托尼·法德尔谈 AI 设备、苹果未来与 OpenAI 策略

今天的嘉宾是 Tony Fadell，硬件设计教父，他在 Apple 和后来的 Google（与 Nest 一起）将一些最重要的消费电子产品变为现实。我与他坐下来，听取了他对 Apple 和其他大型科技公司现状的看法，尽管有传言称他可能是 Apple 下一任 CEO 的竞争者。我们讨论了他如果得到那个最高职位会做什么。我们讨论了他为什么认为 OpenAI 正在构建一种“大到不能倒”的策略，以及所有那些不受企业公关部门约束的未经过滤的想法。这里是 Newcomer 播客。Tony Fadell，很高兴你能来参加播客。感谢你加入我。

Today's guest is Tony Fadell, godfather of hardware design, who brought some of the most important consumer electronics to life at Apple and then at Google with Nest. I sat with him and got his thoughts on how Apple and the rest of the big technology companies are looking, even as he's rumored to be a contender for Apple's next CEO. We talked about what he would do if he ever got that top job. We talked about why he thinks OpenAI is building a strategy of being too big to fail and all those unfiltered thoughts without the constraints of a corporate PR department. This is the Newcomer podcast. Tony Fadell, great to have you on the podcast. Thanks for joining me.

很高兴来到这里。很高兴来到这里。

It's great to be here. Great to be here.

我们正处于设备未来的疯狂时刻。所以我非常兴奋能和你交谈。有胸针、笔，我们到底想要什么？所以我真的想直接切入正题。我的意思是，有传言说，我认为这周 Apple 正在考虑一款胸针，因为 Humane 正在开发一款胸针。我有朋友，我认为现在有一些报道说，也许 Jony Ive 和 OpenAI 正在考虑一款笔。我会试着追问你对细节的看法，但从第一性原理出发，你认为最有可能成为 AI 原生设备形态的是什么？

We're in an insane moment for the future of devices. So I'm super excited to talk to you. It's pins, pens, like what do we want? So I really want to get direct into it. I mean there are rumors I think this week that Apple is thinking about a pin in the wake of Humane working on a pin. There's I have you know friends and I think there's been some reporting now that maybe Jony Ive and OpenAI are thinking about a pen. I'll try and press you on what you think about the particulars, but like from first principles, what do you think is most likely to be the AI-native device format?

AI 可以应用于很多事情。

AI can be applied to many things.

是的。

Yeah.

AI 对于某些特定应用以及围绕它的特定硬件组件来说非常棒。所以有带屏幕的 AI 和不带屏幕的 AI。在所有情况下，它都需要音频，对吧？但 AI 最需要的是上下文。所以你必须看输入侧和输出侧。首先我们来谈谈输入侧。你给任何类型的 LLM 提供的上下文越多——我并不是说它们是万能药，还会有很多超越 LLM 的模型。我想确保我说得非常清楚——所以你首先从输入开始。如果我们知道一切，你一直在听到上下文窗口变得越来越长，现在我们正在对你所有的数据进行 RAG。Gemini 昨天刚刚宣布，你可以引入你所有的 Gmail、所有的 Google Photos 和 Google Drive。

And AI is great for certain applications with certain hardware components around it. So there's AI with a screen and there's AI without a screen. In all cases, it needs to have audio, right? But what AI needs mostly is it needs context. So you have to look at the input side and you have to look at the output side. So first let's talk about the input side. The more context you give any kind of LLM — and I'm not saying they're the panacea, there's going to be lots of models beyond LLM. I want to make sure I'm very clear — so you first start with the input. If we know everything, and you've been hearing about context windows getting longer and longer, and now we're RAG-ing all your data. Gemini just announced yesterday that you can bring in all of your Gmail, all your Google Photos, and Google Drive.

我知道我感觉自己有点鲁莽。我已经把所有东西都连接到 Anthropic 了。就像，是的，搜索我的邮件。去吧。

I know I feel like I'm being reckless. I've connected everything to Anthropic already. It's like, yeah, search through my email. Go for it.

对，没错。所以上下文越多越好。所以有一种是关于你做过什么的上下文，对吧？那会让它更好。但还有关于你正在做什么的上下文，对吧？并且非常清楚地知道这一点。你的智能手机拥有的是你想知道你的位置，就像之前的广告和所有其他网站和社交媒体想要的那样。他们想要你的位置。他们一直想要你的音频。他们一直想要你的摄像头。他们想要你输入的内容、你附近的人、有哪些 Wi-Fi 网络和蓝牙设备、你周围的其他设备，甚至是什么时间、光线如何、天气如何、你在室内还是室外。为什么？你在当下拥有的上下文越多，结合它对你和周围环境的了解，当你提出查询时——无论是语音查询还是文本查询——你不需要告诉它，它就知道，哦，你在此时此地，和这些人在一起，拥有这些资源，天气是这样的。当你问一个简短的问题时，就像 Bing 一样，而回到 Siri 和 Alexa 时代，你必须告诉它一切，因为它没有上下文。所以在这种情况下，界面变得越好，它就必须能够拥有更多的历史上下文以及实时上下文。就像我们知道我们现在在做什么。我可以对你说两个词，你可能就明白了，比如，哦，看那边那棵植物。这就是它需要知道的，以便能够给你更快的答案，就像它在房间里一样。

Right. Exactly. And so the more context the better. So there's one which is context about what you've done, right? And that makes it better. But then there's the context of what you're doing, right? And knowing that really, really clearly. And what your smartphone has is you want to know your location, just like advertising and all the other websites and social wanted before. They wanted your location. They would love your audio all the time. They want your camera all the time. They want what you're typing, who you're nearby, what Wi-Fi networks and Bluetooth things are there, what other devices around you, and even what time of day, what's the lighting, what's the weather, are you indoors or outdoors. Why is that? The more context you have in the moment, combined with what it knows about you and what's around you in the moment, you don't have to ask it when you make a query — whether that's a voice query or a text query — it knows it's like, oh you are at this place at this time around these people with these resources with this kind of weather. And when you ask a short question, it's like Bing, as opposed to back in the Siri days and the Alexa days you had to tell it everything because it had no context. So in this case, the better the interface becomes, the more historical context as well as real-time context it has to be able to have. It's like we know what we're doing right now. I can briefly say two words to you and you probably understand like, oh look at that plant over there. That's what it needs to know to be able to give you much quicker answers, like it's in the room.

但你认为大多数人类最终会希望某些 AI 能够访问他们当前的音频吗？你认为这是需要……

But do you think most humans will ultimately want some AI to have access to their current audio? Like do you think that is something that needs to be...

我认为显然有这方面的用例，也有视频的用例。我没有看到所有其他传感器套件用于传感器融合来使其工作。所以在某些上下文中，在某些应用中，你会想要这样做。某些人会说不。你知道人们用胶带盖住笔记本电脑的摄像头等等。所以这将是个人的偏好。我记得 Google Glass 时代，就像“你不要录我”，而现在我们戴着 Meta 眼镜。所以这完全是一个社会情境意识的事情，不仅仅是技术何时准备好。而是社会中的特定群体何时准备好采用它。我记得我在 General Magic 的日子，那时我们提前 15 年制造了 iPhone。没有人知道我们在解决什么问题，因为他们没有那些问题，对吧？

I think there's obviously use cases for that and also for video. I don't see all the other sensor suites for sensor fusion out there to make it work. So in certain contexts, in certain applications, you'll want to do that. Certain people will say no way. You know people put tape over their laptop cameras and everything else. So it's going to be an individual preference. I remember the Google Glass days, like you're not recording me, and now we're running with Meta glasses. So it's all a social context awareness thing where it's not just about when the technology is ready. It's about when a particular segment of society is ready to adopt it. I remember this from my General Magic days when we were making the iPhone 15 years too early. Nobody knew the problems we were solving because they didn't have those problems, right?

但像胸针式设备这样的想法，主要是你会买吗？这里面有胸针式设备吗？

But the idea of like a pin type device is that mostly would you buy is there a pin type device in this?

当然，有胸针式设备。现在也有一种胸针式设备。它叫 Plaude。但那是试图解决“我在录音”这个社会问题吗？比如为什么手机不能直接录音？我们已经随身携带了。

Sure, there's a pin type device. There's also a pin type device now. It's called Plaude. But is that trying to solve the social problem of saying I'm recording? Like why can't the phone just do the recording? We already have it with us.

好的。所以我们先谈了输入，我想谈谈输出，对吧？然后是形态。所以我们先谈输入。现在，我们谈谈输出。然后我们谈谈形态。我知道你想谈胸针，但我们先谈谈输出。

Okay. So first we were talking about the input, I want to talk about the output, right? And then there's the form. So let's talk about input. Now, we'll talk about output. Then we'll talk about forms. I know you want to talk about pins, but let's talk about output.

所以，有些时候确实很棒，这就是我在做 Google Glass 这类产品时的真实体会。输出在上下文相关的方式中非常重要。比如，当你在移动中，你会看到人们走路撞到电线杆之类的东西，因为他们盯着屏幕看，或者骑车时、开车时（最糟的情况）也在看。他们全都因此分心。很多时候你只想要音频，想要一个更好的音频界面，而不是显示屏。但呈现视觉信息的最佳方式还是视觉，对吧？你知道，我们听过有人试图沟通：“我要改这个示意图……”然后他们说“我要改这个和这个”，对方就“啊？”然后你直接发一张图过去，他们就说“哦，现在我懂了”。对吧。所以你必须理解你所处的上下文，以及什么才是呈现输出的最佳方式。是视觉吗？是音频吗？也可能是震动，或者很多其他方式。所以当你开始分解应用场景，考虑你所在的环境——你在办公室？在家？在走路？在公交车上？在火车上？在飞机上？等等。然后你才能理解哪些应用和输出方式在那种情况下是有意义的。

So, there's really great times when you have, and this is what I really experienced when I was working with Google Glass and these kinds of things. Output is really important in very contextual ways. So, if you're on the move, you're going — I see people out here walking into poles and stuff because they're sitting here looking at their thing or they're on their bike, doing it or driving, worst case. And they're all distracted from that. And a lot of times you just want audio and you want to have a much better audio interface, right, versus a display. But the best way to represent visual information is visually, right? You know, we've heard people trying to communicate, 'I'm changing this schematic or whatever,' and they say, 'I'm going to change this and this,' and they're like, 'Huh?' And you just send them a drawing. They go, 'Oh, now I get it.' Yeah. Right. So you have to understand the context you're in and what's the best way to represent that output. Okay? Is it visually? Is it audio? It could be vibrations, it could be many different ways. And so when you start to break down the applications, you start to consider the environment you're in — are you in your office? Are you in your home? Are you walking? Are you in a bus? Are you in a train? Are you in a plane? Whatever. And then you start to understand the applications and the ways of representing that output that make sense in that given situation.

我理解像 AirPods 或耳机作为 AI 的输出设备。但我不——我想我之所以觉得受限，部分原因在于别针和笔的形态——我看不到输出是什么。对我来说，它们主要作为输入存在。

I understand like AirPods or earphones as an output device for AI. I don't — I guess part of why I was limited in the thing is like the pin and the pen format — I don't see what the output is. Like to me they mostly exist as input.

