特拉维斯·卡兰尼克回归 | 构建工业人工智能的未来

那时候他们之间确实有过一段感情。

It was there was a love affair there.

对。

Yeah.

我差不多也是这么说的。

And that's almost the way I would put it.

很多人都会提到，说……

And a lot of people talk about, well,

你知道的，你是因为这件事还是那件事生气？

you know, are you mad about this or that?

你对发生的事感到难过吗？

You upset that that happened?

我就在想，当你再次爱上一个人的时候，

And I'm like, when you when you fall in love again,

你其实不会经常想起前任。

you don't think about the ex very much.

是的。

Yeah.

就是这样。

That's it.

所以你就可以说，那段时间挺好的。

So then you can just be like, it was a good time.

是的。

Yeah.

她最后有点失控了。

She got a little crazy at the end.

没错。

That's true.

没错。

That's true.

就像那个疯狂的前女友。

Like, yeah, the crazy ex.

是的。

Yeah.

就这样了。

So there we go.

是的。

Yeah.

你们后来是怎么重新相爱的？

And how did you fall in love again?

或者你们是怎么度过那段中间时期的？

Or did it take you how did you handle the the interim period between sort of the

嗯，没错。

Well, yeah.

我们当时确实遇到了一个挺特别的情况。

I mean, we had a really interesting situation.

想想那时候的新闻头条和那些负面报道，还有那种席卷优步的极端政治正确风潮，试图限制它——你懂我的意思吧。

Think headlines and the negativity in the headlines from that period of time, and and you could really say the extreme wokeness that was sort of coming in on Uber and trying to constrain it, if that makes sense.

中间有一段空窗期，大概七八个月，我离开Uber之后，到创办现在叫Atoms之前。

There's this interim period, it was like seven, eight months between when I left and when I started what's now called Atoms.

那段时间我简直是在拼命自保，因为一堆诉讼、一堆调查，各种麻烦接踵而至，但我始终坚信，我在Uber做的每一个决定都没错。

And I had to sort of fight for my life because there are just a number of lawsuits, a number of investigations, all these things where it was just like, I continue to stand by every decision I made at Uber.

不过说实话，确实有很多时候，你完全可以辩解说：我简直是在走钢丝，差点就越界了。

But, you know, there are a lot of times where you could just make the argument, I just got way too close to the line.

但我能证明，我的鞋底连一点粉笔灰都没沾上。

And then, I could show that there's no chalk on the shoe.

我能证明，但你得用电子显微镜才能看得出鞋底干不干净。

I could show, but you would need a electromagnetic scanning microscope to see that there was no chalk on the shoe.

换个角度、慢镜头回放，问题就在于：一个从底层拼命爬起来的初创公司小子，突然被推上了一个飞速膨胀的大平台，这种落差太难把握了。

You reverse angle, slow mo, and it's just like that's, the problem is is the edge of being a, what's the word, a hard scrabble, small startup kid brought to something that's working and going super big.

是的。

Yeah.

你没意识到，当你变得足够大，游戏规则就完全不一样了。

You don't realize that when you get big, there actually are different rules.

对。

Yeah.

而且这些不只是明文规定。

You And they're not just the rules.

还有那种氛围，

There's the there's the vibes of

规则本身。

the rules.

当海盗变成了海军。

When pirate becomes the navy.

是的。

Yeah.

我们在农场经常聊这个。

We talk about this at the farm a lot.

当你还是新兴势力、在对抗巨头的时候，可以无所顾忌；但一旦你成了那个庞然大物，再这么干就不行了。

When when you're the upstart and you're battling the man, you can go hard in ways that if you do it when you're the big the big dog

这可不合适。

It's not okay.

对。

Yeah.

大家对你的看法完全不一样了。

It's very that you're viewed entirely differently.

是的。

Yeah.

我之前还没接触到这些信息。

And I hadn't consumed that information.

这其实帮了我大忙，因为我们刚成立公司时，正是那个挑战者。

It actually helped me because when we started the firm, we were the upstart.

我当时嘴里没个把门的，把其他风投骂得一文不值，甚至直接叫他们混蛋。

And I talked just fucking crazy shit about the other VCs and this and that and being like, call them bitches.

但到了某个节点，我能感觉到风向变了。

But like, at some point, like, I could feel it turning.

是的。

Yeah.

我想起了他后来的下场，当时我们还真搞过一套完整的操作。

And I remembered what happened to him and I was like and we we actually had a whole thing.

我们已经不是当年的野蛮人了。

Like, we're not the pirates anymore.

别再说这种话了。

Don't say that type of stuff.

放轻松点。

Like, be chill.

我们确实有过。

We had yeah.

我们当时针对Lyft，专门去挖他们的司机，把他们拉到Uber这边，让Lyft很难凑齐司机资源。

We had this thing where we would against Lyft, we would we would recruit the drivers on Lyft to bring them over to Uber so that it was hard for Lyft to build supply.

我们干得特别积极，公司内部还专门有个项目，叫‘偷窃’。

And we would aggressively do it, and we had a program for it internally, which we called shoplifting.

行吧

K?

对

Yeah.

有段时间

At some point

就只是这个？

Just that?

顺便问一下，那个名字？

That name, by the way?

别管那些了，我知道他们用过一些手段。

Like, forget about all the I I know some of the techniques they used.

特别激进。

Very aggressive.

但那个名字，绝对不能用。

But that name, you can't use that name

某种程度上来说是这样。

if you're in a way.

我记得当时谷歌的David Drummond在我们董事会，他就说，你知道的，

And I remember, you know, David Drummond from Google was on our board, and he's like, you know,

Travis，这根本行不通啊，兄弟。

Travis, that's not a thing, dude.

