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。
OpenAI 正在放弃视频生成。
OpenAI is giving up on video generation.
以下是背后的真实故事。
Here's the real story behind it.
苹果将在 Siri 中推出一系列 AI 助手，而 Meta 则输掉了一起具有里程碑意义的诉讼案，这可能预示着更大的麻烦即将到来。
Apple is going to make a bunch of AI assistants available in Siri, and Meta loses a landmark court case that could spell even more trouble ahead.
接下来，在周五的大型科技播客节目中，我们将为您带来这些内容。
That's coming up on a big technology podcast Friday edition right after this.
财政负责、金融天才、货币魔术师。
Fiscally responsible, financial geniuses, monetary magicians.
这些是人们在谈到那些将车险转投 Progressive 并节省数百美元的司机时说的话。
These are things people say about drivers who switch their car insurance to Progressive and save hundreds.
因为 Progressive 为一次性付清保费、拥有房产等提供折扣。
Because Progressive offers discounts for paying in full, owning a home, and more.
此外，当您需要帮助时，可以信赖他们出色的客户服务，让您的每一分钱都花得更值。
Plus, you can count on their great customer service to help when you need it, so your dollar goes a long way.
访问 progressive.com，看看你能否在汽车保险上省钱。
Visit progressive.com to see if you could save on car insurance.
Progressive 伤害保险公司及其关联公司，潜在节省金额因情况而异，并非在所有州或情况下都可用。
Progressive Casualty Insurance Company and affiliates, potential savings will vary, not available in all states or situations.
欢迎收看《大科技》周五版，我们将以一贯冷静而细致的方式解析新闻。
Welcome to big technology Friday edition where we break down the news in our traditional cool headed and nuanced format.
今天为大家准备了一场精彩的节目。
We have a great show for you today.
我们将深入探讨 OpenAI 为何放弃 Sora，以及这背后对该公司雄心壮志的更深层意义。
We're gonna talk all about why OpenAI killed Sora and really the bigger story about what it means for the company's ambitions.
我们还会聊聊 OpenAI 和 Anthropic 最近推出的几款新模型，这些消息在业内引起了热烈讨论。
We're also gonna talk about a couple new models coming out from OpenAI and Anthropic that has people on the inside buzzing.
苹果对 Siri 的这次升级，真的会有所作为吗？
This new upgrade for Apple and Siri, is it going to amount to anything?
我们会简单谈一谈这个话题。
We'll touch on that.
最后，Meta本周在法庭上遭遇重大失利，这可能在未来对其造成负面影响。
And then finally, Meta lost big in court this week, and that may be news that will harm it in the future.
和往常一样，周五我们邀请了Margins的Ranjan Roy来参与讨论。
Joining us as always on Friday to do it is Ranjan Roy of Margins.
Ranjan，欢迎你。
Ranjan, welcome.
Sora已经死了。
Sora is dead.
情色聊天机器人不复存在了。
Erotic chatbots are no more.
苹果可能正在修复Siri。
Apple might be fixing Siri.
这感觉像是美好的一周。
This this feels like a good week.
我现在在犹他州的帕克城，但这里一点雪都没有。
I I'm currently in Park City, Utah, and there's no snow.
所以我对这件事不太高兴。
So I'm not very happy about that.
但我对这些新闻感到高兴。
But I'm happy about all this news.
所以
So
好吧，我们会深入探讨一下。
Well, we will dig into it.
显然，支线任务的时代已经结束了。
And, clearly, the side quest era is over.
到目前为止最大的损失是Sora的消亡，那个我们曾多次讨论、不久前还登上App Store榜首的视频平台。
And the biggest casualty so far has been the death of Sora, the video platform that we talked about so much that went to number one on the App Store, not long ago.
不仅如此，它的API也被放弃了。
And not only that, but the API as well.
这是来自《华尔街日报》的消息。
This is from the Wall Street Journal.
OpenAI计划终止其Sora视频平台，该产品去年发布时备受瞩目，但此后已从公众视野中消失。
OpenAI is planning to pull the plug on its Sora video platform, a product that released to great fanfare last year and has since fallen from public view.
这一举措是OpenAI为聚焦商业和编程功能、以便在今年第四季度前可能进行IPO而采取的多项措施之一。
The move is one of a number of steps OpenAI is taking to refocus on business and coding functions ahead of a potential IPO as soon as the fourth quarter of this year.
萨姆·阿尔特曼于周二向员工宣布了这些变更，称公司将逐步停止使用其视频模型的产品。
Sam Altman announced the changes to staff on Tuesday, writing that the company would wind down products that use its video models.
除了消费者应用外，OpenAI还将终止面向开发者的Sora版本，并且不再在ChatGPT中支持视频功能。
In addition to the consumer app, OpenAI is also discontinuing a version of Sora for developers and won't support video functionality inside ChatGPT either.
拉詹，让我先从微观角度来看看，然后我们再宏观展开。
Ranjan, let me just throw let's let's look at it micro and then we can zoom out.
让我跟你分享几个关于Sora为何失败的想法。
Let me throw a couple ideas at you in terms of why Sora died.
一个是，也许视频AI的吸引力并没有那么大。
One is maybe it's just that the appeal of video AI isn't that great.
当你生成一段视频时，起初确实很震撼，但也许大多数人只是想观看，而不是创作。
Like, there's an initial thrill when you generate a video, but maybe most people just want to watch instead of create.
其次，也许所有生成的视频最终看起来都一模一样，对吧？
And then secondly, maybe it's just that all the videos ended up looking the same, right?
你用AI生成的其实是平均值的平均值，结果就是那样。
That you take AI, it generates the average of averages, and that's what you've got.
然后，突然之间，它的实用性就急剧下滑了。
And then all of a sudden, the utility of this just nosedived.
是的。
Yeah.
我认为在这里把Sora和视频AI的整体前景或当前状态区分开来很重要。
I think separating out Sora specifically from just the promise or kind of current state of video AI is important here.
我得说，因为我整天都在和零售消费品行业的营销人员交流，视频仍然非常有趣。
I have to say, I mean, given I'm talking to retail consumer goods type people on the marketing side all day long, video is still very interesting.
而且，似乎谷歌的VO正逐渐成为行业标准。
And everyone has been VO from Google seems to be kind of becoming the industry standard.
所以实际上，每个人都对它非常感兴趣，尤其是从真正的制作角度来看。
So so everyone is actually very interested in it, especially from, like, a, like, a true production standpoint.
所以我认为视频领域，我仍然很惊讶他们竟然完全放弃了这个方向，连API都不再提供，也不再提供任何视频功能。
So I think video, I don't think like, I still am surprised that they're actually completely seeding that area and not even making it available via API anymore, not making any video functionality.
但Sora这个产品，说实话，我不确定。
But Sora the product, I mean, I don't know.
我已经很久没打开过它了。
I I have not opened it in a long time.
你到目前为止最出色的Sora作品有哪些？
I what what were what were your greatest Sora hits to date?
是我和杰克·保罗热情地扶一位老奶奶过马路的那段。
It was me and Jake Paul walking old lady across the street very enthusiastically.
我其实刚刚专门回顾了一遍Sora的创作历程。
I I I went through actually just in memoriam of Sora.
我儿子七岁，我会让他给我提供创意，比如有一条鸡和一匹马围着马桶转的场景。
And so my son, I would actually who's seven years old, like, we would I would ask him to give me prompts, and I had like a chicken and a horse running around a toilet bowl was one of them.
基本上都是类似这种东西。
It was like, it was basically a lot of stuff like that.
这很有趣，他的朋友们都笑得很开心。
And it was funny, and the his friends, they would all get a good laugh out of it.
我想这未必是能证明市场潜力、满足目标市场规模以向投资者路演或筹备IPO所需的那种验证。
I guess that was not necessarily the market validation one needs to address the TAM of whatever they're gonna have to pitch investors for the IPO.
但确实如此。
But Yeah.
一些七岁的孩子可能会对此感到不开心。
Some seven year olds are gonna be unhappy about this.
是的。
Yeah.
你为了得到那匹在马桶周围奔跑的马，烧掉了一整片雨林。
You burned an entire rainforest just to get the
那匹在马桶周围奔跑的马。
the horse running around toilet bowl.
是的。
Yeah.
但这个笑话确实切中了问题的核心，我确实有一些内部消息，关于Sora被搁置的真实原因。
But but that that joke really goes to the heart of the matter, and I actually do have some intel on the real reason why Sora has been Oh.
搁置。
Shelved.
这周我人在旧金山。
And I will say this week, I'm in San Francisco.
昨天，我见了OpenAI的总裁格雷格·布罗克曼。
And yesterday, I was meeting with Greg Brockman, the president of OpenAI.
我们有一场长达十五小时十五分钟的访谈，下周播出。
And we have an hour, ten hour, fifteen long hour and fifteen minute long interview coming next week.
所以敬请期待周三的节目。
So stay tuned for that on Wednesday.
当然，我们先从这个转向企业与编程的新方向说起。
And of course, we begin with this sort of new pivot to enterprise and coding.
我不会全盘透露，但我会分享格雷格关于为什么视频生成被降级、被视为OpenAI内部的支线任务的这段话。
I won't give it entirely away, but I will share this bit from Greg about why video generation became deprioritized and was looked at as a side quest within OpenAI.
所以首先，我要说，我之前就认为是这样，上周我也提到过。
So first, I'll say, I thought this was and I spoke about this last week.
我以为这是个面向消费者和企业的事情。
I thought this was a consumer enterprise thing.
所以也许他们觉得Sora应用更偏向消费者，而现在真正聚焦的是企业市场。
So maybe they thought that the Sora app was more consumer y and they're really focusing on businesses.
但实际上完全不是这么回事。
It's actually not what it is at all.
基本上，OpenAI发现这些GPT风格的模型已经奏效，还有其他方法可以追求更强大的AI。
So basically, OpenAI has seen that these GPT style models are working, and there are other ways to try to pursue the most powerful AI.
而目前最著名的另一种方法就是世界模型。
And that most famous other method right now is the world models.
也就是那些真正理解物理规律的模型，而这正是Sora内置的一部分。
So models that actually understand physics, and that is part of what was baked into Sora.
所以我跟Greg聊了聊，问他：这到底怎么回事？
And so I spoke with Greg and said, well, what's going on here?
他说，关键是要意识到，从技术上讲，Sora 模型——顺便说一句，这些模型非常出色——与核心推理型 GPT 系列属于技术树的不同分支。
And he said the important thing to realize is tech technologically that the Sora models, which are incredible models by the way, are a different branch of the tech tree than the core reasoning GPT series.
它们的构建方式完全不同，在某种程度上，我们实际上是在说，同时推进这两个分支对我们来说非常困难。
They're built in just a very different way, and to some extent, we're really saying that pursuing both branches is very hard for us.
因此，我认为 OpenAI 目前展现出的这种专注方向是必要且非常有趣的，他们基本上是说：我们也要尽可能打造最强大、最强大的模型。
So I think this is needed and and very interesting focus that we're seeing from OpenAI where they're basically like, we wanna build the strongest power most powerful models as well, that we can.
我们正在看到这些文本 GPT 类模型取得的成果：它们能调用工具、进行推理、运用思维链，完成各种任务，而我们必须决定将计算资源投入到哪里。
We're seeing the results in these text GPT style models that do all the things, that call the tools, that do the reasoning, that do chain of thought, that are getting things accomplished, and we have to decide where to put the compute.