你说到点子上了。你完全正确——这正是我的下一个观点，也就是形态。那么，笔、别针或其他什么形态是怎么回事？比如说 OpenAI 的笔，因为我听说过同样的事，我也认识一些以前为我工作的人在做这类东西。你觉得是笔？我和 Johnny 也共事了很长时间，十年吧，对吧？所以我想我理解它，我可能觉得它像一支笔，但也可能是眼镜或其他东西。但考虑到 Johnny 的倾向，他必须做一些非常不同的事，对吧？所以他必须那样做，因为他会有一个故事说它必须是一支笔，也许它会像 Notto 笔。你还记得 Notto 笔吗？Notto 笔实际上有一个纸垫，你可以——它就像一支数字笔。你只需在纸上写，Notto 会保存所有内容，然后生成一个 PDF，你可以用它做各种事情。AI 如何让我们回归人性。当然，会有那些花哨的言论：“哦天哪，我们正在摆脱手机，等等等等”，全是废话。但实际情况是，它之所以是笔，是为了讲故事。但它没有输出，对吧？因为归根结底，最好的输出形式是你已经随身携带的东西，也就是你的智能手机，对吧？它提供最好的视觉输出。你不会再带另一个屏幕。你可能会在这个其他设备上装几个 LED 或一个非常小的东西来显示状态，但它不会取代手机。它无法取代。Humane 的那套“我们要干掉手机”是我听过的最愚蠢的营销信息，因为你根本干不掉它。所以你必须成为它的伴侣。就像 AirPods、Apple Watch 一样，它们是手机的伴侣，对吧？所以它采用这种形态的原因是他们需要所有的传感器套件。为什么需要传感器套件？因为苹果不会在手机上给他们访问权限。

You hit the nail on the head. You absolutely — and that's my next point, which is the form. Okay. And so what's going on with a pen, with a pin, with whatever — in the case of let's say it's a pen at OpenAI because that's I've heard the same things and I also know some people working on things like that who used to work for me. You think it's a pen? So I also worked with Johnny a long time for a decade, right? So I think I understand it and I probably think it's something like a pen, but it can also be glasses and other things. But given Johnny's predisposition for how he has to do something very different, right? So he's got to do that because he'll have some story around it's got to be a pen and maybe it'll be like the Notto pen. Do you remember the Notto pen? The Notto pen actually had like a pad of paper and you could just — it was like a digital pen. You just write on paper and the Notto would keep all the things and it would then create a PDF of that thing that you could then do stuff like that. How AI is bringing us back to our humanity. Of course, there's going to be this whole big flowery, 'Oh my god, we're getting away from the phone and blah blah blah' and all that diarrhea. So what really happens though is that the reason why it has — it's a pen for a story. Okay? But it doesn't have output, right? Because at the end of the day, the best form of output is the thing you already have with you, which is your smartphone, right? Okay. It does the best visual output. You're not going to carry yet another screen. You might have a couple of LEDs or a very very tiny thing on whatever this other thing is for status or whatever, but it's not going to replace the phone. It can't replace. That was the thing of Humane: 'We're going to kill the phone.' It was like that was the stupidest marketing message I've ever heard because you're not going to kill it. So you have to be a companion to it. Just like AirPods, just like Apple Watch, it's a companion to a phone, right? So the reason why it's in this form factor is because they need all the sensor suites. Why do they need the sensor suites? Because Apple's not going to give them access to them on the phone.

所以他们只是不想被苹果去中介化。

So they just don't want to be disintermediated by Apple.

不，不，不。他们说你必须买这个设备，因为他们需要获取所有上下文，获取所有实时输入，以便反馈给服务器来给你——因为蓝牙安全、视频安全、点、所有 GPS 坐标——当你——你必须向苹果请求权限，以及用户对插座的权限，你会想“天哪，他们在过度追踪我们”，对吧？而实际上，它只是一个网络链路，无论是蓝牙还是 Wi-Fi——很可能通过蓝牙连接到你的设备，再回到网络——这是一回事。是的，就像我连接耳机一样，但它内置了 GPS 等一切，所以它只是吸走你所有的上下文，并消除所有那些隐私担忧。你看，这让事情变得非常简单，人们会说“哦，是的，它就这么做了”，而你不明白它在尽可能吸走一切，对吧？为了回到 OpenAI，为了绕过所有限制。

No, no, no. They're saying you have to get the device because they need to get all the context to get all the real-time input so they can feed it back to the servers to give you the — because all the security of Bluetooth security, video security, the dots, all the GPS coordinates — when they — you have to ask Apple for permission and the user permission for outlets things, you go, 'Oh my god, they're hyper-tracking us, right?' as opposed to it's one network link whether it's a Bluetooth or Wi-Fi thing — you know, it's most likely Bluetooth to your device that then goes back to the network — it's one thing. Yeah, I'm just connecting like my headphones, but it's got GPS everything else in it, so it's just sucking up all your context and throwing away and getting rid of all those privacy concerns. You see, so it makes it real simple and people go, 'Oh yeah, it just does it,' and you don't understand it's sucking everything it possibly can suck, right? To get back to OpenAI, to get around all the restrictions.

如果你是一个 AI 的忠实信徒，你想要所有的——你也有点希望它拥有你所有的上下文。如果我们开始与自己的聊天机器人建立关系，比如“我真的依赖你。我想让你知道我在做什么。”你不会觉得这种冲动——这种缺陷不是——它缺乏关于我在做什么的上下文，而且它会是——你没有这种感觉。

If you're an AI true believer, you want all the — you too sort of want it to have all the context you have. If you — if we start to have a relationship with our own chatbot where it's like 'I really depend on you. I want you to know what I'm doing.' You don't feel this impulse that it's sort of the deficit isn't — it's lack of context on what I'm doing and that it would be — you don't feel that way.

你是早期采用者，对吧？

You're an early adopter, right?

但你觉得——

But do you feel —

世界上大多数人——大多数人不是。大多数人会说：“你到底在用我的数据做什么？”对吧。

Most of the world — most of the world isn't. Most of the world is like, 'What the hell are you doing with my data?' Right.

对。所以，我是——我是一个晚期的早期采用者。

Right. So, am I — I am a — I'm a late early adopter.

好的。

Okay.

我了解这些产品是如何制造的。我观察过这些东西，对吧？

I have learned how these products are made. I've watched this stuff, right?

对。

Right.

我没有采用 IQain。我知道它有多糟糕，因为当我第一次开会时他们只有四个人，我就想：“这东西成不了气候。”其他各种设备也发生了同样的事。我想：“我没时间搞这些破玩意儿。”Rabbit 也一样。我想：“这是什么？”和 Jesse 见了面。我想：“不。”所以问题在于，你想成为一个足够晚的早期采用者，而不是疯狂到浪费你的时间。所以我站在那一边。然后我也关注所有数据相关的东西，因为我也相当明智，我知道这些数据是怎么用的，我处于中间状态。

I didn't adopt IQain. I knew how bad it was because when I had the first meeting when they were four people, I was like, 'This thing's going nowhere.' The same thing happened with various other devices. I'm like, 'I don't have time for this crap.' Same thing happened with Rabbit. I was like, 'What is this thing?' Met with Jesse. I'm like, 'Nah.' So the issue is you want to be a late enough early adopter, not so crazy that you're just wasting your time. So I'm on that side. So and then I also look at all the data stuff because I'm also pretty wise and I know how this data stuff is used and I'm in between.

是啊。你知道，你刚才说你有了第一个孩子，一个女儿，对吧？第一个孩子。

Yeah. You know, you just said you have your first child, a daughter, right? Just your first child.

是的。三个半月。

Yes. Three and a half months.

恭喜。

Congratulations.

我记得有孩子之前和之后，我对世界的看法有显著差异。包括 Larry、Sergey 和 Mark Zuckerberg 在内的多数创始人，他们在有孩子、结婚之前，看待世界的方式完全不同。我什么都想要，随时都要。我愿意放弃所有隐私，我不在乎这些数据会发生什么。一旦你有了孩子，开始听到缺陷、社会工程学这些东西，你就会改变对数据收集量的看法，对吧？以及你如何受到保护。而且我确切知道，因为我和这些人共事过，其中一些创始人现在看待世界的方式不同了，他们希望自己当初能做出一些不同的举措和决策。是的，他们想回头但回不去了。所以我想说，关于数据这件事，取决于你的人生经历，差别很大。对我女儿来说，我们用 Tiny Beans，那是一个只限家人使用的私人应用，所以你看，你开始以不同的方式使用产品。

I remember a marked way difference in the way I thought about the world before I had kids and after I had kids. And most of these founders, including Larry and Sergey and Mark Zuckerberg, all those guys think about the world before they had kids, before they were married in a very different way. I want everything all the time. I'll give away all my privacy. I don't care what happens with all this data. As soon as you have kids and you start hearing about defects and social engineering and all this other stuff, you start changing your view on how much data you want sucked up, right? And how you're being protected. And I know for a fact because I worked with these guys, some of these founders think about the world differently and they wish they would have made some different moves and different decisions early on. Yeah, and would like to go back but they can't. So I'm just saying, going back on your data thing, it's very different depending on your journey. For my daughter, we have Tiny Beans, which is a private app just for our family, so I can see you start using products differently.

好的，如果这个别针真的是绕开 iPhone 控制的一种方式，那苹果为什么需要别针？他们已经有 iPhone 了。这故事怎么讲？

Okay, so if the pin is really an end run around the control of the iPhone, why would Apple need a pin? Like they have the iPhone. How does that story go?

哦，因为你可能正在使用设备，但很多时候你不想让设备挡在中间，因为你可以说，它也可以一直处于环境状态，因为你的东西会放在口袋里。所以有一些形式，比如拥有这个其他东西，如果你选择去看它，因为我们之前见过这些生活记录器，你知道，10 年前就有这些生活记录器。这只是另一种形式。这就是它的定位。就像不要错过任何时刻。

Oh, because you might be on the device using it, but also a lot of times you don't want to have the device there in the middle because you're saying, I'm like, it can also be ambient all the time because your thing's going to be in your pocket. So there are some like the form of having this other thing that if you choose to want to watch it, because we've watched these life recorders before, you know, 10 years ago there were these life recorders. This is just another form of that. That's how it gets positioned. It's like don't forget any moment.

是的。你知道，Plaude 现在用他们的 voice one 和这个在做同样的事。

Yeah. You know, Plaude's doing that now with their voice one and this one.

我其实不知道那个。Plaude 来自英国，它就像 Granola。

I actually don't know that one. Plaude's out of the UK and it's, you know, Granola.

是的，当然。它基本上就是 Granola，但做成可穿戴设备。

Yeah, of course. It's basically Granola but as a wearable.

你是投资者吗？

Are you an investor in?

不，我不是。

No, I'm not.

好的。你刚提到它，所以我没有……

Okay. You just brought it up so I didn't...

不，不，不，不，不。我不是投资者。

No, no, no, no, no. I'm not an investor.

嗯，你相信苹果在研发别针吗？

Um, do you buy that Apple's working on a pin?

我确信，我向上帝祈祷他们在研发某些东西，你知道吗？他们需要在这方面下点功夫，但以他们的芯片专长，Johnny Srouji 和团队，他们非常出色。了解机械设计，知道他们做出了手表。对。而且他们在做 AirTag 和其他东西。他们有 AirTags。他们有所有这些非常微型的定制芯片，以及所有 AirPod 定制芯片。他们可以比任何人都更容易做出这种东西，因为他们拥有制造出色硬件的所有专长。他们需要把软件做好，但硬件、连接性，他们有自家的网络协议栈。他们可以做出一个很棒的产品，因为他们也有所有摄像头技术。所以，它要么是放在桌上的东西，要么是穿戴的，或者放在桌上之类的。这很明显。或者你把它集成到你的 AirPods Pro 里。想象一下摄像头从你的 AirPods 里伸出来。

I'm sure, I hope to God they're working on something, you know? They need to be working on a little bit, but knowing their chip expertise, Johnny Srouji and the team, they're incredibly awesome. Knowing the mechanicals, knowing that they've made the watch. Right. And they're making, you know, the AirTag, all the other stuff. They have AirTags. They have all this really miniature custom silicon and all the AirPod custom silicon. They can make one of these things really easy compared to anybody else because they have all the expertise to make an incredible piece of hardware. They have to get the software right, but the hardware, the connectivity, they have their own network stack. They could make an awesome one because they have all the camera technology, too. So, it would either be something you sit on the desk or you wear and you sit on the desk or whatever. It's like duh. It's like or you build it into your, you know, into your AirPods Pro. Like imagine cameras coming out of your AirPods.