你得改一下。

You gotta change it.

然后我当然还参加了反垄断培训，当时我还是个初创公司的小角色。

And then I had, of course, like antitrust training, which like at the time, again, I'm like the small startup kid.

我当时心想，反垄断到底是个什么鬼东西？

I'm like antitrust, what the hell is that?

法务部门的人跟我讲了一堆。

Legal people telling me whatever.

他们跟我说，我当时心里想，我巴不得早点遇到这种问题呢。

They're like, look, I'm like, I can't wait to have that problem.

他们说，听好了，关键是这个。

They're like, look, here's the thing.

你们给项目起的名字，得让十岁的小球员一听就懂。

The way you call projects, it has to be appropriate for a 10 year old basketball team.

于是我们就从‘偷东西’改成了‘北美冠军系列赛’，抱歉，是北美冠军系列赛。

So it went from shoplift to the national, sorry, the North American Championship Series.

我们管它叫NACS。

We call it the NACS.

我们还专门搞了个重磅计划，就叫NACS，目标是确保我们拿下北美冠军系列赛的冠军。

And we had a whole very heavy duty initiative called the NACS to like make sure that the North American Championship Series was won by us.

这太搞笑了。

That is hilarious.

我压根不知道还有个北美冠军系列赛。

I I didn't know about the North American championship series.

真有意思。

That's funny.

是的。

Yeah.

好吧。

Okay.

所以你就这么熬过了这一段。

So you go through this Yeah.

持续了八个月。

Eight month period.

你简直身处地狱。

You're in this hell.

还要应付各种诉讼。

You're dealing with the lawsuits.

你回头反思，想着下一步该怎么做，然后就说：咱们再干一次吧。

You're sort of reflecting back, you're thinking about what's next, and you say, hey, let's let's

重新开始。

build again.

后来发生的事特别有意思。

So what happened was it's super interesting.

我有个哥们儿

So I had a buddy

在

who started a

我还在Uber那会儿快结束的时候搞了个项目，叫Cloud Kitchens。

thing right near the end of my Uber tenure, and it was called Cloud Kitchens.

后来发生的事是，他是个房地产天才。

And what happened was is he he's a real estate savant.

这算是他的专长。

That's just a thing.

他当时就在琢磨房地产。

And he was looking like real estate.

他觉得科技和房地产之间肯定能找到结合点。

Like, there's gotta be an angle in tech and real estate.

我们该怎么做呢？

How do we do this?

他找到了一处房产。

And he got a property up.

其实，也许你知道？

Actually, maybe you know what?

让我从头说起吧。

Let me start even before this.

2015年的时候我们推出了Uber Eats。

We got Uber Eats up in 2015.

那时候我们就看到了首批幽灵厨房。

We saw the first dark kitchens.

大概是2015年底到2016年初那会儿。

I don't know, was like late fifteen, early sixteen.

我们看到了墨尔本一些商用厨房的照片，它们根本不是餐厅，只是接了Uber Eats的订单。

We saw pictures of commercial kitchens in Melbourne that were not restaurants, that were on Uber Eats.

所以第一批暗厨房刚开始冒出来。

So the first dark kitchens were emerging.

我们当时就觉得，这太疯狂了。

We're like, this is crazy.

这真的特别有意思。

That's so interesting.

我有个朋友，是个地产高手，他听说了这事，就琢磨开了。

And a buddy of mine, who is this real estate savant, sort of like caught wind of like, what's going on?

他想到一个点子：为什么不搞个多人共用的模式呢？

And he had this idea, which was why don't we have a multi tenant approach to this?

一个地方放三十个厨房。

So 30 kitchens on a single property.

把这些两百平方英尺的厨房租出去，专门做外卖配送。

And basically lease these kitchens, which are 200 square feet, as delivery only locations.

总之，我其实知道一点情况，但也没真弄明白。

Anyways, he I didn't really I mean, I sort of knew what was going on, but not really.

到了秋天，也就是我离开Uber三个月后，我们碰了面，他跟我讲了这件事。

And then in the fall, like three months after I had left Uber, we meet up and he's telling me about it.

他说，嘿，我们现在已经有了几个租户。

He's like, yeah, you know, we've got a few tenants.

事情还挺顺利的。

It's kind of working.

他说打算再搞几个这样的地方。

He's like, I'm gonna do like a few more of these.

他想试试看能不能多做几个。

I think gonna try to figure out how to do a few more.

我问他，你是说几千个吧？

I'm like, you mean like a few thousand, right?

于是我们就合作了。

And so then we partnered up.

我把它收购了。

I acquired it.

我想他们应该刚完成A轮融资。

I think they'd done like their They'd just done their series A.

所以我把它收购了，然后我们就大干了一场。

And so I acquired it, and then we just went to town.

对吧？

Right?

所以这事儿挺有意思的，有时候是你从头开始做一件事，有时候是你中途加入一件事。

And so it's really interesting because sometimes you start a thing, and sometimes you're involved in a thing, and then you come in.

比如，我觉得有个故事挺有意思的，埃隆·马斯克从一开始就参与了特斯拉，这很明显。

Like, think there's an interesting story about, Elon was involved, like, obviously involved in Tesla from the beginning.

他一开始是个投资者，但后来彻底投入了。

Was an investor at first, but then went all the way in.

这些想法是怎么冒出来的，这点挺有意思的。

And it's interesting how these ideas come.

所以我常说，有时候，你脑子里会冒出一些想法。

So I like to say that sometimes, I you you have ideas.

有时候想法会突然冒出来。

Sometimes ideas come to you.

对。

Yeah.

但如果这是命中注定的想法，就像你的灵魂伴侣一样，你一眼就能认出来。

But if it's an idea that's meant to be if it's your it's meant to be your soulmate, you know it when you see it.