如果走世界模型这条路，将严重限制公司在这个他们认为最具前景的领域取得进展的能力，这就是为什么 Sora 被降级了。
And to do that in a world model way would really limit the company's ability to progress to make progress in the area they see as most promising, and that is why Sora is being deprioritized.
天哪。
My goodness.
这真是专注。
That is that is focus.
这确实是真正的专注。
That is that is real focus right there.
我得说，就在上周的节目中，我们半开玩笑地提到，现在你只要说‘世界模型’这个词，它就成了最时髦的术语，而且Meta正在大肆宣传它。
I have to say because just on last week's episode, we were half joking about now all you have to say is world model, and it's kind of the trendiest term, and it's and Meta's making a big deal about it.
谷歌也在谈论它。
Google's talking about it.
我觉得这肯定会成为2026年的热门词汇之一。
Like, that's definitely, I think, gonna be one of the buzzwords of 2026.
所以，当他们真正承认这不会成为另一个投资领域时，我觉得这确实非同小可。
So for for them to actually acknowledge that that is not gonna be another area of investment, I think that, okay, that is a pretty big deal.
没错。
Right.
这就是他们的逻辑，我觉得完全说得通。
So that is the the logic, and I think that makes total sense.
当然，格雷格和我下周三还会继续讨论这个话题，所以大家记得收听。
And, of course, Greg and I will speak about it more next Wednesday, so folks, please tune in for that.
但现在这引出了一个有趣的问题：这场竞赛到底会是什么样子？
But now it sort of goes to something interesting, which is like, what does this race look like?
因为猜猜还有谁根本没有真正做世界模型？
Because guess who else hasn't really done world models?
猜猜谁也没有涉足图像生成和视频生成这些旁支任务？
Guess who hasn't done the side quests of image generation and video generation?
是Anthropic。
It's Anthropic.
因此，你开始看到一场竞赛正在形成。
And so you're starting to see a race.
如果这场竞赛的形态与不久之前大不相同了呢？
So if you the race is taking a very different shape than it had not long ago.
还记得吗？以前是ChatGPT，OpenAI拥有ChatGPT，靠着这些图像、视频和聊天机器人赢得了消费者市场。
Remember, it used to be that, well, ChatGPT, OpenAI had ChatGPT, and it was winning consumer with all these images and videos and the chatbot.
而Anthropic则专注于企业市场，主要在编程、商业和各种应用领域发力。
And Anthropic was enterprise, and it was sort of doing the enterprise thing with coding and business and all these these applications.
但现在发生的变化是，这两家公司都集中到了同一个使用场景上。
Well, what's happened now is that both these companies have centralized on the use case.
也许你可以称之为OpenCLaw风格的使用场景，这正是两家公司似乎都在追求的——让AI访问你的桌面、手机或其他任何设备。
Maybe you could call it the OpenCLaw style use case, which is what they both seem to be going for, which is that you give the AI access to your desktop, to your phone, whatever it might be.
如果你在工作，就让它替你完成工作。
And if you're at work, you have it do work for you.
如果是在个人生活中，它就能帮你整理个人事务并主动采取行动。
If it's in your personal life, you would organize your personal life and take action for you.
我认为他们确实看到了这一点，我再把话题交给你——这种潜在的智能体使用场景，无论是在消费端还是企业端，技术都会主动去满足你的需求，本质上是一回事。
And I think they do see that this, and I'm going to hand it to you again, this potential agentic use case where the tech goes after what you need, whether it's consumer or business, it's the same thing.
它们正在向这种单一的堆栈体系集中。
It's centralizing in this sort of one stack, so to speak.
现在，这简直就像两者之间一场你死我活的较量，看谁能率先实现并抢占这一领域。
And now it's just like almost like a battle to the death here between the two of them to get this right and go after it.
我的意思是，听众看不到，但在YouTube上你可能会看到我现在正在微笑，因为亚历克斯，被人认可总是让人感觉很好。
Well, I mean, one, listeners cannot see, but maybe on YouTube, you'll see me smiling right now because Alex is, it always feels good to be recognized.
但我得承认，当我提到这种自主的知识型工作时，这正是我们在去年六月开始构建这项技术时在Ryder公司所看到的。
But I have to admit, when I said this kind of autonomous knowledge work, and this is what we started seeing at Ryder when we started building this last June.
就像那个神奇的时刻，从一个地方提取文件，对其进行处理，然后推送到另一个系统，接着思考这如何适用于所有事情。
Like, it was that magic moment of, like, pulling files from one place, doing something to it, pushing it to another system, and then thinking about, like, how that applies to absolutely everything.
我要说，当我做出那个预测时，大概是在十月左右，我根本没想到OpenAI在上市前会真的将整个战略集中到这一点上。
I will say when I made that prediction, I think in October ish probably to start, I did not think OpenAI ahead of an IPO would actually be kinda consolidating its entire strategy around that.
所以这次我赢了。
So I'm gonna take the win on this.
我要指出的一点是，这并不是只有两个玩家。
One thing I'll note, this isn't two players.
我这么说是有私心的，因为这正是我们在Ryder正在做的事情。
And I'm saying that again self interested because this is what we're working on at Ryder.
但越来越多的是，Sierra本周刚刚发布了一个名为Ghostwriter的功能。
But more and more so, Sierra just this week released something called Ghostwriter.
Notion也在进入这个领域。
Notion is going into this space.
我真的在近距离亲身观察着这一切。
It really I mean, I'm seeing this very closely firsthand.
每个人都在争夺这个领域。
Everyone is going after it.
所以我认为大家都意识到了这是关键目标。
So I think everyone has recognized that's the prize.
所以这不仅仅是OpenAI和Anthropic的事。
So it's not just OpenAI and Anthropic on this.
甚至更传统的SaaS公司也在朝这个方向努力。
It's it's even more traditional SaaS companies are trying to go in this direction.
所以我很清楚，我在这一点上是对的。
So so I think it's definitely clear that I was right on this one.
所以是的。
So Yes.
但不是。
But no.
但不是。
But no.
不。
No.
这很明显。
I it it's clear.
但现在我在想，比如图像生成和视频，尤其是在消费端。
But but now I am wondering, like, image generation and video, like, does it especially on the consumer side.
我觉得企业端和消费端之间还是会存在很大差异。
See, I I still think there's gonna be a big difference between enterprise and consumer.
Meta会不会开始在这方面采取行动，填补这个空白？
And, like, does Meta start making moves in here and start kind of filling the gap?
谷歌会不会直接掌控它？
Does Google just kind of own it?
因为在我看来，这些功能在消费端仍然会非常重要。
Because to me, those kind of functions are still gonna be very important in consumer.
对。
Right.
嗯，问题就在这里。
Well, this is the thing.
比如，Nano Banana 对谷歌来说一直是个巨大的优势，它是他们的图像生成器。
Like, Nano Banana has been a huge asset for Google, their image generator.
顺便说一下，有一件有趣的事我不会透露太多格雷格采访的内容，但我认为这个情况不会发生。
And by the way, something that was interesting I don't want to give away too much of the Greg interview, but I think this I'm is not going
不会给你太大压力。
to push you too hard.
因为这具有新闻价值。
Because it's newsworthy.
不会消失的是图像生成和ChatGPT。
Guess what's not going away is image generation and ChatGPT.
当你想到这一点时，你的第一反应是什么？
And when you think about that, what's your initial response?
好吧。
Well, okay.
所以生成图像所需的计算量比生成视频少，这是真的。
So creating images doesn't take as much compute as creating video, which is true.
但格雷格说的是，图像生成使用的是与GPT类似的技术，而视频生成则使用完全不同的技术。
But what Greg says is basically like it is it is the image generation is being done in with the same GPT style technology, whereas, like, the video generation is done with this completely different technology.
这就涉及到模型的通用性，比如它们可以处理文本。
So that that and it goes to the generality of the models where, like, they can do text.
它们可以进行图像生成。
They can do image generation.
但视频生成，它们做不到。
Video generation, they can't.
但我认为你说得对，确实存在一个巨大的机会，让某人能够做好视频生成。
But I think you're right that there is this big opening for somebody to do video generation well.
显然，有一些初创公司比如Runway，但谷歌在这里处境非常有利。
And clearly, there's there are, like, some startups like Runway, but Google Google is in great shape here.
我某种程度上支持Runway。
I I'm kind of rooting for Runway in this.
我不确定。
I I don't know.
比如三年前，2023年，我知道，因为我开始测试每一个可用的生成式AI工具。
Like, three years ago, 2023, I know because I started testing every single generative AI tool available.
我记得Runway可能是我第一次进行图像生成的地方。
And I remember runway was probably the first place that I actually did an image generation.
那是六指时代的早期，你知道的，四条腿什么的。
And this is back in the six finger days of like like when, you know, the four legs, all of that.
至于视频，他们可能也是我最早开始测试和尝试视频的地方。
And then even video, they might have been the first place I started testing and playing around with video.
所以这或许打开了大门，但我也觉得。
So so maybe this opens the door, but but I yeah.
不过我还想补充一点，图像生成也不只是简单地创建一张猫的图片。
I think one other note though is image generation also, it's not just like, create an image of a cat.
我不确定。
I don't know.
实际上，我见过的最棒的Sora视频之一，是那只拿着猎枪射击Ring门铃的猫。
Actually, what what was the the one of the best Sora's I saw that was circulating around was I think it was a cat with a shotgun shooting a Ring doorbell.
我不确定。
I don't know.
那就是Sora的巅峰之作。
That's that was Sora at its finest.
我不知道你有没有看到那条推文，但那是他们发的。
I don't know if you saw the tweet, but it was from them.
他们说，我们会感谢所有制作出有意义视频的人。
It was like, we we we will give credit to everyone who made videos that matter.
画面里是动物在马桶里乱跑，猫在射击Ring门铃。
And it was animals running around toilet bowls and cats shooting ring ring doorbells.
但我认为，即使在企业领域，图像生成也很重要，比如生成图表之类的，这本质上仍然是视觉沟通。
But but I think there's also image generation even in the enterprise, like, generating diagrams and slut like, there's a it's still visual communication in many ways.
这不仅仅是给我生成一个搞笑图片。
It's not just make me a funny image.
所以我认为，它们仍然需要参与其中，这一点仍然很重要。
So I think it makes sense too that they they still have to play in that, that they still it's still important.
对。
Right.
而且我认为更重要的是，它属于同一条技术路线，而不是完全不同的东西。
And I think the important part also is that it is along the same tech tree as opposed to something completely different.
但你说得对，即使它不同，你可能还是希望把它纳入你的工具套件中。
But but you're right, even if it was different, you'd probably want it in your suite of tools.
我想回到你刚才说的一点，这不仅仅是OpenAI和Anthropic的事。
I want to go back to something you said actually, that it's not just OpenAI and Anthropic.
是的，还有其他公司，但很多这些公司都在使用OpenAI或Anthropic的技术作为底层支持。
Yes, there are others, but a lot of these companies are working with OpenAI or Anthropic's technology underlying.
所以无论怎样，它们都很可能从中受益，比如说，即使最终是Sierra将这项技术应用于商业场景。
So there's a good chance that they'll see the benefit no matter what, even if, let's say, it's Sierra that ends up being the one that deploys this for business.
你现在正踏入我的领域了，Alex。
You're you're coming into my world right now, Alex.
我认为在莱德，我们有自己的基础模型系列，叫做保罗·迈耶系列。
I think so at Ryder, we have our own family of foundation models, the Paul Meyer family.
