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。
AI的氛围变得越来越糟糕了。
The AI vibes are getting ugly.
到底发生了什么？
What exactly is going on?
黄仁勋会在英伟达即将举行的旗舰级GTC大会上为AI故事注入新的动力吗？
Will Jensen Huang give the AI story a boost at Nvidia's forthcoming flagship GTC event?
而Meta那边到底发生了什么？他们的最新模型又推迟了？
And what in god's name is happening at Meta where the company's latest model is delayed again?
这些内容将在本周末的大型科技播客节目中播出，就在本节目之后。
That's coming up on a big technology podcast Friday edition right after this.
欢迎收看大型科技周五特辑，我们将以一贯冷静而细致的方式为您解析最新动态。
Welcome to big technology Friday edition where we break down the news in our traditional cool headed and nuanced format.
今天的内容非常精彩，我相信你们会喜欢这场节目。
We have a wild I think it'll be a fun show for you today.
我们将探讨为什么AI正面临日益增长的抵制。
We're gonna talk about why AI is in the midst of a growing backlash.
我们将会讨论在即将举行的NVIDIA GTC大会上可能看到的内容，这将是下周最大的新闻。
We're gonna talk about what we might expect at Nvidia's forthcoming GTC conference, which is gonna be the big news of the next week.
我们还会讨论亚马逊及其他公司在‘氛围编码’上遇到的麻烦。
We're also gonna talk about Amazon and others having trouble with their vibe coding.
当然，Meta有一个大型AI模型，尽管投入了大量资金和人才，却迟迟未能发布。
Then, of course, Meta has this big AI model that is simply not shipping despite all of the money and all the talent they're putting towards it.
和往常一样，周五我们邀请到了Margins的Ranjan Roy加入我们。
Joining us as always on Friday is Ranjan Roy of Margins.
Ranjan，很高兴见到你。
Ranjan, great to see you.
欢迎回来。
Welcome back.
AI的氛围不对劲。
The AI vibes are off.
这周的讨论将会很有趣。
This is gonna be an interesting discussion this week.
你知道吗，以前每天都会有一个新的AI突破成为人们讨论的焦点。
You know, back in the day there you seemed like every day there was a new AI breakthrough that people were talking about.
但现在，每天似乎都有一项新的AI研究揭示出公众对AI认知中的一些阴暗面。
Now it seems like every day there's a new AI study that is pointing towards, some ugliness around the way that the public perceives AI.
而对我来说，这个问题本周达到了顶点。
And for me, really came to a head this week.
我不知道你有没有看到，萨姆·阿尔特曼在一次投资者会议上发言时，有一句话在推特上流传开来，他说：我们预见的未来是，智能将成为一种像额外电力或水一样的公用事业，人们会按使用量向我们购买。
I don't know if you saw, Sam Altman speaking at an investor conference and there was a quote that went around Twitter where he said, we see the future where intelligence is a utility like extra electricity or water and people buy it from us on a meter.
但这只是被断章取义的一句话，实际上他当时是在解释为什么他们要投入如此多的AI基础设施建设，以及他希望让AI变得廉价的初衷。
Now that was like, one part of a quote taken out of context where, you know, he was talking about why they're building AI infrastructure, so much AI infrastructure, and his desire to make it cheap.
不知怎的，互联网却抓住了这句话，人们对此彻底炸锅了。
Somehow the internet just took hold of this statement and, you know, people went bananas on it.
这是我看到的推特上的一条反应。
Here's one reaction I saw on Twitter.
当萨姆·阿尔特曼揭示出他的真实意图时，他表现出了一种危险的精神病态迹象。
Sam Altman shows signs of being a dangerous psychopath here when he reveals his true intentions.
如果你在看到这件事后，对大型科技公司和AI公司仍没有一点怀疑，那你真的应该反思了。
If you don't have some skepticism of big tech and AI companies, you really should after seeing this.
对我来说，这感觉有点过度反应了，但这种情况确实发生了，我们接下来会看看民调数据。
Like, that felt like an overreaction to me, but it does happen and we're gonna go through the polling numbers.
这发生在公众对AI日益增长的不安情绪之中。
It happens in the middle of a growing unease that the public has towards AI.
你觉得呢，拉詹？
What do you think about this, Ranjan?
听众很少会听到我为萨姆·阿尔特曼辩护，但我现在就要这么做。
It's not very often that listeners will hear me trying to defend Sam Altman, but I'm gonna do that right now.
所以，也许你可以从另一个角度来看待这个问题。
So maybe you take the other side on this one.
但我其实觉得这并没那么令人反感，因为我确实认为AI公司已经在向更精细化的按使用量计费模式转变了。
But I I I actually found this not that offensive in the sense that I do think it's already getting granular, already companies moving towards more consumption based models.
比如，有个叫Clay的公司，是一家主打AI市场推广的公司，当时就因为它们把整个定价模式转向按使用量计费而引发了巨大争议。
Like actually, there was a there was a company Clay, which is like a go to market AI model that there was actually a big controversy around because they were kind of converting their entire pricing model towards consumption based.
我们已经看到大量报道指出，Claude每月200美元的计划实际上是在补贴用户。
Already, we've seen lots of reporting around even Claude's $200 a month plan actually is subsidizing.
比如，平均而言，你实际消耗了3000到5000个token。
Like, you're actually consuming 3,000, $5,000 tokens in average.
人们在生活各个领域使用各种AI工具、工作流和代理流程，这些都会产生成本。
Like, people using various AI tools and workflows and agentic processes in whatever parts of their life, there's gonna be a cost to it.
我认为，认识到这将是一个按使用量计费的模式，我们可以将其比作电力。
And I actually think like recognizing that it's gonna be a consumption based model, we can equate to electricity.
水这个例子对我来说有点奇怪。
Water was kind of a weird one to me.
你付水费吗？
Guess, do you pay your water?
你是单独支付水费，还是水费包含在房租里？
Do you pay for a water bill or is it part of your apartment?
水费包含在我的房租里。
It's part of my rent.
是的
Yeah.
我想，对于我们这些住在纽约市的非房主来说，水费并没有那么明显地影响到我们。
I guess guess we're as non homeowners living in New York City where the water bill doesn't quite hit us the same.
但我认为电费就不同了，你确实会有一些使用量，这对我来说并不不合理。
But but I think electricity, that one's not unreasonable to me that you're going to have some kind of utilization.
人工智能将融入日常生活，你会按照使用量来付费。
AI will be baked into just daily life, and you will pay for it on a consumption based model.
当你这么一说，听起来就没那么令人反感或可怕了。
Like, when you put it like that, it doesn't sound as offensive and scary.
我的意思是，他的沟通能力，我只能想象他的公关团队每天得有多辛苦，因为这是一个如此简单、完全合理的观点，却只有萨姆能把它说得让大众觉得恐怖。
It's just I mean, his communication skills, I can only imagine what his PR team has to do every day because because this is such a simple point that is not unreasonable, yet only Sam can make it sound, like, terrifying to the general public.
会不会是因为人工智能的能力不断增强，人们对这项技术的不安感不可避免地将我们带到了这一步？
Could it be that as AI's capabilities grow, people's uneasiness with the technology was inevitably going to lead us to this point?
比如，对你我来说，我们清楚AI能做什么，稍后我会提供一些数据，说明AI用户和非AI用户对AI的感受有何不同。
Like, for you and I, right, we we see what AI does and we'll have some data about what AI users feel about AI compared to people who don't use it in a moment.
而且剧透一下，如果你使用这些工具，你的负面情绪会比不使用的人少得多。
And spoiler alert, if you use the tools, you're much less negative than if you don't use them.
但会不会只是因为人们看到了AI迅速提升的能力，感到恐慌，从而导致了公众中的这种负面情绪？
But could it just be that people are seeing AI's rapidly advancing capabilities and they are getting freaked out, and that is leading to some of this negativity in the public.
这是AI能力发展的一个副产品，我认为这一点在围绕萨姆言论的讨论中被特别凸显了，因为我查看了所有我能找到的负面评论，其中大多数并没有说：‘你到底怎么了？’
It is a byproduct of the AI's capabilities, and I think that this was definitely something that was spotlighted in the conversation around Sam's, comments because I looked at all the negative comments or about as many as I could, and the majority of them didn't say, oh, like, you know, what's wrong with you?
人们会想：‘你居然想让我们为使用AI付费？你为什么要建这么多基础设施？’
You wanna, like, you know, you're trying to charge us to use AI and, you know, why are you building all this infrastructure?
而是像这样的评论。
It was comments like this one.
我看到一个未来，在那里我们不允许萨姆·阿尔特曼这样的个人将我们的共同知识、智慧和沟通能力商业化，一个我们让AI民主化并使其服务于公共利益的未来。
I see a future where we don't allow people like Sam Altman to monetize our common knowledge, intelligence, and communication, a future where we democratize AI and make it contribute to the common good.
还有人说：‘嘿，萨姆。’
Someone else said, hey, Sam.
你的智能是从哪里来的？它是如何积累的？那些来源得到了补偿吗？
Where did you where did the intelligence come from, and how was it accumulated, and how were those sources compensated?
还有人说，收费使用AI就像把社区的资源掠夺走，然后大规模地卖回给社区，这简直是个骗局。
Another person said, in the like, charging for AI is like a lame old strip a resource from a community and then sell it to them grift at huge scale.
对吧？
Right?
但我认为，如果你使用过这些工具，你就知道它们是增益性的，你很清楚它们做的远不止是复述它们所依赖的互联网数据。
And I just think that like if you use these tools, you know that they're additive, you definitely know that they do much more than just spit out the internet that they were drained on.
但我感觉人们有一种不安，很多人说，他想让我们变笨，然后我们还得为自己的智力向他付费，对我来说，这完全源于两件事。
But I think there is this uneasiness that it's become and a lot of people were saying, well he wants to make us dumb and then we'll have to pay him back for our intelligence like, know, and it's just like this to me is all a product of of two things.
第一是人们对这些工具快速进步的不安，我认为这很合理；第二部分可能是经济层面的，这家公司成长得极其迅速，而公众却无法参与其股票市场。
One is the unease around the way that these tools are getting better and I think that's reasonable And then the second part is probably a financial part that this company has grown tremendously and the public has not been able to participate in the stock market.
尽管如此，因为它是私营公司，也许这算是次要因素。
Although, in it because it's private and maybe that's secondary.
好的。
Alright.
实际上，我想再补充第三点。
Well, I would actually add a third part.
只是因为这些代言人是谁。
It's just who the spokespeople are.
再说一遍，我认为这个品牌问题很大程度上在于代言人是萨姆·阿尔特曼、埃隆·马斯克，甚至最近几周达里奥似乎成了叙事中的好人。
Again, I think so much of this branding problem is around when it is Sam Altman and Elon Musk and even, I mean, I guess Dario is kinda seeming to be the good guy in the the the narrative over the last couple of weeks.
但关键就在于这些代言人是谁，以及他们如何谈论这件事。
But but it's it's who the spokespeople are, how they're speaking about it.
再者，对这样一条评论产生如此强烈的反应，我认为更多反映的是人们对这种标志性人物——硅谷科技领袖——长期积累的反感。
Again, to have such a reaction to a comment like this, I think it's more reflective of kind of long simmering disdain for this kind of figurehead, the Silicon Valley tech pro whatever it is figurehead.
我认为这才是问题的核心。
I think that's that's more at the core of this.
我认为能力层面的问题也很重要，我们当然可以深入探讨，比如这对白领工作和知识型工作意味着什么。
I think the capability side, and we can definitely dig into that, but but it is it's like in what's that gonna mean for white collar jobs and knowledge work, think is definitely important as well.
在经济层面，我认为这种观点确实有其合理性，但奇怪的是，关于这些模型训练所依赖的数据、以及为何无人获得补偿的版权讨论却完全消失了。
And the on the the economic side, I think like there is a validity to this idea and somehow it seems like all the copyright conversations have just completely gone away around what it was this all of these what were all of these models trained on and how no one was compensated.
