他事件触发——它就会运行这段代码，并且只在运行时收费。
In effect, whether you give it its code, it runs the code when certain trigger events happen, be it a web request, be it a time being hit, be it something like some other event hitting, and it only charges you for while it runs.
而且能够大规模扩展，你完全不需要操心围绕它的基础设施的维护和管理，这
And massively scales up, and you don't have to worry about any of the care and feeding of the infrastructure around it, which
这真的很酷。
That's really cool.
确实很酷。
It is.
以前很酷。
It was cool.
它在2015年还处于测试阶段。
It was beta ed in 2015.
但现在它已经成为
But now It's one the
亚马逊推出的真正伟大的产品之一，彻底改变了我对这种方式的理解。
last truly great things Amazon put out that changed the way I thought about how this could work.
但如今亚马逊不再推出令人兴奋的新功能了。
Except now Amazon doesn't really put out fun new updates.
就是这样。
It's just.
渐进式的更新。
Incremental updates.
我们有了一个新东西。
We have a new thing.
我们推出了自己的大语言模型，名为Nova。
We've launched our own large language model called Nova.
它没有你目前使用的那些模型好，但更便宜。
It's not as good as the ones you currently use, but it's less money.
是的。
Yeah.
当它还处于概念验证阶段时，他们并不在意。
When it's still a POC stage, they don't care.
很厉害的是，如果你看看Nova现在的位置，哇。
What's great is if you look where Nova is, wow.
它排在Cat Coder Pro、Kimmy 2.5、Grok 4.1、Mimo、Mini Max、Gwen Thinking、Kimmy K2之下，位列前25名，几乎其他所有模型都比它强。
It's in the top 25 below cat coder pro, Kimmy two point five, Grok 4.1, Mimo, Mini Max, Gwen Thinking, Kimmy k two, basically every other model.
对。
Right.
没错。
Exactly.
就像Ed一样，他们和Frontier AI实验室的Ed的标本制作在齐头并进。
It's like Ed like, they're they're racing neck and neck with Ed's taxidermy in Frontier AI Lab.
但事实确实如此。
There's it really is that, though.
它就像是某种叫……是的。
It's like it's something called yeah.
百度的文心一言。
Ernie from Baidu.
好像每个中国模型都在哭。
Like, every Chinese model appears to cry.
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。