对我们来说，实际上还有一个非常有趣的互动平台，现在叫Finn。
So for us and actually, there was, like, a very interesting intercom, which now has Finn.
他们本周也宣布，他们基本上已经训练了自己的基础模型。
They announced this week as well that they basically have trained their own foundation model.
所以我认为，我们开始看到一些结合的迹象，比如Cursor使用DeepSeek——他们虽然没有明确说在这么做，但实际上这是一种非常深思熟虑的做法。
So so I think starting to see some kind of combination of using whether it's like cursor using deep seek, which they just basically didn't say that they were doing, but then were, but is actually a very, like, a very thoughtful approach to this.
我实际上认为，越来越多的公司将会采取这种做法。
I actually think more and more people the companies are gonna start taking this approach.
这不再仅仅是调用OpenAI或Anthropic的API。
This is not just gonna be an API call to OpenAI or Anthropic.
所以我认为，像Notion、Cursor和Sierra这样的公司，越来越多地会采用这种方式；到目前为止，很多工具都只是简单地调用API。
So I do think like the and I say this with the Notions and the Cursors and the Sierras of the world that like more and more, it's that I think a lot of tools to date, it was just that API call.
我认为，越来越多的人将开始在基础模型层面进行定制或完全训练。
I think more and more people are gonna start either customizing or fully training on the foundation model side.
预测将会
Prediction gonna
我持怀疑态度，如果我不得不认错，那我就认错。
come in skeptical here, and if I have to eat crow again, I'll do it.
但我确实认为，基础模型公司毫无疑问将成为这里的重要参与者。
But but I do think that the the foundational model companies are going to be well, without a doubt, they'll be big players here.
让我再跟你谈谈这个。
Let me let me take this to you again.
你知道，如果OpenAI和Anthropic之间的竞争不再像以前那样，而是直接在同一个使用场景上正面交锋——而他们以前并没有这样过。
You know, if the battle between OpenAI and Anthropic shapes up to be not the way that it was previously, but, like, going head to head on the same use case, which they hadn't been previously.
比如，Claude之前乐于不吸引大量终端用户，而OpenAI也乐于不进军企业市场。
Like, Claude was happy to not have lots of consumer users, and OpenAI was happy to not go after enterprise.
现在，他们真的开始正面竞争了。
Now they're really going head to head.
你认为这对这场竞争意味着什么？或者你如何看待这种态势的发展？
What do you think that means for this race, or how do you see that shaping up?
你认为谁会赢？
And who do you think is gonna win?
我不认为这是对OpenAI来说正确的方向。
I don't think this is the right idea for OpenAI.
我认为，他们原本在消费端已经有了立足点。
I think, like, they had a foothold in consumer.
我知道消费端的商业模式还没搞清楚，但那确实是他们的优势所在。
I know the business model for consumer has not been figured out yet, but that was still where they had the edge.
他们本可以开始朝这个方向发展。
Like, they could start to go after this.
我们上周还讨论过这个。
We talked about last week.
他们和微软之间的关系对我来说是个非常大的问题，因为你要知道，当你说企业市场时，这本来就是微软的地盘。
What that means between them and Microsoft is a very big question in my mind because remember when you say Mike enterprise, this is Microsoft's world.
而且现在已经有紧张迹象了，比如OpenAI开始与AWS合作，还有传闻称FDA曾报告过潜在的诉讼。
And already there's tension around, you know, the OpenAI starting to do deals with AWS and, like, potential rumors the FDA had reported on a potential lawsuit.
所以，这让他们置身于一个完全不同的领域，与以往截然不同。
So, like, it just puts them in such a different space than they have been.
老实说，这让我有点惊讶，因为他们基本上就是在说：我们想成为Anthropic。
And it's honestly kinda surprising to me that it's like, basically, we wanna be Anthropic is what they're saying.
他们所做的一切，就是努力追赶Anthropic。
Everything they're doing is let's just try to catch up to Anthropic.
是的，Anthropic去年表现非常出色。
And yes, Anthropic had a very good year.
但别忘了，一年半前，人们还以为Anthropic要完蛋了。
But like, remember, a year and a half ago, people were leaving Anthropic for dead.
也许这有点夸张，但我们当时甚至拿出消费者使用率下滑的图表，开玩笑说我们不再是‘云头’，而是‘Gemini派’了。
Maybe that's an exaggeration, but we were even, you know, pulling up charts of declining consumer usage and joking that we're no longer clod heads and we're Gemini guy.
你知道，曾经有过那么一个时刻，他们真正做到了，而我们当时就预见到了。
You know, like, there was this moment and then they really nailed and we called this.
我们说过这是个风险和赌注，但全力投入编程意味着完全不同的东西。
We said it was a risk and a bet, but going all in on coding meant something very different.
但我只是不认为他们现在转向这个方向已经太晚了，这更多是被动应对，而不是打造我们自己独特的业务。
But I just don't think I think it's too late for them to make this switch, and it's reactive rather than we are our own unique business.
我们有八亿用户。
We have 800,000,000 users.
我们即将达到十亿。
We're gonna get to a billion.
我们将投放广告。
We're gonna run ads.
人们会搜索日常生活相关的内容，这有很多变现方式。
People are gonna search about everyday life, and there's a lot of ways to monetize that.
我只是，是的。
I just yeah.
我不确定。
I I don't know.
我需要亲眼看到它实际运行，而不仅仅是被动应对。
I'm I I need to see this in action versus just being reactive.
这很有趣。
That is interesting.
但让我给你提出一个相反的观点，我们都看到了那个‘开放爪子’的时刻。
But let me put the counterpoint to you here, which is we all saw the open claw moment.
对吧？
Right?
我认为，包括我自己在内，很多人还没有完全理解这一点还能应用在其他什么地方。
And I think many of us still aren't haven't fully including myself, haven't fully wrapped our heads around what that can be applied to elsewhere.
因为OpenAI基本上就像是：你会创建一个虚拟机，或者买一台Mac mini，安装这个AI代理，让它为你控制这台机器，接入你使用的几个服务，然后让它成为一个拥有持久记忆的助手，帮你完成一些事情。
Because OpenAI is basically like you're gonna put your you know, create a virtual machine or get a Mac mini, put this AI agent, allow it to control that machine for you, plug it into a couple of services that you use, and then basically have it be a assistant with persistent memory that gets some gets stuff done for you.
所以想象一下——当然，我必须强调，这并不是说OpenAI只针对企业而不面向消费者。
And so think about let's say and and, again, I think it's important to say that this is not gonna be a breakdown of, like, you know, that OpenAI goes after enterprise and not consumer.
想想这种使用场景。
Think about this type of use case.
想象你正在与医院或保险公司打交道，你想申请某项费用报销，或者想了解自己的数据与其他人的相比如何。
Imagine you're dealing with a hospital or dealing with an insurance company and you're trying to get something covered or you're trying to understand what your data looks like compared to others.
拥有一个始终在线、具有持久记忆的助手，它可以代表你与保险公司谈判，或监控你的健康状况，这算是消费者应用还是企业应用？
To have this always on assistant with persistent memory that can go out and negotiate on your behalf with the insurance companies or go out and monitor your health situation, like is that a consumer or is that an enterprise?
这是消费者应用，但它仍然属于智能代理的范畴。
That's consumer, but it's still it's still in this agentic world.
所以，也许把这一点排除在外，这看起来像是一个糟糕的商业决策。
So maybe to to sit that out, which seems like it's a bad business decision.
好的。
Okay.
我的意思是，同意，毕竟现在是你在提出这些广泛的智能代理应用场景和愿景。
I mean, agreed given now you're you're the one saying that kinda, like, coming up with these broad agentic use cases and visions.
而且，再次强调，我认为这正是让人兴奋的地方——为什么大家都如此激动，因为一旦你感受到这种力量，就会忍不住想象所有可能的应用。
And and, again, remember, like, I think this is the exciting part why everyone's so fired up is, like, once you feel that power, just imagining all the possibilities.
我觉得你讲得很好，始终在线、连接你的数据，并能采取行动。
And and I think you put it well, like, always on connected to your data and able to take action.
这些可以说是整个体系的三大基础。
That those are kind of the three foundations of this whole thing.
而且，我们还不知道它该叫什么，还没人给它命名。
And again, I I we don't know what no one has named it yet.
我一直在绞尽脑汁地想。
I've been, like, racking my brain.
这就像自主的知识工作，OpenClaw。
It's it, like, autonomous knowledge work, OpenClaw.
我不知道这个名称会不会被采纳。
I don't know if it's gonna stick.
也许这可以叫OpenClaw，但这就是工作本身。
Maybe that could be the OpenClaw, but, like, that is the work.
Harness IV。
Harness IV.
Harness IV 就是让我们接管。
Harness IV is the let's take over.
我觉得，但我的确觉得，是的。
I think but but I do think yeah.
我完全同意，这不是消费者与企业之间的对立。
It's I I fully agree it's not consumer versus enterprise.
我认为，每个人都能用它来构建和完成很多东西。
Every person, I think, will have a lot of things that they will be able to build and do with it.
所以我同意，这是这场战役的核心部分。
So I agree it's the central part of the battle.
我只是想说，我们接下来要谈到那个色情聊天机器人的关闭问题。
I just mean more, like and we're gonna get into the shutdown of the erotic chatbot.
但就连广告业务，我这一周都在拉斯维加斯参加ShopTalk，OpenAI那边也大量讨论了商业相关话题。
But, like, even the advertising business, I was just at ShopTalk in Las Vegas all week, OpenAI, ton of talk around commerce.
再次强调，这变成了一件有趣的事：它既是消费者导向的，也是企业导向的——你有零售商，但也有最终消费者可能在上面购物。
Again, it becomes like one of those interesting things that it's consumer, it's also enterprise, be you have the retailers, but you also have the end consumer who will be shopping on it potentially.
所以我认为我同意这一点，但作为一个组织，如果一开始就推出这么多面向消费者的功能，你还能否专注于并推广一个庞大的广告业务？当你试图做所有事情时，这才是我认为存在问题的地方。
So I think it I I agreed there, but I just think that overall as an organization to start cutting these very consumer friendly things, are you gonna be able to focus on and advertise a large advertising business when you're trying to do everything else, that's where I think there's issues.
我认为你会成功的。
I think you will be.
我的意思是，ChatGPT的年化收入已经达到了1亿美元。
I mean, ChatGPT is already at a $100,000,000 annualized run rate.
哦，这个简直让我崩溃。
Oh, that one killed me.
它已经上线六周了，我能说一句吗？
It's been out for six can I can I can I make one?
说吧。
Do it.
没人
No one
去吧，拉詹。
Go off, Ranjan.
产品才上线六周，就说年化经常性收入。
Say annualized recurring revenue when a product has been out for six weeks.
这根本算不上ARR。
Like, it's just not ARR at that point.
别让我们额外增加负担，别做外推。
Let it don't make us extra don't do the extrapolation.
就直接说它才上线六周，我们赚了多少钱？大概是多少？
Just say it's been out for six weeks and we've made what would that be?
比如九亿一千万美元之类的。
Like, $910,000,000 or whatever it is.
现在就只是这么点而已。
Like, that just that's all it is right now.
但它可能会变得更大。
And it could be much bigger.
我会的，那对他们来说会很棒。
And I'll be I'll be that'll be great for them.
但记者们，请别随便用ARR，除非有某种有意义的趋势。
But reporters, please don't use ARR unless there's some kind of meaningful trend.
就这样。
That's all.