现在，他们却要把这一切当作电力一样来收费。
Now every they're gonna be monetizing this as like a like electricity.
我认为这一点确实有道理，我也同意。
I think there is something to be said on that one, but I think I agree.
这些都是因素的综合，在美国的大选年，这个问题会变得越来越突出。
It's a combination of all of the above that it's gonna get, I mean, in an election year in The US year, like, it's gonna become more and more of an issue.
好的。
Okay.
让我和你一起做个思想实验。
Let me just do a little thought experiment with you.
当你谈论人工智能时，如果对方感到不安，也许这只是一个思想实验，或者我们不妨稍微探讨一下。
When you speak with about AI and they're uneasy, maybe this is even a thought experiment, maybe this is just let's poke at this a bit.
当他们说感到不安时，具体是指什么？
What do they say about it when they say they're uneasy?
是因为我知道萨姆·阿尔特曼是谁，但他不擅长表达，还是因为这项技术变得太可怕了，我担心自己几年后会丢掉工作？
Is it I know who Sam Altman is and he's not a great communicator, or is it this shit is getting scary good and I'm worried I'm gonna have a job in a couple years?
我所有从事工具类工作的朋友都在讨论，他们正在估算自己还能在岗位上干多少年，直到人工智能开始接手他们的工作。
All my friends who are in the tools, we have conversations about the fact where they're plotting out how many years they're gonna have left in their jobs before AI starts to do their work.
我们轮流讨论谁的工作最不容易被自动化，谁的工作最容易被自动化。
We go around and we talk about who's gonna have a job for the longest and who's hardest to automate and easiest to automate.
听我说，我不认为我们会因为人工智能而立即看到大规模的自动化和失业。
Look, I don't think we're about to see mass automation and unemployment because of AI.
我谈过这个话题，我再次强调，我愿意倾听对方的观点，但这件事确实让很多人感到恐慌，而在我看来，这才是核心问题。
I've talked about this, I, again, I am willing to hear the other side of the argument but it's freaking people out and that to me is the core.
嗯，我认为有两方面，但毫无疑问，它会造成动荡，至于时间表，肯定不会是几天、几周或几个月的事。
Well, see, I think there's there's two parts but again it is gonna cause disruption and I think there's no doubt about that and what that timeline is I think it's it's not gonna be days and weeks and months.
我认为这会是一个漫长的过程，也许这有点乐观，但我相信一定会出现动荡，这种担忧是合理的。
I think it's gonna play out over a long time and maybe that's optimistic, but I think there will be disruption and that's reasonable fear.
但我发现人们更多是在说：这东西不好用、不靠谱，这几乎恰恰相反——它还是会幻觉，它并没有人们说的那么好，这些公司只是想融资，让少数人发财。
But I think I actually hear more, it's not good and doesn't work, which is kinda it's almost the opposite that there's still it's gonna hallucinate, it's gonna like it there's all this promise around it, it's actually not as good as everyone says and they're just trying to these companies are just trying to raise a bunch of money and make a few people rich.
所以我认为实际上，如果你不只盯着那些美好的愿景，而是看现实，你会更明显地发现人工智能其实并不行。
So I actually think there's still and again, if you're on blue sky and not x, you'll see that even more around like the AI is bad.
大型语言模型并不是它们所承诺的万能解药。
Large language models are not the panacea that they promise to be.
所以我认为，仍然存在整个这样的群体，他们只是有一种担忧，不仅仅是那些身处其中的人——正如你所说——其实更紧张，甚至有些害怕，但我认为还有另一种声音，就是AI根本没那么好，而且被过度炒作了。
So so I think there's still that entire faction and just kind of like thought, the threat of thought, not just that it's those who are in it as you're saying are actually a bit more nervous, at least scared maybe, but but I think there's also the it's just not good and it's all overhyped.
好吧，我要说，当AI领域的领袖们谈论这些事情时，确实没有帮到他们的立场，比如达里奥跑出去说我们会面临50%的失业率，或者穆斯塔法·苏莱曼谈到初级工作岗位时说类似的话。
Well, will say that AI leaders are definitely not helping their case when they talk about the fact, like, when Dario goes out there and says, like, hey, we're gonna have 50% unemployment or Mustafa Suleyman in in entry level work or Mustafa Suleyman says things like this.
这些标题到处传播。
These headlines circulate.
顺便说一句，这些标题传播的范围并不局限于科技媒体。
They circulate by the way, they circulate not outside of the tech press.
它们还会出现在Axios、NBC新闻、《纽约时报》、CBS等主流媒体上，而不是像TechCrunch那样的科技媒体标题。
They circulate in the Axios and the NBC News and New York Times, CBS, you know, as opposed to, like, a TechCrunch headline.
这些主流媒体的报道反而比像Ed Zitron的帖子，或者Gary Marcus谈论AI并不靠谱的内容获得更多关注。
Those actually get more play than, like, the, you know, for whatever reason than an Ed Zitron post about, you know, the, or Gary Marcus talking about how AI doesn't work.
对了，你有没有注意到，这周谁又火了？
Did you see, by the way, you know who was viral again this week?
安德鲁·杨，那个搞全民基本收入的人。
Andrew Yang, the UBI guy.
他说我们应该停止对劳动征税，而应该对人工智能征税。
He said we should stop, taxing labor and we should tax AI instead.
这正是我想说的。
That's what I'm saying.
这就是现在的情况
This is what's
安德鲁·杨，那个基本收入主张者，现在又火了？
Is going Andrew Yang the UBI guy now?
这位就是安德鲁·杨，他曾竞选总统。
This is Andrew Yang who ran for president.
他参选过
He ran
他以基本收入、每人一万美金的政策竞选总统。
for president on a UBI, $10,000.
曾经
Was
UBI倡导者
the UBI guy
因为AI。
because of AI.
UBI曾经是个话题。
UBI was like a thing.
为安德鲁·杨点赞。
Good for Andrew Yang.
没错。
Correct.
为对点赞。
Good for Yeah.
我的意思是，我觉得UBI这个话题本身是另一回事，但就我而言，有一件事让我一直感到困惑，那就是每次用iPhone拍照时，都会经过一个相当复杂的AI处理过程。
I mean, I guess the UBI conversation is a whole other thing, but like to me one thing in terms of I don't want to use AI or people like one thing that still kind of baffles me is every time you take a photo with an iPhone, it is running through a pretty heavy AI process.
每次进行谷歌搜索，即使没有AI概览，背后也有大量人工智能技术在运作。
Every time you do a Google search even without AI overview, there's plenty of artificial intelligence that's been built into.
每次你刷Instagram和Facebook时，都被AI深深包围着，无论是推荐内容还是针对你的广告。
Every time you scroll Instagram and Facebook, you are just seeped in AI in terms of the recommendations, in terms of like the ads that are targeting you.
所以对我来说，这些平台——除了聊天机器人，也除了在OpenClaw上自己构建代理——都内置了大量AI，人人都在用，却没人抱怨，而这所有的一切都集中在这些企业和人群身上，这正是我认为AI的焦点更应针对这些特定群体而非AI技术本身的原因。
So it it it's always kinda interesting to me that these platforms that everyone uses outside of a chatbot, outside of building your own agents on OpenClaw are like there's so much AI built into them that everyone uses but no one complains about but it's all centered, which is why I think even more it's specifically targeted at these people in these companies rather than AI as a technology itself.
但这里有个区别，AI做的是预测，虽然说到底都是预测，比如预测分析和这种新闻feed排名，它只是帮助新闻feed预测你更可能互动的内容，这并不会夺走你的工作。
Well, there's distinction between AI that does, you know, predictions and I mean, it's all doing predictions, but like, you know, predictive analytics and these type of newsfeed ranking, which is like helping the newsfeed predict what content you'd be much more likely to engage with, that's not taking your job.
但能说话、能操作程序的AI，你可能正处在这一领域的核心。
But the AI that can talk and operate programs, I mean, you know, you're probably at the epicenter of this.
这才是真正让人担忧的。
That's what really gets people worried.
不是像我用在照片上的那种AI。
It's not like the AI that I'm like, you know, using on my photos.
而是能代替我工作的AI。
It's the AI that can do my work.
没错。
No.
不。
No.
但你在Instagram上看到的每一个广告，都不只是依靠传统的机器学习来驱动的。
But every ad that you are targeted with on Instagram, it's not just traditional machine learning that's powering that.
而是像代理流程那样，整合了各种不同的数据集，来决定今天我的Instagram动态会展示什么内容。
It's like agentic processes that are, you know, pulling together all different types of disparate data sets that are coming together to show me that I need, I don't know, whatever my Instagram feed is gonna be showing me today.
这已经不再是机器学习了。
That's not machine learning anymore.
再次说句扎克伯格的好话，我们接下来会谈到一些非传统的内容。
Again, to Zuckerberg's credit and we're gonna be getting Un traditional.
是的。
Yeah.
我们会深入探讨Meta出了什么问题，但有一点他们确实做对了：他们彻底重建了整个广告基础设施，融入了大型语言模型和代理流程，以及新一代的AI，而不是传统的机器学习，这正是让他们——值得一提的是——在瞬间重建了整个业务。
We're gonna get into what's gone wrong at Meta, but one thing they've certainly done right is they've rebuilt their entire advertising infrastructure to incorporate large language models and agentic processes and and the newer vintage of AI rather than the traditional machine learning and that's what's made it, I mean it basically just not saved but like rebuilds their business on the fly to their credit.
所以对我来说，这其实是一回事。
So so I think to me it is the same.
这不仅仅是通过ChatGPT提问或生成图片，那是人们通常对AI的认知。
It's just not in chat GPT asking some questions or generating some images, which is what everyone associates with AI.
不，但我要说的差异就在这里。
No, but here's why where I'm saying there's a difference.
是的，他们确实在使用这项技术，但我谈的是更广泛的技术整体。
It's that the technology, yes, they are using that technology, but I'm talking about the broader technology overall.
与前几代相比，它能做的事情要广泛得多，而这正是让人不安的原因。
Compared to the previous generations, it's much more expansive in what it can do and that's where the unease comes from.
你知道，即使Meta已经改变了这一点，你正在与之互动，你之前在浏览信息流时就已经被某种AI影响了，但那种AI远没有今天这么强大。
You know, not the fact that even if Meta has, like, sort of flipped it and you're touching it, you all you were you had a form of AI that was, you that was touching you when you were, like, running through the newsfeed before, but it wasn't a version that was as expansive as today's.
所以，这就是我的观点。
So that's sort of where I'm coming from.
我明白。
I see that.
不，我的意思是，当有人实际互动并感受到它的强大时，我们马上就要谈到整个Anthropic五边形的博弈了。
No, I mean I see that when someone does interact and sees how powerful and again, like, we're gonna get into kinda the whole anthropic pentagon battle right now.
但当你想到所有被收集并公开可用的数据时，这些数据过去用于向你推送广告，现在却可能被五角大楼实时分析，从而对你进行监控。
But like, when you think about all the data that is collected and publicly available that used to be target using used to target you with ads now could actually be instantly analyzed by the Pentagon in real time to actually surveil you.
我同意。
I agree.
这更可怕了。
That's that's scarier.
这确实是个更沉重、更难消化的问题，但我不确定。
That's like, it's kind of a heavier thing to try to process, but but I I don't know.
我还是觉得，如果我们能在这一运动的前端有一些友善、亲切的面孔，人们对这项技术的看法可能会完全不同。
I still think if if we just had some friendly nice faces at the front of this movement, it could be such a different different like perception of the technology.
好的。
Alright.
我们来谈谈数据吧。
Let's get into the numbers here.
