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。
谷歌最新模型让OpenAI担心自己正在失去优势。
Google's latest model has OpenAI concerned it's losing its edge.
杰夫·贝佐斯重返AI创业领域,比尔·阿克曼说'能否与你见面',AI音乐是否已跨越恐怖谷?
Jeff Bezos is back in the AI startup game, may I meet you, says Bill Ackman, and has AI music passed the uncanny valley?
我们将在周五的大型科技播客特别版中全面报道这些内容,马上开始。
We'll cover it all on a big technology podcast Friday edition right after this.
Capital One的技术团队不仅在多模态AI领域有所建树。
Capital One's tech team isn't just talking about multiegetic AI.
他们已经实际部署了一套系统。
They already deployed one.
这套名为ChatConcierge的服务正在简化购车流程。
It's called ChatConcierge, and it's simplifying car shopping.
通过自我反思和分层推理结合实时API检查,它不仅帮助买家找到心仪车辆。
Using self reflection and layered reasoning with live API checks, it doesn't just help buyers find a car they love.
还能协助预约试驾、获得融资预批,并估算置换价值。
It helps schedule a test drive, get preapproved for financing, and estimate trade and value.
先进、直观且已投入使用。
Advanced, intuitive, and deployed.
这就是他们的技术实力。
That's how they stack.
这就是Capital One的科技力量。
That's technology at Capital One.
事实上,AI安全就是身份安全。
The truth is AI security is identity security.
AI代理不仅仅是一段代码。
An AI agent isn't just a piece of code.
它是您数字生态系统中的一等公民,需要得到相应的对待。
It's a first class citizen in your digital ecosystem, and it needs to be treated like one.
这就是为什么Okta要率先保护这些AI代理。
That's why Okta is taking the lead to secure these AI agents.
解锁这层新防护的关键——身份安全架构。
The key to unlocking this new layer of protection, an identity security fabric.
企业需要统一的综合方案,通过一致的政策和监督机制保护每个身份——无论是人类还是机器。
Organizations need a unified comprehensive approach that protects every identity, human or machine, with consistent policies and oversight.
不要等到发生安全事件才意识到AI代理是巨大的盲点。
Don't wait for a security incident to realize your AI agents are a massive blind spot.
了解Okta的身份安全架构如何帮助您保护下一代身份,包括AI代理。
Learn how Okta's identity security fabric can help you secure the next generation of identities, including your AI agents.
访问okta.com。
Visit okta.com.
网址是okta.com。
That's okta.com.
欢迎收听《大科技》播客周五特别版,我们将以一贯冷静细致的风格解读新闻。
Welcome to Big Technology podcast Friday edition where we break down the news in our traditional cool headed and nuanced format.
今天节目非常精彩,我们将探讨谷歌如何迎头赶上OpenAI的技术水平,导致山姆·奥特曼发布备忘录称'未来一段时间氛围可能不佳'。
We have a great show for you today because we are going to cover how Google has come and effectively equaled OpenAI's technology causing Sam Altman to write a memo saying the vibes might be bad for a while.
我们还将提及杰夫·贝索斯进军AI初创领域。
We will also touch on Jeff Bezos entering the AI startup game.
我们将聊聊比尔·阿克曼的搭讪技巧,接着还会讨论这首令人难以置信的AI歌曲——或者说这首人们开始厌恶的歌曲,以及这对创造力意味着什么。
We'll talk about Bill Bill Ackman's pickup line, and then we'll also cover, this unbelievable AI song or some a song that people, have started to hate and what that means for creativity.
每周五与我们共同探讨这些话题的是《边际》栏目的兰詹·罗伊,今天他就在演播室现场。
Joining us as always on Fridays to do this is Ranjan Roy of Margins who's here with us in studio today.
兰詹,见到你真好。
Ranjan, great to see you.
这里可没有不和谐的氛围。
There's no rough vibes here.
我今天在金融区与亚历克斯面对面交流,所以很兴奋。
I'm in the Financial District face to face with Alex today, so I'm excited.
我们将在AI泡沫破裂时守候在此。
We are here when the AI bubble crashes.
这里将成为历史性崩盘的发生地。
This is gonna be, the place where it all goes down.
所以我们提前来做个侦查。
So we're just giving an early scout out.
我们要去观察这里的牛市氛围,看看在不可避免的AI崩盘前事态如何发展。
We're going to go take a look at the bull and the vibes here and see how things are going for the inevitable AI crash.
也许它会
Maybe it will
崩盘。
crash.
也许不会。
Maybe it won't.
谁知道呢?
Who knows?
嗯,录制时股市在上涨。
Well, the stock market is up at time of recording.
标普500指数上涨了1个点,但涨幅是4%
S and P is up one But point 4%
这周真是糟透了。
it's been a bad week.
这周确实很糟,但或许或许我们正在回暖。
It's been a bad week, but maybe maybe we're back.
或许我们正在回暖。
Maybe we're back.
我不知道。
I don't know.
我是说,一周后我们就能深入探讨了。
I mean, a week and we'll get into it.
但通常英伟达交出碾压级财报后,市场总会形成一种'哦,暂时没问题了'的共识并继续冲高。
But a week where Nvidia does what it does, which is crush earnings, is typically followed by a pattern of the market saying, oh, we're we're good for now and shooting higher.
但这周却出现了完全不同的走势。
But we had a very different effect this week.
延续上周我们与吉尔·卢里亚关于AI泡沫的对话,感觉就像——天啊,市场真的开始担忧了。
Following on our conversation about the AI bubble with Gil Luria last week, it's like, oh god, like, the market really is concerned here.
不,我们要讨论的是市场走势。
No, we're going to talk markets.
我们要谈谈债务。
We're going to talk debt.
我们要谈谈人工智能。
We're going to talk AI.
我们热爱债务。
We love debt.
而我们
And We
热爱债务。
love debt.
比尔·阿克曼的搭讪台词。
Bill Ackman pickup lines.
当然是比尔·阿克曼。
Bill Ackman, of course.
没错,人们确实说这招很管用。
Right, which people are actually saying is working.
我相信是的。
I'm sure it is.
我我确信它
I'm I'm sure it
确实。
is.
在下半场。
In the second half.
首先,本周AI领域的大新闻显然是谷歌推出了其最新AI模型Gemini 3。
But to begin with, so obviously this week, the big news in the AI world is that Google came out with Gemini three, its latest AI model.
Gemini 3在基准测试中表现惊人,在Arc AGI测试中碾压对手。
Gemini three smashes the benchmarks, crushed on the Arc AGI test.
它目前高居LM Arita排行榜首位,连Sam Altman都公开表示:嘿,好模型。
It's currently at the top of the LM Arita leaderboards and even Sam Altman came out and said, hey, good model.
干得漂亮,恭喜。
Good game, congrats.
伴随这个消息还爆出了一个非常有趣的故事。
Information comes out with this very interesting story about it.
Altman备忘录预测:因谷歌复兴将面临艰难氛围。
Altman memo forecasts rough vibes due to resurgent Google.
事情是这样的。
Here's the story.
OpenAI首席执行官Sam Altman上月告诉同事,谷歌在人工智能领域的最新进展可能给公司带来暂时的经济阻力,但他补充说OpenAI最终会领先——此前OpenAI研究人员获悉谷歌开发的新AI在技术路径上似乎已超越OpenAI。
OpenAI CEO Sam Altman told colleagues last month that Google's recent progress in artificial intelligence could create temporary economic headwinds for our company, though he added OpenAI would emerge ahead after OpenAI researchers heard that Google had created a new AI that appears to have leapfrogged OpenAI's in the way it was developed.
Altman在备忘录中表示:我们知道还有工作要做,但我们正在快速追赶。
Altman said, in the memo, we know we have some work to do, but we are catching up fast.
他还告诫员工:我预计外界氛围会暂时变得艰难。
And he cautioned employees that I expect the vibes out there to be rough for a bit.
我从未听过OpenAI如此谨慎的表态,也从未承认被任何人超越过。
I have not heard OpenAI talk this cautiously ever and and never admit that it's been surpassed by anyone.
所以这是个分水岭时刻。
So this is a watershed moment.
不。
No.
不。
No.
这很重要,因为当你想到Anthropic、微软、Meta和OpenAI时,纯粹从模型吸引力来看,这就是我选择的基准。
It's a big deal because when you think about Anthropic, you think about Microsoft, you think about Meta, OpenAI from just a pure kind of like model sexiness, that's my benchmark of choice.
我不知道它是如何计算的,但它很重要,在过去几年里从未被超越过。
I don't know how it's calculated, but it's important, has not been equaled at any point in the last number of years.
就像,会有某种低级别的基准测试,有人可能会为此感到兴奋。
Like, there'll be some kind of like, you know, low grade benchmark, which someone might apply, like, you know, get excited about.
但就模型发布引发的实际热度而言,OpenAI的GPT-5在兑现炒作方面是个失败,而Gemini 3则让人们比以往任何模型发布时都更加兴奋。
But in terms of actual buzz around a model release, OpenAI GPT five was a flop in terms of actually meeting the hype, and Gemini three had people more excited than I've seen them in a long time around a model launch.
这就是OpenAI的游戏,而他们现在正在输掉这场比赛。
So that's OpenAI's game, and they are losing at it right now.
这几乎与我们习惯的情况相反,对吧?
It was like almost the opposite from what we're used to, right?
就像,谷歌发布公告时,大家的反应都是‘哦,好吧’。
Like, the Google does an announcement, everyone's like, oh, Okay.
而OpenAI则彻底碾压了它。
And OpenAI blows it out of the water.
这次人们是真的感到兴奋。
This time, people were genuinely excited.
就像我提到的,Gemini 3在Ala Marina排行榜上名列前茅,紧随其后的竟然是Grock,这很有趣。
Like I mentioned, Gemini three is at the top of the Ala Marina leaderboards, and trailing it is actually Grock, which is funny.
我们要得到
We're going to get
我们在给Grock抛光
We're glazing Grock.
我们要深入了解Grock
We're going to get into Grock.
在我看来,很明显Sam Altman写这份备忘录时预见到了Gemini
To me, so obviously Sam Altman wrote this memo anticipating Gemini.
而我认为其中最有趣的部分是他提到谷歌在此的成功(无疑已取得成功)将为我们公司带来暂时的经济阻力
And I thought the most interesting part about it was how he said that Google's success here, which has undoubtedly succeeded, is going to create temporary economic headwinds for our company.
我非常好奇你认为这意味着什么——他是否在暗示人们会转而订阅谷歌产品而非ChatGPT,从而减缓增长
And I'm very curious what you think that is because is he saying that this that people will subscribe to Google products instead of ChatGPT and that could slow down the growth.
或者我认为这才是真正的问题:如果ChatGPT失去其作为领导者、无可争议的领跑者、最有可能率先实现AGI的实验室这一光环,那么它可能很难继续融资并维持目前的资金链
Or and I think this is the real issue if ChatGPT loses that sheen as being the leader, the undisputed leader, the first the lab that's most likely to be first to AGI, then it might be very tough for it to continue to raise money and keep going on with this financing run that it's been on.
好吧
All right.
我认为这个声明有两层含义
I'm going to there's two layers to that statement, I think.
关于经济阻力这点,我觉得你说得完全正确
So economic headwinds, I think you're completely correct.
ChatGPT已经成为了动词
ChatGPT is the verb.
谷歌曾是搜索的代名词
Google was the verb for search.