我觉得我会继续在节目中这么做，就是为了惹你生气，拉詹，但我已经非常明确地跟她沟通了。
I think I'll keep doing this do it doing it on this show just to annoy you, Ranjan, but I I I messaged her loud and clear.
让我用一句话结束这个环节：这一切在理论上听起来都很好，但我和很多人面临的问题是信任问题——我希望AI能为我做所有这些酷炫的事情，但我并不信任它能访问我的Gmail、日历和桌面等所有东西。
Let let me let me end this segment with one thing, which is this all sounds good in theory, but the problem that I have and the problem that many people have is a trust problem, where I want the AI to do all these cool things for me, but I do not trust it to have access to my Gmail and calendar and desktop and all these things.
显然，像一些领先的模型和提供商，甚至像OpenCloth这样的公司，我们也建议你不要在没有采取一些预防措施的情况下这么做，比如在独立的机器上运行。
And obviously, like some of the leading models, like the leading providers, even OpenCloth, like like, we don't recommend you do this without you know, some some precautions like running on a separate machine.
你认为这个信任障碍能被克服吗？
Do you think that that trust barrier is ever gonna be overcome?
是的。
Yep.
百分之百。
100%.
我的意思是，我自己也看到了。
I mean, I I I see it myself.
等等，实际上。
Well, actually, hold on.
为了更细致地说明，我每天晚上7点都会收到一个清单，列出所有今天发送且超过24小时未回复的邮件。
To add nuance to it, like, I I have something that just 7PM every day, here's all the emails that you have not answered that are, like, from today and then greater than twenty four hours old.
并根据你整个Gmail历史，提供一个建议的回复内容。
And here is a suggested response based on your entire Gmail history.
所以我确实会收到这个提醒。
So I get this.
但我还没让它自动发送邮件。
I don't have it send it the email yet.
也就是说，我还没到那个地步——按下按钮，把邮件发给所有人。
Like, so that so I've not gotten to where I'm actually, like, press button, send the email to everyone.
但当你开始看到它逐渐影响你对回复结构的偏好时，我相信确实存在一种可能：我会直接让它自动发送回复。
But as you start seeing it start actually kind of, like, tweaking you the how you want the responses structured, I do think there is a world where I would just have it send the response.
所以我认为，信任是随着时间推移和质量提升而建立的。
So I think that there like, the trust comes with time and quality.
而且，你看到得越多，越能明白哪些问题不该问，哪些数据会出问题、导致回答质量低下，这种能力将成为最重要的技能之一，也正是人们建立信任的方式。
And, like, the more you start to see and the more you also start to understand what questions not to ask or where data is gonna be bad and you're gonna get, like, a subpar answer, that is kinda gonna be one of the most, I think, important skills, but also that's how people will build trust.
这就是为什么我总是在早上7:10、7:12左右收到你的邮件。
This is why I always get emails from you at about 07:10, 07:12.
对吧？
Right?
这个说法，比如
The clause, like
这太棒了。
That's wonderful.
来吧，拉詹。
Come on, Ranjan.
我们一定要在这个时间安排一下。
Let's make sure to schedule at this time.
好吧，我们马上就能看到这会走向何方了，因为我们将分别从OpenAI和Anthropic各获得一个主要模型。
Well, we're about to find out really where this is gonna go because we have two major models coming from both, one each, one from OpenAI and one from Anthropic.
我们先从Anthropic开始。
Let's start with Anthropic.
这周《财富》杂志上有一则非常有趣的故事。
There's this very interesting story in Fortune this week.
Anthropic 承认在一次意外数据泄露后，证实了正在测试一款新 AI 模型，该模型代表了能力上的重大突破。
Anthropic acknowledged test acknowledges testing a new AI model representing step change in capabilities after accidental data reveals, data leak reveals its existence.
该公司表示，在一次数据泄露暴露了该模型的存在后，Anthropic 正在开发并已开始让早期访问客户测试一款比以往发布过的模型更强大的新 AI 模型。
Anthropic is developing and has begun testing with early access customers a new AI model more capable than it has released previously, the company said following a data leak that revealed the model's existence.
Anthropic 的发言人表示，该模型代表了 AI 性能的重大飞跃，是我们迄今为止构建的最强大的模型。
Anthropic spokesperson said the model represented a step change in AI performance and was the most capable we've built to date.
该公司表示，该模型目前正在由早期访问客户进行试用。
The company said the model is currently being trialed by early access customers.
一份草稿博客文章曾在周四晚上之前存在于一个未受保护且可公开搜索的数据库中，文中称该模型名为 Claude Mythos。
A draft blog post, that was available in an unsecured and publicly searchable database prior to Thursday evening said the model is called the Claude Mythos.
现在，该公司认为它带来了前所未有的网络安全风险。
Now the company believe it poses unprecedented cybersecurity risks.
Mythos 也被称作 Capybara。
Mythos has also been called Capybara.
在文件中，Anthropic 表示，Capybara 是一个新层级的模型，比我们之前的 Opus 模型更大、更智能，而 Opus 模型至今一直是我们最强大的模型。
In the document, Anthropic says Capybara is a new name for a new tier of model larger and more intelligence than our Opus models, which were, until now, our most powerful.
与我们之前的最佳模型相比，Capybara 在软件编程、学术推理和网络安全等测试中得分显著提高。
Compared to our previous best model, Capybara gets dramatically higher scores on tests of software coding, academic reasoning, and cybersecurity among others.
哇。
Wow.
所以，我们可能正在见证一种新型模型的出现。
So we could be seeing a new class of model.
你知道，Anthropic 当然有它的三个模型：Sonnet、Opus，还有一个叫什么？
You know, Anthropic, of course, has its three, Sonnet, Opus, and what's the other one?
Haiku。
Haiku.
哦，是的。
Oh, yeah.
对。
Yeah.
是的。
Yeah.
所以现在我们可能正在
So now we might be
变得便宜了。
getting cheap.
是的。
Yeah.
我们可能正在获得Capybara。
We might be getting Capybara.
你对这个有什么快速的反应？
What do you what what's your your quick take reaction to this?
我的意思是，现在模型出现了一个跨越式进步，自从我们讨论这个单独的节目以来，大家都大致明白了这场竞争的本质，我认为这将非常有趣。
I mean, I am a step change in models now that when we've been talking about this solo episode that now we all kinda know what the battle is, I think will be very, very interesting to see.
再说一遍，当你一方面担心网络安全或识别这些风险，另一方面又说它在软件编码、学术推理和网络安全测试中的得分大幅提升时，仍然很难真正理解他们到底想走向哪里。
Like, again, it but it's still hard to understand when you both worry about say that you're worried about cybersecurity or recognizing these risks, but then say it gets dramatically higher scores on tests of software coding, academic reasoning, and cybersecurity, it's still just, I don't know, kinda difficult to actually parse out where they're going with this.
你认为，当你第一次用Capybara产出一些作品时，会是什么感觉？
What what do you think what's your prediction on what it will feel like the first time you crank out some some work on Capybara?
我其实一直在想这个问题，因为你知道，我们一直坐在这里，逐个审查每一个看似微小的更新。
Well, I have been thinking about this because, you know, we've been, like, sitting here and we review, like, every update that seems incremental.
对吧？
Right?
比如，它在这方面的表现好了一点，那方面的表现也好了一点，又在别的地方进步了一点。
Like, oh, it got a little bit better at this, got a little bit better on this, got a little bit better on this.
但这种进步开始让人感觉是在累积了。
And it's starting to feel like it compounds.
你明白我的意思吗？
You know what I'm saying?
我们从2022年的ChatGPT开始，那时候它还有各种各样的缺陷。
Like, we started with ChatGPT in 2022, and it had all these sorts of flaws.
但随着时间推移，这些问题都被逐步修复了。
And over time, they've been patched up in a way.
展开剩余字幕（还有 480 条）
所以我这周才开始琢磨这件事——所有人工智能领域的CEO都在说，当前行业正处在指数级发展的阶段，这就是我开始思考的契机。
And so, like, I just started to think about it this week in terms of, like, all the AI CEOs say that there is this, like, exponential happening.
而或许，这些一点一滴的积累，正是实现这种指数级增长的路径。
And maybe that's just the way that you get that exponential.
对吧？
Right?
举个例子来说，就好比复利的道理，是这么回事。
Like, you know, with your with interest, for instance, like, alright.
就像是你的投资先赚了6%，然后就连上季度赚的那6%也一起，又给你带来了6%的收益
There's, like, 6% on on your investment and then another 6% on your investment and that 6% that you got last quarter.
然后紧接着，资金就突然开始真正地快速增长起来
And then and then all of a sudden, that that starts to really grow.
而现在这些AI模型，似乎也正在经历同样的发展过程
And it seems like that might be what's happening with these AI models.
可能我这么想实在是太乐观了吧
I Maybe I'm being overly optimistic.
不。
No.
不。
No.
不。
No.
但那甚至算不上乐观。
But but that's not even optimistic.
那是现实。
That's realistic.
那就像
That's like
对。
Right.
复利积累价值的增加，实际上正是这种情况的发展方式。
Like compounding accrual of value coming is actually the way this is playing out.
但营销是在那之前做的，比如GPT-3和GPT-4。
But the marketing was done before that, like, GPT three and GPT four.
所以，一切都必须是革命性的飞跃，然后当事实并非如此时，人们就感到失望。
So, like, that everything had to be revolutionary in a step change, and then people were disappointed when it wasn't.
所以我认为这可能是看待它的正确方式，而人们可能并不这么想，也许这仅仅是因为营销上必须大肆宣传的局限性。
So I do think that's that could be the right way to look at it, and and people don't and maybe it is just a marketing limitation that they have to make a big deal.
但如果每个人都能这样谈论它，那就更好了。
But it would be kinda nice if everyone actually just spoke about it like that.
比如，这是我们的更新日志。
Like, here's our release notes.
它确实稍微好了一点。
It's it's definitely a little better.
然后你能做更多事情，这才是我们真正应该关注的。
And then you can do a lot more, and that's all we should really focus on.
他们永远都不会那样说。
They will never ever speak that way.
不。
No.
不。
No.
不。
No.
当然不是。
Of course not.
肯定的。
Guaranteed.
是的。
Yeah.
这是另一个。
Here's another one.
OpenAI，新模型。
OpenAI, new model.
它即将发布。
It's calling coming out.
它叫Spud。
It's called Spud.
OpenAI首席执行官萨姆·阿尔特曼表示，公司已经完成了其下一代主要AI模型的初步开发，该模型代号为Spud。
OpenAI CEO Sam Altman said the company has completed the initial development of its next major AI model code named Spud.
他告诉员工，公司预计几周内将推出一个非常强大的模型，团队相信它能真正推动经济发展。
He told staff that the company expects to have a very strong model in a few weeks that the team believes can really accelerate the economy.
他还补充说，进展比我们许多人预期的要快。
He added things are moving faster than many of us expected.
你知道有趣的是什么吗？
You know what's interesting?
这就像是，现在我要说得特别乐观了，但我得提醒自己别太乐观。
It's like just as the and now I'm gonna really sound optimistic, and I'm trying to check myself.
但就在世界开始应用当前的模型，并逐渐发现它们能完成以前几代模型无法做到的事情时，这正引发一种可能性的爆炸式增长。
But, like, just as the world implements today's models and is starting to find that they can do things with them that they really couldn't do with the previous generations, and that's leading to this, like, explosion of possibilities.
等等，
It's like, wait.