没错。

Right.

然后然后你就是，对。

And And you just Yeah.

有时候这感觉就像谈恋爱一样。

And sometimes you just it's like a love affair.

你就直接冲上去。

You just go for it.

有些人会花很长时间去考虑一些重要的事情

Some people have these long important

这中间有真正的区别，和那种所谓的职业首席执行官不一样。

distinction because there's a real difference between that and, like, what's known as the professional CEO.

对吧？

Right?

除了少数有点古怪的政治人物，根本没人会把马斯克当成职业CEO。

Where nobody would ever consider nobody smart other than, like, a few, like, weird political people would ever consider Elon a professional CEO.

确实如此。

It Yeah.

这就像是一个想法。

It's like, oh, that's an idea.

我没创办这家公司，但它属于我。

That's I I didn't start the company, but it's mine.

它就是我的。

That's mine.

它就是我，我知道该怎么去做。

Like, it is me, and I know how to do it.

这些人在起步阶段做得很棒，但后续的发展他们根本搞不定。

And these guys did a great job in starting it, but there's no way they can do it.

这完全是两码事，性质截然不同。

And that's that's a different, a very different animal.

对。

Yeah.

所以，你知道的，当你真正爱上某件事，觉得它此刻就是为你准备的，那你就会义无反顾地去做。

So, you know, you fall in love with something, and it's meant to be for you at that time, and you just go for it.

对我来说，复杂性挺吸引我的。

And for me, it was like, man, complexity is interesting to me.

那些表面上看起来不性感、不吸引人的东西，反而有种奇怪的吸引力。

Things that are naturally not, they're not sexy on the surface is also weirdly interesting.

然后呢，就像你能看出它其实很有魅力，但大家还没意识到这一点。

And then how like, if you can see that it's sexy, but people don't understand yet.

这时候你眼里会有光，心里会想：你们没看到我看到的东西，但我敢肯定我是对的。

That's where the the sparkle in your eye, you kinda go, you don't see what I see, and I'm pretty sure I'm right.

这也是最有趣的地方。

And that's the fun part.

没错。

Yep.

但这事儿特别复杂，因为我们得先买房、搞装修，然后还得有个销售团队去跟餐厅客户谈，而且做的是纯外卖的模式。

But this was a very complex thing because we're buying property, we're doing construction, we are then have a sales team that's selling to restaurants, a delivery only location.

所以他们是在扩张餐厅连锁品牌。

So they're expanding the restaurant chain.

这可不是我随便找个供应商合作那么简单。

This isn't like I'm just getting into a a vendor flow.

我们可是经过深思熟虑的，这是一步战略性的扩张，目的是让你把业务做大。

I'm actually strict we are strict we're making a strategic sell, which is you should expand.

你要扩大业务规模，而且要用一种全新的方式来扩张，这相当于同时迈出两步战略棋。

You should expand your business, and you should expand it in a new way, which is two strategic moves at the same time.

对。

Yeah.

顺便说一句，你喜欢处理复杂的事情。

And By the way, you like complexity.

对。

Yeah.

没错。

Yeah.

是的。

Yeah.

这就让事情

Which makes it

没错。

Exactly.

是的。

Yeah.

所以你就觉得

So then you're like

所以是这样。

So yeah.

你也很适应。

You're also okay.

我们已经把技术融入了整个房地产业务。

So we've woven technology through this whole real estate thing.

我们有一套软件系统，当然能在这家门店运行，也能推广到你所有的实体门店。

We have a software stack that, of course, will sell will work in this facility, but will also sell it to all your brick and mortars everywhere.

再顺便上点机器人吧，因为对餐饮老板来说，人力问题才是最大的痛点。

And shit, let's just get some robotics going at the same time because, like, labor's your biggest problem if you're a restaurateur.

只要解决了这个问题，意义就非常大。

If you solve that, it's really big.

我本来可以细说背后的战略逻辑，但光是这个项目本身的复杂性就挺有意思。

I can get into the strategy of why that is the whole thing, but just the brass tacks of the complexity of the project was interesting

对我来说。

to me.

所以这个项目更像是慢慢摸索着出来的，而不是一开始就有一个宏大的蓝图，知道最后会变成什么样？

So it was more kind of emergent or iterative than, hey, I've got this master vision for what it's all gonna turn into?

不是的。

No.

我觉得那时候，如果我看不到它最终能带来多大的价值和意义，那为了复杂而复杂，我是不会去做的。

I I think at that time, like, I if it if I don't see how it it's big and meaningful, then the complexity for, like, complexity complexity's sake is not a thing.

总得有个盼头，对吧？就像人们常说的，彩虹尽头有宝藏。

There there has to be, you know, maybe somebody would say a pot of gold at end of the rainbow.

这得是有实际意义的事情。

It has to be something meaningful.

Atoms Food或者大家常说的云厨房，其核心价值在于：能不能把一顿优质餐食的制作和配送做到极致高效，让成本逼近去超市买菜的水平？

The the meaningful thing about Atoms Food or what folks know as cloud kitchens is can you make the preparation and delivery of a meal, a quality meal, so efficient that it approaches the cost of going to the grocery store?

如果真能做到这一点，你就相当于用Uber颠覆出租车的方式，彻底颠覆了厨房行业。

If it does, you do to the kitchen what Uber did to the car.

这时候你就会意识到，这确实是个大动作。

And you go, okay, that's kind of a big deal.

但接着你就会问，要达到那个目标，得做些什么？

But then you say, what do you need to do to get there?

对。

Okay.

首先得做线上餐饮配送的电商，这就意味着我需要像亚马逊那样的仓储中心。

Well, you need to do ecommerce for online food delivery, which means I need warehouses like Amazon style.