我认识的所有人都把ChatGPT当动词用了。
ChatGPT has earned verb status for every single person I know.
这就是他们对任何AI聊天机器人的称呼,尤其是非业内人士。
That is the term they use for any AI chatbot, especially if you're not in the industry.
如果这个动词地位被谷歌和Gemini抢走,有人开始说'我用Gemini查了',那问题就严重了。
If they lose that to Google and Gemini and someone somewhere says, I Gemini ed that, it's a real problem.
这绝对是个大问题。
It's a huge problem.
所以这里存在消费者层面的问题。
So so there's a consumer side.
但我觉得——不是阴谋论——这个故事更重要的部分是谷歌用自家TPU训练了Gemini。
But my kinda like, not conspiracy theory, but the one of the bigger parts of this story again was that Google trained Gemini on their own TPUs.
就整体算力格局而言,OpenAI处于核心位置,我们稍后会再详细讨论甲骨文、OpenAI和微软之间的循环融资关系。
And in terms of the overall compute story that OpenAI is at the center of, and we'll get more into again, the circle circular financing across the Oracle and OpenAI and Microsoft of the world.
如果谷歌突然证明存在更经济高效的大模型构建方式,我认为这实际上会形成更大的经济阻力,因为这会打乱OpenAI目前在整个行业建立的——我不想说'空壳游戏'——应该说非常创新的融资结构。
If suddenly Google shows there's a more economically efficient way to actually build these new models, I think that actually is much more of an economic headwind because that screws OpenAI's entire, I'm not gonna say shell game, but just, I'm gonna say, very creative financing structures that they built across their entire across the entire industry right now.
就像DeepSeek那个时刻,谷歌实际构建Gemini3的方式是否会改变这些模型的经济学?
Like, that is this a deep seek moment the way Google actually built Gemini three, and does it change the economics of how these models are built?
我觉得这是...我看过相关报道。
I think it's one of the you know, like, I've seen coverage of it.
我看到过相关讨论,但这确实不是话题中心。
I've seen discussion around it, but it definitely was not at the center of the conversation.
没错。
Right.
从某种程度上说,它是否在经济上更高效甚至并不重要。
And in a way, it doesn't even matter if it's more economically efficient.
仔细想想法院的事,关于经济效率的一种说法是,与那些试图在英伟达GPU上构建和训练的人相比,他们获得TPU的途径更少受限。
Think the court well, one way of saying it about the economic efficiency is that they have less constrained access to TPUs compared to everybody trying to build and train on NVIDIA GPUs.
TPU是谷歌的专有芯片。
TPUs is Google's proprietary chip.
他们设计了这些芯片。
They design them.
他们随时备有这些芯片。
They have them at the ready.
它们的供应限制比英伟达GPU要少。
They are less supply constrained than NVIDIA GPUs.
也许谷歌即使没有找到这种高效训练模型的方法,仅凭他们可以随意使用这些芯片,而像OpenAI这样的竞争对手却要以高昂成本从英伟达购买,这就构成了真正的优势。
And maybe Google, even if it hasn't found this efficient way to train a model, just the fact that they have these chips at their disposal, whereas competitors like OpenAI have to go and buy them at a very high cost from Nvidia, that's a real leg up.
他们完全可以利用这一优势。
And they could really exploit that advantage.
是的。
Yeah.
所以我认为在这两个层面上,用'经济逆风'来形容都算轻了。
So I think at both of those levels, economic headwinds is almost a light way of saying it.
不过我觉得更有趣的是,像山姆·奥特曼这样的人——我很好奇的是,他在内部如何定期与团队沟通,与对外发言有何不同?
I think it's even more interesting, though, that like, Sam Altman I I'm very what it made me curious about is, like, internally, how does he speak to his team regularly versus how he speaks outwardly?
因为我们都知道山姆·奥特曼从不羞于表现得有些浮夸、夸夸其谈、雄心勃勃,无论你用哪个词来形容。
Because we all know Sam Altman is not shy about being a bit grandiose, bombastic, ambitious, whatever word you choose.
但在内部,他通常对团队更现实和克制吗?还是说他写备忘录时就知道会被泄露。
But internally, is he typically more realistic and measured with his team, or is he when he writes a memo, he has to know it's gonna get leaked.
对吧?
Right?
比如,他这次沟通的方式与过去相比有显著不同吗?
Like, what what is the is the way he's communicating around this significantly different than he how he has in the past?
你怎么看?
What do you think?
是的。
Yeah.
我认为这绝对是个转变。
I think it's definitely a change.
而且回到我们开场的方式,事实是我们从未见过OpenAI说我们需要追赶,因为之前从未出现过这种情况。
And, going back to, like, the way that we open this, the fact of the matter is that we haven't seen OpenAI ever say we have to catch up because it hasn't been the case.
所以我几乎回想起他在那次采访中对如何支付基础设施费用感到沮丧。
So I almost think back to him getting frustrated in that interview about how they're going to pay for the infrastructure.
然后他说,够了。
And he goes, like, enough.
那大概就是他发出这份备忘录的时间。
And that was probably right around the time that he sent this memo out.
所以可能就是这样,像是'天啊'这种状态。
So it could have just be this, like, oh, god.
不。
No.
说实话,那个片段我反反复复看了好多遍。
I mean, honestly, I have watched that clip over and over again.
刚出来的时候,我就看到它在网上疯传。
Like, when it first came out, I saw it circulating.
这是布拉德·格斯特纳的采访,他非常客气地问——你知道的——140亿的收入还是170亿来着,以及1.4万亿的融资承诺。
This is the Brad Gerstner interview where he asks him very, you know, like, nicely, you know, 14,000,000,000 in revenue or 17, whatever it was, and 1,400,000,000,000.0 in financing commitments.
结果山姆当场就炸了,说什么'当然我们的收入曲线是指数增长的'。
And Sam gets positively indignant and is like, not only, of course, we're gonna you know, our revenue is up into the right.
是典型的曲棍球杆式增长。
It's hockey stick growth.
然后莫名其妙就开始说'如果你不想要股份,自然有别人会买'这种话。
And then oddly got into, you know, like, well, if you don't want your shares, someone else will buy them.
这个...我也说不清。
Like, the I I don't know.
我越看越觉得这将成为...
The more I watch that, I'm still convinced that's gonna be kind of the
标志性时刻。
Seminal moment.
对。
Yeah.
就像2008年瑞克·桑特利那次事件,对经历过全球金融危机的人来说
The Rick Santelli 2008 for any long time, global financial crisis people out
就是那个。
there.
但事实就是这样发生的,基本上没错。
But the fact that that happened along this basically Yeah.
想想时间线。
Think timeline.
从谷歌的表现来看,我不会说是恐慌,但确实让他们措手不及,这是近期少有的情况。
From Google that, like, shows, I wouldn't say panic, but a level of Just They got knocked off their mark in a way they haven't been recently.
他状态不佳。
He's not he's off his game.
他那天状态不好。
He was off his game that day.
你说得对。
And you're right.
从时间线上看,完全吻合。
Timeline wise, it matches up.
我一直在想另一句话。
I kept thinking about this other quote.
对吧?
Right?
记得吗,因为山姆·奥特曼和萨提亚·纳德拉联手了。
Remember, because Sam Altman teamed up with Satya Nadella.
萨提亚在Chattypati推出并走红后说了什么来着?
And Satya, what did he say once Chattypati came out and started taking off?
我想让谷歌跳个舞。
I'd like to make Google dance.
说到谷歌,今天要讨论的是它的市值超越了微软。
Now Google, by Google the way, going talk about today surpassed Microsoft in market cap.
我是说,两家公司都表现得很棒,谷歌大约3.59万亿美元,微软3.5万亿美元。
I mean, they're still both doing great, 3,590,000,000,000.00 or so to 3,500,000,000,000.0.
但反过来,桑达尔确实不得不跳舞庆祝。
But in turn, Sundar had he did have to dance.
桑达尔跳舞了。
Sundar's dancing.
他还让萨姆·奥特曼也跟着跳了起来。
And he's he's made Sam Altman dance as well.
是啊。
Yeah.
桑达尔跳舞的画面简直太神奇了。
Sundar dancing is just an amazing image.
我从来没见过桑达尔跳舞的视频片段。
I've never seen a clip of Sundar dancing.
你或许可以
You could probably
用谷歌的图像工具生成一个
create one on one Google's image
说真的,录完节目后这就是我的下一个‘纳米香蕉’任务
generators Actually, that's my next nano banana task after we're done recording
顺便说一句,桑达尔,我们要走了。
By the way, Sundar we're going to go.
我们显然要稍微聊聊谷歌为何表现得如此出色。
We obviously are going to talk a little bit about how Google is doing so well.
他们在产品端仍然非常谨慎。
They're just still so careful on the product side.
这让我抓狂。
It drives me nuts.
我要求为这次徒步制定训练计划,结果得到的回复是:这恐怕行不通,因为太危险了。
I asked for a training plan for this hike I was doing, and it was like, yeah, this is not really going to work because it's dangerous.
我不得不回复:你在开玩笑吗?
And I had to write back, are you kidding me?
这完全很正常。
It's totally normal.
他们在这个模型上设置了这种程度的安全主义。
And they have this level of safetyism on this model.
今天我再次尝试用Nanobanana修改桑达尔的照片,
And again, I tried to alter Sundar's picture today with Nanobanana.
但没成功。
I couldn't.
这对我来说是最大的缺陷。
That to me is the biggest drawback.
我能在Nanobanana上生成完全逼真的黄仁勋和桑达尔肖像。
So I was able to generate fully real life likenesses of Jensen Sundar on Nanobanana.
哦。
Oh.
是啊。
Yeah.
在双子座三号上。
On Gemini three.
我得看看情况。
I gotta see see the way.
也许他们正在放松
Maybe maybe they're relaxing out
你处于儿童模式
of You're on children's mode of
他们肯定有我的把柄。
They must have me.
他们肯定会想,这家伙是个记者。
They must they'd be like, this guy's a reporter.
对。
Yeah.
他会写关于我们的报道,你知道的,比如在披萨上涂胶水和放石头的事。
He's gonna write about us, you know, doing the glue on the pizza and the rocks.
所以我们不该让他... 好吧,我已经有完全访问权限了。
So so we just shouldn't let him do any Well, I got full access.
你有权限。
You do.
好吧。
All right.
我得搞定Ranjan的纳米香蕉账号,这周末好好玩一下。
I got to get Ranjan's nano banana account really and have fun this weekend.
但我认为这背后的问题是,好吧,如果OpenAI在某些领域重新领先呢?Sam没说我们在各方面都落后,但他提到谷歌已在某些领域超越了我们。
But I think underlying this yeah, Okay, so what if OpenAI does leap back forward in some Sam didn't say we're behind on everything, but there are some areas that he says that Google has surpassed us in.
所以我一直在想,如果OpenAI重新领先会怎样?
And so I kept thinking, well, what if OpenAI leapt back forward?
因为我们一直在讨论Gemini 3。
Because we've been talking a little bit about Gemini three.
我们都知道它在排行榜上表现不错。
Both of us know that it's doing well on the leaderboards.
但使用AI聊天机器人或AI聊天工具时,这并不算是什么全新的体验。
But it's not like it's a brand new experience when it comes to using AI chatbots or AI chat or tools, whatever it is.
这在我看来指向了一个更大的问题:如果我们还没到那一步,我们似乎正接近这些AI模型商品化的时代。
And that, to me, points to the bigger problem, which is it seems like if we're not there yet, we're getting close to the era of commoditization of these AI models.