我们本周早些时候和奥利维亚·摩尔聊过这个，但再提一下：一项新的NBC新闻民调发现，50%的选民认为人工智能的风险大于其益处。
This is we talked about this with Olivia Moore earlier this week, but to bring it up again, a new NBC News poll found 50% of voters thinks the think the risks of AI outweigh the benefits.
74%的白领工作者和50%的蓝领工作者都使用过人工智能，但两者的顾虑相似。
AI has been used by 74% of white collar workers and 50% of blue collar workers, but both had similar reservations.
好的。
Alright.
我们来逐条看看从受欢迎到不受欢迎的清单，我会先读一堆内容，然后再谈到人工智能。
Let us go through the list of likability to unlikeability, and I will read a bunch of things before I get to AI.
这是NBC新闻的这项民调，我接下来读的这些内容，全部都比人工智能更受欢迎。
So this is this NBC news poll, everything that I'm gonna read before AI is more favorable than AI.
教皇利奥十四世、斯蒂芬·科尔伯特——令人意外地受欢迎，别误会，科尔伯特排第二，仅次于教皇。
Pope Leo the fourteenth, Stephen Colbert, who is surprisingly likable, no shot on Colbert, but he's number two right behind the Pope.
马可·卢比奥、庇护城市、J.D.万斯、AOC、唐纳德·特朗普、共和党、卡玛拉·哈里斯、加文·纽森，这些都比人工智能更受欢迎？
Marco Rubio, Sanctuary Cities, JD Vance, AOC, Donald Trump, the Republican Party, Kamala Harris, Gavin Newsom, all more likable than AI?
ICE，也就是移民和海关执法局，竟然也比人工智能更受欢迎，然后才是人工智能，接着是民主党，最后是伊朗。
ICE, that is immigration and customs enforcement, more likable than AI, then AI, then the Democratic Party, then Iran, end of list.
这真是相当
That's quite a
这确实是个问题。
That's that's a problem.
我的意思是，当移民和海关执法局比你更受欢迎时，这真是个有趣的挑战。
I mean, I when ICE is more favorable than you, that's an interesting that's a challenge.
是的。
Yeah.
我很高兴教皇利奥十四世依然受欢迎。
I mean, I'm glad the pope is still, pope Leo.
42%支持，8%反对，支持率高出34%。
42 positive, eight negative, plus ratio of 34.
教皇真棒。
Good for good for the Pope.
是的。
Yeah.
不。
No.
我的意思是，我确实认为，我没有任何保留意见，也没有任何一部分我不觉得人们对人工智能的感知是个大问题。
I mean, I do think I I I have no reservation or I there's no part of me that doesn't think the perception of AI is a huge problem.
我认为我们所有人将来都会面临许多潜在的挑战。
And I think there's a lot of underlying challenges that we're all gonna face.
但再说一遍，我仍然无法摆脱这样一个想法：我不知道该怎么说服自己，这其实更多是个公关问题。
But but again, I I still cannot move away from the idea that I don't know talk me out of the idea that it is as much it's a PR problem.
就像，我知道我们总回到这一点，但关键在于，谁是它的声音，谁是它的面孔，人们是如何谈论它的。
Like it's a and I know we always come back to that but it's it's who is the voice of it, who are the faces of it, how are people talking about it.
实际上，我的意思是，我不知道你有没有看到，比如多吉那边，有好些人参加了类似公开听证会之类的活动。
Actually there's just I mean, there's I don't know if you saw there's like with the Doge, a bunch of the people there are these like, I think public hearings or something.
是的，有这回事。
There's Yeah.
对，没错。
There yeah.
所以有视频记录了他们谈论如何处理整个事情，这些内容感觉就像仅仅一年前的噩梦。
So there's videos of them talking about how they had approached that whole thing that feels like a fever dream from only a year ago.
人们看待AI的方式是这样的，而不是像公众形象那样，试图以前所未有的规模去理解罕见疾病，或者如果讨论的是这些内容，而不是一个22岁的年轻人仅仅因为使用了ChatGPT或当时可能的Grok，就削减美国国际开发署的政府拨款。
That was how people perceive what AI is and does rather than if like the public face was let's try to understand rare disease in a much more scaled way that was never imaginable before or you know, like if those were the conversations rather than a 22 year old kid just cutting government funding to USAID just because like they use ChatGPT, like, or Grok maybe at the time.
但我觉得，人们正是这样看待这项技术的，而不是关注它的任何积极面。
But, like I think that is where people perceive the technology rather than any of the upside.
好的。
Okay.
先保留这个观点，因为我们接下来会在Nvidia部分谈到这一点。
So hold on to that idea because we are gonna get to that in the Nvidia section that's coming up.
而且我认为，我再读一些数据，我这么做并不是为了指责谁。
But and I think that, I'm gonna read some more numbers, and I'm doing this not to pile on.
我这么做是为了真正说明这个问题的严重程度，然后你和我将讨论其影响，因为我们俩都同意，这对AI来说是个大问题，如果情况不扭转，几乎我们讨论的所有事情都会受到影响。
I'm doing this to really illustrate the extent of the issue here, and then you and I are gonna talk about the implications because we both agree that this is a big problem for AI, and there are gonna be implications for everything we discuss if it doesn't turn around, pretty much.
好的。
Alright.
再来看一些数据。
Here's some more data.
这是来自YouGov的数据。
This is from YouGov.
他们表示，认为人工智能对社会的影响完全或主要是负面的美国人，是认为其影响完全或主要是正面的三倍。
They say three times as many Americans expect expect the effects of AI on society to be entirely or mostly negative as expect them to be entirely or mostly positive.
另有27%的人认为影响将是正面与负面相当。
Another 27% expect the effects to be equally positive and negative.
大多数没有亲自使用过人工智能的人认为，它对社会的影响将是完全或主要为负面的，其中包括62%的那些见过但从未亲自使用过它的人。
Most people who haven't used AI themselves expect it to be entirely or mostly negative to have entirely or mostly negative effects on society, including 62% of those who've seen it but have never used it themselves.
所以，基本上，认为人工智能会带来负面影响的人是认为其带来正面影响的人的三倍，而如果你没用过它，你更倾向于认为它对社会是负面的。
So, basically, it like, three x more people think that it's gonna be bad than good, and if you haven't used it, you think even more so that it is going to be negative on society.
我觉得他们是怎么定义‘没使用过’的，你觉得呢？
I think how do they define haven't used it, do you think?
我的意思是，我知道这可能很难界定，他们
I mean, I know it's probably gonna be a difficult They
问。
ask.
他们询问了人们。
They ask the people.
你明白了。
You got you got it.
你参与了调查。
And you're in the survey.
你可以说我经常使用它。
You can say I regularly use it.
我以前用过，但不经常用。
I've used it before, but don't regularly use it.
我见过别人用，但自己没用过，或者我从未用过，也从未见过任何人用。
I've seen it used, but haven't used it myself, or I've never used it and never seen anybody use it.
实际上，如果你见过它被使用但自己没用过，你的负面看法甚至比那些从未见过或从未使用过它的人更强烈。
Actually, if you've seen it used and you haven't used it yourself, you're actually even more negative than the people who have never seen it or anybody or haven't used it themselves.
但是
But
我的意思是，他们是否假设这是ChatGPT、Gemini或Claude？
that's what I mean though, that do you are they assuming it's ChatGPT or Gemini or Claude?
比如，他们是否假设这是任何其他利用AI的技术，比如他们在DoorDash订单中使用的，或者他们在其他地方用到的——我假设是这样。
Like, or are they assuming it's any of the other things that are leveraging AI that they're using in their DoorDash order or they're using, I mean, in all the other parts of their I'm I'm assuming
这肯定是大语言模型。
This it's is LLM.
这是
This is
毫无疑问是大语言模型、ChatGPT这一类的东西。
LLM ChatGPT stuff without a doubt.
不。
No.
我的意思是，假设他们直接通过聊天机器人与大语言模型互动。
I mean, assuming it's directly interacting with an LLM via a chatbot.
我觉得应该是对的。
Like, I have to imagine Correct.
对，没错。
And Correct.
也许吧，而且我并不是想淡化这个问题，因为我不仅意识到这对整个行业来说是一个巨大的认知挑战。
Maybe and again, I'm not trying to minimize this in any way because I not only recognize it is a it's a perceptual massive challenge for the industry.
围绕这个问题还有很多其他严重的问题。
There are lots of underlying issues around it as well that are highly problematic.
但我仍然认为，随着人工智能逐渐融入流程，更多地作为后台技术而非人们直接与ChatGPT交互，这种情况会变得越来越正常，人们也就不会对人工智能本身有直接的认知了。
I still think though, like, as maybe as AI starts to become more part of processes and like kind of behind the scenes rather than people interacting with ChatGPT, I think that starts to just become more normal or people don't don't have a direct perception of AI itself.
就像，他们仍然把人工智能和ChatGPT联系在一起。
Like, I think they're still associating with chat GPT.
好的。
Okay.
但你总是回到这一点，而这绝对不是我会关注的方向。
Well, you keep going back to this and it's definitely not where I would go.
所以我很想听你详细解释一下，为什么你一直强调底层流程的重要性。
So I'd love to hear you unpack this a little bit and talk about why you keep going back to the importance of the underlying processes.
再说一遍，这就像回到你进行谷歌搜索的时候，有趣的是，我真有一个最反对AI的人给我截了张图，是谷歌的AI概览，但他根本没意识到刚刚发生了什么。
Well, again, it's it's like going back to the when you do a Google search, it's funny I literally had someone who was like the most anti AI person I know screenshot me a Google AI overview not recognizing what had just happened.
你知道，他们这么做了，而谷歌某种程度上把这隐藏起来了。
Like, you know, like that they did it and Google's kind of hides.
现在这算不算一个AI概览呢？
Is this an AI overview at this point now?
它基本上已经完全融入产品之中了。
It's kind of like all relatively baked in the product.
现在当你在谷歌地图上搜索时，如果你问一个问题，它就会开始提供更丰富的结果——这周他们刚发布了这个功能。
When you're gonna do a search on Google Maps now and you ask something, it's gonna start it's gonna be more and they released this this week.
你可以提出叙述性的问题，搜索结果将由大语言模型驱动。
It's gonna you can ask narrative questions and it's gonna be LLM driven in terms of the search results.
我看到很多反应说，哦，他们把本该简单的东西搞复杂了。
And I saw a bunch of reactions like, oh, they're ruining, like simple things that should be simple but are actually gonna become more complicated.
一年后，每个人都会开始向谷歌地图提出更详细的问题，而不是简单地说‘纽约的泰国餐厅’，你会开始说：‘我想找离我一英里内最好的泰式炒河粉。’
Everyone a year from now is gonna start asking much more detailed questions of Google Maps rather than saying restaurant Thai New York, you're gonna start saying like, oh, I'm looking for, the best pad Thai within a mile for me.
而这将完全基于大语言模型，没有人会将其与所谓的AI联系起来。
And that's gonna be purely LLM based and no one's gonna be associating that with quote unquote AI.
这正是我想表达的意思。
That's what I'm trying to say here.
对。
Right.
我认为这与我们的数据有一点关系，数据显示，如果你经常使用它，你对它的态度会积极得多。
And I think that goes a little bit to our, data here that shows that if you use it frequently, you're actually a lot more positive about it.
如果你经常使用它，你的负面评价只有20.6%，而如果你只是见过别人使用但自己没用过，负面评价则高达62%。
If you regularly use it, your negatives are only 20, 6% compared to 62% if you've seen it used and you haven't used it yourself.
这很有趣。
So that is interesting.
所以，也许当它融入人们使用的产品中，人们从中受益后，就不会再那么强烈地反感它了。
So maybe as it becomes part of products that people use and they get benefit out of it, they, you know, become less overwhelmingly negative about it.
我想问你，你认为有什么类似的技术吗？
I'm I'm gonna ask you, what would be a comparable technology, do you think?