没错。

Yeah.

只不过这些地方不只是用来物流的。

Except they're not just for logistics.

不只是拣货、打包、分拣，然后装上车就完事了。

It's not just picking, packing, sorting, and then putting in a car.

你还得在那儿搞生产。

You also have to have manufacturing there too.

生产，说白了就是餐厅。

Manufacturing, also known as a restaurant.

人们没想到这一点，但根据劳工统计局的数据，餐厅其实属于制造业。

People don't think of that, but you know the labor statistics, bureau labor statistics, restaurants are manufacturers.

所以你得把制造和物流的基础设施放在同一个地方。

So you have to have manufacturing and logistics infrastructure in the same place.

这其实就是房地产。

That's real estate.

接着你还需要实现自动化生产。

Then you actually have to have automated production.

这属于制造环节，说白了就是做食物的机器人。

That's the manufacturing, so food robots, essentially.

食物机器人技术。

Food robotics.

然后你还得有无人配送机器人。

And then you need to have robotic couriers.

如果你能把这三件事都做到，那就能突然间提供超高品质的餐食了。

And you do those three things, and then all of a sudden, are getting super high quality meals.

你能想到的所有美食。

Everything you could ever imagine.

我们管这叫网络美食广场。

We call it the Internet food court.

而且价格跟去超市买菜差不多。

And it gets to you at the price of going to the grocery store.

所以你就明白了。

So you go, okay.

不错。

Nice.

这故事挺有意思。

That's a nice story.

那咱们换个角度，我先问个简单的问题：二十年后，机器人会普及吗？

Then you go, well, in 20, will robotics I start with a really easy question.

二十年后，人们还会经常自己做饭吗？

In twenty years, will people be cooking food regularly?

到时候会有一整套机器人技术，基本上能做出质量更高的饭菜。

And there will be a suite of robotics that basically will make higher quality meals.

它们做得会比我们任何人都好。

They'll be doing a better job than any of us could do.

我觉得大多数人，尤其是搞机器人领域的，肯定都觉得这显而易见，这事儿迟早会发生。

And I think most people, especially those in the robotics space, we're of course, that's obvious that that's gonna happen.

然后你会想，那二十年后，我们是不是会有那种全自动的墨西哥卷饼？

And you're like, in twenty years, are we gonna have, I call them autonomous burritos?

比如那种能保温、像盒子一样装在轮子上直接送到你家的机器人？

Like boxes on wheels that hold food at temperature that come to your house?

对。

Yes.

大家都会说，绝对会。

Everybody will say yes, absolutely.

你看，Waymo的车现在已经在路上跑着呢。

We see Waymo's cruising around already.

我们只需要一个非常便宜的版本，便宜得多的那种。

We just need a very cheap, a much cheaper version of that.

所以二十年后，这事儿已经算简单了。

So twenty years is like the easy one.

那再想想，十年呢？

Then you go, well, ten?

七年呢？

What about seven?

五年呢？

What about five?

这时候现实就来了。

And so that's when reality.

你得把时间线往现在拉一拉。

You're bending it towards now.

大家都同意这事迟早会发生，现在的问题只是什么时候、谁来实现。

So everybody would agree that this is going to happen, and now it's about when and who.

然后你就喊一声，冲啊。

And then you say LFG.

有时候你确实会稍微早了一点。

Now sometimes you're a little bit early.

我觉得我们在云厨房这块起步稍微早了点，但不管怎样，你把这些基础都打好了，就是我刚才说的那些东西。

And I would say on the we were a bit early on cloud kitchens, but but you build those bricks, all those things I talked about.

我们在30个国家拥有数百个设施，真的是很多很多。

We have hundreds of facilities in 30 countries, many hundreds of facilities in 30 countries.

餐饮业，因为在线外卖市场一直在增长，餐厅的生意越来越好，所以成功的机会也越来越多。

Restaurants, because the online delivery market keeps growing, the restaurants are doing better and better, so success is happening.

对。

Yeah.

是吧？

Right?

餐饮业的淘汰率很高。

Restaurant world has high churn.

你的流失率开始下降，因为订单量在上升。

Your churn starts coming down because volume's going up.

我们有应用能保证这些餐厅的订单量，这样就能吸引成功的租户，从商业模式来看，我这个房地产商也就成功了。

We have applications that guarantee volume into these restaurants, so that we can have successful tenants, which means I'm a successful real estate guy from a business model perspective.

机器人开始发挥作用了。

The robotics start to work.

我们将在第四季度上线机器人的生产线。

We're getting a manufacturing line for our robots up in Q4.

各个部分开始串联起来了。

So the pieces start to come together.

你会发现，生产成本直接降了一半。

You're like, production is now 50% cheaper.

等我们用上机器人配送员，每单餐的配送成本就能从12美元降到50美分到1美元。

Once we get the couriers, the robotic couriers, you go from $12 a drop for each meal to 50¢, a dollar of distribution cost per meal.

现在你省下了六美元的人工成本，再加上配送和场地成本，每单又能砍掉十到十一美元，因为你订单量上去了，价格自然能压下来。

Now you're taking, okay, six out in labor, you're taking, let's call it $10 or $11 out on the courier, on the real estate because you're gonna get higher volume because you're driving the prices down.

你大概能省下一点钱，每顿饭在租金上可能省个两三美元。

You probably save a few bucks, maybe let's call it 2 or $3 on occupancy per meal.

这样一来，一份原本要八九美元甚至十美元的餐食，送到你手里全包价可能只要几块钱。

And all of a sudden, you've got, like, a 8 or $10 meal that's delivered to you all in.

而且你也不用去招司机了，还不用担心有人顺手牵羊。

Plus you don't have to recruit drivers, you know, and get shoplifted.