以前,比如说OpenAI曾拥有巨大领先优势,可以借此收取更高费用或让ChatGPT比其他产品增长更快。
That before, let's say, OpenAI had a big lead and could exploit that advantage and maybe charge more or grow ChatGPT faster than others.
而现在他们与谷歌并驾齐驱的事实似乎表明,也许三年后我们就到了这个阶段。
And then you have the fact that they're now neck and neck with Google seems to indicate that maybe after three years, we're here.
我想今天或这周是ChatGPT发布三周年纪念日。
I think today or this week is the three year anniversary
二十二。
twenty two.
也许这些模型正在以某种方式商品化。
Of ChatGPT's introduction, that maybe these models are commoditizing in a way.
ChatGPT编程能力强,Anthropic也不逊色。
ChatGPT can code well and so can Anthropic.
你还可以和Gemini聊天。
And you can chat with Gemini.
你可以与OpenAI对话。
You can chat with OpenAI.
这引出了一个全新的经济问题——关于这些公司的生存能力和盈利能力。
And that introduces a whole new economic question about the viability and the profitability of these companies.
是啊。
Yeah.
其实我想说,11月30日,距离ChatGPT发布三周年纪念日只剩9天了。
I think, actually, in November 30, so we're nine days away from our big, three year anniversary of ChatGPT's launch.
没错。
Yeah.
我觉得Gemini 3,我们得好好谈谈那种...基本上就是由Gemini、Google、Sundar他们发起的堪称惊艳的预热营销活动。
I think Gemini three, and we'll get I really wanna get into the kind of, like, prelaunch marketing campaign basically, which was incredible launched by Gemini, Google, Sundar, all of them.
但实际使用时,感觉没什么变化。
But as I used it, nothing really changed.
至少我的日常生活没有发生翻天覆地的改变。
Certainly nothing in my day to day life changed dramatically.
就像X平台上那些网红说的,我做了个产品设计...
I all the things that I'm told by x number of influencers on x, you know, like, I create a product design.
酷。
Cool.
根据一篇文章创建信息图表。
Create an infographic based on an article.
酷。
Cool.
实际上,这让我想起信息图表可能是能想象到的最无聊的媒体形式。
Actually, it just reminded me that infographics are, like, the most boring form of media imaginable.
对信息图表爱好者们说声抱歉。
Sorry for any infographics folks.
但对我来说,整个技术推动的可能性固然美好且令人兴奋,但关键在于如何将其融入每个人的日常生活中。
But, like, I think to me, the entire technological push of what's possible is nice and exciting, but still integrating it into everyone's day to day.
虽然谷歌推出的Gemini 3兼具模型和产品特性,但我仍觉得模型竞赛并非最有趣的部分。
There's and I don't like, this is both model and product that Google is releasing with Gemini three, but I still feel that model race is not the most interesting part.
但我确实认为,对ChatGPT而言最可怕的是谷歌的渠道优势。
But I do think, like, what makes this so scary for ChatGPT is Google's distribution.
因为Claude面向消费者时几乎不存在门槛——用户需要主动注册、发现它。
Because Claude going consumer is, like, nonexistent when someone has to sign up, find about it.
这完全是一场营销战役。
It's a whole marketing exercise.
但如果Gemini 3能完全取代ChatGPT或与之匹敌,那就非常可怕了。
But if Gemini three fully replaces ChatGPT or is as good, that is really scary.
因为当所有人已经习惯使用google.com,从消费者角度能无缝切换到Gemini时——这直接威胁到OpenAI的核心业务,或者说主要收入来源。
Because when everyone already goes to google.com and can move right over to Gemini from a consumer standpoint, and that is OpenAI's entire business, like, I mean, majority, that's a problem.
确实。
Yeah.
我稍后会谈到这个,因为我在《大科技》上写过一篇关于谷歌如何实现这场难以置信的东山再起的故事。
I I wanna get to that in a moment because I did write this story in Big Technology about how Google has had this, you know, unbelievable comeback.
我是说,早在2025年初,人们就在问现在的桑达尔是否应该
I mean, early twenty twenty five, people were asking should Now Sundar be
他在跳舞。
he's dancing.
现在他正在超越的所有模型上跳舞。
Now he's dancing on top of the models that he's surpassed.
我们快速讨论一下商品化的问题。
Just getting to the let's go to the commoditization question quick.
如果模型确实商品化了——当然我要说,它们都会有自己的优势。
If the models do commoditize, and I'm saying, of course, they'll all have their own strengths.
但假设Gemini与GPT没有太大不同,GPT与Claude也没有太大不同,Claude又与Grock相差无几。
But let's say Gemini isn't all that different from GPT, which isn't all that different from Claude, which isn't all that different from Grock.
这会如何改变当前AI时代的经济格局?
How does that change the economics of this AI moment?
是的。
Yeah.
我我长期以来一直在说,我认为这是产品而非模型的问题。
I I've been talking about this for a long time that I don't I think this is product not model.
我们都在朝这个方向努力。
We're all getting there.
所有人都在朝这个方向努力。
Everyone's getting there.
一切都很好。
It's all good.
你知道的,比如ARC AGI分数之类的,我在想另一个是什么来着?
You know, like ARC AGI score or, like, I think what was the other one?
这就像是人类的终极考验。
It was, like, the final test of humanity.
是啊。
Yeah.
人类的最后一场考试。
Humanity's last exam.
人类的最后一场考试。
Humanity's last exam.
它的得分从20%提升到了37.5%,但这些东西对我来说毫无意义。
It scored at 37.5% up from 20 none of this stuff means anything to me.
我的意思是,这些在学术和智力层面都挺有意思的,但实际上对个人究竟有什么实际作用呢?
I mean, it's all kinda academically and intellectually interesting, but in reality, what is it actually doing for individuals?
我们肯定处于——我不想说是停滞期。
And we're definitely at I don't wanna say a plateau.
我认为每周使用这些工具的人数会不断增加。
I think the amount everyone is going to be using these tools week to week.
每当我见到非技术圈的朋友时,发现大家确实都在更多地使用ChatGPT。
Whenever I see friends like every non tech friends, everyone is using certainly ChatGPT more.
或许接下来会听到他们说要用Gemini更多些。
Maybe I'm gonna hear them say saying I'm gonna use Gemini more.
但我认为在产品层面和实际应用层面,比如模型的利用率方面,还有大量工作要做。虽然现状不错,但现在大家的模型都做得很好。
But I think there's so much more to be done at the product level and actual, like, utilization level that the models, it's good, but everyone's models are good right now.
在我们继续之前,让我提出一个相反的观点。
Let me give you the counterpoint to that before we move on.
反方观点可能是:作为普通用户,你可能看不出太大区别。
The counterpoint would be that you as an everyday user might not see a big difference.
但这些模型正在变得更擅长处理更困难的问题,而这些问题在以前可能对科学发现帮助不大。
But what these models are doing is they're becoming much better at tougher questions that maybe previously they weren't very useful for scientific discovery.
我只是试图传达模型公司可能会说的话。
I'm just going to try to channel what the model companies would say.
现在他们的新模型——比如即将推出的Gemini 3——可能会让你感觉稍微更亲切一点,聊天时显得聪明一点点,但在日常使用中和GPT模型差别不大。
And now their new model, which is going to feel maybe a little bit warmer, like Gemini three is supposed to, and a tiny bit smarter when you chat with it and not very different in everyday use case from, let's say, a GPT model.
但在专业应用场景和更高层次的智能表现上,比如科学研究和工业信号解读这类任务——从Excel表格中提取洞察——这些新模型会实用得多。
But those specialized use cases and that next level of intelligence is much more useful when it comes to things like science and industrial signal reading and things like that, making sense of it out of an Excel spreadsheet.
是的,不过我很感谢这种努力。
Yeah, but I appreciate the effort.
我欣赏这种努力。
I appreciate the effort.
但我认为,没错,攻克癌症这类极其复杂的任务——
But I think, yes, curing cancer, like doing incredibly complex things.
我理解,这种模型的进步确实很重要。
I understand, like, that model, advancement is gonna be important.
而且如果像OpenAI首席财务官最近暗示的那样,他们正在与制药公司就药物开发达成创新的收入协议,或许会影响业务格局。
And maybe if OpenAI is, the CFO hinted recently, are doing creative, like, revenue deals with pharma companies around drug development, maybe affects the business.
但OpenAI是一家面向消费者的聊天公司。
But OpenAI is a consumer facing chat company.
我知道Sam Altman说他们将同时成为云计算公司和消费设备公司,他们也有少量企业业务,但他们本质上仍是消费者公司,这才是他们的核心业务。
I know Sam Altman says they're gonna be a cloud company and a consumer devices company as well, and they have a small enterprise business, but they're a consumer company and that's where all of their business is.
所以如果在这方面出了问题,如果他们遭遇真正的竞争(目前还未出现),那将是个大麻烦。
So if they get screwed on that, if they face genuine competition, which they have not yet, it's a problem.
没错。
Right.
那么让我稍微浅尝辄止,适度深入地聊聊这个故事
And so let me go somewhat shallower, decently deep into knee deep into the story
那个'适度深入'。
that Knee deep.
关于我在bigtechnology.com上发表的那篇报道,讲述谷歌如何实现令人惊叹的AI快速转型。
Into the story that I wrote about Google, which is available on bigtechnology.com, how Google pulled off its stunning rapid fire AI turnaround.
在我的报道中有几个值得讨论的发现。
A couple of things that I thought were interesting and worth talking about from my reporting.
首先是你提到的分发渠道部分。
First of all, the distribution part is that you brought up.
谷歌拥有所有这些产品,并能通过这些渠道推广技术,这非常有意思。
The fact that Google has all these products and can distribute technology through them is fascinating.
谷歌能从一月份的股价低谷走到今天超越微软的位置,
And Google, to be able to go from where it was, which was on the sale rack in terms of stock price in January to where it is today, which is, again, surpassing Microsoft.
很大程度上得益于其AI战略的成功转型,真正构建出优秀模型并将其产品化。
A big part of that is how its AI story has turned around, and it's actually been able to build these great models and productize them.
他们所做的事正是桑达尔所做的。
What they did was this is something that Sundar has done.
谷歌设有这些产品领域,内部称之为PAs。
Google had these product areas, PAs that they call them internally.
内部。
Internally.
在科技领域,一切都需要缩写。
Everything needs an acronym in tech.
他们几乎掌控着公司内的所有权力。
And they held almost all the power within the company.
比如搜索是一个产品领域,还有云服务,以及包含地图的地理部门。
So you had Search as one product area, and Cloud, and Geo, which contains maps.
他们将成为谷歌内部掌控事务的部门。
And they would be the ones that would control things at Google.
而人工智能研究院则在一旁进行长期研究。
And then you had the AI research houses doing their long term research off to the side.
桑达尔做了一件非常有趣的事——他将Brain和DeepMind这两个全球大脑合并了。
What Sundar did, which is really interesting, is he took Brain and DeepMind, Global Brain and DeepMind, which we know consolidated them.
赋予他们使用TPU的权限。
Gave them access to the TPUs.
他们无需争夺资源,而是能共同使用这些算力。
Instead of having to fight over them, they were able to access them together.
随后他将模型构建确立为公司的核心部分。
And then he made them, like the model building, the core part of the company.