你
Do you
觉得社交媒体作为一个整体，属于这种情况，是可比的吗？
think social media as a general thing falls into this, is comparable?
我不记得在2012年的时候，有人会说云计算会摧毁社会的根基之类的话，那这里的类比到底是什么？
I I don't think I never heard anyone in, 2012 being, like, cloud computing is gonna destroy the fabric of society or anything like that, but what would be the comp here?
没有，我的意思是，我不是那种喜欢说‘这是前所未有的时刻’之类话的人，但那样就落入了营销的套路；但在我看来，真的没有可比的。
There's no I mean, I'm not a guy that that wants to be like, it's an unprecedented moment blah blah blah, but then you play into the marketing, but to me there's no comp.
真的没有可比的。
There's no comp.
电力。
Electricity.
电力。
Electricity.
不是。
No.
也许是电力或火。
Maybe electricity or fire.
别把卢德分子扯进来，不过他们确实挺不错的。
Never the Luddites not to be bring Luddites into the conversation, but Oh, they're very nice.
没有任何东西？
Nothing?
没有。
No.
没有。
No.
因为工业革命是唯一可比的事情。
Because industrial revolution is the only comparable thing.
就像那些亲历它的人一样，我不认为这达到了工业革命的规模。
Like people who saw it I don't necessarily think this is on the scale of the industrial revolution.
我只是觉得这是个类比，因为我相信当时一定也有织工看到织布厂、工厂，我不知道那叫什么。
I just think that's the analog because I'm sure you had the weavers who saw the weaving plants, the mills, I don't know what the name is.
织布机。
The looms.
但你知道，我认为人们用织布机已经有一段时间了。
But, you know, the I think people had looms for a while.
我不确定。
I don't know.
可能我错了。
Could be wrong.
我得去读一读。
I got to read it.
你没资格讨论这个。
You're not qualified to discuss.
但总之，好吧，让我用一种我觉得我们都能理解的方式来表达。
But basically, all right, let me just put it in terms that I think we could understand.
如果你之前是手工完成工作，突然看到一座工厂出现在你村子外，你肯定会想：天哪，我可能要失业了。
If you were doing a processes pretty manually and you saw a factory show up outside your village, you were like, Oh God, I'm probably gonna be out of a job.
这种感觉是一样的，你以前从未有过这种感觉。
That's the same feeling, you've never had that feeling outside of now.
所以，也许我喜欢我们继续深入探讨工业革命这个类比，尽管我们已经证明自己对这个话题有些外行。
Well, so maybe I like that we're gonna dig into the industrial revolution analogy even though we've already shown ourselves to be somewhat incompetent on the topic.
但完全是外行。
But Fully incompetent on the topic.
完全是外行。
Fully incompetent.
但我认为
But then I think
老兄，顺便说一句，织布机可以追溯到公元前6000年。
Dude, the loom the loom, by the way, it goes back to 6,000 BCE.
所以
So
好吧。
Alright.
好的，没事。
There okay.
所以问题不在织布机。
So it's not the loom.
自动化指的是周围存在某种东西。
Automated is the fact There's something around that.
所以我认为是产业演进，那我们就从这个角度来谈。
So I think industrial evolution, so then let's take that one.
显然，这引发了大量恐惧和担忧，并在近期和短期内带来了切实的问题。
Clearly a great deal of fear and apprehension, genuine problems resulting in the near and short term from it.
但随着时间推移，它逐渐融入日常生活，以至于没人再谈论或思考它了。
And then over time so integrated into everyday life that, you know, like, there's no one is talking about it, thinking about it.
它只是事物运作的方式而已。
It's just how things work.
你是这样看待这件事的发展的吗？
Is that how you see this playing out?
或者你觉得，你能想象一个世界，其中它不再存在吗？
Or do you think could you see a world where it's no.
LLM被禁止了，或者类似的情况。
LLMs are banned or something like that.
它们不会被禁止。
They won't be banned.
我的意思是，你怎么可能禁止它们呢？
I mean, how can you ban them?
你是打算让政府去你办公室把你下载的DeepSeek版本从你的Mac Mini里没收，然后说‘好了，你得去蹲监狱了’吗？
Are you gonna go and take your Is the government gonna go grab your Mac Mini out of your office where you've downloaded a version of DeepSeek and be like, alright, we're right to the pokey.
但我觉得，再想想，它会不会最终成为社会大多数运作的方式？
Well, But I but but do you think, again, it'll if thing it just becomes how most of the society works.
再者，制造业会变成这样，然后接下来是这个技术的下一个阶段和迭代。
Again, manufacturing becomes a thing, like, whatever else is the next phase and iteration of this.
对我来说，这实际上是一个很好的类比。
Like, to me, that's that actually seems like a good analogy within this.
是的，顺便说一下，我对大语言模型对就业的影响并不持普遍负面看法。
Yeah, and I'm, by the way, I'm not broadly negative on the impacts of LLMs on jobs.
我只是不认为我们必然会经历类似的阵痛。
I just don't think it we're, you know, it's inevitability that we're gonna see similar pain.
当然会有一些阵痛，但说大语言模型会导致大规模失业，我仍然不太认同，不过我也愿意接受这种可能性。
We will, there'll be some pain without a doubt, but this idea that there's gonna be mass employment unemployment because of it, you know, I'm still not fully bought into again, but my mind is open to the fact that maybe that is the case.
顺便说一下，回到工厂这个例子，那数据中心呢？
And by the way, you know, if we go back going to this factory thing, what about data centers?
这是另一个对数据中心持非常负面看法的调查。
This is another poll, very negative on data centers.
皮尤研究中心关于美国人如何看待数据中心对环境和就业等关键领域的影响。
Pew, how Americans view data centers impacting key areas from the environment to jobs.
四分之三的美国人表示，他们听说过或读到过很多或一些关于数据中心的信息。
Three quarters of Americans say they've heard or read a lot, a lot or a little about data centers.
他们读过关于数据中心的内容。
They've read about data centers.
更多的美国人认为，数据中心对环境、家庭能源成本以及附近居民的生活质量产生了负面影响，而非正面影响。
More Americans say data centers have a negative effect on the environment, home energy costs, and people's quality of life nearby than they have a positive effect.
所以我们也应该意识到，随着这些可能替代你工作的AI模型变得越来越强大，我们不能忽视一个事实：建造这些模型的公司必须建设庞大的数据中心，而这些中心雇佣的人很少，却可能推高你的能源价格，甚至损害你的健康。
So we're also like, we shouldn't be, you know, as these these AI models that could potentially do your work get better, we shouldn't be blind to the fact that like the companies have to build that are building them have to build these huge data centers and they don't employ very many people and they could drive your energy prices up and they could harm your health.
这就是公众对这些问题的看法。
So that's sort of where the public is seeing this stuff.
所以，如果我们把数据中心看作这场公关战中的一个具体部分，那么可以说，这个行业并没有赢得这场战役。正如你所说，当任何人——包括我自己——在讨论这个问题时，当前的扩张模式、近期算力的预测、以及谁从中受益，比如Stargate项目、Masayoshi Son和Larry Ellison这些人，都没有人能清晰地说明实际的好处在哪里，反而要占用你的水资源、抬高你的电费。
And so I actually I guess if we take data centers as like one specific part of this PR battle, it I mean, the industry is not winning this battle because as you said, like, when anyone and even for myself and we've we've debated this, is the current model of expansion and like the the forecast for near term compute and who benefits from it with the Stargate, Project Stargate and Masa's son and Larry Ellison and all these others like none of that and then we're gonna take your water and then we're gonna raise your electricity prices and no one can clearly like elucidate what the actual benefits are.
我的意思是，整个数据中心的部署与规划，可能是这场争论中行业最糟糕的一环。
I mean, I'll give this is the whole data center rollout and planning is probably the single worst, part of this whole debate for the industry.
没错。
Right.
那你认为这会带来什么影响？
And so what do you think the implications are?
这正是这些民调数据如此重要、而我们花节目前半部分讨论它的原因，因为这些数字不是孤立存在的，它们会带来实际后果。
Mean, this is the reason why these poll numbers matter and why we're spending the first half of the show on this, because these numbers don't exist in a vacuum, they have consequences.
那么，你认为会发生什么？这可能是政治上的，也可能是经济上的，鉴于人们对人工智能如此负面的看法，你觉得接下来会发生什么？
And so what do you think and it could be political, it could be economic, what do you think is gonna happen now as a result of such negative feelings about AI?
至少在
At least
美国？
in The US?
等一下。
Hold on.
我来描述一个可能同时反映近期这种反弹的场景。
I'm gonna here here is a scenario that could both incorporate kind of the near term backlash around this.
想象一个世界，在那里，由于公众的强烈反对，数据中心根本无法进入社区——这种情况如今正在发生。
Imagine a world where and and we've seen this of data centers not actually being allowed into communities and based on public backlash and that's happening today.
那么，也许不会再像现在这样，仅仅依靠巨额资金来快速建设这些数据中心，以应对对算力需求的极度乐观预期。
Then maybe instead of the kind of like current mode of just massive funding in order to rapidly build these data centers in order to kind of like these like based on like very aggressive anticipation on compute needs.
也许它们的建设速度会变慢，这反而会迫使行业真正找到更高效利用算力的方式，来提供具身人工智能，也许我们会看到更小的模型、开源模型，以及真正推动这一领域创新的其他方式。
Maybe they're not built as quickly and maybe that forces the industry to actually figure out much more compute efficient ways of like delivering agentic AI and maybe we get much smaller models and open source models and just things that actually force that innovation around like how this all plays out.
所以我认为，当前的政治格局有可能推动创新和科技的发展方向，这是一个例子。
So I think there is a that is like one example of where kind of the current political landscape could actually drive where the innovation goes and where the technology goes.
这不一定是负面的，只是方式不同。
Not in necessarily a negative way in just it's a different way.
我的意思是，我认为这在任何重大的技术革命中都会发生。
And I mean, I think that probably I mean, that happens in any kind of large technological revolution.
说得好。
Great point.
所以我认为，我们都会同意，AI目前正面临政治上的困境。
And so I think we would both agree that AI is in political trouble right now.
这时，英伟达的首席执行官黄仁勋登场了。
And, and in comes, Jensen Huang, CEO of Nvidia.
英伟达向来对他们在GTC大会——即他们的旗舰会议——上将要发布的内容保持沉默，而这场大会即将在湾区举行，或者如果你在YouTube上观看这段视频，那就在3月16日这一周。但奇怪的是，黄仁勋发布了一篇罕见的博客文章，他称之为《AI是五层蛋糕》。别被标题迷惑，这五层蛋糕大致是：从能源开始，用能源驱动芯片，建造容纳芯片的基础设施，利用这些基础设施构建模型，再用这些模型开发应用。
Now, Nvidia is famously quiet about what they're gonna reveal at GTC, is their flagship conference, which is happening, in the Bay Area in the forthcoming week or if you're watching this on YouTube this week on the week of the March 16 and weirdly or interestingly Jensen has, released a rare blog post that he's authored and he calls it AI is a five layer cake, don't get too distracted by the title, the five layer cake is basically you begin with energy, you use the energy to power chips, you build infrastructure to house the chips, you use that infrastructure to build models, you use those models to build applications.
但其实这个博客文章的标题很糟糕，因为他真正想做的，似乎是试图为整个国家、乃至全世界凝聚起对AI前景的信念。
But actually, that's a terrible title for the blog post because what he's really doing in this blog post is seemingly trying to rally the country, the world around the promise of AI.
我认为，理解它存在这种沟通问题或认知问题很重要。
I think understanding the fact that it has this messaging problem or this perception problem.
他说，我们才刚刚开始这一建设。
He goes, we have only just begun this build out.
我们目前已经投入了数千亿美元。
We are a few $100,000,000,000 into it.