确实是这样。

That's true.

没错。

So that's right.

你没法偷走机器人的工作。

You can't shoplift a a robotic career.

对。

Yeah.

是的。

Yes.

我们该停止使用这个词了。

We should you to stop using this term.

你完全错了。

So you're so wrong.

好吧。

Okay.

三十大国的一百个设施，八年间数千名员工，你们却一直低调行事。

100 facilities in 30 countries, thousands of employees in a over eight year period, and you're in stealth.

是啊，老兄。

Yeah, dude.

是啊，这简直难到极致了。

Yeah, that's just super hard mode.

没错。

Yeah.

关键是，你看，从2017年往后，每天都有上百篇负面报道。

The thing is, is like, look, coming out of '17, things were it was 150 articles a day of negativity.

所以我想在构建产品时，能把那些负面声音隔绝在外。

And so I wanted to be able to build without bringing that negativity into the system.

我也希望团队能安心做事，不用天天担心《纽约时报》明天又会写些什么。

And I wanted the team to be able to build without worrying about what the New York Times was writing tomorrow.

对。

Yeah.

就是这么简单。

Simple as that.

而且这是正确的决定。

And it was the right thing to do.

接下来我们得开启地狱模式，也就是从零开始招人。

Now we had to do hard mode, is we had to recruit cold.

而且每次都是那些在领英上挂着保密协议、签名里也注明保密的猎头来联系。

And it was always from a recruiter who has a stealth thing in LinkedIn and on their signature.

对。

Yeah.

明白吗？

Okay?

你得用LinkedIn悄悄地去拉客户，而你们公司规模有好几千人。

And you had to go get customers with like stealth in LinkedIn, like and you're thousands of people big.

这可是高难度模式。

This is hard mode.

这事儿挺有意思的。

And it was interesting.

我们也不知道这种状态会持续多久。

We didn't know how long it was gonna go.

你得相信

And you have to trust

没人会泄露出去。

that no one's gonna leak.

你知道吗？如果你能坚持这么长时间的保密，这里面有个特别妙的地方。

You know what happens is, here's the cool part about stealth if you go that long.

我不建议任何人这么做。

And I don't recommend this for anybody.

这属于非常特殊的情况。

It's very special case.

对，其实我也没太深入，但你最终会发现，你的故事线出现了断层。

Yeah, you don't really, I don't, But you eventually, there's this break in your narrative.

就像原本有一个完整的故事脉络，一直在推进，然后突然就断了。

Like there was a story arc and it was going, going, going, and then it's discontinuous.

事情就戛然而止了。

It stops.

然后大家就一阵惊呼。

And there's this whoops.

不好意思啊各位。

Sorry about that, guys.

接下来有一段时间，什么新消息都没有了。

There's this long period of time now where nothing is filling in.

结果媒体就不愿意报这类新闻了，因为要讲一个几年前出现断层的故事，实在太难了。

And what happens then is the media doesn't want to cover a story because they it's so hard to discuss a thing that has a gap a few years before it.

所以这故事没法讲，也没法报道。

So it's too hard to do the story, and it's too hard to cover it.

于是就形成了一个自我实现的循环。

So you get to this place where it becomes self fulfilling.

对。

Yeah.

随着时间推移，保持低调反而更容易，因为就算有人想爆料，也没人知道该怎么处理这些信息。

Where it's easier to be stealth over time, because even if somebody tries to leak, nobody even knows what to do with it.

他们只会说：我根本搞不懂这是怎么回事。

They're just like, I don't even understand.

顺便说一句，我们当年干过不少疯狂的事。

And by the way, we did crazy stuff.

在美国，人们管这叫云厨房，但在拉美，这叫赌场续集。

So people know it as cloud kitchens in The US, but it's Casino Sequelte in Latin America.

我们有不同的叫法。

And we have different names.

其实就是一种神奇的厨房。

That means like magical kitchens, essentially.

韩国那边叫厨房山谷。

We have like Kitchen Valley in Korea.

中国有闪厨，还有另外三个名字。

We have Flash Kitchen and three other names in China.

伦敦叫Food Stars，中东那边我听说叫厨房公园之类的。

We have, I mean, Food Stars in London, like it goes, You know, I think it's a kitchen park in Middle East.

我们本来打算用Yala Kitchens这个名字。

We were gonna go with Yala Kitchens.

类似这样的名字有一大堆。

Like, you know, it's like a bunch of things like this.

所以很难把这些事情串起来。

So it was very hard to connect the dots.

对。

Yeah.

你还是很忙啊。

You're extra still.

是的。

Yeah.

总之。

Anyways.

所以在某个阶段，这家公司浮现出一个更大的愿景，那就是将物理世界数字化。

And so at some point, an even bigger vision for for this company emerges to to digitize the physical world.

你们正在开创一个全新的品类，也就是所谓的工业人工智能。

We're category creating the the the you're creating a category of the term industrial AI.

对。

Yeah.

来聊聊这个更大的愿景是从什么时候开始萌芽的。

Talk about when this bigger vision starts

最终实现。

to culminate.

说实话，我当初在Uber的时候，就开始有了数字化物理世界的愿景。

So look, the vision for digitizing the physical world for me started at Uber.

明白吗？

Okay?

把实体物品当作数据来处理，也是从Uber开始的。

And treating atoms like bits started at Uber.

那时候听起来简直像科幻片。

You were it was sci fi at one point.

我发誓，真不是吹的。

I promise.

你只要点一下屏幕，车就自动开到你面前。

You would touch glass and a car would come to you.

是的。

Yeah.

你还能看到卫星地图，眼睁睁看着那辆车朝你开过来。

And you had satellite view, you're watching this car come to you.