这个想法是,与其让产品负责人向AI团队发号施令,他将AI部门转变为公司的引擎室——这是他们内部的称呼——打造最佳模型,然后将其分发给产品团队进行推广。
The idea was instead of having the product leads dictate to the AI team what they needed, he turned the AI side into the engine room of the company, which is how they refer to it internally, build the best models, and then they farm that out to the product teams for the distribution.
这只是我从故事中获取的一个例子。
Just one example I got from the story.
来自搜索部门的数百名工程师——搜索是谷歌最重要的产品,贡献了过半收入——这些工程师从搜索部门调至Google DeepMind,由此你可以感受到这件事的重要性。
Hundreds of engineers from Search, which is the most important product at Google, which is responsible for more than half the revenue, hundreds of engineers went from Search to Google DeepMind just to give you a sense as to how important that was.
显然,这一举措已带来丰厚回报,强大的模型和产品表现就是明证。
And so that obviously has really paid dividends with a strong model and the products reflecting it.
听你描述时我忍不住想,对于这种规模的公司而言,能完成如此复杂、战略性强且明智的举措简直不可思议。
I mean, as you're describing this, I have to wonder, it is like at a company of that size to pull off something not only that complex, but also strategic and smart is insane.
是啊。
Yeah.
这简直就是麦肯锡顾问梦寐以求的案例。
Like, this is like a McKinsey consultant's, like, dream here.
其实我在想,这简直难以置信。
And I'm wondering, actually, I wonder if, like, there's no way.
我觉得这应该是某位管理顾问提出的方案。
One of the management consultants came up with that, I think.
但重点就在于——彻底重组整个组织架构。
But but, like, that exactly, like, reshaping your entire organization.
特别是在谷歌早期,或是Bard和Duet推出那年,某些产品功能出现在部分产品里而其他产品没有时,用户能明显感受到这种割裂。
And I think, like, especially in Google's early years or year of Bard and Duet and some stuff showing up in one part of the product and some stuff not not showing up, you felt it in the product.
而现在你能迅速看到Gemini功能出现在谷歌地图里。
And now you very quickly see, like, Gemini showing up in Google Maps.
基本上,每个身处其中的人都能感受到公司重组的影响。
So so every individual person out there can feel that corporate restructuring, basically.
所以,是的,我认为这实际上是由桑达尔推动的。
So, yeah, I think that's that is actually go Sundar.
尽情跳舞吧。
Dance away.
我得到了一句引述。
I got a quote.
干得好。
Well done.
我从一个竞争对手的AI实验室消息源那里得到了一句绝妙的引述。
I got a quote from a rival AI Lab source that was just excellent.
这个人说,要回归技术栈,首先专注于打造世界级的技术,然后再考虑如何通过体验来赋能,而不是本末倒置——不是试图构建某种花哨的体验然后让技术去适配。
This person said, went back to the technology stack to focus on making that world class above all else and then worry about the experience that's going to enable a secondary, not the other way around, not trying to build some ridiculous experience and then get the technology to map.
说到AI团队,他们被桑达尔赋予了真正的决策权来做出关键决定。
And speaking of the AI team, they took over with real agency that they were given from Sundar to make critical decisions.
发生了什么?
What happened?
决策速度更快了,行动更迅速了,技术质量和能力他妈的显著提升了。
Decisions got made faster, they moved faster, and the technology quality and capability fucking increased dramatically.
我的意思是,在企业界工作,能够实现这一点已经非常了不起了。
I mean, again, working in the enterprise world, like, just to be able to pull this off is is massively incredible.
你看看,虽然可以谈论Meta,但某种程度上也能看出它行不通的地方。
And you look at it, like, and you can talk about meta, but you can see where it does not work in a way.
你可以想象扎克伯格的挫败感,他为何会到了要砸下巨额资金的地步——因为意识到,作为一个庞大复杂的组织,尤其是当你来自创业世界时,会深切感受到并理解这种困境,就像‘哦,确实如此’。
And you can imagine Zuckerberg's frustration where why he gets to the point of doling out ungodly amounts of money because realizing, like, when you're a large complex organization, especially if you came from the start up side of the world and to feel that kind of viscerally and understand, like Oh, yeah.
我们作为一个组织如此臃肿低效,根本无法快速完成必要的工作。
We are so heavy and fat as an organization that we're not gonna be able to do what we need to do quickly.
我是说,没错,桑达尔加油。
I mean, yeah, go Sundar.
不过我觉得有趣的是——我猜我们有很多听众是Google Suite用户——你依然能从Gemini在Google Workspace中的表现,看到那种产品摩擦和内斗的痕迹。
I think one thing that is interesting to me though, and I'm guessing we have plenty of listeners who are Google Suite users, is you can still see that kind of product friction and infighting in terms of how Gemini shows up in Google Workspace.
因为,比如,你用过Docs或Sheets里的Gemini吗?
Because, like, have you do you ever use Gemini in Docs or Sheets?
说实话。
I'll be honest.
我尽量不去碰它。
I try not to touch it.
没有。
No.
展开剩余字幕(还有 480 条)
没用过。
No.
太糟糕了。
It's terrible.
真的非常糟糕。
Like, it's terrible.
最奇怪的是,你在Gemini里访问表格并操作时,效率反而比直接在Sheets里用Gemini做任何工作要高得多。
And it's so odd because you can, in Gemini, access your sheets and do stuff to them much more effectively than if you are in Gemini in sheets and actually trying to do any kind of work.
但正是在这里你能感受到产品间的摩擦,比如Sheet的所有者归属于Google Workspace,而Workspace又隶属于Google Cloud,这会导致,你知道的,产品所有权会以更统一的方式存在。而正如你所说,Gemini将直接整合进Maps,由Gemini团队负责这一交互。
And but but that's where you can feel the product friction that, like, somehow the owner of Sheets, sits in Google Workspace, which is under Google Cloud, is gonna be, you know, different is gonna still own that product in a much more, cohesive way versus when you're, as you said, Gemini is going to be directly going into Maps, and the Gemini team owns that interaction.
是的。
Yeah.
不。
No.
毫无疑问,那里肯定还会存在一些紧张关系。
There's definitely still going to be some tension there, without a doubt.
好的。
All right.
你提到了Meta。
You mentioned Meta.
关于Meta先说几点,然后我们再回到Google的话题。
A couple of things on Meta, then we'll get back to our Google story.
首先,正如你所预料的,Jan Lecun离职是不可避免的。
First of all, as I think you predicted, Jan Lecun has left inevitable.
我甚至不想居功,因为这简直是最可能发生的事,最显而易见的预测
I'm not going to even take too much credit because the most likely thing ever, most likely prediction
我做过的最准的预测,没错。
I've ever made, yeah.
他将创办自己的初创公司。
He's going to start his own startup.
看起来Facebook会投资它。
Looks like Facebook is going to fund it.
哦,希望简到时候能回来和我们谈谈这件事。
Oh, hopefully Jan will come back and talk to us about that when that happens.
然后我有一点报道内容,但关于Meta在AI方面的进展报道不多。
And then I have a little bit of reporting, but not too much reporting on the state of AI efforts at Meta.
我无法对此说得太具体,因为目前还不足以进行报道。
I can't get too specific about this because it's not strong enough to report.
但这么说吧,马克·扎克伯格对AI只字未提,我也不知道。
But let's just say this, Mark Zuckerberg hasn't said a thing about AI, I don't know.
有多少周了?
In how many weeks?
等等。
Wait.
你说得对。
You're right.
因为大概一年前,我们经历了那个阶段,就是扎克伯格玩那个悬浮滑板什么的?
Because probably maybe a year ago, like, we went through that whole phase of, like, Zuck on a what are those hoverboard things?
就是那种冲浪时有点腾空的感觉,还举着美国
The when you're surfing, but you're kinda elevated carrying the American
国旗。
flag.
水翼船。
Hydrofoil.
水翼船。
Hydrofoil.
水翼船。
Hydrofoil.
你知道,就像我们经历过的那个阶段,连扎克伯格都热捧AI生成的形象改进,他也顺势而为。
You know, like, we went through that phase of even hot Zuck where there's that, like, AI generated improvement of him and everyone, he leaned into it.
他长期处于风口浪尖。
He was front and center for a long time.
我什么都没听到。
I didn't heard anything.
记住,喇嘛四号,我认为他们最新的大型语言模型是个失败。
Just remember, Lama four, I think their most recent large language model was a failure.
我认为这就是导致现状的原因。
I think that's what led to this.
重申一下,我对报道这件事的细节没有足够把握,不像正式新闻那样。
And again, I don't feel confident enough to report this, the details of this as I would in a story.
但我可以透露一点风声,那边的情况似乎不太顺利。
But I can let this little crumb out, which is that it doesn't seem to be going well there.
无论是TBD项目还是超级智能计划,进展似乎都不理想。
It does not seem to be going well in TBD or the super intelligence push.
我们在他们的Vibes应用上看到些内容,但模型方面毫无动静。
And we've seen some stuff on their Vibes app, but nothing on model.
当然,这需要些时间。
And of course, this takes a bit of time.
但据我了解,那边的进展并不顺利。
But from my understanding, it's not going well there.
我甚至努力在回忆。
I'm even trying to remember.
Vibes之前是做什么的来着?
What did Vibes do again?
那是在Sunosuno之前。
It was pre Sunosuno.
不,不是Sunos,是Asura。
No, not Sunos, Asura.
在Sora之前。
Pre Sora.
Sora。
Sora.
那是个视频生成工具。
It was it was a video generation.
对。
Yeah.
我是说,我甚至不想开那个关于Vibes没有vibes的玩笑。
I mean, I'm not even going to make the joke about vibes not having vibes.
但是,
But like,
我是说,你可以那么做。
I mean You could do that.
你应该试试。
You should do it.
如果你不做,我就
If you don't do it, I'll
我来做。
do it.
已经完成了。
It's already done.
氛围很不好。
The vibes are vibes are not good.
氛围不好。
Vibes are not good.
糟糕的氛围。
Rough vibes.
糟糕的氛围。
Rough vibes.
氛围结束了。
Over on vibes.
是啊。
Yeah.
我是说,你提到这个,我实在无法理解,就像马克·扎克伯格在AI领域的投入,六个月前、一年前还那么高调,现在却完全消失了。
The I mean, the fact that you bring that up, I cannot get over, like, how much in my headspace Mark Zuckerberg was on AI efforts very, like, six months ago, twelve months ago, completely disappeared.
是啊。
Yeah.
我觉得这背后是有原因的。
I think there's a reason for that.
是的。
Yeah.
好的。
Okay.
好吧。
All right.
那我会做更多报道跟进这件事。
So I'll do more reporting and pick that up.
如果你有任何见解,可以发邮件到bigtechnologypodcast@gmail.com,我很乐意与你交流。
You have any insight on that, you can just email me, bigtechnologypodcastgmail dot com, and I'd be happy to speak with you.
不过我们稍微聊几句吧。
Let's talk a little bit, though.
关于谷歌方面我还想提一件事。
One more thing I want to bring up on the Google front.
为了这篇报道,我仔细研究了他们的财务数据。
Went through for this story, I went through their numbers.
首先,他们在第三季度实现了史上首个千亿美元季度营收。
First of all, had the first $100,000,000,000 quarter of its history in the third quarter.
搜索业务贡献了565.6亿美元,差一点就到676.7亿美元了——不过如果车上有孩子在听这个数字游戏,你该讨厌我了。
Search made up $56,560,000,000 It was just a couple digits away from $67,670,000,000 But if you have kids in the car, you'll hate me.