你在用，可能，或者，你觉得苹果能交付多少？我猜，你知道，你在跟一个深度使用苹果生态的人说话：我的笔记本电脑、AirPods、手表、手机、蜂窝网络。但后来他们，我想我买了那款以 AI 为卖点的手机，这可能是营销史上最严重的误导之一。当我看到 AI 第一台笔记本电脑、AI 第一部手机时，我差点要发一条非常恶毒的推文，但我忍住了，因为我觉得，你知道，这……苹果以前从没做过营销，我看到那完全是，对吧？你知道，我忍住了。所以，你知道，希望他们找到新的信仰，回到过去真正聪明的做法，那就是低承诺、高交付。

You're using, could, or like, how much do you expect Apple to deliver? I guess, you know, you're talking to somebody. I'm deep in the Apple ecosystem: my laptop, AirPods, watch, phone, cell. But then they, I think I bought the phone that was sold on AI and it's one of the most egregious misleads perhaps in ever marketing. When I saw AI first laptop, AI first phone, I was ready to tweet something that was really, really nasty and I pulled back because I'm like, you know, it's... Apple has never done marketing before and I saw that was complete, right? You know, and I was like I was holding back. So, you know, hopefully they're finding new religion and going back to the old real smart days, which is underpromise and overdeliver.

你觉得他们现在能做好什么？又是什么在拖后腿？

What do you think they can do well right now? And what do you think is holding them back?

他们现在能做好什么？哦，芯片、底层软件、设备。你知道，Apple Vision Pro 是一个彻底的失败，但从技术角度看，它是一个不可思议的技术奇迹。所以他们能做出很棒的东西。问题在于他们是否还有过去那种产品能力，能够打造完整的堆栈，创造不可思议的用户体验，并交付所有必要的中间环节，把一切整合在一起。而这正是苹果能走到今天的关键原因。

What can they do well right now? Oh, chips, low-level software, devices. It's really, you know, Apple Vision Pro is an abject failure but it's an incredible technical marvel from a technology point of view. So they can make great stuff. It's do they have the product chops like they used to to be able to create the full stack, create the incredible user experience, and deliver all those pieces of the puzzle in the middle necessary to pull it all together. And that's the thing that has been really important for the reason why Apple is where it is today.

你觉得在多大程度上是 AI 的底层能力拖了苹果的后腿？比如，为了出色的用户体验，你希望达到 100% 的准确率，对吧？很多人谈论一些吸引人的 AI 能力。这几乎就像自动驾驶的 90% 情况。你知道，它能做到你想要的大部分事情，但越深入一个问题，你遇到的就越多。所以如果你试图建立这些消费者循环，比如每次我走进家门，你就做这个，AI 有 90% 的时间做对。能指望苹果交付人们想要的这种消费者体验吗？

To what extent do you think just the underlying capability of AI is holding Apple back, right? Like for a great user experience, you want sort of the 100% case, right? A lot of people talk about some of the catchy AI capabilities. It's almost like the self-driving 90%. Like, you know, it can do most of what you want, but then the more you drill into a problem, the more you start hitting problems. And so if you're trying to set up these consumer loops where it's like every time I step into my house, you do this and the AI gets it right 90% of the time. Can Apple be expected to deliver that consumer experience that people want?

嗯，首先要理解的是，大语言模型有根本性问题，对吧？它们有根本性问题，这源于它们的创建方式，这就是它们产生幻觉的原因。除非它们被缩小范围，真正在非常特定上下文的语料上训练，比如医疗、法律或一般消费者内容，而不是这些极其庞大的模型，它们会产生幻觉，因为它们不知道你在谈论医疗、法律还是别的什么。它试图成为一个万事通。你在现实世界中永远不会雇用一个万事通。我们受不了，对吧？我承认我不知道的东西，而 ChatGPT 从不说不知道。它总是给你一个答案，而且它不知道答案是对是错。所以，你知道，我们投资了医疗领域的各种公司，比如 Knobla 和其他公司，它们有非常具体的上下文和非常具体的护栏，规定它们会做什么、回答什么，不会回答什么。

Well, the very first thing to understand is LLMs have fundamental problems, right? They have fundamental problems and that's in the way they're created and that's why they hallucinate. And unless they're descoped and they're really trained on very contextually specific information like medical or legal or general consumer stuff instead of these incredibly huge models that can hallucinate because they don't know whether you're talking medical, legal, whatever. It tries to be a know-it-all. You would never hire a know-it-all in the real world. We can't stand, right? I admit what I don't know and ChatGPT never says. It always gives you an answer and it doesn't know whether it's right or wrong. And so, you know, we're invested in various companies in the medical domain like Knobla and other things that have very specific context and very specific guardrails for what they will do and what they will answer and they will not answer stuff.

问题在于当你试图制造这些通用模型的时候。所以回到 LLM 本身，如果苹果为特定场景下的消费者提供非常具体的模型，那效果会很好。但当你应用这些非常宽泛的模型，不知道会输出什么时，问题就来了。还有 LLM 之外的模型和其他东西，即使是高度上下文约束的 LLM、经过训练的 LLM，它们也将解决很多这类问题。但现在每个人都在谈论 LLM、AGI、越来越大的模型，我觉得这……

It's when you're trying to make these general purpose ones, that's when it's a problem. So getting back to the LLM specifically, if Apple has very specific models for consumers in certain contexts, it can work really well. It's when you apply these very broad models and you don't know what's going to come out, then that is where the problem is. There are still models and other things beyond LLMs, even highly contextually constrained LLMs, trained LLMs that are coming that are going to get around a lot of those things. But right now, everybody's blah blah blah LLMs, AGI, bigger and bigger models, which I think is...

接下来会有什么？

What's coming?

嗯，世界模型要来了，对吧？所以是世界模型与 LLM 的结合。那就是理解物理、理解化学，所有这些我们已知的东西，因为存在真实的规则和数学。你知道，现在的 ChatGPT 甚至画不出钟表的面盘，对吧？或者看不懂钟表显示的时间，因为它没有图形视觉。对我们来说，图形钟表很容易理解，数字钟表则不然。

Well, there's world models coming, right? So world models combined with LLMs. That's understanding physics, understanding chemistry, all this stuff that we know because there are real rules, mathematics around those things. You know, today ChatGPT can't even draw a face of a clock, right? Or understand what a clock says because it doesn't graphically see. It's easy for us for a graphical clock, not a digital one.

所以世界模型会这样运作：‘哦，那是图形。我要用这个模型。你知道，它就是这样工作的。’你会把这些东西构建到模型工具中，对于某个问题，它会调用这个软件或技术来解决。

And so it's like though that a world model would go, 'Oh, that's a graphic. I'm going to use this model. It, you know, this is how it works.' You would build that stuff into models tools where it says for this question you go off to this software or this technique to solve something.

是的，完全正确。然后语言模型擅长它擅长的领域，而不是试图去推理化学，因为它没有真正的基础化学模型。

Yeah, absolutely. And then the language model is good for what the language model is good for, as opposed to trying to think it reasons about chemistry because it doesn't have a real basic chemistry model.

我们开始对话时，我觉得你提到音频和语音是 AI 想要消费的核心。这公平吗？我的意思是，显然监听我们做的一切是这些设备想要的。在 AI 设备的故事里，你怎么看 AirPods 和耳机这类产品？

When we started this conversation, I think you were saying audio and voice is sort of at the core of what AI wants to consume. Where, or maybe is that fair? I mean clearly listening to everything we're doing is part of what these devices we've talked about want. Where do you think about sort of the AirPods and earbuds in terms of the AI device story?

嗯，有边缘设备，比如 AirPods、手机这类东西。我首先考虑的是边缘的边缘。边缘的边缘就是最外围的层面：比如你在说话吗？我如何做降噪？我如何做说话人聚焦？这类事情。有 AI 可以很好地完成这些。你的耳朵就在做这些。你也在做。你转头、聚焦。所以我们可以把 AI 应用到每个边缘的边缘设备上，实现更好的智能，帮助我们过滤噪音……完成设备的核心任务。最基本的就是从信号中分离噪声，无论是视觉、听觉还是其他类似的东西。所以我们需要达到那个水平。这是 AI 应用的第一层。然后在其之上是标注、解释这类 AI，有些情况下可以在设备上完成，或者由伴侣设备完成，很可能是你的智能手机。这是另一套 AI。再往上一层是：好的，推理是什么？获取上下文，然后弄清楚发生了什么。所以有所有这些层次。我们需要把 AI 分解成不同层次，就像你的大脑有不同中心来处理不同事情。你会开始看到 AI 变得更加多样化，而不是一个单一的庞然大物。我们在自动驾驶汽车上已经看到了这一点，对吧？

Well, there's edge devices which are AirPods and phones and these kinds of things. I think of the edge of the edge first. So the edge of the edge is the very fringe of like: are you speaking a word? How do I do noise cancellation? How do I do speaker focusing? Those kinds of things. There is AI that can do that stuff really well. Your ear does that. You do that. You turn your head. Focus. So we can apply AI into each of these at the edge of the edge devices to do much better intelligence to help us be able to filter out noise from... to do the device's core task. The very basic task of separating signal from noise, whether that's visually or auditorially or other things of that nature. So we need to get there. That's the first level of AI applied. Then there's the AI above that which is labeling, interpretation, those kinds of things, and that can be done on device in some cases or it'll be done with a companion, which would most likely be your smartphone. So that's another set of AI. Then there's another one on top of that which is like: okay, what's the reasoning? Let's get the context and then figure out what's going on. So there's all these levels. We need to break down the AI into levels just like your brain has many different centers in your brain to do this, this, this, this. You're going to start seeing AI becoming much more diverse and not just one big thing. And we're seeing that with self-driving cars already, right?

我知道。我们聊得有点散。

I know. We're all over the map.

不，我知道。我喜欢这样。某种程度上，AI 浪潮造成了这种破碎的思维，因为有些公司想要证明估值合理，所以需要卖点东西，他们的故事就是 AI。所以在某些方面，我也参与了他们正在做的事情，那就是从对 AI 设备的需求出发，而不是从人类问题出发，然后正确地构建设备。这就是为什么这场对话最终变得混乱，因为我们在技术和……之间切换。就像我们在寻找一个……

No, I know. I love it. To some degree the AI wave creates this broken thinking because you have companies that want to justify valuations so they need to sell something and they have the story which is AI. And so in some ways I'm complicit in what they're doing, which is like think from the need for an AI device out, rather than some human problem and then build the device correct up. And that's why this conversation ends up being chaotic because we're shifting from the technology to... It's like we're in search of a...

但你不觉得现在正是向消费者推销新设备的时机吗？他们听说过这项技术，愿意接受新的行为方式，所以人们在寻找我们能提供什么设备……对我来说，这些设备——别针、笔或其他什么——是使能器，不是解决方案。我们之前说到，我们需要更多上下文特定的 LLM，这样消费者才能获得无幻觉的可靠体验。这意味着你必须提出应用，并约束应用以更好地理解你想要什么。它不是一个万事通，什么都去尝试。所以当你切换到某些应用时，你希望这样，为什么？归根结底，我们还没有达到我们想要的状态。我们还没有达到《她》那个水平。每个人都想要《她》，但我们还没有技术。所以我们必须慢慢来，这意味着更多针对特定工作流的、应用特定、上下文特定的 LLM，而不是一个通用的任何设备。

But don't you think there is this moment to sell the consumer on a new device? They've heard about this technology, they're sort of open to adopt a new way of behaving, and so people are searching for what device can we deliver them in this... To me these devices—a pin, a pen, or whatever—are enablers, they're not a solution. And where we were going was I was saying that we need more context-specific LLMs so that consumers can get a reliable experience without hallucinations. What that means is you have to come up with the applications and constrain the application to better understand what it is you want. It's not just a know-it-all that's going to go and try everything. So when you switch into certain applications, you want that because why? At the end of the day, we're not where we want to be. We're not at 'Her' yet. Everybody wants 'Her', but we don't have the technology yet. So we have to creep into it, which means more application-specific, context-specific LLMs for given workflows, and not just a general anything device.

好的。所以我们会看到很多……这有点像 Web 1.0 到 2.0 再到 3.0。我们必须做出这些跳跃，因为我们需要技术来推动它。

Okay. And so we're going to see much... It's kind of like Web 1.0 to 2.0 to 3.0. We've had to make these leaps because we have to have the technology to move along with it.

但我们多年前就有了 Siri。我们有了 Alexa。你觉得我们离那种人们通过语音交流的个人助理还很远吗？

But we had Siri many years ago. We've had Alexa. Like do you think we're still far away from the sort of voice-communicated personal assistant that people talk to to engage with a computer?