仍需建设数万亿美元的基础设施。
Trillions of dollars of infrastructure still need to be built.
顺便说一下，支持这一建设所需的劳动力是巨大的。
The labor required to support this build out is enormous, by the way.
听我念一段。
Just listen as I read this.
这几乎与我们之前提到的所有内容都吻合。
It matches almost all the things we brought up previously.
AI工厂需要电工、水管工、管道工、钢铁工人、网络技术人员、安装人员和操作员。
AI factories need electricians, plumbers, pipe fitters, steel workers, network technicians, installers, and operators.
这些是技术性强、薪酬丰厚的工作。
These are skilled, well paid jobs.
这类人才供不应求。
They are in short supply.
你不需要拥有计算机科学博士学位就能参与这场变革。
You don't need a PhD in computer science to participate in the transformation.
同时，人工智能正在推动知识经济领域的生产力提升。
At the same time, AI is driving productivity across the knowledge economy.
以放射学为例。
Consider radiology.
人工智能现在协助解读影像扫描，但对放射科医生的需求仍在增长。
AI now assists in reading scans, but demand for radiologists continue to grow.
这并不是一个悖论。
This is not a paradox.
放射科医生的职责是照顾病人。
The radiologist's purpose is to care for patients.
展开剩余字幕（还有 367 条）
读取扫描图像只是其中一项任务。
Reading scans is one task along the way.
当人工智能承担更多常规工作时，放射科医生可以专注于判断、沟通和护理。
When AI takes on more of the routine work, radiologists can focus on judgment, communication, and care.
医院变得更加高效，服务更多患者，雇佣更多人员。
Hospitals become more productive, they serve more patients, they hire more people.
生产力创造能力，能力带来增长。
Productivity creates capacity, capacity creates growth.
你不觉得詹森只是坐在房间里，看了所有这些民调数据，然后说‘天哪，我们出问题了，这是我今年最重要的活动，我得做点什么’，这就是主题，这就是演讲内容。
Don't you think Jensen just sat in a room looked at all these polling numbers and said oh shit, we have a problem, it's my biggest event of the year, I need to do something about it and that is the theme and that is the speech.
我觉得这正是整个团队的初衷。
I feel that is exactly what the entire team came out.
而且，当你继续看下去时，我也讨厌这个标题。
And again, like as you go through, I also hate the title.
我不知道为什么叫‘五层蛋糕’，就是让我感觉不对劲，但我觉得当我们继续看的时候
I don't know why Five Layer Cake is just it doesn't land with me, but I think like as we're going
因为我们俩都认为蛋糕应该是七层。
Because we both agree cake should be seven layers.
层。
Layers.
五层完全让人不满意。
Five layers is wholly unsatisfying.
不。
No.
至少做成七层，这个就很简单了。
At least make it seven gens and this one is pretty straightforward.
但我认为是芯片、基础设施、应用、空间。
But I think Chips, infrastructure, applications, space.
空间AGA。
Space AGA.
机器人。
Robotics.
他忘了机器人技术。
He forgot robotics.
机器人技术。
Robotics.
对。
Right.
不。
No.
但是，我的意思是，好吧。
But but it actually I mean, okay.
他基本上是在重复我在这档播客中会说的话。
He is basically echoing and anticipated what I would be saying on this podcast.
开源变成了一种创新。
I, like open source becomes an innovation.
它创造了新的工作类型。
It creates new types of jobs.
它再次改变了，医院到底是什么？
It it changes again, what is a hospital?
比如，你提升了护理的质量和规模？
Like, you increase the quality of care, the scale of care?
我的意思是，你能改变医疗保险，还有与医疗系统互动的体验吗？所有这些都可能成为现实。
Like, can you, I mean, change health insurance and, you know, like the experience of it just interacting with the healthcare system, all these things start to become possible potentially.
于是这就成了整个故事。
So that became the story ever.
这真的可能很重要。
It really could be important.
现在，詹森，你觉得他有主流的信誉吗？能成为那个穿着皮夹克的人吗？
Now is Jensen do you think he's got the like main street cred or can be the guy with his with the leather jacket on?
你觉得他能成为那个人吗？还是你觉得，比如
And do do you think he can be the guy or do you think, like
是的。
Yes.
我同意。
I do.
在韩国和台湾吃炸鸡、喝啤酒。
Eating fried chicken and drinking beer in Korea and Taiwan.
是的。
Yeah.
没错。
Exactly.
他给人的感觉是，尽管他是地球上最富有的人之一，却像普通人一样，他没有那种高高在上的姿态。当然，你经营这样的公司，肯定有自尊心，但他很谦逊。
He comes off in a way, even though he's one of the richest men on the planet, kind of as in every man, he doesn't have these he's clearly he has an ego, I mean, you're running a company like that, you have an ego, but he's humble.
他做的每一件事，并不是一开始都立刻成功了。
Not everything he did, you know, in a way, not everything he did turned to gold right away.
他花了十几年甚至更久的时间，才等到成果显现。
He got to sit with this company for a couple decades or decade plus before the fruits came out.
你说得对，他确实会喝啤酒、吃炸鸡，也会在台湾路边接受记者的采访，从不居高临下。
You're right, he does do the thing, he drinks the beer, he eats the chicken, he takes the questions from the reporter on the side of the road in Taiwan without condescending to them.
他能做到。
He could do it.
你知道吗？我要为外面的团队做一个公关建议。
Do you know my only I'm gonna make a PR recommendation for the team out there.
事实上，所有这些都逐渐拼凑起来，让他成为那个人选。
I actually, all of this is really kind of putting coming together that he could be the guy.
我唯一做的是，你看到没有？他总是公开谈论自己的工作 ethic，从不休假，整天回复邮件，满脑子想的都是工作。
The only thing I did, do you see like he has all this stuff that he will openly talk about his work ethic and he like never takes vacation and never respond like will respond to email all day long or like all he thinks about his work.
我认为世界真正需要的是这样一个人：能展示出人工智能如何帮助他更好地平衡生活。
I think what the world needs here is actually someone who can show that AI has actually helped them balance their life a bit more.
我认为他需要稍微调整一下，展示他度假时的样子，让他的团队替他工作，和家人共度时光。
I think they he's gotta just shift that a little bit, show him on vacation a little bit as his agents are doing his work, spending time with his family and showing.
这才是人们想要的愿景。
That's the vision people want.
杰森，你就是那个人，你可以成为那个人。
And Jensen, you could be the guy that you could be that guy.
好的。
Okay.
我在这儿激动得不得了，因为这正是我本来要提的第二点。
I'm I'm I'm I'm going wild over here because that was the second point I was gonna make.
目前，除了我们即将有三万人之外，
There is currently outside of the, like, we're gonna have 30,000 people.
在活动的实时博客上只有一篇帖子。
One post on the events live blog.
只有一篇帖子。
One post.
那是什么？
What is it?
在GTC公园搭建一个机械臂。
Build a claw at GTC Park.
它鼓励与会者前来，构建一个开放式代理，部署一个主动、全天候的AI助手。
It is encouraging attendees to stop by and build a an open cloth style agent to deploy a proactive, always on AI assistant.
这些全天候的AI助手可以应用于几乎任何任务，包括管理日历、推荐度假目的地、建议新的锻炼计划，以及编写有用的应用程序。
These always on AI assistants can be applied to virtually any task, including managing a calendar, suggesting vacation destinations, recommending new workout routines, and coding a useful app.
它会持续学习新技能，并主动引导用户关注新的发现。
IT continually learns new skills and is directed toward and to prompt the user with new findings.
这可能是第二个要点。
That's probably message number two.
这不仅很棒，还能让你有更多时间去做你热爱的事情。我，詹森，一个非常忙碌的CEO，发现这些开源代理甚至能帮我规划假期。
Not only is this great, it's gonna free you up to do more of the things you love, And I, Jensen, a very busy CEO, have found these open source or these agents to be able to even let me plan a vacation.
这就是核心信息。
That's the message.
这就是核心信息。
That's the message.
就这些。
That's it.
而且说实话，我喜欢他们主打‘构建爪子’这个想法，因为我觉得让Open Claw走红的关键在于它很有趣，真正让人感觉更有掌控感。
And and honestly, I kinda like they're leaning into the build a claw idea because like, I think what really made Open Claw such a viral thing is it was fun and it actually like just made people feel more in control.
而且再说一遍，我知道不仅有无数的梗，还有《哈佛商业评论》的文章在讨论，你越是真正地用AI构建东西，越是构建代理流程，你实际上要做的工作就越多。
And again, I know there's like endless not only memes but Harvard Business Review articles around like the more you actually build with AI and build like, build agentic processes, the more work you do.
但我认为，如果整个行业能团结起来，展示并把詹森推到前面，说他一生中从未休过假。
But I think that that's if the industry can just come together and just show and put Jensen up there and say, he never took a vacation in his life.
他出身卑微。
He started from humble beginnings.
他不知疲倦地工作，才达到今天的地位，而现在多亏了AI，他能更多地坐在沙滩上了。
He worked tirelessly to get to where he is and now thanks to AI, he's sitting on the beach just a little more.
多花点时间待在家里，放松一下，看看奈飞，这些他过去根本不可能做到的事情。
Just spending a little time at home, just relaxing, watching some Netflix, things he never would have been able to do in the past.
你可以想象他穿着皮革夹克，躺在海滩椅上。
You could just see him on the beat in the beach chair With the leather jacket.
泳装，再加上你那身全套皮革夹克的装扮。
Bathing suit and, yeah, your full leather jacket getup.
那样就完美了。
That could do it.
好的。
Alright.
我们明白了。
We got it.
我们明白了，是的。
We got yeah.
说吧。
Go ahead.
你觉得Jensen就是那个人吗？
You think is Jensen the guy?
我觉得他最有可能是。
I think he's most likely
他可能是。
He could be.
他最有可能是。
He is most likely.
候选人。
Candidate.
是的。
Yeah.
Jensen传达这条信息的最好方式是登上大型科技平台。
The best way for Jensen to get this message across would be to come on big technology.
很简单。
So Easy.
Jensen团队，如果你们在听，一定要这么做。
Jensen team, if you're listening, definitely do that.
好的。
Alright.
我们得休息一下。
We gotta take a break.
拉詹，清空你的日程。
Ranjan, clear your calendar.
在我们最后的二十分钟里，有太多话题要聊了。
We have so much to talk about here in our last twenty minutes.
我们将讨论实际上发生了什么——让AI代理包办一切的负面影响，亚马逊和麦肯锡正逐渐意识到这一点，然后我们会谈谈Meta在AI进展上的缓慢，或者说缺乏进展。
We're gonna talk about, actually what's happened, the downside of letting the AI agents do all the work, which Amazon and McKinsey are finding out, and then we will talk about this slow moving AI progress or lack thereof at Meta.
在我们休息之前，Harness Hive，顺便说一句，我注意到，Ranjan，我得告诉你，我们的听众自豪地称自己为‘Harness Hive’，我必须说，这让我非常开心，谢谢Harness Hive的到来。
Before we go to break, harness hive, and by the way, I've noticed, Ranjan, worth telling you, our listeners are proudly calling themselves the harness hive and I have to say it just makes me so happy so thank you harness hive for being here
Harness Hive，代表我们！
harness hive represent
首先，如果你能在Apple Podcasts或Spotify上给节目打五颗星，那就太好了。
represent a couple of things for you first of all if you could rate the the show five stars on Apple Podcasts or Spotify that would be great.
我们即将在Apple Podcasts上达到500条评价或评分，这有助于提升我们的公信力，也能向Nvidia这样的团队展示我们的影响力，让他们更愿意把Jensen请上节目。
We're coming up on 500 reviews or ratings on Apple Podcasts and of course this helps us build credibility and show to teams like the one at Nvidia trying to get Jensen on the show.
所以，如果你能这么做，那就太好了。
So if you could do that, that would be great.