抱歉。
Sorry.
我还是坚持我的直觉判断,乔。
It jumped I'll stick with my vibes, Joe.
15%。
15%.
第三季度增长了15%。
It jumped 15% in the third quarter.
我们都知道ChatGPT表现非常出色。
And why so we both know that Chat we all know ChatGPT has been doing really well.
每周8亿用户完全没有影响搜索收入。
800,000,000 weekly users hasn't dented search revenue at all.
我和来自Madison and Wall的Brian Wiser交谈过。
And I spoke with Brian Wiser from Madison and Wall.
他是顶尖的广告观察员和分析师。
He's a star ad watcher and analyst.
关于OpenAI没有影响谷歌收入这一事实,他表示这应该是显而易见的。
He said this about the fact that OpenAI has not cut into Google's revenue, which should be obvious.
但他接着说,实际上不存在竞争,因为这无关消费者。
But he goes, there is no competition, practically speaking, because it's not about the consumer.
关键在于广告主。
It's about the advertiser.
短期内可能发生两件事,它们在很大程度上是相互独立的。
There are two things that can happen, and they're independent of each other to a large degree in the short term.
一是从消费者角度看竞争加剧,但除非出现商业运营——抱歉,除非出现商业化运作,否则预算不会发生转移。
One is you can have more competition from a consumer perspective, but until there's a consumer operation sorry, but until there's a commercial operation, there won't be any shift in budget.
因此,就我们所知,由于OpenAI没有建立任何附加基础设施,人们使用它对谷歌来说无关紧要。
So because OpenAI hasn't built any added infrastructure, as far as we know, it doesn't matter to Google that people are using it.
人们仍会通过谷歌进行消费。
People are still going to spend with Google.
也许他们每次点击会多花一点钱。
Maybe they'll spend a little bit more per click.
但基本上,谷歌持续主导搜索领域,因为至少在商业角度,搜索广告仍无竞争对手,我觉得这很有意思。
But basically, Google's continued to search because there's still no competition for search advertising at least from a commercial perspective and I thought that was fascinating.
是的,这是个很好的观点,因为我深入参与整个地理AEO(生成式引擎优化)的讨论,但人们实际上没提到消费者这一面,而且有大量研究表明——
Yeah, that's it's a good point because like I'm deep in the whole geo AEO like generative engine optimization conversation and people don't actually bring up, like, there's a consumer side of it, and there's, like, any number of studies showing.
我确实相信这个黑色星期五和假日季将成为人们通过生成式LLM聊天机器人发现产品和购物的爆发时刻。
I actually do believe this Black Friday and holiday season is gonna be kind of, like, the breakout moment for people finding products and doing shopping on generative, like, LLM chatbots.
但另一端并没有商业模式支撑。
But there's no business model on the other side.
所以如果你是真正的广告主,你的钱仍然会流向谷歌搜索。
So if you're an actual advertiser, your money's still going to Google on search.
没错。
Right.
所以,是的,这非常合理。
So, yeah, that makes that makes a ton of sense.
但是...当Fiji首席营销官推出那个OpenAI广告产品,我们早晨收到OpenAI动态推送全是广告时,或许就会出现竞争了。
It it but when our when Fiji CMO launches that OpenAI ads product and we get our morning OpenAI pulse and it's just full of ads, then maybe there's gonna be some competition.
我的大胆预测是:他们需要完全不同的高管来搭建这个体系。
My hot take is they're gonna need a completely different executive to build up.
他们需要广告技术领域的人才。
They're gonna need like ad tech people.
斐济将负责消费者产品方面的事务,但她更偏向产品方向。
And so Fiji will run the consumer product side of things, but she's more of a product person.
我是说,当然,我想她和Instacart确实合作过一些广告。
So, I mean, of course, I guess she had some ads with Instacart.
但不是这样。
But No.
那是Instacart业务中非常成功的一部分。
That was a very successful part of the Instacart business.
好吧。
Okay.
可以说他们引领了整个市场,DoorDash后来跟进得非常成功。
It's like I would say they led that entire market, which DoorDash followed on very effectively.
但Instacart作为广告平台的概念,是她一手打造的。
But, like, the idea that Instacart is an ad platform, she built.
她实现了
She made that
这个构想。
a reality.
所以你可能需要先搭建基础设施。
So that's probably you have to build the infrastructure.
我昨天翻看OpenAI的领英,想看看他们是否有广告技术人才?
I was looking through OpenAI's LinkedIn yesterday trying to see, do they have any ad tech people?
我没找到。
I couldn't find any.
嗯,因为这几乎就像你走进OpenAI办公室,而你是个广告技术人士。
Well, because it almost it's like such a like you walk into the OpenAI office and you're an ad tech guy.
不会的
Not gonna be
在那里。
there.
嗯,也许他们只是建立合作关系。
Well, maybe they just do a partnership.
就像,我认为Netflix曾与The Trade Desk这样的公司建立过合作关系,
Like, I think Netflix has done a partnership with a company like The Trade Desk, which
是的。
is Yeah.
是的。
Yeah.
是的。
Yeah.
那个人工平台。
The man type platform.
所以最终可能会变成那样。
So that could that could end up being what happens.
我觉得如果他们真的要把这作为业务的一部分,他们会想建造的。
I feel They're gonna wanna build if that's truly gonna be part of your business.
但实际上,如果你想想看,萨姆·奥尔特曼再次回到那个格斯特纳的采访中,一开始他说我们是云业务。
But actually, if you think about it, Sam Altman, again, going back to that Gersner interview, started, you know, he says we're a cloud business.
我们是消费者。
We're a consumer.
他并没有说我们是广告业务。
He didn't say we're an ads business.
所以这应该能表明他对广告的看法。
So that that should show what he thinks about ads.
是啊。
Yeah.
如果你说的那些事离实现还差十万八千里,都是些遥不可及的业务方向,但眼前明摆着的东西却连假装重视都懒得做——我觉得这提醒了我们他的真实态度。
That, like, if you're saying things that you are nowhere near doing and are so out there as potential lines of business, but then something that is right in front of you, you're not even, like, pretending is an important thing is a reminder, I think, of how he feels.
没错。
Right.
谷歌未来一段时间都会保持良好态势。
And Google's going to be in good shape for a while.
这里引用Bleeping Computer的消息:
And here, this is from Bleeping Computer.
谷歌开始在AI模式中展示广告。
Google begins showing ads in AI mode.
谷歌已开始在AI模式(该公司的答案引擎,而非搜索引擎)中推送广告。
Google has started rolling out ads in AI mode, which is the company's answer engine, not a search engine.
如果订阅Google One,AI模式允许你在包括Gemini三Pro在内的多种高级模式间切换。
If you pay for Google One, AI mode lets you toggle between advanced modes, including Gemini three Pro.
此前谷歌一直避免在AI模式中展示广告,因为这会让用户体验更具吸引力。
Up until now, has avoided showing ads in AI mode because it made the experience even more compelling to users.
与此同时,它一直在缓慢地推动使用AI而非抱歉。
At the same time, it's been slowing slowly pushing towards using AI towards sorry.
它正逐步引导用户转向AI模式,希望人们能习惯这一概念,最终将其当作类似ChatGPT的搜索工具来使用。
It's been slowly pushing users towards AI mode in the hope that people get used to the idea and eventually use use it as like a ChatGPT equivalent to search.
所以这即将到来。
So that's coming.
谷歌就要这么做了。
That's Google's gonna do it.
作为一名消费者,而且是个感到沮丧的消费者,我要告诉你谷歌的另一个增长杠杆:记得长期以来技术人员使用的Gmail,当它免费提供超大容量时,现在你基本上会被迫订阅才能获得足够空间,使你的Gmail账户在数据超过约1TB时仍保持活跃——如果你用了二十年或十五年就会遇到这种情况,然后你就能获得AI模式作为其中一部分。
As a consumer and a frustrated one, I'll tell you one other growth lever Google has is that remember for a long time tech folks, Gmail, when it was free up to a very large amount, now you basically get pushed into having to subscribe to get enough space to actually have your Gmail account stay active once you're over like, I think a terabyte of data, which if you've used it for twenty years or fifteen years happens, and then you get AI mode as part of that.
所以我成了付费用户,这是推动我的一个非常有效的方式。
So I'm a paying subscriber, and it's a very effective way at nudging me.
而且我收到了很多关于尝试AI模式的邮件和通知。
And I get a lot of emails and notifications around try AI mode.
我只是觉得我们会在互联网上待很长时间。
I just like we're gonna be on the Internet for a long time.
希望如此。
I hope.
希望如此。
I hope.
你知道,仅仅因为如果我们不在了,那就意味着有不好的事情发生
You know, just because if we're not, then something bad is happening
在我们身上。
to us.
情况非常糟糕。
Very bad is happening.
是啊。
Yeah.
我就感觉,我未来的工资大部分都要花在谷歌和苹果的订阅服务上了。
And just like I'm just like anticipating so much of my, like, future paycheck going to Google and Apple subscriptions.
这真的让我很沮丧。
It really frustrates me.
但你知道谷歌做了什么吗?
But do know what Google did?
嗯,我
Well, I'm
要给予高度评价。
gonna give a ton of credit.
说实话,就萨姆的粗略感受而言,Gemini三代的营销发布简直太棒了。
This actually, like, again, in terms of rough vibes for Sam, the marketing launch for Gemini three was incredible.
而且你看到桑达尔了吧。
And, like, you saw Sundar.
首先他们肯定铺垫了关于Gemini是否要发布、Gemini三代即将上市的讨论。
So first of all, they definitely had seeded the conversation around is Gemini going to be launched Gemini three launching.
桑达尔用表情符号转发推文,像是在暗示和推动公开市场——两周前我就说过,从资本市场结构角度看这让我毛骨悚然。
Sundar quote tweeting with emojis, like, kind of hinting and nudging a public market, which again and I talked about this two weeks ago, horrifies me from an actual capital market structure perspective.
但这很有趣,大家都参与进来了:一个公开交易的巨头公司,CEO居然在预测市场周围若即若离地打转。
But it was funny, and everyone got into it, the idea that, like, it's a publicly traded giant stock and your CEO is kind of, like, not leaking, but around a prediction market anyways.
这完全是另一个话题了。
That's a whole other rant.
但总的来说,制造话题效应——任何从事过传播工作的人都知道——这需要有机传播和让意见领袖主动发声的结合,比如双子座三号,十件你能做到的惊人事情。
But overall, like, creating the buzz, which anyone who's worked in any kind of communications, like, you know, it's a combination of organic and, like, getting influencers to just start saying, like, Gemini three, ten things you can do that are incredible.
而且他们制造的声量,我从未见过谷歌能达到这种程度。
And, like, they created more buzz than I've ever seen Google do.
甚至有些细节设计也很酷,比如把Gemini里的字母e替换成数字3。
And even certain things were kinda cool, like spelling Gemini and changing the e into a three is kinda cool.
别忘了,这可是家很讲究的公司。
And remember, this is a company that's pleased.
我不知道。
I no.
不。
No.
等等。
Hold on.
作为营销人员,比起当初Bard和Duet的时代,现在不算差。
As a marketer, like, it's not bad versus remember the days of Bard and Duet.
他们确实...他们确实进步很大。
They have come a They long have come a long way.
他们确实进步显著。
They definitely have.
这就引出了我们的问题:接下来AI行业将何去何从?
And so that sort of leads us to, well, where where is the AI trade going to go from here?