那时候真觉得不可思议。

That was crazy.

现在嘛，就觉得稀松平常了。

Now it's just like, whatever.

明白吧？

Okay?

但那时候真的挺神奇的。

But it was crazy at the time.

每个人第一次用Uber，都像经历了一场魔法。

Everybody's first Uber experience was a magical moment.

把现实世界数字化，这才是核心。

Digitizing the physical world was the thing.

我在Uber快结束的时候，开始琢磨一种我称之为‘基于原子的计算机’的概念——CPU处理比特，存储器保存比特，网络把比特从A传到B。

And this idea of what I would call an Atoms based computer was something I sort of was working on in terms of her framework at the end of Uber, which is CPU manipulates bits, storage stores bits, network moves bits from point a to point b.

好的。

Okay.

CPU处理的是比特，那什么处理的是原子呢？

Well, CPU manipulates bits, what manipulates atoms?

那是制造业。

That's manufacturing.

存储保存的是比特，那什么保存的是原子呢？

Storage stores bits, what stores atoms?

这就是房地产。

That's real estate.

网络负责把比特从A点传到B点，那原子靠什么移动呢？

Network moves bits from point A to point B, what moves atoms?

那就是运输物流了。

Well, that's transport logistics.

这三样现在就是你基于原子的计算机里的三大核心计算资源。

And those now are your three core computing resources in an atoms based computer.

你知道吗？

And you know what?

你在计算机科学课程里学的一切，还有工程方面的知识，到了实体世界全都用得上。

Everything you learn in your computer science curriculum, everything you learn in engineering, it plays in the Atoms world.

同样的框架。

Same frameworks.

所有的数据结构、所有算法，运作方式完全一样。

All the data structures, all the algorithms, it works exactly the same.

比如路由算法。

I can Routing algorithms.

没错。

Totally.

完全没错。

Totally.

实际上，我们用的是TCP SPF。

In fact, we use TCP SPF.

我们用的是TCP锯齿波算法，简单来说，就是根据容量和可靠性来管理向特定厨房发送的需求量，举个例子。

We use TCP sawtooth to basically manage how much demand to send to certain kitchens based on capacity and reliability, as an example.

但就像我平时琢磨的时候，比如我在跟懂技术的人解释，想让他们理解那个一万平方米的设施，我会把它比作一个一万平方米的半导体芯片。

But like we look at when I go and I'm messing around, like when I'm pitching this to a tech person who wants to understand, that 10,000 square foot facility, I call a 10,000 square foot semiconductor.

它相当于一个30核的处理器，也就是30个厨房。

And it's a 30 core processor, 30 kitchens.

它们处理的是原子，而不是比特。

They're computing atoms, not bits.

食物移动的那些通道，其实就是网络总线。

The corridors where the food moves, that's a network bus.

加工中心就像是一级缓存，或者类似的东西。

The processing center's like an L1 cache or something similar.

他把这个类比延伸得挺远的。

He's taking this analogy very far.

还有那个冷储存区，就是放冰淇淋和果汁的地方，对吧？

The cold storage where you have ice cream and juice, okay?

那些是预先计算好的项目，放在边缘节点上，就像你们的Akamai服务器一样。

That's pre computed items that sit on the edge, those are your Akamai servers.

那个来取餐、在路上跑、然后把饭菜送到你手上的外卖员，就是一个TCP数据包。

The courier who comes and picks up the food and goes on the road and brings the meal to you, that's a TCP packet.

这玩意儿没完没了地跑。

This shit goes forever.

这比喻太棒了。

That's excellent.

好的，所以这个愿景最初是在优步提出的。

Okay, so the vision started at Uber.

所以重点是，基于原子的计算确实存在。

So the point is that Atoms based computation is a thing.

说实话，它一直都存在。

And honestly, it's always been a thing.

有人觉得，优步是物理世界的网络，实现了交通的数字化。

And the idea that, oh, okay, Uber was network for the physical world, digitized transportation.

顺便说一句，快讲完了。

And by the way, almost done.

为什么说是快讲完了？

Why almost done?

因为几乎完全是软件实现的。

Because it's almost fully software.

对吧？

Right?

Waymo的车正在路上跑。

Waymo's are cruising.

这几乎完全是软件在驱动。

It's almost fully software.

那物理世界的CPU和存储又该怎么理解呢？

But what about what about CPU for the physical world and storage for the physical world?

说白了，就是数字化的制造业和数字化的房地产。

Basically, digitized manufacturing and digitized real estate.

那里要做的事太多了，简直让人难以置信。

There's just so much to do there, it's crazy.

所以那里有大量的创新，不只是Atoms在做的，未来还会有更多超越我们目前成果的东西。

And so there's a huge amount of innovation there, not just with what Atoms is doing, but like there's gonna be lots of stuff even beyond what we were doing.

如果你在投资，这其实是一种组合策略。

It's a portfolio strategy, really, if you're investing.

这非常有意思，因为你得出了我们从AI中学到的结论——几乎任何东西都能用计算来建模。

It's very interesting because you kind of got to the conclusion that we've now all learned from AI, which is you can model just about anything with computation.

而且

And

你其实早就开始用计算来建模这个物理世界了，哪怕只是隐喻性的，现在这些正逐渐变成现实。

you basically were ahead of that in modeling this physical world, even metaphorically with computation, and now kind of manifesting that.

这大概完全是建立在你的计算模型之上的AI应用。

It's, you know, probably entirely with AI on top of your computational

模型。

model.

挺有意思的是，这个框架其实是我还在Uber快结束的时候开始的，那时候我们正深入进入深度学习和高强度的机器学习。

It's funny because the framework started near the end of my Uber time, really when we were getting deep, we were going into deep learning and heavy duty ML.