他们正在开发比这些更好的模型？
And they're building better models than these?
他们说这些是巨大的飞跃。
Some that they say are sizable leaps.
这真是一个让你坐下来惊叹的时刻，太疯狂了。
Like, it is one of those moments where you, like, sit back and just go, this is crazy.
我的意思是，确实有这种感觉。
I mean, it feels it definitely feels like that.
但不好意思。
But sorry.
我得停一下，说说这个叫‘Spud’的名字，难道你们没觉得OpenAI在搞什么鬼吗？
I I just have to stop for a moment and say the name Spud did not jump out to you as what the hell is going on in OpenAI sorry.
Anthropic那边可是藏着Mythos和Capybara呢。
Anthropic sitting on Mythos and Capybara.
我不知道。
I don't know.
那Spud是他们模型的代号吗？
That's kinda like but Spud is the code name for their model?
这消息从哪儿来的？
Where is this coming from?
一点也不令人振奋。
Not exactly inspiring.
不。
No.
不。
No.
他本人真的被认为能够极大推动经济发展。
He's literally, the team believes can really accelerate the economy.
他甚至没说，你能多做一些多步骤推理。
He's not even saying, like, you'll you'll be able to do a little bit more multistep reasoning.
他又在说，就像我们刚才讨论的，一切都得宏大无比，能真正推动经济发展，还叫Spud。
He's being again, as we were just talking about everything has to be grandiose and can really accelerate the economy, and it's called Spud.
我不知道。
I don't know.
也许Can
Maybe Can
你可以想象一下，你在上班，经理走过来对你说：得告诉你个事，我们要用Spud取代你。
you imagine you're at your job and your manager walks over, and they're like, gotta tell you, we're replacing you with Spud.
这会伤我自尊，但也许真会发生。
That would that would hurt my feelings, but it might happen.
也许你觉得这太糟糕了，反而让我永远记住了这个名字，而不是那些神话之类的东西。
Maybe do you think this is it's so bad that it actually makes me I will never forget the name versus mythos and stuff.
也许它就会自然而然地变成一首俳句。
Maybe it'll just kind of fall into haiku.
我一时想不起来，但叫Spud。
I couldn't remember off the top of my head, but Spud.
你会记住的。
You'll remember.
别管它叫GPT六。
Don't even call it GPT six.
直接推出我们新一代的模型。
Just come out with our new class of models.
就叫Spud系列模型。
Just Spud family of models.
还有Spud。
And Spud.
Spud。
Spud.
但你看。
But look.
咱们说清楚点。
Let's let's let's be clear here.
这只是一个代号。
It's just a code name.
OpenAI 不会发布什么 Spud。
It's not like OpenAI is gonna release Spud.
不是。
The No.
不是。
No.
我说的是产品
I'm saying the product
的名字叫 Spud。
name is Spud.
而代码名呢，但再说一遍，我所接触或参与过的大多数科技项目，都有一个相对雄心勃勃的名字，或者说，那种名字通常很宏大、很霸气。
And then I code name, but, again, most tech project like code names that I've ever come across or been part of, everyone has a somewhat ambitious name or, like, it's kind of, like, strong, grandiose, big name.
所以这个名称真的让我特别在意。
So that's why this one really jumped out at me.
但我现在还挺喜欢它的。
But I kinda like it now.
那些为五角大楼项目命名的人，现在也开始起代码名了。
The same people who are branding the Pentagon operations have come and started to code names.
不。
No.
但是这个
But the the
OpenAI 的产品。
The OpenAI products.
不。
No.
“史诗怒火”这样的名字才更符合神话风格。
The epic fury would be more in line with mythos.
“Spud”？我不确定。
Spud, don't know about that.
我不知道。
I don't know.
好吧，我们很快就会知道的。
Well, we'll find out soon enough.
好的。
Alright.
各位，如果你们喜欢看兰詹生气的样子，那就等等吧，下半场我们会聊到Siri。
Folks, if you are enjoying, seeing Ranjan riled up, well, just wait till the second half where we talk about Siri.
我们将会回来，希望能带来一些关于Siri发展方向的更好消息，但我不能保证任何事情。
We'll be back with hopefully some better news about Siri's direction, but I can't promise anything.
广告后马上回来。
And we're back right after this.
广告后我们再次回到这里，欢迎收听科技大讲堂周五特别版。
And we're back here on big technology podcast Friday edition.
Siri正在朝着新的改进迈进。
Siri is on a way on its way to a new improvement.
这是来自彭博社的消息。
This is from Bloomberg.
苹果计划在iOS 27更新中开放Siri，使其支持竞争对手的AI助手。
Apple plans to open up Siri to rival AI assistance in iOS 27 update.
该公司正准备在即将推出的iOS 27操作系统更新中，对Siri进行重大改造以实现这一变化。
The company is preparing to make the change as part of a Siri overhaul in its upcoming iOS 27 operating system update.
目前，Siri已能通过与OpenAI的合作接入ChatGPT，但苹果现在将允许其他竞争性服务也实现同样的功能。
The assistant can already tap into ChatGPT through a partnership with OpenAI, but Apple will now allow it allow competing services to do the same.
苹果正在开发新工具，使通过App Store安装的AI聊天机器人能够与Siri助手集成。
The company is developing new tools to allow AI chatbots installed via the App Store to integrate with the Siri assistance.
这些聊天机器人还将与即将推出的Siri应用以及Apple Intelligence平台的其他功能协同工作。
The chatbots will also work with an upcoming Siri app and other features in the Apple intelligence platform.
我先简单分享一下我的看法，然后你接着发挥吧，Ranjan。
I'll just quickly share my perspective here and then, you know, let you riff off it, Ranjan.
人们原本以为，这或许意味着Siri得救了，但随后我意识到，这不过是延续了以往那种令人失望的体验——你依然得费劲去找ChatGPT在Siri里的入口。
People were, and I initially took this as maybe Siri is saved, but then I realized that it's just gonna be the same disappointing user experience that you use to find ChatGPT in Siri.
换句话说，还是同一个Siri，只是能通过它访问一些新功能，而不是一个真正经过精炼的COD之类的版本。
So in other words, same Siri, new stuff you can access with it, not a Siri that is actually, you know, a a distilled version of COD or something like that.
这让我对iPhone上AI助手的体验更加失望了。
And that just makes me even further disappointed in what we're gonna get on the iPhone in terms of AI assistant.
我会尽量克制自己，努力保持冷静，但看到这个描述，我觉得‘Spud’这个名字简直堪称诗意——因为我读完简介后，完全同意。
I will refrain and try to remain from I will try to remain calm, but this was this made Spud look like pure poetry of a name because when I'm reading the synopsis, I agree.
之前有很多关于Siri的令人期待的消息，让我一度质疑自己对它的厌恶，甚至觉得它还有可能迎来转机。
There's been a lot of, like, very promising things about Siri that have made me question my hatred for it and think there is a possible future again.
这反而更让人困惑了，因为如果Gemini成为Siri的基础核心并真正让它变得有价值，那确实令人兴奋。
Like, this got even more confusing because Gemini being the kind of base foundation on Siri and actually making it valuable is kind of exciting.
你上一次实际使用Siri中的ChatGPT集成功能是什么时候？
This not only like, when's the last time you actually used the ChatGPT integration in Siri?
我不用。
I don't use it.
我的意思是，我为什么不去直接用ChatGPT应用呢？
I mean, why would I do anything but go to the ChatGPT app?
是的
Yeah.
对
Yeah.
但然后，实际上这里的科技，比如Siri真正连接了应用，或者他们可能称之为技能的东西。
So but then, like, the actual technology here, like Siri actually having connected apps or kind of I think they called them maybe skill.
多年来，一直存在试图在Siri中集成特定应用功能的尝试。
There was something for many years, like, trying to actually integrate specific functionality within apps has existed within Siri.
所以，像这样直接以查询方式操作，如果意味着Siri能直接为你朗读答案或显示结果，而不是打开目标应用，就会减少一些摩擦。
So, like, just having that as the query, I guess it'll be a little less friction if it means that Siri will then be able to directly read out the answer to you or have it appear rather than opening up the target app.
但，我不确定。
But, like, I don't know.
这确实是。
This is yeah.
我同意。
I agree.
这太可怕了。
This is horrifying.
如果苹果的Siri还停留在这个水平，那它必须足够优秀才能与这些应用竞争。
Like, if this is still where their heads Siri has to be able to be good enough to compete with those apps.
它必须做到这一点。
Like, that it just has to.
它不应该触发其他东西，我以为他们正在解决这个问题。
It should not trigger some other thing, and I thought that's what they're working on.
所以，这意味着这将成为某种功能。
So the fact that this is gonna be some kind of thing.
实际上，最后一点，这种方法应该能让苹果通过App Store从第三方AI订阅中赚取更多收入。
And actually, last line, this approach should allow Apple to generate more money from third party AI subscriptions through the App Store.
这才是最令人沮丧的部分。
That was the most depressing part.
如果这只不过是：通过App Store购买你的Gemini，然后我们抽成。
That if this is just like, well, buy your Gemini through the App Store, and then we're gonna take a cut.
这确实令人担忧。
This is this is troubling.
我不知道。
I don't know.
我们走着瞧吧。
We'll we'll we'll see.
我仍然愿意给他们一次机会，但并不乐观。
I'm still I'm gonna give them give them a chance here, but not positive.
距离WWDC还有两个月。
Two months till WWDC.
那就等等看吧。
So let's see.
我们再看看。
But Let's see.
目前我对正在发生的事情并不怎么热情。
Not very enthusiastic right now about what's going on.
不过，你先说吧。
Although Do you Go ahead.
我本来想问，你觉得WWDC上，苹果会全面展示Apple Intelligence和Siri的愿景吗？
I was gonna ask, do you think WWDC, they're gonna have, like, a well fleshed out vision of what Apple intelligence is, what Siri is?
不会。
No.
如果今年完全听不到关于Apple Intelligence的消息，我也不觉得意外。
I wouldn't be surprised if we didn't hear anything about Apple intelligence this year as well.
但即使是Siri的全面升级之类的内容呢？
Well, you but even the Siri overhaul and stuff.
对吧？
Right?
你觉得他们不会吗？
Like, you think they're no?
什么都没有。
Nothing.
不。
No.
也许几分钟。
Maybe a few minutes.
他们必须发布产品。
They have to release products.
他们必须让产品自己说话。
They have to let the products do the talking.
他们不能再告诉我们未来会有什么了。
They cannot tell us again about what's coming.
这似乎看起来是
That seems to be
我同意。
I agree.
我同意。
I agree.
是的
Yeah.
好的
Okay.
我同意
I agree.
让我们来看看我们在硅谷的其他朋友，也就是Meta和YouTube。
Let's go to our other friends over here in Silicon Valley Meta and YouTube.
这实际上是一件大事。
This is a pretty big deal, actually.
加利福尼亚州一家法院裁定，他们因通过其产品损害年轻用户而负有责任。
The a case a court in California found them both liable for harming a young user with their futures with their futures.
这是来自《纽约时报》的报道。
This is for The New York Times.
Meta和YouTube在具有里程碑意义的社交媒体案件中被认定存在过失。
Meta and Youtube found negligent in Landmark social media case.
它说Meta必须支付赔偿，是的。
It says Meta must pay so they yeah.
他们认为，这一里程碑式的判决可能会让社交媒体公司面临更多关于用户福祉的诉讼。
They've that they the Landmark decision could open up social media companies to more lawsuits over users' well-being.