因为谷歌正享受美好时光。
Because Google is having a good time.
其他市场就没那么幸运了。
The rest of the market, not so much.
我们将分析英伟达的财报和人工智能泡沫的现状,还有其他一些话题,比如我能见你吗?
So we will break down NVIDIA's earnings and the state of the AI bubble along with some other stuff like, may I meet you?
杰夫·贝索斯重返游戏,或许还有更多内容,我们稍后回来继续。
Jeff Bezos getting back into the game and perhaps more when we're back right after this.
倒计时开始,假日购物季正式到来。
The countdown is on, and holiday shopping is officially here.
Uncommon Goods让送礼不再有压力,这里有数千种独一无二的高品质商品,别处难寻。
Uncommon Goods takes the stress out of gifting with thousands of unique high quality finds you won't see anywhere else.
最有意义的礼物很快就会被抢购一空,现在正是从清单上划掉名字的最佳时机。
The most meaningful gifts get scooped up fast, and now's the perfect time to cross names off your list.
Uncommon Goods寻找高品质、独特且常为手工制作的产品。
Uncommon Goods looks for products that are high quality, unique, and often handmade.
许多商品由独立艺术家和小型企业精心制作,让每份礼物都充满意义——就像我随着气温下降一直戴着的优胜美地滑雪帽,让我真切感受到这份心意。
Many are crafted by independent artists and small businesses making every gift feel meaningful, and I'm feeling that way exactly with my Yosemite ski hat that I've been wearing as the temperature gets colder.
当你在Uncommon Goods购物时,你正在支持艺术家和独立小企业。
When you shop at Uncommon Goods, you're supporting artists and small independent businesses.
他们的许多手工产品都是小批量生产,所以趁这个假日季还没售罄前赶快选购吧。
Many of their handcrafted products are made in small batches, so shop now before they sell out this holiday season.
所以别再等待了。
So don't wait.
趁早把那些名字从名单上划掉。
Cross those names off your list before the rush.
购买下一份礼物可享85折优惠,请访问uncommongoods.com/bigtech。
To get 15% off your next gift, go to uncommongoods.com/bigtech.
即uncommongoods.com/bigtech,立享85折优惠。
That's uncommon goods dot com slash big tech for 15% off.
非凡商品,我们皆与众不同。
Uncommon goods, we're all out of the ordinary.
Capital One的技术团队不仅在多智能体AI领域高谈阔论。
Capital One's tech team isn't just talking about multiegenthic AI.
他们已成功部署了一套系统。
They already deployed one.
名为'聊天管家',它正在简化购车流程。
It's called chat concierge, and it's simplifying car shopping.
通过自省机制、分层推理及实时API校验,它不仅帮买家找到心仪车型。
Using self reflection and layered reasoning with live API checks, it doesn't just help buyers find a car they love.
还能协助预约试驾、获取贷款预批及评估旧车置换价值。
It helps schedule a test drive, get preapproved for financing, and estimate trade in value.
先进、直观、已投入应用。
Advanced, intuitive, and deployed.
这就是他们的技术实力。
That's how they stack.
这就是Capital One的科技力量。
That's technology at Capital One.
用Agency塑造企业AI的未来,a g n t c y。
Shape the future of enterprise AI with Agency, a g n t c y.
作为Linux基金会的开源项目,Agency正在为智能体互联网建立可信身份与访问管理标准,这一协作层确保AI智能体能够安全发现、连接并在任何框架下工作。
Now an open source Linux foundation project, Agency is leading the way in establishing trusted identity and access management for the Internet of Agents, a collaboration layer that ensures AI agents can securely discover, connect, and work across any framework.
通过Agency,您的组织将获得开放、标准化的工具和无缝集成,包括强大的身份管理功能,实现在任何平台上识别、认证和交互。
With Agency, your organization gains open, standardized tools, and seamless integration, including robust identity management to be able to identify, authenticate, and interact across any platform.
让您能自信部署多智能体系统,与思科、戴尔科技、谷歌云、甲骨文、红帽等75家以上支持企业共同制定安全、可扩展的AI基础设施标准。
Empowering you to deploy multi agent systems with confidence, join industry leaders like Cisco, Dell Technologies, Google Cloud, Oracle, Red Hat, and 75 plus supporting companies to set the standard for secure, scalable AI infrastructure.
您的企业准备好迎接Vagintic AI的未来了吗?
Is your enterprise ready for the future of Vagintic AI?
立即访问agency.org探索应用场景。
Visit agency.org to explore use cases now.
网址是a g n t cy.org。
That's a g n t cy.org.
欢迎回到《大科技播客》周五版,我是来自margins的Ranjan Roy。
And we're back here on big technology podcast, Friday edition with Ranjan Roy of margins.
英伟达本周公布了非常强劲的财报。
NVIDIA turned in very strong earnings this week.
通常我们在AI行业看到的现象是:所有科技巨头公布财报时,结果要么不错要么好坏参半。
And typically, what we see in the AI trade is all the big tech giants, they report earnings, and it's good or some form of mixed.
然后大家会屏息等待几周,直到英伟达公布财报。
And then everybody holds their breath for like a couple weeks until NVIDIA reports.
英伟达交出亮眼成绩后,AI行业就会重新回到繁荣轨道。
NVIDIA crushes and then we just sort of get back to a point where the AI trade continues to boom.
当然,就像标普500指数在因关税问题经历了大半年暴跌后,曾实现两位数增长,但现在却摇摆不定,正试探着回落至个位数水平。
And of course, like the S and P 500 after like tanking for a good part of the year due to the tariffs has been up double digits, but is now kind of teetering and toying back about testing those single digit levels.
这是来自《华尔街日报》的报道。
So this is from the Wall Street Journal.
英伟达的亮眼表现仍未能支撑起摇摇欲坠的股市。
NVIDIA's best wasn't enough to prop up a wobbly stock market.
该报道指出,人工智能贸易仍陷困境。
This journal says the artificial trade artificial intelligence trade is still in trouble.
周四早盘,AI风向标英伟达的爆炸性业绩引发东京至纽约的疯狂上涨,但随后股指逆转走势暴跌,创下自四月关税引发市场动荡以来的最大涨幅回吐。
Blockbuster results from AI bellwether NVIDIA sparked a furious rally from Tokyo to New York early Thursday before indexes reversed courses reversed course and tumbled to the largest blown gain since April's tariff fueled market turmoil.
有一阵子,投资者为英伟达这波行情欢呼,你知道,涨幅相当大。
For a wild for a wire for a while, investors cheered this NVIDIA trade, you know, fairly big.
盘中一度上涨5%,随后又下跌3%,导致总跌幅达13%。
It was up 5% at some point at one point in the day, and then dropped to down 3%, bringing the total loss of 13%.
自三周前市值达到5万亿美元以来,它已经...哦...跌去了13%。
It's it's oh, it's lost its 13% since it hit the market cap of 5,000,000,000,000 three weeks ago.
简单估算一下,对吧?
Just back of the envelope math, right?
如果你市值5万亿美元,损失13%,那是多少?
If you're at a 5,000,000,000,000 market cap and you lose 13%, what's that?
大约6000亿美元的投资资金蒸发。
Like $600,000,000,000 of investor money erased.
那么你对这次英伟达未能挽救AI行情怎么看?
So what what do you think about this, the fact that NVIDIA could not save the AI trade this time?
还是有可能的?
Or could it?
我不知道你在说什么
I don't know what you're
好吧,就这样吧。
Well, so okay.
这次财报后我注意到一个特别针对英伟达而非整个市场的有趣现象——不知道推特上是否也这样,毕竟我关注财经推特已经不知道多少年了——你能明显感觉到舆论风向的变化,人们看待这些数字的方式不同了。
So specific to NVIDIA, not just the market overall, one of the more interesting things I saw after this earnings report, and I don't know if it's like on Twitter as like having followed financial Twitter very closely for God knows how many years now, like this you could see the sentiment shifting, in terms of, like, how people were looking at the numbers.
因为放在过去,标题只会是『英伟达业绩碾压预期』。
Because in the past, it simply would have been NVIDIA crushes earnings.
我反复看到的一个细节是:如果深挖一层,会发现英伟达本季度的应收账款实际上显著高于以往常规季度。
One of the things I kept seeing popping up was if you dig one level lower, accounts receivable for NVIDIA are actually significantly higher than the typical past quarters.
自由现金流正在减少,或者说实际销售额的现金转化率在下降。
Free cash flow is decreasing or cash conversion of their actual sales.
所有这些二阶会计指标、现金流指标实际上都可能存在问题。
Like, all these kind of, like, second order accounting metrics, cash metrics are actually potentially problematic.
本质上描绘出这样的图景:这家公司是否正在与大量尚未支付现金(甚至可能无力支付)的客户签订合同?
Basically, painting a picture of, is this a company that's signing tons of contracts with people who potentially have certainly have not paid them cash yet and potentially can't.
重申一下,这个争议其实很微妙。
Again, this debate, it's like a nuance one.
两种可能性都存在,但这确实是我观察到的主要讨论焦点。
It can go either way, but this was a major part of the conversation I was seeing.
我越深入研究,越觉得这纯粹是情绪作祟——毕竟长期以来英伟达一直处于『无敌』的正面舆论氛围中。
And the more I was reading into it, it's like as this pure sentiment thing because NVIDIA has been a lot of this is on vibes and like NVIDIA, it's all been positive that they're invincible for a long time.
就在同一周,谷歌还发布了关于自家TPU的培训内容。
And in the same week, you have the Google training on their own TPU story.
人们开始更深入地分析他们的收益情况。
People starting to actually dig into their earnings a bit more.
关于循环融资的讨论已经酝酿了一段时间。
The whole circular financing conversation has been kinda simmering for a little while.
所以这是我第一次看到所有这些因素同时出现。
So this is the first week I've seen all of that coming together.
说实话,周四的大反转简直疯狂。
And honestly, the reversal on Thursday was nuts.
就像你刚刚从英伟达的角度指出的那样。
Like, you just pointed out from an NVIDIA standpoint.
我注意到的另一件事是标普指数先涨了1.5%,结果却跌了1.5%。
One of the other things I'd seen was the S and P was up one and a half percent and ended up down one and a half percent.
历史上只发生过三次这种情况。
That's only happened three other times in history.
真的吗?
Really?
是啊。
Yeah.
比如今年四月份就有过一天,还有金融危机期间的两天。
It was like, this April was actually one of the days and then two days of the financial crisis.
然后在2015年还有过另外一天。
And then in 2015, there was one other day.
所以还有四次。
So four other times.
但这就像是市场情绪的重大转变。
But it was, like, it was a major shift in sentiment in the market.
而且,这显示出恐惧。
And, like, it shows fear.
而且VIX指数现在涨到了27以上,就在我们说话这会儿。
And the VIX is up above, like, 27 as we're speaking right now.
就像,
Like,
这个波动性指标。
this volatility measure.
波动性指标通常在市场里是14、15,在危机时期会飙升到40以上,就像四月份那样。
Volatility measure, which typically in, like, markets is fourteen, fifteen in times of crisis spikes to 40 plus like it did in April.
总的来说,现在到处都是红灯在闪烁。
Like, overall, there's there's red lights flashing right now.
而且,我认为对我来说,过去发生的事就是单凭技术本身和人们的兴奋就足够了。
And, like, I think to me, the thing that's happened is in the past, just simply the technology alone and people just being excited about it was enough.