那时候我们一直在追逐最顶尖的AI人才。

And I was, you know, we would go after some of the best AI.

我们还建了一个AI实验室，可能是当时全球最顶尖的之一。

We had an AI lab that was probably one of the best in the world as well.

我总是强调，对这类技术来说，物理世界比数字世界有趣多了。

And I was always pitching that the physical world is more interesting for this stuff than the digital one.

说实话，我这也在推销我的书，但你想想，物理世界里的变量和随机分布的维度要多得多。

And you know, look, I'm selling my book, like that's what we do, But you have way many more variables, axes of stochastic distribution.

是的。

Yeah.

这意味着问题复杂得多。

Then means Much

更难应对。

harder problem.

这意味着你没法靠一个算法来解决它。

Which then means you're not going to solve it with an algorithm.

你得通过实际观察，去理解它到底是怎么运作的。

You're going to solve it by empirically understanding how it works.

是的。

Yeah.

没错。

Right.

那我们直接跳到Atoms的更大愿景吧，看看这些公司是如何结合在一起的。

So let's fast forward to how bigger vision for Atoms comes together in terms of the combination of

公司。

the companies.

我老是打断你。

I keep stopping you.

我知道。

I know.

你们打算怎么达成合作？

How do you want to get to the partnership?

因为这也确实如此。

Because it also Yeah.

一切都好。

It's all good.

好吧。

So okay.

你得有个制造和物流中心。

You've gotta have your manufacturing logistics hub.

这属于不动产。

That's the real estate.

你得有自动化生产线，还得有自动驾驶的墨西哥卷饼。

You've gotta have your automated production, and you've gotta have your autonomous burritos.

大概一年前，我开始研究自动驾驶的墨西哥卷饼，就去了中国，考察了所有做自动驾驶技术的公司。

The minute about a year ago, I started looking into autonomous burritos, and I went to China and checked out all of the autonomy guys.

我当时在美国跟一堆搞自动驾驶的人聊，消息就传开了。

I was here in The US talking to a lot of the autonomy guys, word got out.

大家都说，特拉维斯又在关注自动驾驶了。

They're like, Travis is looking at autonomy again.

不少人就开始问，这到底怎么回事？

People started going, what's going on?

是的。

Yeah.

对。

Yeah.

像优步这样的公司，还有其他不同领域的合作伙伴，也都表示感兴趣。

And, you know, folks like, you know, folks like Uber, other partners across different categories were like, we're interested.

不管出于什么原因，你想想，你可能是搞食品供应链的。

Like for one reason or another, because it's like, you could even be do you could be like supply chain guy on food.

你也会觉得，为了业务连续性和安心，得有个备选方案，来解决这些货运卡车和重型运输的移动问题。

You're also saying, I need to have I for business continuity and just like peace of mind, I need to have another alternative for how these freight how these trucks, heavy transport are gonna move.

我需要一个替代方案，也很想找一个合作伙伴，比如这种类型的合作。

And I need an alternative, and I'd love to have a partner, as an example, this kind of stuff.

所以我们找了几家合作伙伴，干脆新成立了一家公司来做这件事，因为想靠一家主营食品的公司来投资自动驾驶，一开始根本行不通。

And so, okay, so partnered with a couple folks, actually created a new company to do it because funding autonomy through a sort of core food company, it just wasn't gonna work at first.

原因在于，这两类业务的定位完全不同。

And the reason why is because it's like, these are just different profiles.

作为一家房地产公司，自动驾驶这种项目需要的是重型软件投入，而且资金量巨大。

As a real estate company, this is like heavy duty software, big money going after it.

这就像是一家电脑公司想顺手搞个网络公司，根本搞不定。

It's like a computer company trying to build a network company off of the side.

不如直接收购一家网络公司。

Gotta buy a networking company.

所以我们干脆成立了这家独立公司，就在去年底左右，但接下来的问题是，我得先把团队搭起来。

So we just got, and these partners wanted a pure play, And so we created that sort of near the end of last year, but what happened is, it's like you gotta get, okay, well now I need to get my team together.

我得把2016、2017年那会儿一起搞自动驾驶的老班底重新召集回来，重新启动起来。

I need to get my guys from way back when, 2016, 2017, we were running hardcore at autonomy, we need to get that going.

因为如果你要涉足专业机器人领域，就必须掌握自己的自主技术。

Because if you're gonna be in the specialized robotics game, you have to have autonomy for yourself.

这是必须的。

You have to.

你的机器人是在物理世界中移动和行动的，你不能只依赖那家目前只有概念验证的公司。

Your robots are act are moving and acting in the physical world, and you cannot depend of on only one company that has an existence proof now.

当然，特斯拉肯定能搞定，你知道的，也许很快，或者需要一段时间。

Of course, Tesla's gonna get there, you know, sometime soon or maybe some amount of time.

我们还不太清楚。

We don't know exactly.

你不能在那种情况下产生依赖。

You gotta you you you can't be dependent in that way.

所以，好吧，把你的团队组建起来。

And so, okay, get your team together.

之前负责ATG的Erik Maihofer，现在已经在负责我的食品机器人部门了。

Erik Maihofer, who ran ATG, is already running my food robotics division.

那是优步的高级技术团队在为我们工作。

That's Uber's advanced technology group is working for us.

然后还有安东尼·莱万多夫斯基，他是我在优步时负责自动驾驶项目的核心骨干之一，后来他掌管了Pronto，那是一家自动驾驶采矿公司，我们可以称之为非道路自动驾驶。

And then Anthony Lewandowski, who is one of my top leads on autonomy back at Uber, was running Pronto, which was an autonomous mining company, let's call it off road autonomy.