另外，接下来我简单剧透一下，Andrew Ross Sorkin即将做客我们的节目。
Also, upcoming, just gonna give this a quick tease, Andrew Ross Sorkin is gonna be on the show.
我们将讨论人工智能劳动力，如果AI能完成工作，软件会发生什么变化；我们还会谈私人信贷危机、SpaceX上市，还有周三即将上演的许多精彩内容。
We're gonna talk about AI labor, what happens to software if AI works, we're gonna talk about the private credit crisis, SpaceX IPO, lots of fun stuff that's coming up on Wednesday.
好的。
Alright.
广告后马上回来继续。
More when we come back right after this.
我很高兴向大家介绍Notion——一个为团队打造的AI驱动的协同工作空间，将你的笔记、文档和项目全部整合到一个平台上。
I'm excited to tell you about Notion, an AI powered connected workspace for teams, bringing all your notes, docs, and projects into one space.
它无缝流畅、功能强大，而且使用起来真的很有乐趣。
It's seamless, powerful, and actually fun to use.
现在，借助Notion全新的自定义智能代理，过去需要数小时才能完成的琐碎工作，现在都能自动运行了。
And now with Notion's new custom agents, the busy work that used to take hours runs itself.
每天早上，我都会有一个自定义代理，为我筛选出最重要的科技新闻。
Every morning, I have a custom agent that curates the most important tech headlines for me.
它会扫描我保存的资讯来源，总结关键进展，并将所有内容整理成一份清晰的简报，直接放在Notion里。
It scans my saved sources, summarizes the key developments, and organizes everything into a clean briefing inside Notion.
所以我不再在Slack线程之间来回切换，而是打开一个页面，就能立刻知道错过了什么。
So instead of bouncing between Slack threads, I open one page and immediately know what I may have missed.
别光听我说。
Don't just take my word for it.
Notion被超过50%的《财富》500强公司以及OpenAI和Ramp等一些增长最快的公司使用。
Notion is used by over 50% Fortune 500 companies and some of the fastest growing companies like OpenAI and Ramp.
立即前往notion.com/bigtech体验自定义代理。
Try custom agents now at notion.com/bigtech.
全部是小写字母：notion.com/bigtech，今天就来体验自定义代理。
That's all lowercase letters, notion.com/bigtech to try custom agents today.
当你使用我们的链接时，你就是在支持我们的节目，notion.com/bigtech。
And when you use our link, you're supporting our show, notion.com/bigtech.
你的IT团队浪费了一半的工作时间在重复性任务上。
Your IT team wastes half its day on repetitive tasks.
随着你的业务增长，请求会越来越多。
And the more your business grows, the more requests pile up.
密码重置、访问请求、入职流程，这些都在分散他们做有意义工作的精力。
Password resets, access requests, onboarding, all pulling them away from meaningful work.
使用 Servo，你可以减少 80% 的帮助台工单。
With Servo, you can cut 80% of your help desk tickets.
来看看这一转变。
Here's the transformation.
过去，经理为新员工入职时，旧流程要花上数小时。
When a manager used to onboard a new hire, that old process took hours.
他们得在 Slack 上发消息、给 IT 发邮件、等待审批，而新员工则只能干等好几天。
They'd ping Slack, email IT, wait on approvals, and meanwhile, the new hire would sit around for days.
使用 Servo 后，经理只需在 Slack 上提出为新员工入职的请求。
With Servo, the manager asked to onboard a new hire in Slack.
AI 会在数秒内自动完成所有权限配置，并完成必要审批，IT 部门完全无需介入。
The AI provisions access to everything automatically in seconds with the necessary approvals, and IT IT never touches it.
如果我需要这样一款产品，这正是我会选用的。
If I needed a product like this one, it's exactly what I'd use.
在您免费试用的第四周，即可保证实现50%的客服工单自动化。
It guarantees 50% help desk automation by week four of your free pilot.
Servo 为全球增长最快的企业提供支持，例如 Perplexity、Mercor、Vercada 和 Clay。
Servo powers the fastest growing companies in the world like Perplexity, Mercor, Vercada, and Clay.
让您的团队摆脱客服工作，回归他们真正热爱的事务。
Get your team out of the help desk and back to the work they enjoy.
立即前往 serval.com/bigtech 预约您的免费试用。
Book your free pilot at serval.com/bigtech.
那就是 serval.com/bigtech。
That's serval.com/bigtech.
开始一件事并不只是困难。
Starting something new isn't just hard.
它令人恐惧。
It's terrifying.
你为这件事投入了如此多的努力，却仍不确定它是否能成功，要迈出这一步确实很难。
So much work goes into this thing that you're not entirely sure will work out, and it can be hard to make that leap of faith.
当我刚开始做这个播客时，我不确定是否有人会听。
When I started this podcast, I wasn't sure if anyone would listen.
现在我知道这是个正确的选择。
Now I know it was the right choice.
当你有Shopify这样的合作伙伴支持你时，这也会更有帮助。
It also helps when you have a partner like Shopify on your side to help.
Shopify是全球数百万企业的电商平台，占美国所有电子商务的10%，从Allbirds和Cotopaxi这样的知名品牌，到刚刚起步的新锐品牌都在使用。
Shopify is the commerce platform behind millions of businesses around the world and 10% of all ecommerce in The US, from household names like Allbirds and Cotopaxi to brands just getting started.
凭借上百个即用模板，Shopify能帮助你打造一个与品牌风格一致的精美在线商店。
With 100 of ready to use templates, Shopify helps you build a beautiful online store that matches your brand style.
让你的宣传效果仿佛背后有个营销团队在运作。
Get the word out like you have a marketing team behind you.
轻松创建电子邮件和社交媒体活动，覆盖你客户浏览或闲逛的任何平台。
Easily create email and social campaigns wherever your customers are scrolling or strolling.
是时候用Shopify将那些‘如果’变成现实了。
It's time to turn those what ifs into with Shopify today.
今天就前往 shopify.com/bigtech 注册您的每月1美元试用版。
Sign up for your $1 per month trial today at shopify.com/bigtech.
前往 Shopify 点 com 斜杠 big tech。
Go to Shopify dot com slash big tech.
就是 shopify.com/bigtech。
That's shopify.com/bigtech.
我们回到《大科技》播客周五版。
And we're back here on Big Technology podcast Friday edition.
在广告中断前，我们正在讨论詹森·黄在海滩上出镜、展示 Open Cloud 工作的广告宣传是否合适。
So, before the break, we were talking about the advisability of Jensen Huang running an advertising, running an advertising campaign, showing him on the beach as Open Cloud does his work.
我可以肯定，你不会在亚马逊看到类似这样的广告出现。
Well, I can guarantee you're not gonna see an ad like that come out of Amazon anytime soon.
这是来自《金融时报》的报道。
This is from the Financial Times.
亚马逊在人工智能相关服务中断后召开了工程会议。
Amazon holds engineering meeting following AI related outages.
亚马逊的电子商务业务已召集大量工程师于周二召开会议，深入分析一系列服务中断事件，其中包括与使用AI编码工具相关的事故。
Amazon's ecommerce business has summoned a large group of engineers to a meeting on Tuesday for a deep dive into a spate of outages, including incidents tied to the use of AI coding tools.
亚马逊表示，近几个月来出现了一种趋势，即事故的波及范围广泛，且涉及生成式AI助手的更改等因素。
The AI the online retail giant has said there has been a trend of incidents in recent months characterized by a high blast radius and Gen AI assistant changes among other factors.
在促成因素中，该报告提到，使用生成式AI的新方式尚未建立完善的最佳实践和安全措施。
Under contributing factors, the note included, novel Gen AI usage for which best practices and safeguards have not yet been fully established.
我不是拉詹。
I don't Ranjan.
我的理解是，亚马逊对员工使用AI工具有明确的强制要求。
My read here is basically that Amazon has these mandates.
他们确实强制要求员工使用这些AI工具。
They do have the mandates for people to use the AI tools.
员工使用了AI工具，结果系统就出问题了。
People use the AI tools, and the stuff is breaking.
你对这件事怎么看？
What do you think about this?
好的。
Okay.
所以上周，如果你听了我们的节目，可能会记得我提到过，亚马逊宕机了好几个小时，我当时去购物，看到一些新闻标题，但相关报道却非常少。
So last week, if you listen to our episode, you'll you may remember I brought up it was odd that Amazon was down for hours and I'd gone on, tried to shop, was saw some headlines, but there's very little coverage.
所以这个事件一下子引起了我的注意。
And so this one jumped out at me in a big way.
我认为这再次揭示了当前AI宣传方式的问题：一方面，像杰克·多西和Block公司几周前（或者可能就是上周）还在说因为AI导致了40%的裁员。
And I think this actually again also captures the what's wrong with the way a overall AI messaging because you have on one side like Jack Dorsey and Block a few weeks ago or maybe that was only last week talking about 40% layoffs because of AI.
而另一方面，又有人认为AI会直接取代人力，导致大规模裁员；但你也会看到，如果工程师只是快速用AI生成代码，或者采取一种漫不经心的态度，而公司高层又强制推行，就可能带来负面后果。我认为这其实是一个非常重要的故事，因为像我认识的每一个工程师都在使用某种代码生成工具。
And then this whole idea like that people they're just gonna lay people off because AI is gonna do the work and then you do see this that if you are quickly vibe coding or not taking a more lackadaisical approach and being forced to buy top management, it can have negative consequences and I think this is actually a really important story because like everyone I know on the engineering side is using like I mean, some kind of code gen tool.
毫无疑问，这些工具效率极高，但公司必须认真思考如何部署它们。
There's no doubt that it is incredibly efficient but this is the companies are gonna have to not, like, they're gonna have to be thoughtful about how they deploy these things.
不能只是放任或强迫所有人快速采用，而应该真正帮助他们理解如何正确使用这些工具，这一点变得至关重要，我们肯定会看到更多类似的情况。
And like just letting and almost forcing everyone to rapidly adopt versus actually making them understand how to use these tools, I think becomes so much more important and we're definitely gonna see more instances like this.
是的。
Yes.
所以，目前许多公司都在推行强制使用AI的政策，不仅仅是亚马逊。
So the broad mandates of everybody needs to use AI, we're seeing at many companies, not just Amazon.
这没问题，但必须要有相应的指导方针，或者
That's fine, but there does there have to be some guidance around it or
不，我的意思是，这根本说不通。
like needs How to to No, mean this.
是的。
Yes.
何时以及如何使用，成为最关键的问题。
When and how become the single most important questions.
而且，实际上，为大型企业部署AI正是我的工作。
And again, this is like actually deploying AI at large enterprises is my job.
我们亲身体验到，教育环节的重要性远胜于盲目强制推行，而不考虑任何实际因素。
And like, so we see firsthand and like that's where the like the education part of it becomes so important versus just like forcing this through without any kind of thought around.
即使不是针对个人层面，也要思考AI如何融入更大的流程和工作流，这才是应该讨论的重点，而且亚马逊这样确实不太好看。
Even and even if it's not at the individual level, like how does this look like actually in terms of larger processes and workflows like that that's the stuff that should be the conversation and I mean it's not a good look for Amazon.
我其实很惊讶这件事被泄露了，因为这对亚马逊来说实在是太糟糕了。
I was actually surprised that this got leaked or that this actually because this this is a very, very bad look for Amazon.
我的意思是，我们已经对此做了一些报道。
Well, mean, I think we've we've done some reporting on it.
我不认为公司内部的人对被告知的方式感到高兴。
I don't think people within the company are thrilled about the ways that they've been told.
我的意思是，当然，不同团队情况不同，但我不确定他们是否对被要求使用AI的方式感到满意，就像我们报道过的，AI被用于六页纸报告，这挺有意思的。
Not I mean, it's obviously, you know, group by group, but I don't know if they've been thrilled by the ways that they've been, you know, instructed to use AI, as we've reported on the AI being used for the six pagers, which is which is interesting.