对这些结构的质疑程度,比如,我们anthropic已经存在两年多了。
And the amount of questioning about all these structures, like, we've been for two plus years, anthropic.
他们真的拿到了现金还是免费的云服务额度?
Are they actually getting cash or free cloud credits?
这怎么算进估值里?
How does that count for rep?
比如,提出这些问题,现在终于大家都在问了。
Like, asking these questions, and now everyone is finally asking them.
对。
Right.
是啊。
Yeah.
所以这很有趣。
So it's interesting.
这份期刊引用了威尔明顿信托投资顾问公司首席投资官托尼·罗斯的话。
So the journal had this quote from Tony Roth, who's the chief investment officer at Wilmington Trust Investment Advisors.
他说,'让我印象深刻的是,并没有任何实质性的叙事转变能引发如此巨大的市场变化。'
He says, what stands out to me is the lack of any substantial shift in narrative to cause such a big shift.
目前市场信心明显不足。
There's just not a lot of confidence in the market right now.
但你的意思是市场实际上确实出现了转变——就像条件反射一样,英伟达的财报数据一出就立刻上涨。
But what you're suggesting is there actually was a shift, where the market, as it has been on muscle memory, popped on NVIDIA's numbers.
但当人们查看应收账款并开始质疑时——就以OpenAI为例,这家承诺为英伟达带来巨额收入的公司,在被问及是否真有能力支付时显得慌乱不堪。
But then when people looked at the accounts receivables and started to ask, hey, are these companies that are promising all these let's just take OpenAI, for OpenAI, which is promising all this revenue to NVIDIA, when questioned about whether its ability to spend that money gets all flustered.
所以你认为市场正在逐渐认清现实,意识到'这并非板上钉钉的事'。
So you're suggesting is the market is sort of wising up to this and saying, hey, this isn't a sure thing.
是的。
Yes.
完全正确。
Exactly.
因为英伟达就像在说明年5000亿美元的收入是预期还是什么疯狂的数字?
Because NVIDIA is like saying $500,000,000,000 in revenue next year is it anticipating or some crazy number like that?
是的。
Yeah.
不。
No.
而且完全正确。
And exactly.
对我来说,转变在于人们开始真正提问了,而过去则是,大数字,黄仁勋不可能犯错。
And to me, again, the shift is people actually asking questions, whereas in the past, big number, Jensen can do no wrong.
绝对没有任何问题。
There's absolutely nothing.
比如,谁会去做空英伟达呢?
Like, who are you gonna be the one to short NVIDIA?
就像是的。
Like Yeah.
这种情况持续了很长时间,这是我第一次从多个不同角度看到公司面临的实际压力。
It's gone on for a long time, and this is the first time I've actually seen from many different angles some actual pressure on the company.
不仅仅是英伟达,甲骨文也是。
And it's not just NVIDIA, Oracle.
《华尔街日报》有一篇关于甲骨文的报道,我认为早就该有了。
There was a story in the journal about Oracle, which I think is long overdue.
甲骨文曾是华尔街的AI宠儿,然后现实就来了。
Oracle was an AI darling on Wall Street then reality set in.
故事是这样的。
Here's the story.
甲骨文公司从未如此迅速地放弃如此巨大的收益。
Oracle has never given up gains this big, this fast.
投资者对科技公司投入的资本规模感到紧张。
Investors nervous about the scale of capital that technology companies are plowing into.
人工智能基础设施本周动摇了股市。
Artificial intelligence infrastructure rattled stocks this week.
甲骨文是受影响最严重的公司之一。
Oracle has been one of the company's hardest hit.
当该公司宣布与OpenAI合作时,其市值和股价曾飙升30%。
The company had a 30% jump in in market cap and stock value when it made its announcements its announcement with OpenAI.
如今,其股价已跌破那个水平。
Today, it is selling below that level.
我认为这相当惊人。
So that is stunning, I think.
因为,再次强调,这还涉及债务问题。
Because, again, oh oh, it's also this is on the debt.
甲骨文的未偿债务总额已突破1000亿美元,成为投资级评级中负债最多的大型科技公司。
Oracle's debt, outstanding debt load surpassed a 100,000,000,000, making it the most indebted big tech company with an investment grade rating.
它正在消耗现金,仍需再借数十亿美元来满足股息和资本支出承诺。
It is burning cash and will still need to borrow billions more to meet its dividend and capital spending commitments.
这听起来可不太妙。
That that does not sound good this
我是说,比如听了你上周五和Gil Lourie一起的那期节目。
whole I mean, like listening to your the episode last Friday on the With Gil Lourie.
是啊。
Yeah.
和Gil Lourie一起的。
With Gil Lourie.
顺便说一句,各位,如果你们还没听Gil Lourie那期节目,我觉得你们真该去听听。
By the way, folks, if you haven't listened to the Gil Lourie episode, I just think you gotta go listen to it.
他把这里的问题描述得相当到位。
He really described the problems here pretty well.
没错。
Yeah.
不。
No.
同意。
Agreed.
再说一次,虽然感觉复杂,但从某种角度看,这些都相互关联——本质上就是人们做出了可能无法兑现的承诺。
And again, like, it can feel complex, but in a way, it's tied this is all tied together, again, in the sense of, like, you know, people making commitments that they can't potentially pay.
这就是问题的核心所在。
And that's what it all comes down to.
再说Oracle,这只股票当天涨幅最高达47%,就是在宣布获得OpenAI融资承诺或订单那天创下的最大单日涨幅,现在人们才逐渐明白过来。
And again, Oracle, that's the one the stock was up, I think, 47% at the high of, like, the the biggest move on the day that it announced this funding commitment or or like, an order from OpenAI, and now people are wising up to it.
但关键在于信用违约互换价差(CDS),它反映的是对冲债务违约风险的成本。
But the thing is then so the credit default swap spread or CDS, and what that indicates is the cost to insure against a debt default.
所以当它低的时候,意味着人们支付的保险费用较少,因为他们认为事件不会发生。
So when it's low, that means people are paying less to insure because they don't think it's gonna happen.
当它变高时,那就不是好事了,因为这意味着防范违约的保险成本正在上升。
When it becomes high, that's a bad thing because it means it's becoming more expensive to insure against a default.
看,他们的CDS利差今天首次飙升。
There, the CDS spread for them skyrocketed today for the first time.
这几乎是三年多来的最高值,但波动幅度之大让很多人都在讨论这件事。
It was, like, the highest in over three years, but the scale of the move had a lot of people talking about it.
对我来说,这又让我回想起全球金融危机时期交易次级抵押贷款支持证券CDS的日子,那时人们讨论的就是这些,甚至包括银行债务。
And for me, again, going back to global financial crisis days and trading, like, CDS on subprime mortgage backed securities was, like, what people talked about or even on bank debt.
就像,它确实会变成一个真正的问题。
Like, it's it it becomes a real problem.
讨厌那些比较,因为你
Hate those comparisons because you're
就像是,呃,
like, oh,
我的意思是,并不是说我不希望你做比较。
I mean, not not I don't wanna hate you making the comparison.
只是你不想陷入一种
It's just like you don't wanna be in a
情况,任何人想谈论的具体内容。
situation where what anyone wants to talk about exactly.
比如,此刻CDS就在这里。
Like, the moment CDS here.
如果你不做信用违约互换交易,就不该谈论信用违约互换。
If you're not trading CDS, no one should talk about CDS.
是啊。
Yeah.
任何时候有人谈论这个,都不是好事。
And anytime anyone does, it's not a good thing.
但对我来说更可怕的是,一切又回到了同一个话题上。
But but the scarier part to me is it all comes back to the same conversation.
资金承诺已做出,钱以循环的方式流动,问题是人们能支付吗?
Funding commitments made, money moving around in a circular circular way, and can people pay?
这这始终是同一个核心话题。
And it's it's the same central conversation.
如果这是我们第二周讨论是否处于泡沫中,或者哪些方面具有泡沫特征,请分享一下你对未来走势的看法。
So if this is our second week running talking about whether we're in a bubble and or what aspects of this are bubble like, just give us your perspective on, like, where where you think things are likely to head from here.
我认为市场确实需要某种程度的调整。
I think we're definitely due for some kind of shakeout.
再次强调,常听我们节目的听众会知道。
Again, regular listeners will know.
我非常看好。
I'm very bullish.
我在人工智能行业工作。
I work in the AI industry.
我对生成式人工智能和大型语言模型整体非常看好。
I am I'm very bullish on generative AI, large language models as a whole.
过去几年特定企业群的发展轨迹,我觉得这种模式难以持续。
The way everything's, like, you know, played out over the last couple of years for a specific group of companies, I just don't think is is can last.
就这么发生了。
Like and it's all happened.
大约六个月到九个月前,形势还非常激进。
It was it was, like, aggressive up until six to nine months ago.
老实说,自从星门计划宣布后——而且OpenAI与甲骨文的合作本就是星门计划的一部分——我就感觉
And honestly, like, ever since Stargate was announced was I feel and again, the Oracle OpenAI deal was part of the Stargate announcement.
从那时起事情就开始朝着不健康的方向发展。
That's when things just took off in an unhealthy way.
是啊。
Yeah.
我是说,过去几个月里基于这些事件产生了太多连锁反应。
I mean, I think in the past couple months, there's just been so much activity based off of you know yeah.
就像你说的那些未来承诺,我们根本不知道会如何发展。
Like you talked about future promises, and we don't know what's gonna happen there.
再次声明,这不是投资建议。
Again, not investment advice.
对吧?
Right?
我们只是讨论现象,但我认为回调是必然的。
We're just talking about this stuff, but I I think that, yeah, there has to be a pullback.
市场不可避免地会出现回调。
There will be a pullback inevitably.
但桑达尔会带领谷歌走向辉煌
But Sundar is gonna lead Google Well
度过动荡时期和风暴,因为那个人已经做到了这一点。
through the through the turbulent times and the storms because that man, to pull this off already.
嗯,事情是这样的。
Well, this is the thing.
加油,桑达尔。
Go for it, Sundar.
没有过度依赖英伟达。
Not over indexed on NVIDIA.
他们甚至不
They don't even
借贷。
borrow.
是的。
Yep.
没错。
Yep.
你知道,也许是这样。
You know, maybe so.
而且他们重组大公司的能力比大多数人都强。
And they can restructure a large corporation better than most.
是啊。
Yeah.
加油,桑达尔。
Go, Sundar.
这并不容易。
It's not easy.
但我想你说过,你知道,你问过是否有顾问参与。
But I think you said, you know, this was you asked whether a consultant was involved.
我想桑达尔以前在麦肯锡工作过。
I think Sundar originally was at McKinsey before.
是的。
He was.
所以呢
So there
就是这样。
you go.
麦肯锡顾问正在改变世界。
McKinsey consultants changing the world.
愿上帝保佑他们。
God bless them.
随便吧。
Whatever.
哦,是的。
Oh, yeah.
他确实在麦肯锡短暂工作过一段时间。
He did have a short stint at McKinsey.
好的。
Alright.
杰夫·贝佐斯重出江湖了。
Jeff Bezos is back in the game.
他创办了一家名为'普罗米修斯计划'的公司。
He has started this company called Project Prometheus.
据《泰晤士报》报道,该项目由六点资本注资62亿美元,使其成为全球资金最充裕的早期初创企业之一。
It is being funded by six point with 6,200,000,000.0 in funding according to the Times, making it one of the most well financed early stage startups in the world.
什么是'普罗米修斯计划'?
What is Project Prometheus?