Meta必须支付420万美元的综合补偿性和惩罚性赔偿，YouTube必须支付1180万美元。
Meta must pay 4,200,000.0 in combined compensatory and punitive damages, and Youtube must pay $11,800,000.0.
这起具有风向标意义的案件由一位如今20岁的女性提起，她指控社交媒体公司设计了像香烟或数字赌场一样上瘾的产品，导致她出现焦虑和抑郁，法院在本案中对此予以认定。
The bellwether cage switch was brought by a now 20 year old woman had who had accused social media companies of creating products as addictive as cigarettes or digital casinos, and they led to anxiety and depression in the court found in this person's case.
还有成千上万起类似的诉讼正在陆续提起。
And there are thousands more of these lawsuits coming through.
就在这件事发生时，我刚在CNBC上露面，我当时就说：你们可别输掉这些案子，否则后面还有更多。
Now I was on CNBC just as this happened, and I was like, you don't wanna lose these cases because you're gonna have others.
和我一起上节目的另一位嘉宾说，本质上，这对他们来说是个胜利，因为赔偿金额太小了。
And the other panelist that was with me was like, basically, you know, this is a win for them because the amount was so small.
只有，我想，几百万美元，总共600万美元。
It was only, I think, just a few million dollars, 6,000,000 total.
但当你输掉一场官司时，你就为其他败诉打开了大门。
But when you lose, you open the door for other losses.
有些人看到这个赔偿金额后可能会想，既然他们能拿到这么多，我们也能拿到更多。
And some might see the award and say, if they got that, we can get even more.
果然，Meta的股价在本周剩余时间里暴跌了。
And lo and behold, Meta stock has just tanked over the rest of the week.
那么，拉詹，你如何看待Meta在此次败诉后可能面临的连带责任问题？
So how do you read this, Ranjan, in terms of the the potential for successive liability for Meta after this problem, after this loss?
但等一下。
But hold on.
我正想说，因为我还看到他们被裁定支付3.75亿美元的民事罚款。
I'm trying to because I had also seen that they're ordered to pay 375,000,000 in civil penalties.
是的。
Yes.
所以，这是他们在新墨西哥州本周输掉的另一桩案子，但不知为何，这起案件没有被广泛讨论，我对此有个猜测。
So that's a separate case they also lost this week in New Mexico, which for some reason wasn't talked about as much, and I have a theory as to why it wasn't talked about.
等等。
Wait.
好吧。
Okay.
等等。
Hold on.
让我理一下，顺便提一下，Meta的公关负责人安迪·斯通，我看到他发过一条推文，基本上说，这钱没多少。
Walk me through because and just to note, like, Andy Stone, the head of comms over at Meta, I'd seen a tweet where, like, basically, he was like, it's not that much.
这还不到州政府要求的3.75亿美元的一小部分，那金额简直吓人。
It's a fraction of what the state sought even for the 375,000,000, which was just terrifying 2 to me.
十亿。
Billion.
但重点是，如果你输了，就会让自己陷入更多损失的风险。
But the part of the point is that, like, if you lose, you open yourself to self up to further losses.
所以，这对我来说就是这样。
So that's that's to me.
所以，我想我可以从我的角度来谈一谈。
So I guess I can sort of give my perspective here.
这里的问题是，这项裁决基本上是在告诉他们，不能再把第二十三条当作挡箭牌了。
The issue here is that the ruling is basically telling them that you can't use section two thirty as a shield anymore.
这并不一定意味着你们因为平台上发布的内容而被起诉。
It's not necessarily that you're being sued over the content on your platforms.
你们是直接因为平台的设计方式而被起诉，法院认为，是的，你们不能再以第二十三条为借口来逃避责任，这条法规原本是保护论坛所有者免于承担用户发布内容的责任的。
You're being sued directly over the way that you design them, and courts are finding that, yeah, that you can't hide behind section two thirty, which protects, like, forum, owners from the content that people put on top of them.
现在我们可能会看到成千上万起类似的案件。
And now we could potentially see thousands of similar cases.
这确实是个问题。
That is a problem.
是的。
Yeah.
我认为，老实说，自2015、2016年以来，我就一直希望这一点能被认识到。
I think I mean, and honestly, this is something since twenty fifteen, sixteen I've been hoping would be recognized.
所以，虽然晚了十年，但我对此仍然非常乐观。
So it is, I mean, ten years late, but still, I'm very positive about it.
但我觉得，尤其是不仅责任问题打开了大门，就连这个较小的案件也认定，报告和发现证实了一种新颖的法律理论：社交媒体平台或应用可能导致人身伤害。
But I think, like, especially, yeah, the it not only being liable opening the door, the fact that the the smaller case still found that and it said, like, the reporting, the finding validates a novel legal theory that social media sites or apps can cause personal injury.
我的意思是，这意义重大。
I mean, that's a huge deal.
是的，就是这样。
Like, it's Yeah.
真正有趣的是，责任并不在于实际伤害本身，而是其他方面。
It's that the actual being liable not, like, for the actual injury side of it, I think, gets really interesting.
你觉得这件事会如何发展？
Where how do you see this playing out?
至少在过去十年里，每一个这样的案件都来来去去，而Meta始终在回避。
Like, for for how many year for at least a decade, every one of these has come and gone, and Meta keeps meta ing.
还有Instagram，我认识的每个人整天都在刷它。
So and Instagram, every single person I know just all day long scrolls on it.
他们仍然在做着一直以来都在做的事，并且做得非常好。
Like, they're still they're still doing what they've always done and have been doing a very good job at it.
那你如何看待这件事呢？
So how do you see this?
你认为这真的会影响他们的业务吗？
Do you see this actually affecting their business?
这周我看到一个法学院教授的精彩访谈，这位教授的名字我想不起来了，抱歉，他基本意思是，Meta必须上诉，而且他们确实会上诉，因为这实际上为全国设定了一个先例，即第230条是否能为你提供保护。
I saw a great interview with a law school professor this week where the law professor, and his name escapes me, so apologies, was basically like, Meta has has to appeal this, and they will appeal this because this is effectually effectively setting precedent in the country for whether section two thirty can work to protect you or not.
而法院刚刚裁定：不行。
And the court just found no.
这位教授认为，这件事最终会一路上诉到最高法院，然后最高法院会明确界定第230条的边界。
So the way that this professor sees it going is that it goes all the way to the supreme court, and then the supreme court rules, you know, specifically on the boundaries around section two thirty.
如果最高法院——我只是假设，假设事情按照这位教授的想法发展的话。
And if the Supreme Court and I'm just saying, like, let's say it goes out the way that this guy thinks.
如果最高法院裁定第230条不具有保护性，那影响的就不只是眼下这数千起诉讼了。
If the Supreme Court rules that section two thirty is not protective, it's not just the thousands in action now.
可能还会涌入更多。
It could be even more that come in.
而且，你知道，我的意思是，这在某种程度上令人担忧，比如，好吧。
And, you know, I mean, I guess that's somewhat concerning in terms of, like, alright.
如果你是一家内容公司，你现在是否要承担责任？
If you're a content business, are you now liable?
比如，我们有Discord，幸运的是，大家在那里都表现得很好，积极贡献。
Like, you know, we have the Discord, and luckily, everybody's pretty well behaved and happily contributing there.
但我现在是否要对所有发言负责？
But am I now liable for everything said?
在我们的Discord平台上，这打开了一个潘多拉魔盒，可能对互联网，尤其是对Meta的业务造成损害。
In our Discord instance, it just opens up this Pandora's box that could cause damage to the internet and especially to Meta's business.
关于Meta的业务，我再补充一点想法。
And on the Meta's business side, just one more thought here.
今年，他们在人工智能基础设施上的支出高达1150亿到1350亿美元。
They are spending a lot of money this year, 115 to 135,000,000,000 on AI infrastructure.
而市场之所以表面上允许他们这样做，是因为他们赚了这么多钱。
And the reason why the market has, quote, unquote, allowed them to do that is because they're making all this money.
如果你开始看到利润率被这些诉讼一点点蚕食，就像千针刺痛一样，那么你投入创新的能力就会下降，因为你的利润率降低了，而市场也不会再像以前那样给予你宽容。
If you start seeing your margin trimmed by a, like, sort of death by a thousand needle pokes from these lawsuits, then all of a sudden, your ability to spend on innovation goes down because your margin comes down and the market doesn't give you the leeway that it might have otherwise.
所以，这本身就是一个潜在的问题。
So that's a that's a potential problem there.
好的。
Okay.
还有几点想法。
So a couple of other thoughts.
我觉得，他们一直没受到任何影响，已经太久太久，因此我几乎难以相信他们真的会或能被影响到。
See, I I guess it's just been so long where they have not been affected that it's just still it's almost unbelievable for me to think that they will or could be.
我的意思是，这已经好多年了，马克·扎克伯格上次在国会听证会上亮相是什么时候？
I mean, that's just it's been years and years of when was Mark Zuckerberg in front of congress way back?
那大概是二十年前了吧。
Well, that was, like, twenties
2018年，2017年。
Twenty eighteen, 2017.
2017年，2018年。
Twenty seventeen, eighteen.
是的。
Yeah.
还记得当年那些经典的梗吗？
Like, remember good memes back in the day.
但我觉得
But, like, I think
我当时在场。
I was there.
哦，你没事啊。
Oh, you're okay.
所以是的。
So yeah.
你看，那件事发生得太久以前了，但我记得有几件事让我印象深刻，比如香烟这个类比让我觉得挺有意思，因为我看到一些有趣的观点，这真的是个恰当的类比吗？
See, that and it was so long ago, but I think a couple of things that jumped at one, the cigarette analogy is interesting to me because, like, I saw, like, some interesting you know, like, is that really the right analogy?
我觉得《纽约时报》上有一篇关于这个的评论文章。
I think there was an op ed in the New York Times on this.
基本观点是，有好的一面，也有坏的一面。
Basically, the idea is, like, there is good and there's bad.
所以这不像香烟，我的意思是，也许你还能辩称香烟有好处。
So it's not like cigarettes, which, I mean, maybe you can argue there's good.
但总的来说，我认为大多数人甚至都不假装香烟有什么真正的益处，而社交媒体却可能带来整体的社会正面影响。
But, like, in general, I think most people are not even pretending there's any true good out of it versus, like, social media can be a net so societal positive.
它也可能带来很大的负面影响。
It can also be very negative.
但对我来说，我对此写过很多内容。
To me, though and I've written about this a lot.
我觉得算法才是真正的香烟或烟草。
Like, the algorithm is the cigarette or the tobacco.
不是内容本身。
Like, it's not the content.
甚至不是发布照片的核心技术。
It's not even, like, the core technology of posting a photo.
只是基于算法的内容。
It's just algorithmic based content.
如果这件事最终能让人重新关注这种危险，我会非常非常高兴。
And I think if this finally gets people back to talking about that as a danger, I'm very, very happy about it.
我觉得这是好事。
I think it's good.
但确实。
But yeah.
我要看到Meta的业务受到实质性影响，才会相信像修复Siri这样的事真的有作用。
I'll I'll just like fixing Siri, I'll believe there is a a material impact to Meta's business when I see it.
我知道我们没在法庭上，但我很好奇，你是否同意这里的判决结果。
I know we haven't been in the courtroom, but let's just I'm curious if you agree with the with the verdict here.
因为Meta的论点是青少年心理健康问题很复杂。
Because the meta argument is teen mental health is complicated.