但在某些工作流程或领域中，情况远不止是“这个东西需要移动”这么简单。

But in a workflow or category where it's like much more than just like this thing needs to move.

而是“它该如何移动”的问题。

It's like, how does it move?

它是在什么场景下移动的？

What is the context for which it's moving?

这属于非常垂直细分的领域。

It's like a very specialized thing.

我是 Pronto 最大的投资人。

I was the biggest investor in Pronto.

我说，干吧。

I'm like, let's go.

收购了Pronto，现在我涉足矿业了。

Acquired Pronto, and now I'm in the mining business.

但这不是说我之前毫不知情。

But it's not like I didn't know about it.

我从一开始就是最大投资人，这背后有充分的理由。

I was the biggest investor in it to begin with, for a good reason.

这是最出色、最有意思的自动驾驶应用场景之一。

It's one of the most beautiful and interesting autonomy categories or use cases.

在矿业业务上，我们的使命是让矿山更高效，为全球工业提供动力。

And on our mining business, our mission statement is more productive mines to power earth's industries.

我们现在可以直接去找金矿的CEO，问他们：想不想每年多产20%的黄金？

We can go to a gold mine CEO right now and say, would you like to get 20% more gold per year?

他们从不会说不。

We don't get no.

他们会说：你得证明给我看。我们就说：行，咱们走。

He says prove it, we're like, let's go.

实际情况是，Pronto 已经独立在矿业领域深耕了八年。

But that's what's happened is so separately, Pronto is working on mining for eight years on their own.

他们自己干了七八年了。

Seven, eight years on their own.

请继续。

Go ahead.

这也是自动驾驶技术的一个绝佳应用场景，因为对工人来说，采矿恐怕是最辛苦的工作了。

Also a great great target for autonomy because, maybe the worst job for humans there is, is the mining jobs.

所以我们的采矿客户，因为我们正在全球各地的矿山运营。

We have So our mining customers, because we're operating in a bunch of mines around the world.

我们的采矿客户面临着生存级别的安全问题。

Our mining customers have existential safety problems.

或者我们称之为安全问题、安全顾虑。

Or let's call safety issues, safety concerns.

确实。

Yeah.

每天都有可能发生意外。

Like where bad things can happen on any given day.

那些矿工真的在拿自己的生命冒险。

And you've got real people that are putting their lives on the line when they're at these mines.

所以一旦自动化部分采矿作业，安全水平就会大幅提升。

And so when you automate certain parts of mining, you dramatically improve safety.

我是说，这是实实在在的改善。

And you I mean, in a real way.

当你看到他们是怎么干活的，真的让人心里很不是滋味。

That's like when you look at the work that's being done and how it's being done, it's a little bit heart wrenching.

当你亲身接触到采矿现场，看到实际的工作方式时，就会明白。

When you are exposed to it and you see how mines work.

这不是人的错，而是这份工作本身的性质决定的。

It's not the mind's fault, it's just the nature of the work.

安全性的提升和生产效率的提升是同步的，这一点特别重要。

And there's a massive safety upside that goes along with the productivity upside, which is really special.

是的

Yeah.

所以实际情况是，Pronto现在的生产效率已经超过了人工。

So so what happened is is that Pronto just now is coming past human productivity.

这就跟任何一家企业软件公司一样。

So it's like you do it's like any enterprise software company.

是的

Yeah.

你会先和一些大公司做一些很棒的试点项目。

You do some cool pilots with big companies.

是的

Yeah.

是的

Yeah.

你已经有了一些成果。

You got stuff.

你会说，哎，我碰到个特殊情况。

You're like, dude, I got like corner case.

我这家公司有一万人，但我只需要八个席位。

I've got like eight seats at this 10,000 person company.

然后你就会想，行吧，也许可以试试。

And you're like, okay, maybe.

结果突然之间，效果变得更好了。

And then all of a sudden, it's better.

在自动驾驶领域，它的表现要比人类更高效。

In autonomy world, it's about it's better than human productivity.

这意味着它能处理你的大脑和那些复杂的情况。

That means your minds and the quarries.

而且它

And it's

很安全。

like safe.

就是这么回事。

It's like, exactly.

该行动了。

Go time.

这时候就得踩油门了。

That's when you hit the gas.

我们会看到超指数级的增长。

So we're gonna see super exponential.

我们已经在看到了。

We're seeing it.

现在正处在其中。

We're in the middle of it right now.

采矿领域的增长是超指数级的，这特别关键。

Super exponential growth on the mining side, which is super special.

这些人长期保持精简运营，现在他们的客户都在催我们加快步伐，尽快推出更多产品。

These guys ran a very lean startup for a long time, and now I'm going to their customers who are begging us to move faster and get more out there.

我们得生产套件。

We have to manufacture kits.

抱歉。

Sorry.

我们得搞定供应链，还要生产套件。

We have to do supply chain and manufacture kits.

我们得把这些套件装到那些机器上，不管这些机器是在亚马逊雨林深处，还是在沙特和伊拉克的边境，或者是其他那些荒郊野岭的地方，而且必须得快。

We have to install these kits on these machines that are in the middle of the Amazon, or on the border of Saudi and Iraq, or like in these crazy wild places, and do it fast.

然后在每一个矿场都做好变革管理。

And then do the change management at every one of these mines.

所以他们觉得是时候行动了，赶紧上，一直在催我们。

So they're like, it's time to go, let's go, and they're pushing us.

所以我们跟他们沟通的方式是，听着，我们之前是个精简的初创团队，现在要变得强大起来。

So how we talk to them is like, look, we were a lean startup and we're going to muscular.

我跟团队也是这么说的，因为可不想让他们变得臃肿。

And that's how I talk to the team too, because you don't want them to get fat.