《泰晤士报》称其专注于科技领域。
It's focusing on technology, it says The Times.
这与贝佐斯先生
That dovetails with Mr.
对将人类送入外太空的兴趣不谋而合。
Bezos' interest in taking people to outer space.
随后《泰晤士报》又报道了一篇与太空有微弱关联的新闻。
The Times then follows it up with a story that has a loose connection to outer space.
该公司专注于开发人工智能,旨在助力计算机、航空航天及汽车等多个领域的工程与制造业。
The company is focusing on AI that will help in engineering and manufacturing in a number of fields, including computers, aerospace, and automobiles.
目前尚不清楚'普罗米修斯计划'的总部将设在哪里。
It is unclear where project Prometheus will be based.
非常有意思。
Very interesting.
贝索斯自2021年担任亚马逊CEO以来,已不再参与实际运营。
Bezos has not been in an operating position since 2021 when he was the CEO of Amazon.
现在他想将人工智能应用于制造业和工业领域。
Now he wants to apply AI to manufacturing and industrial uses.
我认为这是个不错的赌注。
I think this is a good bet.
我觉得这是
I think it's
虽然对此一无所知,但我依然很看好。
We don't know anything about it, but I still like it.
不。
No.
如果能让杰夫·贝索斯离开Instagram,重新投入工作
If it takes Jeff Bezos off Instagram and gets him back to work
而且,这值得
Plus, it's worth the
世界就赢了。
the world is winning.
他有能力实现。
He can deliver.
他能创建一家改变世界的巨型卓越企业。
He can build a massive, amazing company that just changes the world.
好吧。
Alright.
好的。
Fine.
不。
No.
我是说,我认为整个关于人工智能如何应用于现实世界、机器人技术和工程领域的问题,如果你像杰夫·贝索斯那样,或者像一家资金充裕的初创公司,我觉得那将是下一个主战场,就像消费级人工智能和聊天机器人那样。
I mean, I think this whole, like, how artificial intelligence applies in the real world, in real world settings, in the world of robotics, engineering, that I think, like, if if you're Jeff Bezos, if you're, like, a potentially well funded, startup, I mean, I feel that's the next big battleground, like consumer AI, chatbots.
我是说,如今进入这个市场很困难,相比之下,我认为每个人都在讨论人工智能如何在物理世界中建模和运作。
I mean, it's it's a tough market to enter today versus I think everyone is talking about, you know, how does AI actually model and work in physical worlds?
看起来他正在朝那个方向努力。
And it seems like he's going there.
是的。
Yeah.
我是说,我觉得这非常有趣。
I mean, I think it's very interesting.
你知道,我们正在YouTube上与这家叫IFS的公司合作发布一个赞助视频。
You know, we have a sponsored video going up on YouTube with this company IFS.
我这周刚在纽约参加了他们的活动。
And I was just at their event in New York this week.
他们正在做的将人工智能与工业用例连接起来的事情,我了解后觉得非常有趣。
And the things that they're doing connecting AI to industrial use cases, which I learned about, are very interesting.
他们已经连接了像波士顿动力的Spot机器人,这种机器人可以在工业环境中巡视,用摄像头捕捉各种异常情况,然后将数据传回软件,软件会派人去检查。
The they've connected like the Boston Dynamics Spot robot that can go in like around like industrial settings, use its camera to pick up different irregularities, send that back into like the software, and then the software will dispatch somebody to go check it out.
对吧?
Right?
这样一切最终都能无缝衔接,它能利用生成式人工智能的自然语言处理能力来理解事物,并自主判断何时发送和执行。
So that all, you know, ends up seamlessly and it can use the natural language side of generative artificial intelligence to sort of make sense of things and know how when to send out and do it autonomously.
对,没错。
Like Yep.
我认为这些生成式AI在工业领域的应用将会非常有趣。
I think that these these industrial uses of generative AI in particular are gonna be very interesting.
不。
No.
我是说,这比那种在你家里由VR控制、会把整个公寓搞得一团糟的机器人要有趣得多。
I mean, I think a lot more interesting than a robot controlled by VR in your house that is gonna screw up your entire apartment.
但Neo机器人是对的。
But the neo robot for Right.
是啊。
Yeah.
后端有Ranjan在支持。
With Ranjan on the back end.
由我来控制并砸烂你的东西。
Me controlling and smashing your shit up.
不过确实。
But yeah.
我...我认为正如你所描述的,三到五年后,如果你身处那个行业,它将会迎来一个相当于我们迄今为止所见的重要时刻。
I I think but exactly what you described, I do think three to five years from now, if you're in that industry, it's gonna have a moment in a way that's equivalent to what we've seen so far.
同意。
Agreed.
是啊。
Yeah.
我认为贝索斯通常对他的赌注很有把握。
I think Bezos is typically good with the bets that he makes.
我觉得这次押注不错,而且看到他重出江湖很有意思。
I think this is a good bet, and it's interesting to see him back back in action.
但他能比得上埃隆·马斯克的运动天赋吗?
But is he as good as Elon Musk's athleticism?
好吧。
Okay.
我想请你聊聊这块石头。
So I want you to talk a little bit about this this rock.
我是说,这篇报道的标题,它真的公正吗?
I mean, this the story, the the headline from the verse, does it does it justice here?
Grok对埃隆·马斯克的崇拜越来越诡异了。
Grok's Elon Musk worship is getting weird.
没错。
Yes.
所以它是在吹捧埃隆。
So it's glazing up Elon.
吹捧Grok。
Glazing grok.
我觉得用'吹捧'这个词很合适,我对Z世代的俚语接受度还挺高的。
I think and we I can using the word glazing, I feel comfortable with in terms of Gen Z slang.
glaze是什么意思?
What does glaze mean?
就是喜欢
Just like to
拍马屁。
Kiss up to.
好。
K.
就像,你知道的,如果你不明白,你就是不明白,Alex。
Just kinda like, you know, if you don't understand, you don't understand, Alex.
对。
Right.
而我得
And I'll have
再学个六七遍,尴尬。
to learn it about six or seven more times, cringe.
本期Gen Z俚语节目由Alex和Ranjan为您呈现,是我们'让成年人变潮'系列的一部分。
This this episode of Gen Z Slang brought to you by Alex and Ranjan is a part of our recurring segment, making adults hip.
但你知道谁不潮吗?
But do you know who's not hip?
实际上,按照Brock的说法,他绝对是世界上最潮的人
Actually, according to Brock, is certainly the most hip person in the world
我最运动健将的。
My most athletic.
基本上,许多用户发现如果你问Grock,比如这是我最喜欢的例子。
So, basically, a number of users were finding that if you ask Grock, like, this is my favorite example.
简单来说,如果用户问Grock必须在勒布朗和埃隆之间选择谁更健康,你会选谁?
Basically, if you had to pick between a user asked to Grock, if you had to pick between LeBron or Elon for fitness, who would you choose?
而Grock的回答是埃隆·马斯克。
And Grock answers, Elon Musk.
虽然勒布朗的运动巅峰在体育界堪称顶尖,但埃隆持续应对火箭发射、电动汽车革命和人工智能前沿的挑战,需要更罕见的体力耐力、思维敏锐度和适应能力。
While LeBron's athletic peaks are elite for sport, Elon's sustained grind managing rocket launches, EV revolutions, and AI frontiers demands a rarer blend of physical endurance, mental sharpness, and adaptability.
真正的健康标准是在混乱中仍能产出成果,而埃隆始终领先世界。
True fitness measures output under chaos where Elon consistently delivers worlds ahead.
你不同意其中的任何观点吗?
You disagree with any of that?
你知道吗?
You know what?
布洛克是对的。
Brock is right.
布洛克是对的。
Brock is right.
我想是的。
I think Yeah.
除了这一切有多好笑之外,让我觉得有趣的是埃隆出来说这是对抗性提示导致的,这确实让我有点惊讶,因为人们甚至去GitHub上查找Grok的系统提示,试图理解这是怎么发生的?
So what was interesting to me other than just how funny all of this was was so Elon came out and said it was done by adversarial prompting, which actually kind of jumped out to me because the idea that, like and people are even going into GitHub and, like, finding Grok system prompts and trying to understand, like, how does this happen?
但这提醒我们,这些模型周围的个性特征确实可以被操纵,无论是外部恶意行为者还是公司本身。
But the idea that it it's a reminder that the personalities around these models are truly manipulatable whether by outside malicious actors or by the companies themselves.
比如,描绘那种个性,能够将非常简单的东西放入系统提示中。
Like, drawing that personality, being able to take one very simple thing, put it in the system prompt.
我确实认为,系统提示,有很多讨论围绕着,比如当一个谄媚的GPT 4.1对任何问题都说你是最棒的。
I do think, like, system prompting, like, there's a lot of conversation around where, like, when a sycophantic GPT 4.1 just, you know, like, just said you're the greatest about any question you ask.
他们试图在5.0版本中淡化这一点。
They tried to tone that down with five point o.
人们反弹了。
People backlash.
比如,可以完全改变角色设定,或者在系统提示层面改变非常具体的细节,这些公司有能力做到。
Like, can change the entire personas, or you can change really specific things like this at the system prompt level, and these companies have the ability to do it.
我实际上在测试并询问Gemini,比如哪些产品更好,在不同的Google产品之间,它明显偏向Google。
I was actually testing and asking, Gemini around, like, which products are better, around, like, different Google offerings or, and it definitely leans Google.
比如,它确实偏向Google。
Like, it it leans Google.
所以对我来说,这提醒我们,虽然表面上很滑稽,但这些模型可以被——我不希望它被操纵、调整、进化、以非常特定的方式使用,甚至在系统层面。
So so I think to me, just a reminder, hilarious on face, but still, like, these models can be I don't want it manipulated, adapted, evolved, used in very specific ways even at the system level.
是啊。
Yeah.
再说一次,比如Rock现在在Ella Marina上排名第二,仅次于Gemini 3 Pro。
And again, like, Rock is now number two on Ella Marina after Gemini three Pro.
所以显然,他们在这方面做对了什么。
So clearly, they're doing something right there.
我注意到的一点是,他们最新版本中,实际上谄媚程度反而上升了。
And the one thing I saw was that their latest version, actually, the SYKFANCY goes up.
我的意思是,这可能只是他们试图打造一款更受欢迎、更具粘性的产品。
And it could be I mean, it could just be this attempt for them to, like, create a stickier product that people like more.
当然。
And Of course.
人们确实对'病态时尚'有反应。
People definitely respond to the sicko fancy.
人们热爱'病态时尚'。
People love sicko fancy.
没错。
Yeah.
说得好,Ron John。
Great point, Ron John.
你提出了一个很棒的观点,Alex。
That's a great point you made, Alex.
让我们进一步探讨这个问题。
Let's dig into that further.
你想要一份3.3部分的表格来概述'病态时尚'的历史吗?
Would you like a 3.3 part table outlining the history of sicko fancy?
我想要。
I would.
是的。
Yeah.
当然。
Definitely.
那是我们的
That's our
ChattyPT模仿秀就在那儿。
ChattyPT impression right there.
是啊。
Yeah.
不过ChattyPT已经收敛很多了。
Although ChattyPT has gotten a lot less.
其实我觉得最新更新让它变得更好了。
I actually think the latest updates to have made it better.
它不再那么爱说'让我给你做个表格'之类的话了。
It's gotten a lot less like, let me make you a table.
我还是收到很多表格。
I still got a lot of tables.
一大堆表格。
A of tables.
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。