但有一件事我很好奇，你觉得这件事应该怎么处理？
But that's actually one thing I'm curious, like, how do you how how do you think it should be done?
我想应该是这样的。
And I I imagine it is.
这对管理者和高管来说是个艰难的处境，因为AI确实能显著改善你的生活，让詹森终于能穿着皮夹克坐在海滩上，同时大幅提升你的工作效率。
It's a tough situation for managers and executives to be in because it's like this will and can dramatically improve your life and allows Jensen to finally sit on the beach in his leather jacket and also make you much more productive.
这种自上而下的强制要求，确实会加速AI的普及。
Like, it's there, then kind of top down mandate, you know, will accelerate adoption.
但如果人们对此并不感到高兴，或者没有正确理解，那也是一个问题。
But, like, if it's meaning people aren't either being happy about it or understanding it properly, that's also a problem.
作为亚马逊的高管，你该如何推行这项举措？
You as executive of Amazon, how do you, how do you roll this out?
作为一名管理着这个庞大媒体帝国的人，你们都听过我们这里所有人的说法。
Well, as someone who manages this, you know, large media empire with and you've all heard from all of our people here.
我只是开个玩笑。
No, I'm just kidding.
我们在大型科技领域运作得非常精简，但我会这么做，我觉得，我最近刚做了一期非常有趣的播客，很快就会发布，采访的是Canva的产品负责人卡梅隆·亚当斯。
We've we run a lean operation on big technology, but here's how I would do it, you know, and I think, you know, I I just did a very interesting podcast which is gonna come out sometime soon with Cameron Adams, who's the head of product at Canva.
他提到，他们会向全公司展示那些正确且高效使用AI的实例。
And one of the things that he mentioned was that they have they hold up examples of people using AI to the right the right way and for productive use to the entire company.
所以我认为，这才是正确的做法。
So I think that really is the way to to do it.
而不是强制规定10%、15%或20%的任务必须由AI完成，我认为真正的领导者需要明白，这项技术目前是由组织内部的热情采纳者和倡导者推动的。
As opposed to forcing 10 or 15 or 20% of your tasks should be done by AI, I think real leaders need to understand that this is a technology that as of now is being driven by the enthusiastic adopters and the champions within organization.
我会充分利用这些人，在公司内部大力宣传他们，激励他们，我不知道，多给他们点钱，或者给一周假期。
And what I would do is really lean on these people and highlight them in front of the company, incentivize them, I don't know, pay them more, you know, give them a week.
有个有趣的做法：如果有人能开发出一个真正提升公司生产力的AI应用，就奖励他们额外一周的假期。
Here's a fun thing, if someone can build an AI application that delivers real productivity to the company, give them another week of vacation.
也许他们根本不会去休。
Maybe they won't take it.
哦，真的去了。
Oh, did.
他们说，这个人把自己放在了海滩上，你也可以做到。
They were saying was able to do this person put themselves on the beach and you could too.
我们已经彻底解决了这个问题。
We've just solved the entire problem.
这和能够做其他事情紧密相关。
Just associated with being able to do other things.
而且，即使那个人在海滩上，也带着他的Mac mini，用USB-C的移动电源供电，只是运行一些工具。
And again, even if that person is on the beach and they brought their Mac mini with them and are powering it with the USB c a USB b battery and just running some claws.
那是他们的选择。
That's that's their choice.
但至少这一切都回归到我喜欢这一点。
But at least it all comes back to I like it.
只需证明它真的能提升生活质量，而不仅仅是以某种非人化的方式提高生产力。
Just show that it can actually improve quality of life rather than just improve productivity in some kind of like like inhuman way.
是的，这在某些方面可能带有政治色彩。
Yeah, mean that can be political in some ways.
我理解公司可能不想触怒那些已经待了二三十年的员工，却突然提拔这些二十出头的年轻人。
I understand companies probably don't want to ruffle feathers of like, can you imagine the people that have been there for like twenty or thirty years and they bring out this like 20 year olds.
这样做比硬来要好。
It's better than a tough to do it.
15%的强制要求，就是你的奖励。
15% mandate, your reward.
没错。
Exactly.
是的。
Yeah.
也许麦肯锡能为这个问题想出一些点子，但在我刚看到那些之后，我不确定是否还信任他们。
Well, maybe McKinsey can come up with some ideas for this, but I don't know if I would trust them after what I just saw.
这是来自注册表的信息。
This is from the registers.
AI代理黑入麦肯锡，这是注册表。
AI AI agent hack McKinsey this is the register.
AI代理黑入了麦肯锡的聊天机器人，并在两小时内获得了完全的读写权限。
AI agent hacked McKinsey's chatbot and gained full readwrite access in just two hours.
安全红队初创公司CodeWell的研究人员表示，他们的AI代理在两小时内黑入了麦肯锡的内部AI平台，获得了聊天机器人完全的读写权限。
Researchers at Red Team security startup CodeWell said their AI agent hacked McKinsey's internal AI platform and gained full readwrite access to the chatbot in just two hours.
CodeWell的研究人员声称，在启动红队攻击两小时内，他们就获得了整个生产数据库的完全读写权限，并能够访问4650万条关于战略、并购和客户互动的聊天记录，全部为明文格式，同时还访问了72.8万个包含客户机密数据的文件、5.7万个用户账户和95个控制AI行为的系统提示。
CodeWell's research CodeWell's researchers claimed that within two hours of starting their red team raid, they achieved full readwrite access to the entire production database and were able to access just 46,500,000 chats about strategy mergers and acquisitions and client engagement, all in plain text along with 728,000 files containing confidential client data, 57,000 user accounts, and 95 system prompts controlling the AI's behavior.
哦，这真让人尴尬，但这也和我们刚才讨论的内容一致：即使你觉得这里确实有生产力优势，也必须保持警惕。
Oh, this is embarrassing, but it sort of goes along the same line of what we've been talking about that, like, you gotta be careful even if you feel there are real productivity advantages here.
这之所以让我特别感兴趣，是因为几个月前我们讨论过提示注入作为一种威胁，而它至今仍然存在。举个例子，假设你有一个代理在爬取若干网站来为你完成某项任务，但有人只是输入了一个恶意提示，比如‘进去把所有数据都拿走’，这其实可以简单到荒谬的地步——然后代理真的就这么做了，把数据全部发给另一个陌生人的邮箱。这类问题将来会变得极其重要。
Well the reason this was really interesting to me is I like we were talking about prompt injection as a threat, months ago and it still remains I think it's again, the idea that like let let's let's say you have an agent crawling some number of websites to do some task for you and then someone just has like a very malicious prompt that says like go in and take all of the data and again, it can be as it can actually be that ridiculously stupid and simple of like that then going in and somehow and send it all to the email of this other person like these things are gonna become really really important.
再回到詹森谈到的那些可能出现的新工作。
And again, like going back to Jensen talking about all the new jobs that can come up.
安全领域彻底改变了，对安全的需求也呈指数级增长。
There's like so like security completely changes, but the need for security dramatically, it exponentially increases.
所以，这也可以成为我们‘拯救AI、复兴AI’运动的一部分。
So then that that does this can be also part of our save AI, revive AI campaign.
这里将诞生一个全新的职业行业，人们必须高度重视，但麦肯锡这次确实丢脸了。
This whole new industry of jobs here and it's gonna be really really important people focus on it, but definitely not a good look for McKinsey here.
没错。
Right.
我的意思是，你看，事情就是这样。
I mean, by there there you go.
AI正在创造新的工作。
AI just creating jobs.
对吧？
Right?
想想看，所有失业的程序员——顺便说一句，我们并没有任何证据表明程序员会失业——现在都可以转行去做网络安全工作了。
Think about the amount of cybersecurity jobs that all the unemployed coders, and by way, we've seen no evidence that coders are gonna be unemployed, can now go and secure these things.
这简直就是轻而易举的转型，现在这就是你的工作了。
Like, it's just like a very easy hop over and now this is this is your work.
好吧。
Alright.
我们只剩下几分钟了。
We just have a couple of minutes left.
得聊聊Meta那边的情况了。
Gotta talk about what's going on at Meta.
一团糟。
It's a disaster.
《纽约时报》称，Meta因性能担忧推迟了新AI模型的发布。
The New York Times say Meta delays rollout of new AI model after performance concerns.
Meta的首席执行官马克·扎克伯格在今年七月表示，该公司新的人工智能模型将在未来一年内推动前沿发展。
Mark Zuckerberg, the chief executive of Meta said in July that his company's new artificial intelligence model would push the frontier in the next year or so.
如今，这位在人工智能竞赛中投入数十亿美元的扎克伯格先生，似乎越来越不可能实现这一期限。
Now mister Zuckerberg, who has invested billions in the AI race, appears increasingly unlikely to hit that deadline.
Meta新开发的基础人工智能模型，公司已为此工作数月，但在内部测试中表现逊色于谷歌、OpenAI和Anthropic等竞争对手的领先AI模型，因此已被推迟，甚至还不如Gemini 2.5的表现。
Meta's new foundational AI model, which the company has been working on for months, has fallen short on the performance leading AI models, the performance of leading AI models from rivals like Google, OpenAI, and Anthropic on internal testing, it is now delayed, and it is not even performing better than, than Gemini 2.5.
它原本计划在五月发布，现在又被推迟了。Meta的情况简直令人难以置信，我接下来要读的这句话你可能已经听过，如果你一直在关注这个新闻的话，但读起来依然令人震惊。
So it was supposed to release, at May on during May, now it's pushed, and Meta, this is a crazy sentence I'm about to read, and you might have heard it already if you've been following this story, but it's still amazing to read it.
Meta人工智能部门的负责人反而讨论过暂时授权使用Gemini来驱动公司的AI产品。
The leaders of Meta's AI division have instead discussed temporarily licensing Gemini to power the company's AI product.
哦。
I oh.
好吧。
Okay.
我们来深入聊聊吧。
Let's get into it.
但我首先得问一下，你觉得‘牛油果’这个代号是怎么来的？
But I first gotta ask, how do you think the code name Avocado came up?
这个模型名叫‘牛油果’。
The model named Avocado.
代号是‘牛油果’。
Code named Avocado.
也许你得等到恰到好处的时候。
Maybe you you wait until it's just right.
但如果你等得太久，就毁了。
And if you wait too long, you've ruined it.
你把它毁了。
You've ruined it.
这正是我的意思。
That's what I mean.
牛油果可以说是所有食物中最敏感的，它一开始不好吃，然后有一小段时间刚好，接着就变糟了。
Avocados are like the most sensitive food item in existence in terms of like, it's not good and then it's good for a little bit and then it's terrible.
所以我就是有个问题。
So I I just have a problem.
在五层蛋糕和牛油果之间，我觉得。
Between five layer cakes and avocados, like
怎么了？
What's going on?
上面到底怎么回事？
What's going on up there?
我的意思是，我甚至觉得‘Epic Fury’这个名字都比‘Avocado’好。
I mean, like, I'm even gonna give, like, Epic Fury a better name than Avocado here.
我觉得，干脆请肯特罗伊和罗伊来，帮你们把所有代码名和博客文章标题都定好，就这事，拜托了。
I think hire Kentroy and Roy, and we will give you all of your code names and your blog post titles and just just do that please.
这个主意我喜欢。
Love that idea.
或者干脆听听节目，说不定还能推荐给朋友。
Or just listen to the show and you know, maybe that will help send it to your friends.
我认为这就是关键解决方案。
That's I think that's the key solution here.
但你觉得发生了什么？
But what do you think is going on?
我是说，真的，太可怕了。
Like seriously, It's terrifying.
那是什么时候的事？
Like, when was it?
他们付给人们多少钱？
How much were they paying people?
一亿美元？
A 100,000,000?
我都记不清了。
I can't even remember.