是的。为什么？我看到的是你将拥有你的设备联盟。那将包括你的笔记本电脑、电视、游戏机、你随身携带的所有东西或手表，等等。你会拥有这些。然后你会有一个后台的智能代理，真正理解你在哪个设备上、你是谁、所有数据和上下文，然后能够根据这些切换工作方式。今天，我们说的是把所有东西放在一个设备上它就能工作，或者把所有传感器放在这里，把所有智能放在云端，中间的一切都是哑的。对吧？我们将构建一个更加多 AI 的世界，其中边缘的边缘、边缘设备、边缘设备联盟和云端协同工作，产生我们渴望的《她》那样的应用，但根据所有人告诉你的，我们今天还无法达到，仅仅是因为 LLM 的能力。

Yes. Because why? What I see is you're going to have your federation of your devices. That's going to be everything from your laptop, your TV, your game machine, all the things you carry on you or watches, whatever it is. You're going to have that. Then you're going to have a smart agent in the background really understanding which device you're on, which users you are, and all the data and all the context, and then be able to switch how it's working based on those things. Today, what we're saying is put everything on a device and it's going to work, or put all the sensors here and put all the smarts in the cloud and everything else is just dumb in between. Right? We're going to build into this much more multiple AI world where edge of the edge, the edge devices, the federation of edge devices, and the cloud all work together to produce the 'Her' application we're hearing that we would love to have, but we're just not going to get there today from what everyone's telling you, just because of the capability of LLMs.

是的。因为我们还没有……我们还没有世界模型。

Yeah. Because we're just not... we don't have the world models.

我们还没有传感器。当你看这些人形机器人时，每个人都说：“哦，它们会翻跟头、耍把戏，什么都会。”没错，我们能做到。但我们缺乏必要的传感和驱动能力来获取数据，对吧？它们需要世界模型才能做人们期望人形机器人做的那些事。记得吗，自动驾驶汽车是 2008 年底出现的，对吧？我们现在才刚刚达到那个水平，对吧？不，这很棒，对吧？我们投入了数十亿，现在才开始推广，而且还有价值 30 万美元的汽车，配备巨大的传感器套件和算力，这些是巨大的边缘设备，再加上一整套云端协调。对吧？所以这就是我们需要做的，而且我们需要在非常低的成本和高能效水平上做到，这需要时间才能实现。

We don't have the sensors. When you look at these humanoid robots, everyone's like, "Oh, they do flips and tricks and everything." Great, we can do that. But we don't have the necessary sensing and actuation necessary to be able to get the data to be able to, right? And they need world models to do anything like a humanoid stuff that people are hoping it'll do. Remember, self-driving cars came out in the late 2008, right? We're just there now, right? No, it's great, right? And we've spent billions and billions and we're now getting and it's just getting rolled out and there's still $300,000 cars with massive sensor suites and compute and those are massive edge devices plus a whole set of coordination in the cloud. Right? So that's what we need to do and we need to do it at a very cheap level and battery efficient level and right and it's going to take time to get there.

对。智能家居的现状如何？或者——好吧，我知道我们在这里讨论的是大设备问题，而且这符合这个 AI 对话，但你对构建智能家居设备的成败有什么看法？

Right. What is the state of the smart home today? Or what — okay, take — I know we're talking the big device questions here, like, and it fits into this AI conversation, but like, what is your read on just the success or failure of building, you know, smart home devices.

好吧。我打算为自己建造一个大型新智能家居的改造项目。所以，你知道，我在追踪最新动态，但现在我不像以前那样深入其中了。

Okay. I'm going to build a remodel of a big new smart home thing for myself. So I'm, you know, I'm tracking the very latest and now I'm not in it like I used to be.

那包括定制设备，还是你只是采购现有的最好的东西？

Does that include custom devices or you're just sourcing the best stuff that exists?

嗯，我认为现有最好的东西来自设备端，但还有自定义的方式将它们整合在一起。就像我们刚才讨论的自动驾驶汽车一样。你必须——当我设计这个东西时，我必须考虑设备、底层网络、应用层网络、服务器里有什么、云端有什么，等等，而且如果你想实现我们想象中的那种 Jetson 智能家居，仍然非常非常困难。

Well, I'd say that there's the best of what exists is from the device side, but then there's custom ways of putting it all together. It's like what we just talked about the self-driving car thing. You have to — and when I'm designing this thing, I have to think about the devices, the low-level network, the application level network, what's in the server, what's in the cloud, like, and it's still very, very difficult if you want to pull off what again the smart home that the Jetson smart home that we all think about.

那是你想要达到的目标吗？

Is that what you're trying to get?

是的。而且我更清楚该怎么做。比如，我会把所有东西都布线，确保在必要的地方有 Thread 边界路由器。

Yeah. And I just know better. Like I'm like, okay, I'm going to wire everything and make sure I have thread border routers in the necessary places.

你希望房子能做的几件主要事情是什么？

What are the main things you want the house to be able to do?

哦，现在只是最简单的事情，比如我有一个——你知道——在这种情况下，我有一架自动演奏钢琴。对。所以我向它输入 MIDI。我从网络或本地输入 MIDI 歌曲，然后它自动演奏，它配有麦克风，麦克风信号被路由到 Dante 音频网络，然后通过 Dante 分配到整个房子或我想要的任何区域，送到房子里所有这些 Dante 放大扬声器，Sonos 也可以接入，所以我可以有一架自动演奏钢琴，琴键在按动，对吧？琴键在按动。所以，这是一架模拟钢琴，被数字录制并流式传输到整个房子。

Oh, right now just the simplest things like I have a — you know — in this case I have a self-playing piano. Right. So I'm pumping MIDI into it. So I pump MIDI and I pump songs from the net or from locally MIDI into it to then it self-plays and then it has it's miked and the mics are then routed into a Dante audio network which then distributes all the way through the entire house or whatever zones I want through Dante into all of these Dante amplified speakers around the house and the Sonos can pump in and so I can have a player piano and the keys are pressing, right? The keys are pressing. So, it's an analog piano getting digitally recorded and streamed all through the house.

它像——它不是老式的自动演奏钢琴，还是说它就是？

Is it like an — it's not an old school player piano or it is?

不，不，不。这是一架现代普通三角钢琴，但上面装有执行器。它演奏歌曲的方式——因为是从专业人士那里录制的。所以，就像你有一位音乐会钢琴家在你身边。然后你给它拾音，放大，并在整个房子里播放。所以感觉就像家里有一位音乐会钢琴家。

No, no, no. It's a modern regular grand piano, but it has actuators on it. And it plays the songs just as it — because they record it from a professional. So, it's like you have a concert pianist with you. And then you're miking it and then you're amplifying it and playing it throughout the house. So it feels like you have a concert pianist in the house.

这个实现了吗，还是仍然是个梦想？

Executed on that yet or is this still a dream?

哦不，我们——它正在整合中，而且可以工作。绝对可以工作。但然后还有所有的切换，让 Sonos 从一个设备切换到另一个，以及——自动演奏钢琴去了哪个区域——

Oh no, we're — it's in the mix and it can work. It absolutely can work. But then there's all the switching to make the Sonos move from one thing to another and how do — and which zones did the player piano go to —

对。

Right.

而且你是个音响发烧友吗？你有那种花哨的老式音箱，还是用 Sonos 之类的？

And are you like a speaker head? Like do you have the sort of fancy old school speakers or you're using like Sonos?

嗯，不是。我用 Sonos 端口将房子连接到放大扬声器，用于某些——要么是吸顶式扬声器，没有——你看不到格栅。有些情况下我不喜欢它们，但在某些潮湿位置你需要格栅，还有其他东西，比如 L-Acoustics 的 Dolby Atmos，你知道，整套系统，你必须担心所有这些网络层级才能让所有东西工作。你知道，有 AC、AEC，我有这么多不同的东西。另外我还有完整的照明系统，所以有 DMX 灯光控制，就像舞台上用的那样。还有——所以，MIDI over Ethernet 和所有东西都通过以太网运行。所以，有 MIDI over Ethernet、灯光控制 over Ethernet、Dante over Ethernet、灯光控制 over Ethernet。对，对。所以，这是一个完整的下一代系统。

Well, no. I'm using Sonos ports to get the house — to amplified speakers for certain — either in-ceiling speakers without — you don't see the grills. In some cases I don't like them, but there's some cases in certain wet locations you need the grills and other things where it's L-Acoustics like Dolby Atmos, you know, whole thing, and you have to worry about all those levels of networks to make all that stuff work. You know, got AC, got AEC, there's so many different things I have. Plus I also have full lighting so there's DMX lighting control like you do on stages. There's also — so, the MIDI over Ethernet and everything runs over Ethernet. So, it's MIDI over Ethernet, lighting control over Ethernet, Dante over Ethernet, lighting control over Ethernet. Right. Right. So, it's a full-on, you know, next generation system.

还有其他东西吗？你的智能家居。

Anything else? Your smart home.

嗯，当然会有 Thread，因为我们做了——

Well, it's going to have Thread, of course, because we did —

什么是 Thread，先生？

What is Thread, sir?

所以，Thread——我们基本上在 Nest 发明了它，现在 Thread 是用于整个房子的网状网络的蓝牙等效技术。所以它通过网状网络覆盖整个房子，用于低速数据和控制，延迟非常非常低。所以你可以用于灯开关、灯泡、传感器，以及其他所有可以电池供电 10 年的东西，诸如此类。而且它非常有弹性。

So, Thread — we had basically invented it at Nest and now Thread is the Bluetooth equivalent for mesh networking for entire house. So it mesh networks an entire house for low-speed data and control, very, very low latency. So you can do light switches, light bulbs, you know, sensors, everything else that can be battery operated for 10 years, that kind of thing. And it's very resilient.

你有做任何机械类的东西吗——比如我不知道——有自动做饭的吗？

Do you do anything like mechanical — like I don't know — are there meals being made in automated?

有机械照明和弹出式电视，灯光下降，可移动的机器人灯，对吧？嗯，是的，除了机械锁和基于 NFC 的门锁之类的东西，这些就是大部分机械部分了。

There's mechanical lighting and TVs that pop out, things the lights drop down, robo lights that move, right? Um, yeah, that's the most of the mechanical except mechanical locks and, you know, NFC-based door locks, that kind of stuff.

那你用什么来与房子进行实际的语音通信？

And what do you use for the actual like voice communication with your house?

那还有待确定。

That is to be determined.

是比选吗，还是什么意思？

It's a bake off or what would you mean?

嗯，还没有任何东西准备好。对，对。所以你必须——下一代 Alexa 还没有出现。所以它已经设置好——那是所有这些基础设施之上的下一层。再下一层就在那之上，目前待定。

Well, nothing's ready yet. Right. Right. So you have to — next generation Alexa is not there yet. So it's set up — that's the next layer on top of all of this infrastructure. The next layer is on top of that and that's TBD at this point.

但你是用打字吗？会发生什么？你必须做出决定，对吧？

But are you typing it? What's going to happen? You have to make a decision, right?

不，我不需要。

No, I don't.

为什么？好吧，你必须与——我——你会用什么？这房子什么时候准备好？

Why? Well, you're going to have to talk to — I — what are you going to use? When is this house supposed to be ready to go?

你会用什么？你用按钮。就像汽车从屏幕和按钮演变而来。

What are you going to use? You use buttons. Just like cars are coming from screens and buttons.

现在你还没准备好使用语音。

Right now you're not ready to use voice.

不，这是因为——你有像模块一样的东西——我不喜欢每个人都对我大喊大叫，你知道，“开灯。”你完全没有那些。

No, it's because — you have like modules like — I don't feel like having everyone scream at me and, you know, "turn on the lights." You don't have any of that.

没有。

No.