这不能归结为一个应用。
It doesn't come down to one app.
你不能把所有问题都归咎于一个单一的应用。
You cannot blame everything on a single app.
我的意思是，显然在某些情况下，它们确实对青少年的心理健康问题有影响。
I mean, obviously, in some cases, they're contributing to teen mental health issues.
但另一方面，我认为说存在多种因素共同作用，而不仅仅是一个罪魁祸首，是有一定道理的。
But then again, there is some merit, I think, in saying that there's a combination of factors and not just one culprit.
你怎么看？
What do you think?
所以我会说，这已经是一场很长很长的抱怨了，我觉得。
So I will say, and again, this has been a long, long rant, I think.
就像2019年，我们已经从五个方面提出了改善社交媒体的建议。
It's like 2019, we've written the margins five ways to fix social media.
其中一个我仍然很喜欢，但我们永远得不到，那就是时间线默认应该是逆时间顺序排列。
One of them I still loved, which we'll never get, but it's at the timeline should be reverse chronological by default.
因此，不会有算法推荐内容。
So there's no algorithm suggesting the content.
因为在我看来，有一个真正的罪魁祸首。
Because to me, there is one culprit.
那就是内容的算法推荐。
It's the algorithmic recommendation of content.
就这么简单。
That's it.
无论是YouTube、Meta或Facebook、Instagram、TikTok，这整个平台的问题就在于此。
Like, whether it's on YouTube, whether it's on Meta or Facebook, whether it's on Instagram, TikTok, that's the entire platform.
正是它让人们变得极端，让他们感到糟糕。
That's what show it radicalizes people, makes them feel bad.
所以我认为，这里确实有一个罪魁祸首。
So I do think there's one culprit here.
我觉得这很有趣，比如说
I think it is interesting like
等等。
Wait.
稍等一下。
Hold on.
你是说，这个人的心理健康问题完全是由于算法造成的吗？
You're saying that this person's mental health issues, you would say, are entirely due to the algorithm.
我的意思是，这就像说吸烟是否导致肺癌一样？
I mean, that's like saying, is smoking responsible for lung cancer?
或者肥胖、环境因素和空气质量等等呢？
Or could obesity or environmental factors and air quality and come on.
我们都使用社交媒体。
We all use social media.
我们都喜欢，每个人都会，我只是不知道怎么说。
We all like, everyone like, it's just I don't know.
我想我确实很喜欢，很多时候我的朋友都说，我不会被社交媒体影响。
I guess I I do love, like, a lot of the times I have friends who are like, I am not influenced by social media.
我不受广告的影响。
I'm not influenced by the ads.
这些帖子并没有让我觉得错过了什么，或者觉得需要改善我的假期，但对我来说，我不知道。
The posts don't actually make me feel like I'm missing out on something or I need to improve my vacation or but to me, I don't know.
这难道不是对你来说最明显的事情吗？
Does is that not the most centrally clear thing to you?
嗯，我想这正是我想表达的，我明白你的观点。
Well, I guess this is the this is sort of I see your point.
这就像对Meta论点的反驳：吸烟并不会直接导致癌症。
It's like the counterargument to Meta's argument is it's not that smoking cigarettes lead directly to cancer.
而是吸烟是一种已知的致癌物。
It's that smoking cigarettes are a known carcinogen.
你患癌症的几率，因此在很多方面，烟草公司应对他们所导致的额外癌症病例负责，即使你无法一一对应地直接证明。
Your odds of getting cancer, and then therefore, in many ways, the cigarette companies are liable for sort of the additional cancer cases that they cause, even though you can't draw a straight line one to one.
也许今天更多孩子感到抑郁，是因为社交媒体？
And maybe there's there's a similarity with, like, are more kids depressed today because of social media?
如果你能证明这一点
If you can prove that
因为因果关系，尤其是在心理健康这个领域，几乎不可能被证实。
because causality, it's it is new, especially in this case with mental health is like, feels nearly impossible to prove.
我不知道。
I don't know.
也许你可以根据他们的使用数据，设法找出特定用户更直接的关联，尤其是当你关注个体时。
Maybe you could, like, based on their usage statistics, somehow start to, like, draw more of a direct correlation of that specific user, especially if you're looking at individuals.
但我的意思是，任何点开过YouTube推荐视频栏的人。
But, I mean, anyone who has clicked on the YouTube bright rail of, like, recommended recommended videos.
它只是加剧、放大，很多时候还激化了内容。
It's anything like, it just exacerbates, exaggerates, like, radicalizes in many case.
我的意思是，它的设计就是为了让你产生情绪，而让人产生情绪并保持参与的最简单方式，就是让他们感觉不那么好。
I mean, it's just designed to make you feel, and the easiest way to make people feel and stay engaged is to make them not feel great.
也许吧。
Maybe.
但他们一直在回来。
But they keep coming.
但假设是的，人们确实不断回来。
But you let's say, yeah, people keep coming back.
我不知道。
I don't know.
我想称它为‘末日滚动’，而我就是想进行末日滚动。
I wanna I I call it doom scrolling, and I want to doom scroll.
每次都是选择。
Every time choice.
如果你想玩体育博彩，这就是一个很好的成瘾机制。
That's what a good addictive if you want a sports scam, bet on sports.
如果你想 vaping，想抽烟，或者任何你选择的成瘾行为，对我来说都类似。
If you wanna vape, if you wanna smoke cigarettes or whatever your advice of choice might be, like, it's similar to me.
我想知道我的一个热门观点吗？
I don't Do do you wanna know one of my hot takes?
推特在2015年默认从逆时间顺序改为了算法推荐，每个人的默认信息流都变成了算法驱动的。
This is a this is a so Twitter changed from reverse chronological by default in the 2015, and to everyone's default feed was algorithmic.
对。
Right.
那么2015年到2016年间发生了什么？
And what happened through 2015 into 2016?
等等。
Or hold on.
让我确认一下确切的说法。
Let me get the exact Yeah.
这是否就是导致当前全球政治氛围的原因？
That that is that is responsible for the political climate globally right now?
是的。
Yes.
好的。
Okay.
对。
Yes.
首先，有几件事。
First of all, a couple of things.
第一，你描述的那种情况——有人抽烟、吸电子烟、赌博体育赛事、刷滚动页面，这基本上就是我的周末，我手里有电子烟和这个东西。
Number one, the what you described, you know, somebody smoking and vaping and sports gambling and spinning through reels, That's basically my weekend where I've got the vape and this thing.
你们在包装上标了20块。
Y'all putting 20 on the package.
贝德MGM。
Bed MGM.
不。
No.
听好了。
Look.
关于整个算法这件事，我想说的是，这对我来说一直都很奇怪，因为当时我是BuzzFeed的记者，我率先报道了Twitter将转向算法的消息。
I I one thing I'll say about the whole algorithmic thing is that was that's always gonna be weird for me personally because I I was a reporter at BuzzFeed at the time, and I got the scoop that Twitter was moving to an algorithm.
那是在，呃，星期五。
That was on, like, Friday.
等等。
It hold on.
2016年2月。
February 2016.
是的。
Yes.
所以这更符合我的理论。
So even more in line with my my theory here.
2016年2月10日。
02/10/2016.
我的意思是，你看。
Well, I mean, look.
你可以说，特朗普是所有候选人中把社交媒体玩得最好的。
If you you you could say that Trump I mean, out of all the candidates, he played social media well.
但我认为真正让他当选的是那些辩论，他在那里简直……嗯，谢恩·吉利斯对此有个特别好的段子。
But I think the thing that really put him in office were those debates where he just I mean, say Shane Gillis has, like, a pretty good bit on this.
比如，有一位候选人，我是兰德·保罗，我支持学校教育。
Just that, like, you know, he's, like, brought you know, one one candidate's, I'm Rand Paul, and I believe in schools.
然后特朗普就说：你是个彻底的失败者。
And then Trump was like, you're a complete loser.
所有人都觉得：你居然能这么干。
And everyone's like, you could do that.
不对。
No.
但因为它播出的时间，传播得更广了，然后
But but it traveled more because of the time And then
它之所以传播开，是因为‘让美国再次伟大’，但你说得对。
it traveled because of make But you're right.
这很有趣。
That's interesting.
是的。
Yeah.
有可能。
It's possible.
我不是说这是
I'm not saying it's
有可能。
possible.
人们会，那也是一个奇怪的选举年。
People would That was weird election year also.
我的意思是，不把话题扯得太远，但当时我在 BuzzFeed 工作时，确实报道过一些特朗普的集会，还被一个田纳西州共和党账号转发了，他们说主流媒体永远不会展示这些。
I mean, not to bring us all the way back to it, but, like, I definitely you know, I when I was also at BuzzFeed, I did some reporting on Trump rallies and, you know, got retweeted by this Tennessee GOP account that was like, you know, the mainstream media will never show you this.
这挺讽刺的，因为一方面，我当时本身就是媒体一员——我不知道你是否管这叫主流媒体，但当时的媒体确实包括我。
And which was funny because, a, like, I was part of I don't know if you call it the mainstream media, but the media at the time.
而且，那个叫田纳西州共和党的账号，当时在选举期间拥有大量粉丝并极具影响力，但它实际上是由圣彼得堡运营的。
And b, that that account, Tennessee GOP, which was massively followed and influential due during the election was run from Saint Petersburg.
但那是另一个故事了。
But that's a different story.
我们可以改天再聊这个。
Like, we can talk about that another time.
真是美好的时光啊。
What what a good good times.
但你没事就好。
But you're fine.
所以我拿到了独家消息，说周五会推出这个推特算法。
So I got that scoop scoop that that there there was was gonna gonna be this Twitter algorithm on Friday.
然后发生了一件大事。
And then there was a big thing that happened.
它被称为‘杀死推特’。
It was called rip Twitter.
我不确定你是否记得这件事。
I don't if you remember that.
就像，我的报道发布后的一个周末，有一百万人在推特上发布了‘RIP Twitter’。
Like, a million people tweeted RIP Twitter over a weekend after my story came out.
这导致杰克·多西表示：我们从未计划在下周推出算法信息流。
And that led Jack Dorsey to say, we were never planning to introduce an algorithmic feed next week.
随后，我的私信被大量留言淹没，人们说：‘你是个骗子。’
And then my mentions flooded with people saying, you're a liar.
你的职业生涯完了。
Your career is over.
失去公信力的感觉如何？
How does it feel to have no credibility?
我当时觉得自己完全被精神操控了。
And I thought I was totally gaslit.
我以为自己完蛋了。
I thought I was done.
然后他们在接下来的周二宣布了这一消息。
And then they made the announcement that Tuesday, the following Tuesday.
杰克。
Jack.
等一下。
Just a minute.
等一下，报道一下那家公司。
Just a minute reporting on that company.
这太疯狂了。
That was crazy.
好吧。
Alright.
是的。
Yeah.
那么这就是我的观点。
Should we So so that's that's my theory.
我坚持这个观点。
I'm sticking to it.
好的。
Okay.
我们聊聊科技股吧？
Should we talk about the tech stocks?
科技股这一周表现非常糟糕。
Very rough week for the tech stocks.
这是来自CNBC的报道。
This is from CNBC.
科技股经历了近一年来最糟糕的一周，受战争忧虑和Meta法律困境的拖累。
The tech stocks suffer their worst week in nearly a year driven down by war worries, meta legal woes.
我的意思是，微软股价跌了30%。
I mean, Microsoft is 30% off.
这太高了。
It's it's it's high.