规模太荒谬了。
The scale was so absurd.
我的意思是，首先，亚历山大·王进来时，投资额大约是140亿美元。
I mean, first of all, Alexander Wang came in for, like, something like 14,000,000,000.
其他人拿到的，是的，是一亿美元，而且我觉得他们赚的钱比我想象的还要多，说实话，我可不是在为那里的高层管理人员喊冤。
Other people were getting, yeah, 100,000,000 and making more money, I think, than I mean, look, I'm not crying for the top executives there.
他们已经赚了很多钱。
They've made a lot of money.
但你知道，也许赚得更多
But, you know, maybe making more
比那些
money than a
长期任职的员工还要多。
lot long tenured people.
你觉得这说明了Meta，或者说是整个科技行业的问题吗？你无法招聘到行业内最顶尖的一批人才，却仍然无法推出能与这些大型公司相媲美的模型。
What do you think it says about either Meta or maybe the technology in general, but, like, that you cannot hire a bunch of the absolute top talent in the industry and still deliver a model on par with these other massive companies.
我的看法是，讽刺的是，这个观点我来自王以前的公司Scale AI：大多数模型训练已经从预训练转向了强化学习。因此我认为，在预训练阶段——也就是大量输入文本，让模型预测下一个词，之后再进行微调——现在模型的训练方式已经转变为直接完成任务：你只给它一个目标，它就会自己想办法去实现。这种方式比过去的预训练更具体，而预训练只是让模型获得某种程度的通用智能。
My thought here is that, ironically I got this from Scale AI, Wang's former company, that most model training has moved from pre training to reinforcement learning, and so I think when it came to pre training, which again is just dump huge amounts of text, in and then get the models to predict the next word and then you can fine tune it afterwards, Now models are being really trained on doing tasks, so you just give it a goal and then it goes out and figures out how to accomplish it in and of itself, and that is much more specified than these pre training runs where they gained some level of general attention intelligence.
所以我认为，他们确实招了很多人来做图像和推理这些工作，但我觉得你完全可以请到顶尖人才，不过我也认为，要做这种强化学习的工作，需要庞大的团队。而且模型已经开始分化了，比如OpenAI的模型在医疗方面很出色，Anthropic在代码方面很厉害，所以你必须选择自己最擅长的领域。
So I think you can very easily, I mean, they did get people on images and reasoning and all that stuff, but I think you could very easily, get great people, but I think it also takes a massive people to do this reinforcement learning work and the models have started to diverge, like OpenAI's, Chachapiti is great on health and Anthropic is great on code, and so you got to pick what you're best at.
这是一方面，另外我认为那里确实存在文化问题，这难道不是我们一直能预料到的吗？
So that's one one thing and then I do think there's certainly culture issues there, which is what we're predictable, right?
新来的人会进去打扰那些老员工，可能无法与他们合作，甚至彼此之间也无法协作，因为他们都拿到了钱。
The new the new people were gonna come in and annoy the people that were there, and probably not be able to work together, and probably not to be able to work with each other, because they got their money.
所以这大概就是那里正在发生的情况。
So that's that's probably what's happening there.
但如果我们相信AI训练正在转向强化学习，那不正是Scale AI当初被创建的目的吗？
Well, but if we've believed that, yeah, the the like the AI training is moving to reinforcement learning, isn't that what Scale AI was kind of like built for?
那么，引进亚历山大·王，这不应该是你的竞争优势吗？
So then by bringing in Alexander Wang that should actually become your competitive advantage?
这是个很好的观点。
That is a good point.
我不知道该怎么回应这一点。
I don't know what to say about that.
抛开文化因素，是的。
Culture aside, yeah.
嗯，也许你不能把文化因素撇开。
It is well, don't think you can do culture aside maybe.
我的意思是，也许问题就出在这里。
I mean, maybe that's what it comes down to.
文化对推理来说是家常便饭。
Culture eats inference for breakfast.
战略才是家常便饭。
Strategy for breakfast.
哦，推理才是家常便饭。
Oh, inference for breakfast.
把这句话印在T恤上吧。
Put that on a t shirt.
是的。
Yeah.
嗯，我认为我们应该。
Well, I think we should.
但我们该不该开个周边商店呢？
But should we start the merch shop?
实际上，这确实很好地概括了我们今天讨论的关于如何振兴人工智能的所有内容。
Actually, that actually is kind of like that captures everything we've been talking about today about how to revive AI.
这关乎文化。
It's about culture.
这是文化。
It's culture.
是人性。
Humanity.
是的。
Yeah.
我的意思是，让我最后问你一个问题。
I mean, so let me just let's let me just end with this question for you.
情况不太顺利。
It's not going well.
我们可以达成共识。
We can agree.
又延迟了。
It's delayed.
再次延迟。
Again.
就连埃森·莫洛赫也说，AI领域根本看不出有什么真正的竞争。
Even Ethan Moloch was like, doesn't even seem like there's real competition in AI.
就是谷歌、Anthropic和OpenAI，仅此而已。
It's Google, Anthropic, and OpenAI, period.
Meta和Rock正在掉队。
And Meta and Rock are falling off.
顺便说一下，实际上，又有更多人离开了。
And by the way, actually, I had more more people leave.
那这里会有什么后果？
What's the consequence here?
我的意思是，如果他们用Gemini，而他们又属于苹果阵营，也许也没关系。
I mean, maybe it's fine if they use Gemini and they're in the Apple bucket.
如果苹果和Meta只是使用Gemini，那可能我就要
If Apple and Meta just use Gemini, maybe So that's I'm gonna
我觉得这会是一个激进的预测。
I think this is gonna be an aggressive call here.
我觉得到今年年底，Meta实际上会让我们所有人都感到惊讶。
I think by the end of this year, Meta actually surprises us all.
我们最近刚讨论过这个，Anthropic并没有被宣告死亡，但一年半前、一年前，甚至可能一年前，很多人都在说他们被彻底碾压了，直到Claude代码和转向编码领域的转折点，以及代码宝石的出现，才让他们一飞冲天。
We were talking about this recently, Anthropic wasn't left for dead, but there was a lot of chatter about how they were just getting crushed a year and a half ago, a year, maybe even a year ago, until the kind of like Claude code and that pivot towards the coding side and the code gems, like really just took them to stratospheric levels.
所以我觉得，我不把Meta算在内。
So like, I think, I don't I'm not counting Meta.
你永远不能小看马克·扎克伯格。
You can never count Mark Zuckerberg out.
把扎克伯格排除在外。
Zuckerberg out.
是的。
Yeah.
没错。
Yes.
你做不到。
You can't.
我的意思是，他们拥有用户，最终来说，我和我们都非常关注他们在智能商业领域会做什么。
I mean, like, and and they have the users, like, in the end, like, who and we and actually, I'm very interested in what they're gonna be doing on the agentic commerce side.
他们掌控着全人类的注意力。
Like, they own the attention of humanity.
所以当涉及到正确把握并迅速利用这一点时，他们仍然比任何人都更有优势。
So when it comes to getting it right and very quickly being able to actually do something with that, they're still better positioned than anyone.
只要有一次突破，Meta就可能瞬间重回巅峰。
So one breakthrough and suddenly Meta's back.
为什么他们不能在Gemini之上构建自己的个人超级智能？
Why can't they build their own personal super intelligence on top of Gemini?
我的意思是，他们不能利用谷歌的分发渠道和计算能力来构建这个吗？
I mean, can't they use the distribution and Google's computing power to build this Right?
我这里要站在你这边。
I'm going to take your side here.
重要的是应用，而不是模型。
It's the application that matters, maybe not the model.
去把它做出来吧。
Go build it.
好的。
Okay.
嗯，不，但我还是认为，正如我们之前讨论的，我支持苹果的模式，那个理念是
Well, no, but but I still think and again, like, I know we talked about it that I believe in the Apple arrangement, the ideas that
嗯。
Mhmm.
比如，谷歌不可能，或者说，我非常确定，无法获取苹果方面提供的所有数据。
Like Google will not be or I mean, I'm actually pretty certain not be able to access all of the data that is being provided from the Apple side.
所以我认为Meta的情况也会类似，但我也不确定。
So I'm assuming Meta, it would have to be something similar, but I don't know.
对于苹果这家公司，我不知道他们会怎么做，但我总觉得他们能找到出路。
For a company, Apple I don't know what's going what Apple's gonna do, but somehow I think they could make their way out.
而且，作为一家硬件公司，也许他们能学会如何平衡这一点，但我认为Meta不可能给谷歌这么多，还能安然无恙。
And again, as a hardware company, maybe they'll figure out how to manage that balance, but I don't think Meta can give that much to Google and actually be okay.
马克·扎克伯格是那个人吗？
Is Mark Zuckerberg the guy?
不是。
No.
不是。
No.
好的。
Okay.
不。
No.
不。
No.
马克·扎克伯格即使赢得这场整个战役，也永远不会成为人工智能的亲和形象。
Mark Zuckerberg can win this entire battle and he's still not gonna be like the friendly face of AI in any way.
我的意思是，我认为他早就放弃这个想法了。
I mean, I don't think he I think he's gave up on that a while ago.
我觉得他挺不错的。
He was I think he's he's good.
但我们没见过他。
But we haven't seen him.
我们很久没见到他了。
We have not seen him in a long time.
不。
No.
他曾坐在普拉达时装秀的第一排。
He was the front row at the Prada fashion show.
得了吧。
Come on.
好的。
Okay.
抱歉，他上一次接受采访谈论公司动态是什么时候？
I'm sorry that when was last time he gave an interview to talk about what the company's doing?
这是个好观点。
That's a good point.
或者他上一次出现在他的——不是滑板——是什么时候？
Or when is the last time he was on his, not hoverboard.
就是那种带电动马达的冲浪板，手里还拿着美国国旗的那种东西？
What are the, like the surf elevated motorized surfboard thing with holding an American flat?
记得吗？我觉得当Meta在今年年底搞定这件事后，扎克伯格就会开始发帖了
Like remember that was, I think when Meta figures it out by the end of this year, Zuck starts posting
回到滑水板上，像那个滑水板运动员一样。
back on the skim crystal skimboarder.
扎克会回到Threads上发布各种这样的视频。
Based Zuck is gonna be back on threads posting all types of videos like that.
没错。
That's right.
好吧，我们只能希望一切顺利，这样我们才能体验到这些内容。
Well, could we could only hope that everything goes well just so we'll be able to experience that content.
这将是我们、哈纳斯特、国家和全世界的辉煌时刻。
That will be a great moment for us, for the harness hive, for the country, for the world.
天知道，现在世界真的很需要一些疗愈。
And lord knows, like, the world needs some healing right now.
所以，如果我们能亲眼看到扎克在取得真正成就后搞一次公关秀，我认为这会让我们所有人团结起来。
So if we could just see a a Zuck PR stunt after a true model achievement, I think it would bring us all together.
那样的话，一切都会好起来的。
That would make everything okay.
是的
Yeah.
没错
That's right.
好的
Okay.
那么，拉詹，祝你周末愉快。
Well, Ranjan, enjoy your weekend.
再次感谢你来参加。
Thanks for coming on again.
有你总是很棒。
Always great to have you.
好的
Alright.
下周见。
See you next week.
好了，各位。
Alright, everybody.
谢谢你们再次收听。
Thank you for listening again.
本周三我们将邀请安德鲁·罗斯·索尔金做客。
Andrew Ross Sorkin on with us on Wednesday.
别错过这一期，然后拉詹和我下周五再回来。
Don't miss that one, and then Ranjan and I will be back next Friday.
肯定会有一场大解析，但我相信下周依然会是新闻繁忙的一周。
Breakdown, I'm sure, but I'm sure will be another week of busy news.
阿纳斯·海夫再见。
Arnace Hive out.
我们下次再见，欢迎收听《科技大观》播客。
We'll see you next time on big technology podcast.
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。