你知道吗，这是我从 Nest 学到的另一件事。大多数设计智能家居的人都是为自己设计的，就像单身汉那样。我想要一个按钮做这个，另一个按钮做那个。但当你有客人来，有孩子，或者有朋友，或者有妻子或伴侣在身边时，他们会说：‘我不想要那个按钮，我要这个按钮。’每个人都有自己的单按钮界面。所以你必须设计并理解——你和其他人一样，这不只是你一个人的事。当它是一个共享空间时，你必须设计整个环境，对吧？而不是像你自己的 AI 或她的空间那样，那是完全不同的东西。

You know, look, this is the other thing I learned from Nest. Most people who design the smart home design it for themselves. It's like for single guys. I want one button that does this and one button. And then when you have people over, when you have kids or you have friends over or you have a wife or partner with you and they're like, 'I don't want that button. I want this button.' Everyone has their own one button interface. So you have to design and understand you're just like anything. It's not just you. You have to design this whole world when it's a shared space, right? Versus your own AI, her space. It's a very different thing.

我的意思是，你们在苹果生产的设备有一个很棒的地方——显然它们很贵，但本质上它们是大众市场的。你知道，感觉就像亿万富翁的手机，美国中产阶级也能买得起。但你描绘的智能家居不是这样，它有点像老式电子产品。这真的很疯狂，但你可以做很多，对吧？那你觉得怎么才能让家居达到那个水平？在某种程度上，这也是你想要的吧——一个能解决问题的设备，同时我不想家里有更多屏幕。我讨厌屏幕，真的。我讨厌屏幕，我不想把它们放在家里。那么，为什么家居还没有那种大众市场级的出色电子产品呢？

What I mean, one of the things that's amazing about the devices you guys produced at Apple, obviously they were expensive, but they're fundamentally mass market. You know, it feels like the billionaire phone is the phone that sort of the middle class American is able to afford. I mean, the portrait you're painting is not that, you know, where it's like with the home and that's sort of like old school electronics. That's really crazy, but you can make a bunch, right? But how do you think about sort of getting the home to that point? Is that on some level it's what you want too, which is a device that's made that sort of works that solves some of these problems without the sort of I want less screens in my home. I don't want more screens. I hate screens, frankly. I hate screen. I don't want them in the home. So, why hasn't the home had that sort of like mass market amazing electronic work yet?

哦，因为我们还处在早期阶段。你知道，Wi-Fi 直到 2001 年才出现，对吧？Thread 虽然 2011 年就开始了，但直到现在才真正普及。我们还在搭建所有基础层，连接一切。现在我们即将拥有大语言模型和语音识别等所有技术，然后才能把它们整合在一起。但它仍然处于临界点，就像当年的自动驾驶汽车一样。我认为智能家居要达到你想象中那种理想状态，还需要五到六年的时间。你只需要设置好你的家，确保它……

Oh, because we're still in the days of like, you know, remember Wi-Fi didn't even exist until 2001, too, right? So we thread just finally got even though it started in 2011 it's just now there we're still building all the base layers and linking everything and then and we and we're now going to get the LLMs and voice recognition all that stuff to then put it all together right but it's still on the verge just like the self-driving car was when this that smart home like that is you got to give it another I think five to six years for what we think of when you think smart home you want it to be I think it's going to be there just set up your home to make sure it's

你认为哪家公司最有优势？

What company do you think is best positioned?

我不知道。还需要五六年。不，因为我在 Nest 学到的是，所有这些传感器、智能插头之类的东西都不赚钱。绝对不赚钱。这就是为什么你上网买，你知道，因为人们只愿意多付 20% 的钱。比如我墙上有一个开关，其他任何东西，像我这样的疯子才会做别的事。这个市场不够大。真的不够大。所以没人愿意进入这个领域。说实话，宜家终于开始行动了，因为他们用了 Thread 和 Matter 这类技术。宜家正在起步，但也只是刚刚开始。一切都还处于萌芽阶段。所以我认为，最终成功的将是那些真正实现我们 Nest 最初愿景的公司。但话说回来，如果是我，我现在就会重建 Nest，因为三到四年后，你就能实现我们在 2010、2011 年就有的愿景。

I have no idea. It's five six years. No, because the thing that I learned at Nest was all these sensors and smart plugs and all that other stuff there's no money in it. There's absolutely no money in that stuff. That's why you go online and you know and that's why there's you know um because people are only willing to buy like 20% more expensive like I have a switch in my wall anything else and any crazy guy like me doing this other thing, it's just not a big enough market. It's not a big enough market. So, nobody wants to play in that space and so IKEA is finally, you know, to tell you the truth because they're using thread and they're using matter and that kind of stuff. So, IKEA is starting, but it's just starting. It's all nascent. So, I think that the ones that are going to be successful are the ones who are going to finally deliver the nest vision that we really had. But it's still I you know if I was I would be rebuilding Nest right now you know and and because in three to four years you can be there to get what we've been talking about that we had the vision for in 2010 11.

你现在显然在投资，但你还有创业的冲动吗？你觉得你还能再创办一家伟大的公司吗？

You're obviously investing right now, but do you still have the entrepreneur bug? Like do you think you have another great company in you?

哦，我现在脑子里有一百多家伟大的公司。我直接投资了很多项目，也间接投资了很多。所以我现在有超过一百家公司，其中一些我每天都在合作。比如今天我就和一家公司合作，帮他们对接苹果。我在几家公司担任影子 CEO，还在一些公司做产品设计。所以我能在很多不同的领域做这些事。

Oh I have over a hundred great companies in me right now. I I we you know I invest directly in many many things and I invest indirectly in many things. So I have over hundred companies now and something companies where I'm working with them every day. Like today I was working with a company and helping them with Apple. I'm a shadow CEO at a couple of companies. I'm doing product design at companies. So I get to do this in many different domains.

影子 CEO 是什么意思？

What does it mean to be a shadow CEO?

就是有 CEO，然后他说‘请帮帮我’。所以我就在幕后，我会说‘你要不要这样想？’运营、人员编制、融资方式、产品设计等等，我扮演各种角色，他们请求帮助。我就去做，然后参与其中。

So there's the CEO and then he's like please help me. So I get to be behind I'm like would you think about it this way? operations, headcount, how should we should do funding, product design, and so I have various roles like that where they please help. And so I do and I go in and I get to

你现在花时间最多的几家公司是哪些？

A couple companies are spending the most time on right now.

呃，我花了很多时间在 Orionis Biosciences 上。他们做 AI 药物发现，我们有药物进入 FDA 二期临床，而且确实有效，我在帮忙推进。所以我花了很多时间，我在那里不是影子 CEO，而是董事会成员，提供帮助。呃，我还在 Menllo Micro。

Uh I'm spending a lot of time with Orionis Biosciences. um they are doing AI drug discovery and we have drugs in uh FDA phase 2 clinical where they're actually working um and helping with the with that. So I'm spending a lot of time I'm not a shadow CEO there but I'm I'm on the board and helping out. I'm um I'm at Menllo Micro.

好的。

Okay.

Menllo Micro 是做 MEMS 开关的。我们有晶体管和继电器，MEMS 开关介于两者之间。这是一项不可思议的新技术，我们已经研究了 10 年，而他们几乎研究了 40 年。它用于驱动 AI 数据中心、机器人等所有东西。这绝对是一个非凡的东西，我帮忙做业务开发、产品规划、战略、融资之类的事情。

Menllo Micro um is a um a MEMS switch. So we have transistors and we have relays. Mems switch sits in the middle and it's an incredible new technology that we've been working on for 10 years. They've been working on it for almost 40 years um to power AI data data centers and robots and all these other things. um it's a absolutely phenomenal thing and helping with the business development, the product plans, the the strategy, the financing, that kind of stuff.

嗯。

Yeah.

嗯，让我想想，我还在做什么？呃，我今天早上刚做了一些 VR 相关的工作。哦，真的吗？

um let's see, what else am I working with? Uh I'm doing I was just doing some VR work this morning. Oh, really?

是的。

Yeah.

你相信 VR 吗？

Are you a believer in VR?

我相信元宇宙。我相信 VR 在间歇性 B2B 和一些游戏中的应用。但只是间歇性的，你不是生活在里面，不像有些人说的那样会住在里面。不是那样的。但在协作场景中，它很有价值，而且主要是 B2B。当然还有游戏，但那只是边缘领域。你知道它很大，但在我看来仍然是边缘。不过在做在线设计时就不一样了。所以我们投资了一家叫 Gravity Sketch 的公司。

I believe in I believe I the metaverse. I believe in VR for episodic B2B and some gaming. um um but episodic so you're not living in it like you know some people are saying you're going to live in it. It's like no. But where it's collaborative, it's high value and mostly B2B. Sure there's gaming and that's a a fringe. You know it's big but it's still a fringe in from my perspective. but where you're doing design online. So, we're in a company called Gravity Sketch.

是吗？

Okay?

Gravity Sketch 是排名第一的。你可以把它想象成 VR 世界里的 Figma，你在 3D 空间里做协作设计，可以做标记，也可以创作。我们服务所有大公司，比如 Stellantis、耐克、福特，每个人都用我们做汽车设计、鞋类设计、家具设计。所以，嗯，我相信它在这些方面的价值。他们有团队戴着耳机，做一小时左右，进去再出来，或者评审汽车之类的。我绝对相信 VR 在这些场景中的作用。但对于 Meta 和苹果推销的大部分东西，我完全不相信，因为我从 1988 年就开始研究 VR，看到了它的问题。

And Gravity Sketch is the number one. Think of it as Figma in the in the VR world where you're doing collaborative design in a 3D space and you can do markups. You can create in it. And we're at all the biggest companies. We're at Stellantis, Nike, the Ford, everybody uses us to do car design, shoe design, furniture design. So, um, so I believe in it for that. So they have teams of people wearing headsets, doing this for an hour or something, getting in it, coming out, you know, or reviewing cars or whatever. And I believe in VR for that for sure. But for most of the stuff that was pitched by, you know, Meta and Apple, I don't I don't believe in it at all because I've been working on VR since 1988 and I've seen the issues with it.

这又回到了我们对话的一些大主题。你认为哪家公司最有能力推出像 iPad 那样规模的新设备？如果你必须预测一家公司能推出达到那种文化影响力的设备，你会押注谁？

This is circling back to some of the big themes of this conversation. Like what company do you think is best positioned to produce a new iPad-scale device? Like someone like what if you had to make a prediction about a company producing a device that reaches our culture at that scale? Who are you betting on?

希望苹果能做出 iPad 手机或 iPhone 平板，基本上就是可折叠的 iPhone。

Hopefully Apple makes the iPad phone or the iPhone pad, which is basically a foldable iPhone.

哦，好的。好的。对。对。把 iPad 的好处带到 iPad 上。

Oh, okay. Okay. Yeah. Yeah. Bring the benefits of the iPad to the iPad.

是的。只要加入折叠功能，你知道这已经传闻很久了，但操作系统 iPad OS 之类的，新设备或者……我认为这对苹果来说是一个有意义的细分市场，可以继续增长，并确保他们不会因为折叠屏手机而失去市场份额。但我确实认为有很多使用场景，如果你能随身携带并打开它，尤其是在某些 AI 场景下，对吧？嗯，那会非常强大。对。因为 iPad 没法塞进口袋。你想要一个口袋大小的，你知道，类似 iPad 的体验，可以用语音交互，不需要键盘之类的东西。那会非常酷。

Yeah. Just bring the foldable stuff, and you know it's been bandied about and stuff like that, but the operating system iPad OS and stuff, new device or... I think it's a meaningful segment for Apple to continue to grow and to make sure that they don't lose market share to people with foldables. But I do think there's a lot of use cases where if you can carry it with you everywhere and you can open it, especially in some cases with AI, right? Um, that it could be pretty powerful. Right. Because an iPad you just can't slip in your pocket. You want a pocketable, you know, kind of iPad experience where you can use voice so you don't need a keyboard, all that stuff to interact with it. Could be pretty damn cool.

如果你是苹果 CEO，你会专注于这个吗？

Is that what you would focus on if you were Apple CEO?

如果我是，我会做很多事情……

I have many things that I would do if I was...

好吧，跟我们说说其中一些。

All right, give us some of them.