30%。
30%.
你觉得这只是因为战争，还是说，这些公司对AI支出和短期利润不足感到越来越不安和担忧？
Do you think this is just war, or is it, like, a a growing uncomfortableness and unease around the spending and the lack of near term profits from AI for these guys?
所以
So
我认为本周发生的事情非常重要，我认为表面上看，是的。
I do I do think it is very important this week what's happening, and I I think that the kinda, like, headline side of it is, yes.
这周市场确实遭受了重创。
Like, the market's been getting cream this week.
科技股之前本来就已经经历了史诗级的上涨。
Like, tech stocks had been on it, like, epic runs anyways.
所以，稍微回调一下感觉非常自然。
So, like, giving a little back feels a pretty, like, a pretty natural thing.
但对我来说，由于当前AI领域许多活动都建立在循环融资的基础上，一旦科技巨头们真的开始遇到一点麻烦，就会引发一系列连锁反应。
But to me, because of all the circular financing that's at the foundation of a lot of what's happening in AI right now, because of all the kinda, like, follow on effects of just the the tech giants start actually being in a little bit of trouble.
这对整个行业意味着什么？
What does that mean for the industry writ large?
所以，我认为，尤其是微软，它们的表现不如其他公司，这可能是一个完全不同的话题。
So so I think, like and and again, Microsoft, I think, could be a whole other segment just in terms of why are they not doing as well as the others.
但给
But Give
用六十秒给我们讲讲微软的情况。
it give it to us in sixty seconds on on Microsoft.
不，不行。
Well, no.
我的意思是，这很明显。
I mean, I think it's it's clear.
它们已经落后了。
They're they have fallen behind.
没有任何令人兴奋的新产品推出。
There's no nothing exciting coming out.
他们拥有大量用户在使用Copilot，但人们并没有以任何有意义的方式付费并转化为付费订阅用户。
They have the installed base of everyone using Copilot, but people are not paying up and converting to paid subscribers in any meaningful way.
他们只是更换了Copilot的领导层。
They just replace Copilot leadership.
所以，关于整个AI领域，他们曾经拥有OpenAI这件事真是令人震惊。
So, like, on the whole AI thing, it is pretty crazy that they had, like, OpenAI.
他们最初是非常早期的合作伙伴，但现在却并没有取得任何显著的进展。
They were the the the partner at the beginning very early, and now still, they're not really anywhere notable.
你能想到微软在AI领域最近一次令人兴奋的突破是什么吗？
Like, what what's what's the last exciting thing from Microsoft in AI that you can think of?
Bing。
Bing.
我的意思是，他们确实有Bing，但他们本可以率先推动这一进程。
I mean, they had Bing, but, like, they they actually they were the first to they could have pushed this through.
是的。
Yeah.
我觉得大家正在逐渐接受这一点。
I think everyone's kinda coming around it.
也许这会成为一个很好的警醒。
And maybe it'll just be a good wake up call.
我确信，考虑到他们的用户基础和他们的身份，只要他们能搞明白，就一定会成为一股强大的力量。
I'm I'm sure given their installed base, given their, like, who they are, if they figure this out, they will be a force.
但我认为，市场已经开始稍微意识到这一点了。
But I think, like I think there, the market is recognizing it a bit.
好的。
Okay.
让我们以我们传统的‘产品/功能追悼会’来结束这一周。
Let's end this week with one of our traditional product slash feature funerals.
女士们、先生们，今天我们聚集在这里，悼念短暂而充满波折的‘OpenAI成人模式’，它已永久退出我们的世界，似乎短期内也不会回归。
And ladies and gentlemen, we're gathered here today to pay our respects to the short and and quite eventful life of the, OpenAI adult mode, which has left our world indefinitely and doesn't seem like it's coming back anytime soon.
据《金融时报》报道，OpenAI已计划无限期搁置推出情色聊天机器人的计划，因其员工和投资者对性化AI内容对社会影响的担忧，而决定重新聚焦核心产品。
For the Financial Times, OpenAI has shelf plans to release an erotic chatbot indefinitely as it refocuses on core products following concerns from staff and investors about the effect of sexualized AI content on on society.
萨姆·阿尔特曼的初创公司已在内部讨论是否彻底取消该功能的同时，推迟了成人模式的发布。
Sam Altman's startup has already delayed the release of its adult mode amid a internal discussions over whether to scrap the mile entirely.
这款性爱聊天机器人因可能助长用户对AI系统的不健康依恋，以及让未成年人接触不当的性内容而面临日益增长的反对声音。
The sexual chatbot faced growing pushback over how it could encourage unhealthy attachments to AI systems and expose minors to problematic sexual content.
安息吧，ChatGPT的成人模式。
Rest in peace, adult mode on ChatGPT.
或许我们的星球更好，因为你从未真正面世。
Perhaps our planet is is better off that you never saw the light of day.
亚历克斯，你对这件事有什么看法？
How do you feel about this, Alex?
陪伴一直是坎托里德学派关于AI未来的核心理念之一。
Companionship has been one of the cornerstones of the Cantorid School of the future of AI.
我不想当那个道德警察，但像和你的聊天机器人进行网络性行为这种事，我还是觉得不太合适。
Well, I don't wanna be the morality police sensation, be able to, like, have cybersex with your chatbot.
但说到OpenAI不该涉足的业务，这看起来就是其中之一。
But speaking of businesses that OpenAI shouldn't be in, this seems like one of them.
这简直打开了潘多拉的盒子。
It just opens up this whole can of worms.
我认为这是正确的选择。
I think this is the right choice.
你怎么看？
What do you think?
Meta，毫无疑问是正确的选择。
Meta, unquestionably the right choice.
当他们一提出企业版时，我就觉得，你不可能让情色聊天机器人到处乱跑，还假装人们会信任你。
Like, we've brought this up when the moment they said enterprise, I was like, you cannot have erotic chatbots running around and like, then, still pretend that people are gonna trust you.
但再说一遍，我不确定。
But but again, I don't know.
也许他们本可以做一些有趣的事情。
Maybe they could have done something interesting.
也许吧。
Maybe.
这种对行业创造力的集中，是否让我们失去了Sora原本那种怪异的特质，以及现在所有人都在做爪子时可能消失的色情聊天机器人？
This focus all the creativity in the industry, are we losing the the weirdness of Sora and potentially erotic chatbots now that everyone's just making claws?
说到这个，就像你知道的，潜在的竞争确实为其他聊天机器人提供商打开了大门，让他们可以利用底层技术开发自己的色情聊天机器人。
Well, speak of speaking of, like, you know, potential competition, like, it does open up the door for other chatbot providers to use some of the underlying technology and make this erotic chatbot of their own.
即使你不能在ChatGPT界面中使用它，也许你可以用基于GPT的成人模式聊天机器人，这样完全可以打造一个不错的初创公司。
Just because you can't use it within the ChatGPT interface, maybe you can use, like, a GPT based adult mode chatbot, and you can make a pretty good pretty good startup that way.
有个问题。
Question.
需要说明的是，我们不是律师，也不会假装是，但考虑到社交媒体场景中的责任问题，AI公司是否应该对其VAPIs生成的最终内容负责？
And with the disclaimer that we are not lawyers and will not pretend to do so, but given the idea around liability in the social media use case, should AI company be responsible for the end content created with its VAPIs?
他们对此有控制权，而且这实际上又回到了五角大楼和战争问题上。
They have control over the and again, I mean, actually, ties back to the Pentagon and the war question.
但，他们真的应该承担责任吗？
But, like like, should they be responsible?
对成年人？
For adults?
不。
No.
成年人应该签署声明，表明他们不知道这种对话会走向何方，而且他们不应承担责任。
Like, adults should sign off that they don't know where this is gonna go, and they shouldn't be liable.
但对于孩子，绝对应该。
But for kids, absolutely.
你怎么看？
What do you think?
等等。
Well, hold on.
你的意思是，如果其他服务通过API调用OpenAI的模型，然后成年人使用它进行成人聊天机器人对话？
You're saying OpenAI, if some other service is calling them there's a their models via API, and then it's adults having erotic chatbots.
这算是OpenAI在提供这项服务和内容吗？
Is that is OpenAI delivering that service and content?
他们应该为发生的一切负责吗？
Should they be responsible for what whatever happens?
而且，如果孩子在使用这个，那就是完全不同的另一回事了。
And yeah, definitely, if then kids are using this, that's a whole other thing.
服务提供方应该承担责任吗？
And should the service be liable?
OpenAI 也应该承担责任吗？
Should OpenAI also be liable?
好的。
Okay.
所以这是两个独立的问题。
So two separate questions.
是的。
Yes.
很好。
That's great.
这是个很好的问题，因为它和云托管的类比有点不同，但又不完全一样。
That is a great question because it's a little bit different than like, the comparison is cloud hosting, but it's a little bit different than cloud hosting.
对吧？
Right?
因为云托管就是你把东西存在这里，这让你能够做你想做的事。
Because it's like cloud hosting is you store your stuff here, and that enables you to do what you want to do.
而聊天机器人则是你使用这项技术来实现你想做的事。
Whereas, like, chatbot is like you're using this technology to do what you wanna do.
不对。
No.
不对。
No.
而且它是在主动生成新内容。
And it's actively generating new content.
我觉得是那个人，抱歉。
I think it's the person sorry.
我觉得是部署它的人。
I think it's the person that deploys it.
我不认为如果别人使用OpenAI的技术，OpenAI就应该承担责任。
I I don't think OpenAI should be liable if somebody else uses their technology.
我认为他们应该制定服务条款，因为他们希望这项技术拥有良好的声誉。
I think they should have terms of service because they want this technology to have a good reputation.
记住，他们面临的是民意问题。
Remember, they're the polling issues.
但我认为，如果别人以这种方式部署这项技术，他们不应当承担法律责任。
But I don't think they should be legally liable if somebody else deploys it in this way.
好的。
Okay.
那么在服务条款中，你认为他们会阻止别人创建色情聊天机器人吗？
Then in the terms of service, do you think they're going to kind of, like, prevent others from creating erotic chatbots?
大概不会。
Probably not.
我的意思是，如果你想想
I mean, because if you think about
你是不是也在想我正在想的事？
Are it you are you thinking are you thinking what I'm thinking?
我们得自己做一个这样的版本吗？
Do we have to call it, like, make make our our own version of this?
不用。
No.
不用。
No.
不用。
No.
不用。
No.
因为也许。
Because Maybe.
也许。
Maybe.
但是
But
如果我们这么做，你知道我们会叫它什么。
If we do, you know what we're calling it.
等等。
Wait.
什么？
What?
什么，Chat？
What, Chat?
哦，不。
Oh, no.
停下。
Stop.
不。
No.
不行。
No.
不行。
No.
不行。
No.
不行。
No.
不行。
No.
我本来想换个角度说，想象一下，如果这是一个绝妙的策略，他们可以实际看到大量基于API的收入——来自其他人创建的成人聊天机器人，然后把这些收入归入企业收入的 umbrella 下，用增长曲线展示我们的企业与API业务增长有多快。
I was gonna say something in a whole different direction about imagine if it's a brilliant maneuver that they could actually, like, see a ton of API based revenue of basically everyone else creating erotic chatbots and then put that under the umbrella of kind of, like, enterprise revenue and have hockey stick charts about look how fast our enterprise and API business is growing.
因为从技术上讲，这确实是企业业务。
Because that's technically enterprise.
但我现在想不下去了，因为
But I can't think anymore because
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。