好吧，我跟你说一些。嗯，第一件事，我认为他们真的搞砸了的是移动出行。你知道，苹果汽车是真实存在的，他们把它砍掉了。但还有其他方式可以做到。

Well, I'll give you some of them. Um, the first thing I think they really flubbed on was mobility. You know, there was the Apple car and we all know that that was a real thing and they killed it. But there's other ways of doing it.

你认为他们应该继续做下去？

You think they should have kept going?

不，不，不，不。苹果应该……苹果是一家重新定义生活某些方面的公司。它重新定义了桌面出版，对吧？或者整个出版业。它重新定义了音乐。它重新定义了很多东西。所以，他们应该重新定义移动出行。不要做一辆四轮汽车去和其他人以及中国人竞争。

No, no, no, no. Apple should... Apple is a company that redefines certain aspects of life. It redefined what it was to be desktop publishing, right? Or publishing in general. It redefined music. It redefined various things. So, they should redefine mobility. Don't make a four-wheel car that competes with everybody else and the Chinese.

你会如何改变移动出行？

How would you change mobility?

两轮、三轮和轻量四轮车辆？

In two, in three, and lightweight four-wheel vehicles?

你怎么做到？

How do you do that?

是的。那种 14 岁的孩子就能用，并且会想一直用下去的东西。对吧。不是那种 Lucid 之类的……你知道，你得想想……当史蒂夫·乔布斯和我在 2008、2009 年走在苹果园区时，我们讨论过苹果汽车，它会是什么样子？他真的很想，我们需要做……他认为革命性的是大众汽车，对吧？人民的车。他问，什么是人民的车？下一代人民的车是什么？城市里会用什么车？现在你看到了菲亚特 Topolino、Twizy 等等各种东西。我住在欧洲，对吧？所以你能看到这些东西到处跑，而且卖得疯狂。你会想，那是苹果吗？或者两轮苹果？或者三轮苹果？你知道吗？

Yeah. That the kids are like at 14 can use and go, I want to keep using it. Right. Not a Lucid blah blah blah for just... You know, you got to think... When Steve Jobs and I would walk around the Apple campus back in 2008, 2009, we talked about the Apple car, what would it be? And he was really like, we need to do... He thought it was revolutionary was the Volkswagen, right? The people's wagon, the people's car. And he was like, what's the people's car? What's the next generation people's car? What's going to be used in the cities? And now you're seeing the Fiat Topolino, you see the Twizy, you see all these different things. I live in Europe, right? So, you see all this stuff running around and they're selling out like crazy. And you're like, that Apple? Or two wheels Apple? Or three wheels Apple? You know?

所以，你会说继续做那个车辆项目，坚持这个车辆愿景。

So, you would say go keep working on that vehicle, work on this vehicle vision.

哦，我认为那只是一件事。我认为移动出行是应该被解决的问题，苹果可以用一种非常不同的方式来解决，而且它拥有各种品牌。另一件事是，我认为那是个烟雾弹，它只是……或多或少是那种东西。坦白说，我认为他们的配件产品线可以做得更多……他们可以更深入地投入。嗯，在这方面有很多事情可以做。显然还有眼镜……

Oh, I think that's only one thing. I think mobility is something that should be addressed and Apple could address it in a very different way than it has, and it has all kinds of brands. This other thing is I think it's a red herring and it's just... it's more or less a kind of thing. Frankly, I think that their accessories lineup could go much... they could lean in much more into that. Um, lots of things to be done around that. Obviously glasses...

比如更偏设计或美学方面？

Like more designer or the aesthetic piece of it?

不，不，只是可用性。不，不，美学。不，不，不，不，不。我们说的是新市场，他们可以追求的新东西。

No, no, just usability. No, no, aesthetics. No, no, no, no, no. We're talking about new markets, new things that they could go after.

直接跟 Oura 竞争。

Go after Oura directly.

呃，当然。你知道，这还用说。嗯，

Uh, sure. You know, like duh. Um,

你现在看好 Oura 吗？

Are you bullish on Oura right now?

说实话，我有一个。不，不，不。说实话，我在 Oura 之前就投资了那家公司。它叫 Motive。

I got one on, you know, to be truth be told. No, no, no. To be truth be told, I was invested in the company before Oura. It was called Motive.

好的。

Okay.

我们做了第一款智能戒指。我们拥有所有专利、所有基础技术等等。而来自苹果的团队在做出戒指后就想做别的东西。我说，“伙计们，专注于业务。”他们说，“不，我们想做其他可穿戴设备。”我说，“不，专注于业务。”所以他们搞砸了。然后发生了什么？Oura 最终买下了我们所有的专利。

We made the first smart ring. We had all of the patents, all the fundamentals and everything. And the team who was from Apple just wanted to do something other than a ring after they got the ring. I'm like, "Guys, focus on the business." and they're like, "No, we want to make, you know, some other wearable." I'm like, "No, focus on the business." So, they lost it. And then what happened? Oura ended up buying all of our patents.

我不知道这事。

I didn't know that.

所以，对我来说，我们比 Oura 早了四年在做这件事。现在我在 Nest 的团队成了 Oura 的市场团队，我很喜欢那些人，他们做得很好。我只是希望 Motive 能成功而不是……

So, to me, like we were doing it four years before Oura did. And now my team from Nest is the marketing team at Oura, which I love those guys and they're doing great. I just wish we would have Motive would have been it instead of...

苹果现在应该考虑做戒指吗？还是他们没在想？

Is Apple supposed to be thinking about a ring right now or if they're not?

如果他们没在想，那才奇怪。而且对他们来说太容易了。他们拥有所有芯片。就像……他们没有芯片团队？不，他们什么都有。他们拥有制作一款出色戒指所需的一切。这还用说。而且他们拥有所有复合材料，你知道，材料，一切。这简直是显而易见的。就像那个别针一样。显而易见的事情。他们拥有所需的一切，只需要有勇气把它做出来。就这么简单。

It'd be crazy if they weren't. And it's so easy for them to do. They have all the chip. Like... doesn't have chip guys. They don't have all that. They have everything they need to make an incredible ring. Like duh. And they have all the composites for, you know, the materials, everything. Like it is a no-brainer. It's just like the pin. No-brainers to do. They have everything they need, they just need to have the gumption to get it out and do it. It's really simple.

还有其他配件吗？

What are other accessories?

嗯，我认为有更聪明的方式，比如做一个内置 iPod 的 AirPod。所以，我认为他们需要重新推出 iPod。

Um, I think there's smarter ways of, you know, making an AirPod that has an iPod in it. So, I think they need to bring back the iPod.

好的。我喜欢这个想法，因为人们不想一直带着手机。

Okay. I like this because people don't want to have their phone with them all the time.

嗯，有两个原因。一是怀旧价值，每个进入苹果生态的人，很多都是因为 iPod 入门的。对吧。

Well, there's two reasons. One is there's the nostalgic value of everybody who got into Apple. A lot of them got in because of the iPod. Right.

对。所以人们想要……他们喜欢怀旧……

Right. So, people want... they would like the nostalgic...

是的。大多数人是因为在 iPod 时代，美国只有 1% 的市场份额，全球更少。所以大多数人的设备是因为他们没有 Mac，也没买 Mac，因为他们有 PC，所以他们不得不买一个苹果设备，最可能的就是 iPod。所以从怀旧角度看，对于那些说“哦，我记得 iPod”的人来说，他们不再想要干扰了。对吧。他们不想要干扰。总之，第三件事……

Yeah. Most people is because at the time of the iPod, 1% market share in the US only, not around the world. So most people's device because they didn't have a Mac and they didn't buy a Mac because they had a PC, so they had to get an Apple device and it was most likely an iPod. So iPod from the nostalgic point of view for all those people who's like, "Oh, I remember the iPod." But they don't want distractions anymore. Right. They don't want distractions. Anyway, the third thing...

抱歉，抱歉，苹果。谁不想要干扰？

Sorry, sorry Apple. Who doesn't want distraction?

嗯，很多人想要那种纯粹的体验。所以，那些热爱音乐的人……我刚买了一个唱片机，所以我深有体会，对吧？

Well, a lot of people want that pure mystery. So, the music lovers who love... I just bought a record player, so I'm deep in this, right?

你看到索尼这周做了什么吗？

Did you see what Sony did this week?

没有。

No.

索尼砍掉了电视业务，卖给了 TCL，然后回头去做黑胶唱机了。他们刚发布了两款新唱机。砍掉电视，带回唱机。你说呢？对吧。所以 iPad 也有怀旧元素吗？我觉得有。没错。在这个愿景里，iPad 是软件还是硬件设备？你是说它存在于什么形态？

Sony killed their TVs and sold them off to TCL. And they went back to the turntables. They just released two new turntables. They kill the TVs and they bring back turntables. You tell me. Right. So, is there a nostalgia element for the iPad? I think so. I think that's right. Is the iPad in this vision software or it's a hardware device? You're saying where does it exist? It's what format?

这个……我不想把所有想法都透露出来。

It's... I don't want to give away all my stuff.

你留着呢。有人跟你聊过这个吗？你知道，之前有报道说下一任 CEO 的事，你的名字被提及了。有人试探过你吗，还是你觉得得有所保留？

You're saving it. Has anybody talked to you about it? You know, there was obviously reporting about the next CEO and your name was bandied about. Has anyone done any feelers or you're like, I got to hold back?

我收到了很多联系。我只想说谢谢那些前苹果员工和现任苹果员工，他们联系我，试图拉我参与。你知道，苹果会在需要的时候对下一任 CEO 做出正确决定。我爱这家公司，从 1980/81 年我拥有第一台 Apple II 开始，它就融入了我的血液。从 10 岁或 11 岁起，它就一直是我生命的一部分。如果有人打电话来，比如董事会或蒂姆打来，我都会乐意接听，对吧？就这样。我会以他们认为需要的方式提供帮助。所以那个配件，那是个很棒的东西……我对它很着迷。然后还有其他他们应该做的事，还有一整个产品类别。但我们以后再说。

I've gotten a lot of incoming. I want to just say thank you to those ex-Apple employees and those current Apple employees who have reached out and have tried to canvas me and try to get me involved. You know, Apple will make the right decision when they need to about who the next CEO is. I love the company. It's been in my blood since 1980/81 when I had my first Apple II. You know, it's been the thing since I was 10 or 11. If anyone calls, you know, from the board or if Tim calls, whatever, I'll be happy to pick up the phone, right? And that's it. I'll help them anyway they think they would like to be helped. So the accessory, that's a great... I'm fascinated by that. And then there's other things they should do. There's a whole other product category. But we'll leave that for later.

好的，我们以后再说。那基础模型公司呢？你怎么看 OpenAI、Anthropic、xAI？我知道你想聊这个。

Okay. We'll leave that for later. What about the foundation model companies? What's your read on OpenAI, Anthropic, xAI? I know you want to go there.

是的，当然。这是当今最大的技术问题。它们是伟大的企业吗？我想这是第一个问题。它们是好的企业吗？它们必须真正瞄准应用才能成为伟大的企业。平台不是业务，应用才是伟大的业务。这一点已经被证明过了。上一个伟大的平台业务是 Windows 95，或者说 Windows。除此之外没有别的平台业务了。好吧。你也许可以把 AWS 或 Google 之类的算作平台，但那是 B2B 的东西。这其实不是关于应用，而是关于平台。所以当你把 OpenAI、Google、Anthropic、Mistral 等等放在一起看，再聊聊中国甚至韩国的公司。听着，有一个关键点。Dario 在 Anthropic 做得很好，把它变成了一个真正的业务，对吧？而编程，因为它非常以语言为中心，是 LLM 的一个不可思议的应用，对吧？我们看到了，对吧？现在他们的非编程版 Claude 和编程版 Claude 都太疯狂了。我正在用，我在玩，我想，‘天哪，我又开始编程了。我多年没编程了。’我不是程序员，但我下载了桌面版。有点……我觉得我没有文件，所以作为用户我有点困惑，因为我在 Google 上有东西想让它们用，但我觉得我需要在他们的网页浏览器里用。我是说这些……我必须说，他们有一整套应用和训练，因为感觉 Claude 还是 GitHub 模式，非常极客。如果你看这个，它只覆盖了很少的人。他们必须用不同的方式让人们参与进来。它仍然不够亲民。我用的时候还是困惑，甚至安装的时候，想运行什么？‘好的，终端这个。’有多少人会这么做？有其他方法可以绕过这个。我相信他们最终会做到的，他们不可能做不到。但话说回来，他们在专注和解决编程、护栏等所有事情上做得很好。他们正在努力实现盈利，成为一个真正站得住脚的业务，对吧？所以我赞赏他们所做的一切。他们是前 OpenAI 的人，对吧？他们确实在瞄准。这很棒。然后我们来看 OpenAI。OpenAI 现在是在‘散弹枪式’地乱打，对吧？散弹枪式。你看，这周有传闻，而且我确信是非常强烈的传闻，他们试图达到 500 亿美元，对吧？500 亿美元。就在几个月前还是 50 亿，再之前是 5 亿。除非你在市场里并且理解这个，否则你不知道这是巨额资本，连大型财富 500 强公司都拿不到。所以我认为 Sam 在玩一个非常危险的游戏，就是‘大到不能倒’，对吧？他让每个人都给他们钱，包括大型政府等等，所以他们大到不能倒。对吧？这就是 Sarah Frier 说的‘兜底’那回事。是的。那个兜底，就像‘不，不’，然后他们又在往回缩。他们写过一些文章暗示……你不认为 OpenAI 内部没有那些讨论吗？比如，‘是的，这些政府最终会给我们兜底，因为我们会搞垮一切：半导体，一切。’所以我认为他们在玩一个极其危险的游戏，尤其是那些循环交易之类的。那里有‘实质’吗？我相信有实质，但 Google 会赢，对吧？Google 拥有一切，从芯片到数据中心，一直到应用，还有海量数据。我希望他们有一个更好的设备团队。那方面很糟糕，对吧？我会继续说，你知道的。但归根结底，他们拥有所有必要的东西，加上一个每季度能提供数十亿美元资金的广告业务，还有分发渠道。它无处不在。有趣的是，我觉得有那么多 AI 仇恨者，他们显然在用 Google 搜索，然后会得到 Google 的 AI 结果。所以我觉得，或者说 Gemini 从根本上在提供结果。所以存在一个奇怪的世界，人们讨厌 AI，却一直通过 Google 使用它。

Yeah, yeah, of course. That's the big technology question of the day. Are they great businesses? I guess that's the first question. Are they good businesses? They have to really target applications to be great businesses. Platforms are not businesses. Applications are great businesses. And that was proven. The last great platform business was Windows 95, or Windows. There's been no platform business other than that. Okay. So you could call maybe AWS or Google or whatever, but that's a B2B thing. It's not really about applications, not platforms. So when you stack up OpenAI, Google, Anthropic, Mistral, you can pick whoever, and then let's talk about the Chinese and even the Koreans and stuff like that. Look, there is a point. And Dario is doing a great job at Anthropic turning it into a real business, right? And coding, because it's so language focused, is an incredible application for an LLM, right? And we're seeing that, right? It's just crazy with now their non-coding version of Claude as well as the coding version. I'm using it now. I'm playing. I'm like, 'Oh my god, I'm getting back into coding. I haven't done coding in years.' I'm not a coder, but I got the desktop version. It's a little... I feel like I don't have files, so I'm a little confused as a user that I have stuff on Google that I want them to use, which I think I need to use in their web browser stuff. I mean these are... I must say, look, I must say that they have a whole set of application stuff and training because it still feels like it's a GitHub model for Claude. It's very geeky. And if you look at that, it's a very small number of people. They have to come from a different way to get people involved. And it's still not approachable. I'm still confused when I'm using it or when I'm even installing it and trying to run what? 'Okay, terminal this.' Like, how many people do that? There's other ways of getting around that. And I'm sure they'll get there. There's just no way they're not. But again, they're doing a great job of really focusing and figuring out that stuff around coding, around guardrails, all those things. And they're trying to get to profitability so that they're a real standing business, right? So I commend them for all that they're doing. They were ex-OpenAI people, right? And they really are targeting. So that's really great. Then we go to OpenAI. OpenAI right now is spray and pray, right? Spray and pray. Look, this week it was rumored, and I'm sure it was a very strong rumor, they're trying to reach $50 billion, right? $50 billion. Just a few months ago it was $5 billion, and just before that it was $500 million. I don't think, unless you've been in the market and understand this, it's massive amounts of capital that even large Fortune 500s can't get access to. And so what I believe Sam is playing, and it's a really dangerous game, is 'too big to fail', right? He's getting every single person to give them money, including large governments and everything else, so they're too big to fail. Right? Well, this was the whole Sarah Frier 'backstop' thing. Yeah. The backstop which is like 'no, no' and they're backpedaling. They've written stuff that suggests... You don't think they're not having those discussions at OpenAI? Like, 'Yeah, these governments will backstop us at some point because we're going to crash everything: the semiconductors, everything.' So I think it's an incredibly dangerous game they're playing, especially with the circular deals and everything. Is there a 'there there'? I believe there's a 'there there', but Google's going to win it, right? Google has everything from the chips to the data centers to all the way to the applications and tons of data for all this stuff. I wish they had a better devices team. It sucks there, right? I'll go on, you know, whatever. But at the end of the day, they have all the things necessary, plus an ad business that can fund billions of dollars a quarter, and distribution. It's just everywhere. It's funny. I feel like there's so many AI haters that, you know, they obviously Google things and so then they'll get the AI results from Google. And so I feel like there's, or Gemini is fundamentally delivering the results. So there's this weird world where people hate AI and still use it all the time via Google.

我明白你的意思。你知道，这更像是‘我讨厌它会对世界造成的影响。我讨厌……但它对我有用。’

I hear what you're saying. You know, there's this... it's more like 'I hate what it's going to do to the world. I hate... but it works for me.'

所以我觉得这里有点不协调。但我必须称赞 Sundar 和团队，他们经历了一场生存危机，我认为他们把文化清理得足够干净了。我敢肯定还没完全干净，但足以让他们说我们面临生存危机，我们要开始干活了。我们得回到正轨。之前太随意了。你还记得 Gemini 刚出来时的那些东西吗？我们不能这样那样，安全在 MAGA 话语里就是“觉醒”，对吧？太安全、太平衡了。你会想，得了吧。那么，你觉得 Sergey 到底有多少功劳，或者内部到底是谁在真正推动？

So I think there's a little dissonance there. But I have to credit Sundar and the team that they've had an existential crisis and I think they cleaned up the culture enough. I'm sure it's not fully clean, but enough to say we have an existential crisis, we're going to get to work. We got to go back. It was too casual. If you remember the first things of Gemini that came out, we can't blah blah blah safety in the MAGA words, woke, right? Too safe and too balanced. You're like, come on. So, do you have a read on whether how much Sergey really gets credit or what the internal dynamics are there in terms of who's really driving?

哦，我听到一些风声。但我不想多评论。看结果吧。结果确实他们搞砸了几次，但他们现在专注投入了。所以我必须把功劳都给他们，他们扭转了局面。Google 这艘船很大，但感觉它真的在朝正确方向前进。有时候确实需要濒死体验才能激发人们的最佳状态。我说改变文化的唯一方法就是经历濒死体验。

Oh, I hear a little bit. I don't really want to comment. Look at the results. The results are sure they fumbled a few times, but they're locked in and engaged. So I have to give them every credit for turning the ship. It was a big ship at Google, but it feels like it's really going in the direction. Sometimes it does take a near-death experience to get the best out of people. I say that's the only way you change cultures is through a near-death experience.

你认为他们会创新吗？我的意思是，他们写了第一行代码。Dennis 很厉害，他非常聪明。他们是第一个提出这个方向的人。所以现在只是在追赶，但这是大卫对歌利亚。OpenAI 可能有很多现金，但他们的资源远远不够。今天，有一件很有趣的事很多人没意识到：如果你看看 OpenAI 想在哪里赚钱，Google 现在在哪里赚钱，然后你说哦，Anthropic 在做应用或 API。

Do you think they'll innovate? I mean, they wrote the first code. Dennis is amazing. Dennis is incredibly smart. They were the ones who came up with this first course. So now it's just playing a game of catch-up, but it's really David and Goliath. OpenAI might have a lot of cash but they don't have anywhere near the resources. And today, here's the thing that's really interesting and people don't realize this: if you look at where OpenAI is trying to make its money and where Google is making its money today, and you say oh and Anthropic's making applications or API.

是的，API 和编码之类的。他们正在做，而且会成功。OpenAI 什么都做。他们挑挑拣拣，哦，我们要关注健康，然后他们又宣布了这个。他们不惜一切代价让应用更具体。现在他们又宣布了广告。他们是在乱枪打鸟，希望能成功，因为只有那么多人愿意付 20 美元或 200 美元。所以他们必须转向广告模式来获得更多收入。

Yeah, API and coding, that kind of thing. They're doing that and they're going to get there. OpenAI is doing everything. They're picking out, oh, okay, we're going to worry about health, and then they announced this. They're doing whatever it takes to get more application specific. And now they're announcing ads. They're doing spray and pray to hope to go because they can get only so many people to pay $20 or $200 or whatever it is. So they have to get to this ad model to get more revenue.

中国人做得太聪明了，对吧？

The Chinese are doing this so smart, right?

嗯，也许是因为他们没有英伟达的芯片，但我一直喜欢资源受限的环境，因为你会更有创造力。所以你看 DeepSeek，你看其他这些东西，它们都是开源的。OpenAI 今天的利润，或者说不是利润，应该说是 OpenAI 今天的收入，如果那些开源模型或 Meta 的模型成为现实，而且它们足够接近，我知道很多开发者都在用中国模型或其他开源模型。如果 OpenAI 在应用端没有达到那个水平，中间的一切都会消失。他们没有数据中心，没有其他任何持续的技术优势。

Well, maybe because they don't have the Nvidia chips, but I always like resource constrained environments because you're more creative. So if you look at DeepSeek, if you look at these other things, and they're open source. The profits of OpenAI today, or not profits, I should say just revenues of OpenAI today, if those open source models or a Meta model becomes reality and they're close enough, and I know a lot of developers using the Chinese models or other open source models. And if OpenAI doesn't get there at the application side, everything in the middle is gone. They don't have data centers. They don't have anything else on a persistent technological advantage.

没错。你将会看到的是应用和基础设施。这就是最终格局。而在中间，就像每个人都有开源这个、开源那个，里面没什么真金白银。这就是为什么人们说 SaaS 不再赚钱了。所以奇怪的是，基础设施、规模和应用才是获取客户或平台的地方。我相信这就是趋势。而 OpenAI 卡在这里，因为他们没有任何数据。他们一直明确表示不想要数据，但他们不在这里。而 Google 已经在应用端了。还有 Claude 或 Anthropic，也许有一天苹果也会加入。

Correct. What you're going to see is application and infrastructure. That's where it's going to settle. And in the middle, it's going to be like everyone's got open this and open that, and there's going to be no real money in it. That's why people are saying there's no money in SaaS anymore. So you're going to see, oddly enough, infrastructure and scale and application is where capturing the customer or capturing the platform is going to be. And that's where I believe we're headed. And OpenAI is stuck here because they don't have any data. They've been very clear they don't want it, but they're not here. And you already have Google up here at the apps. You have Claude or Anthropic there, and maybe one day Apple up there.

你不觉得通用助手就是一个应用吗？我是说为 OpenAI 辩护一下，他们有关于我的所有上下文。我用得越久，他们拥有的上下文就越多。

Don't you think like general purpose assistant is an application? I mean to defend OpenAI, they have all this context on me. The longer I use it, the more context they have.

Google 多得多。苹果也多得多。我有点跑题了，但这就是为什么他们想要一个设备。这就是为什么他们需要所有这些来拼凑出这个图景。他们需要上下文。人们说他们应该做手机，但那样他们只能做安卓手机，因为他们不够。好吧，也许他们能找到更多钱，但他们正在烧钱。另一件人们没意识到的事是，随着他们增加更多容量、更多数据中心，每个用户都意味着他们需要花更多负钱来留住这些用户。

Google has much more. Apple has much more. And I'm drifting, but that's why they want a device. That's why they need all these things to sort of have this picture. They need the context. People are saying they should make a phone, but then they have an Android phone because they don't have enough. Well, maybe they can find more money, but they're bleeding money. The other thing that people don't realize is as they add more capacity, as they bring on more data centers, each of those users is more negative money that they have to spend to keep those users.

对。所以随着他们扩大规模，却没有收入来源，他们每个用户亏的钱更多，因为用户更多了。

Right. So as they scale up and they don't have a revenue well, they're losing more money per user because they're getting more users.

这就像以前的拼车时代。我们靠量来弥补。单位经济是负的，但靠量能补回来。这就像，不，你补不回来。

It's the old ride sharing days. We'll make it up in volume. Unit economics are negative, but we'll make it up in volume. It's like, no, you won't.

我的意思是，你认为 OpenAI 会破产吗？如果他们现在要 500 亿美元，你能多快烧完？

I mean, do you think OpenAI goes bankrupt? If they're asking for $50 billion now, how fast are you going to burn through that?

对。我们还没聊 xAI。你怎么看？

Right. We haven't talked about xAI. Do you have a view on that?

听着，Elon 玩的是非常不同的游戏。他从特斯拉开始，然后是 Twitter 变成了 X，现在 Twitter 或 X 成了特斯拉的一部分。然后他做数据中心，现在那也要成为特斯拉的一部分。现在他想做……我最新听到的是 SpaceX 不会上市，SpaceX 会被特斯拉收购。

Look, Elon plays a very, very different game. He started with Tesla, then it was Twitter that turned into X, and now Twitter or X is part of Tesla. And then he was doing data center stuff, and now that's going to be part of Tesla. Now he wants to make... And what I heard latest is SpaceX is not going to go public. SpaceX is gonna get bought by Tesla.

是的，我想我可能在这个播客里预测过。对，显然是这样。

Yeah, I think I made a prediction maybe on this podcast. Yeah, I think clearly so.

他只是在整合，变成一个 conglomerate。

He's just rolling up and just becoming a conglomerate.

你是说这有点像纸牌屋，总是下一个东西？

Are you saying it's sort of house of cards, always the next thing?

哦不。我是说这是纸牌屋。SpaceX 是家了不起的公司。了不起。如果 Elon 没有做一年前那些事，特斯拉会好得多。那些事就是失去专注，去做别的事，涉足政治，然后惹恼全世界，每个人都说我不买那个产品了，因为你玷污了品牌。

Oh no. I'm saying it's house of cards. SpaceX is an incredible business. Incredible business. Tesla would be much better off if Elon didn't do what Elon did a year ago. Which is like lose focus and go do different things, be in politics, and then piss off the world, and everyone said I'm not buying that product because you tainted the brand.

对。所以问题是，好吧，我需要钱来维持特斯拉的生存。

Right. So the thing is, it's like okay, I need money to keep Tesla alive.

我要把 SpaceX 放进去，从 SpaceX 那里筹一大笔钱，因为特斯拉的股票会涨，这样我就能继续给特斯拉输血。你懂我意思吧，钱的问题？就是这么回事。他最终需要保护他的投资者，因为这一切只有在投资者觉得“他会以某种方式照顾我”时才行得通。他会以某种方式收购 Solar City，他会做一些交易，让下一件事成功，然后他所有的好点子都会拯救其余部分。但你不觉得 XAI 是很大一部分。

I'm going to put SpaceX in, raise a bunch of money off SpaceX, and make because the Tesla stock's going to come up, and so I'll be able to continue to float Tesla. You know what I mean with money, right? So that's how that works. He ultimately needs to protect his investors because this whole thing only works if he's like, 'Well, he's going to look out for me somehow.' You know, he'll acquire Solar City some way. He'll do some deal that the next thing works out and like whatever good idea he has will save the rest. But you don't see XAI as sort of a big part.

我觉得 XAI，你看，他就是在做他 XAI 那套。它可能又变成什么，你知道，哦，我们现在在做什么？给人下药什么的，然后说“我不知道”，然后就有三百万张非法图片在流传。所以，你知道，这种来回打擦边球，我们就在法律的灰色地带玩，然后退回来。就像自动驾驶十年前就该上路了。他让每个人都付钱，但没人得到他们应得的，对吧？对。所以他 somehow 总能逃脱，就在那条线上玩。

I think XAI, look, he's doing what he does with XAI. It can be something, you know, again. Oh, now we're doing, you know, what is it? Drobing people on like, and then it's like, well, I didn't know. And it's like, now we have three million images of illegal images running around. So, you know, there's this kind of skirting back and forth like, we just play within the gray area of the law and then we pull back. Just like self-driving is going to be up and running 10 years ago. And he makes everyone pay, but yet no one gets delivered what they were, right? Right. So he somehow gets away with everything of playing just the line.

作为领导者，我们有点触及了“觉醒文化”可能如何扼杀了谷歌。现在我们处于这种虚无主义时期。你认为科技应该追求什么，才能让他们的建设有某种意义支撑？

As a leader, you know, we sort of touched on how woke potentially killed Google. Now we're in this sort of like nihilist period. Where do you think tech should aspire to be in terms of having some meaning underpinning what they're building right now?

嗯，你看，我是个乐观主义者，也是个可能玩不了这个游戏的人，因为我更看重高道德标准、高伦理。当我说“你将成为我的合作伙伴”，你就是我的合作伙伴，我们一起做事，而不是像 OpenAI 那样“我要和所有人合作，我们是排他的，但其实不是”，你知道，这有点像今天的社交媒体数据约会套路，对吧？所以，我觉得，我不知道，我总在问自己：如果史蒂夫·乔布斯在这里，他会如何应对这个环境？

Well, look, I'm an optimist and I'm also a guy who, you know, maybe I couldn't play in this game because I'm much more of the high moral standards, high ethics, high, you know, when I say you're going to be my partner, you're my partner and we work together on stuff, not open AI and I'm going to partner with everyone and we're exclusive, but we're not, you know, that's kind of today's social media data dating scheme, you know, right? So, I think that, you know, I don't know if I always ask myself all the time. I'm like, how would Steve Jobs fare in this environment if he was here?

嗯，他会吗？我没有答案。你觉得他会进白宫吗？你有点不得不……

Well, would he? I don't have an answer. Do you think he would be in the White House? You sort of have to...

我不知道。我觉得加文·纽森对蒂姆·库克说了些相当宽容的话，大意是他必须玩这个游戏。但嗯，很难说史蒂夫·乔布斯会不会，你知道，以我对他的了解，以及和他共事十年的经历，我不认为他会玩，但我不确定。我跟着玩了，但我不确定。

I have no idea. I think Gavin Newsom said something fairly charitable about Tim Cook, which is sort of, you know, he has to play the game. But yeah, it's hard to know if Steve Jobs would have, you know, knowing what I know and working alongside him for a decade, I don't think he would have played, but I don't know. I played along, but I don't know.

但说真的，我觉得，你知道，我认为 Dario 做得很好。他在达沃斯站出来说，这全是胡扯，这不对，这不对。Dario 站起来说 LLM 不是关键，还有更多东西要来。所以，我觉得我们需要更多这样的领导者，因为太多价值观有问题的人在运营这些东西，我希望他们能站出来，或者更直言不讳，对吧？因为蒂姆是个好人，对吧？蒂姆有很高的价值观。

But no, really, I think that, you know, I think Dario is doing a great job. He got in front of Davos saying, you know, this is BS, this is not right, this is not right. Dario got up and saying LLMs aren't the thing, you know, and there's more to come. So, I think we need more leaders like that because we have too many, you know, people with questionable value systems, running these things and, you know, I wish they were there or they were more vocal, right? Because Tim's a great guy, right? Tim's got high values, you know.

我同意你。Dario 他们，Anthropic 刚刚发布了一部针对缺陷的宪法，他们非常深思熟虑，我发了条推文。在反智主义盛行的时刻，他们却让哲学博士、学者们以真正有原则的方式思考，我真的很钦佩他们的做法。

I mean I agree with you. I mean Dario, they just Anthropic just released a constitution for flawed where it's like they're being extremely thoughtful and I tweeted something. It's like in a moment of anti-intellectualism here they are it's like let's have some philosophy PhDs academics you know think through you know within a real principled way like I really admire what they're doing.

我觉得当很多投资者和很多人被我们现在所处的这个局面烧伤时，会有很多自我反省，比如“哦，我们做对了什么？做错了什么？”你总是会有那种华尔街狼的心态，每个人都在玩。但我也记得 89 年，当一切崩溃时，像谷歌这样的伟大公司才真正登场，对吧？所以，我们等着瞧吧。我不知道。

I think when a lot of investors and a lot of people get burned by this thing that we're in right now, there's going to be like a lot of naval gazing going 'oh you know what did we do right? what did we do wrong?' You're always going to have that Wall Street wolf full of Wall Street mentality and everybody's playing. But I remember that in '89, too, when it all fell out and then great companies like Google really came into play, right? So, we'll see what happens. I don't know.

这次对话很棒。最后一个问题。我们谈了很多设备。如果你是一个刚毕业的聪明工程师，想在设备领域成名，你会建议他或她往哪个方向？哪个设备领域值得押注未来二十年？

This has been a great conversation. Last one, last question. We've touched on a lot of devices. I mean, if you were, you know, I don't know, this smart engineer kid out of college like I want to make my name in devices, like where would you point him or her? Like what is the sort of device area that said okay make a bet on this for the next two decades?

非常非常深入地理解应用场景，并理解你是否真的需要硬件。你看，我们有足够多的硬件和各种不同的东西。未来将是一个由非常智能的软件连接起来的设备联邦。所以不要只考虑机械和设备，你必须考虑系统，考虑设备的联邦，才能真正达到下一个层次。其他所有东西都只是功能，不是产品。

Understand the application really really well and understand if you even need hardware. See, we have more than enough hardware and all kinds of different things. It's going to be a federation of things with a really smart software thing. So don't think of just mechanical and devices, you got to think system and you got to think federated of devices to really get the next level of what it's going to happen. All this other stuff is all kind of features, not a product.

我知道我说过这是最后一个问题。但我得说，我爱我的 iPhone，我一直在用。但你不觉得我们看它看得太多了吗？你觉得二十年后我们还会用这样的东西吗？

What I know I said that was the last question. I have to like I love my iPhone. I use it all the time. But don't you get we get sick of looking at it or it's like you're looking at the phone too much. Do you think in two decades like we're still going to be using like a here?

我们只需把这些戴在耳朵里，当手机放进口袋时，它会知道上下文，然后开始和我们对话，更多聆听，更少屏幕时间。是的。会有不同的形态吗？这些是开放式、耳罩式、入耳式。降噪效果很好。好的，很好。还会有眼镜。所以，形态已经足够多了。它们需要用更智能的软件和更好的东西整合起来，才能完成它们需要做的事。

We'll just put these in our ears and then when we put the phone in our pocket, it'll know the context and it'll just start talking to us and then more listening, less screen time. Yeah. You know, are there going to be different forms? These are open ear, over the ear, in ear. The noise cancellation is great. Okay, great. You're going to have the glasses. So, we have enough of the forms. They need to be pulled together with smarter software and better things to get them to do what they need.

我不是说……你用什么耳机？

You know, I'm not saying... what headphones do you use?

这些是 Nothing 的。好的。这是 Nothing。我有耳罩式的，头戴式的。还有开放式。我爱它们。绝对爱。

These are Nothings. Okay. These are Nothing. I got the over the ear, the over the headphone ones. Yeah. And also the opens. I love them. Absolutely love them.

Tony Fadell，太棒了。非常感谢你和我做这个节目。

Tony Fadell, this has been awesome. Thanks so much for doing this with me.

嗯。很好。谢谢。非常感谢，Eric。

Yeah. Great. Thanks. Thanks a lot, Eric.

感谢收听本周的播客节目。如果你是新来的，请点赞和订阅。这对频道很有帮助。每周在您获取播客的地方收听新剧集。

Thank you for tuning in to this week's episode of the podcast. If you're new here, please like and subscribe. It really helps out the channel. Listen in for new episodes every week wherever you get your podcasts.