帕尔默·拉奇：自主武器为何古老，以及安杜里尔为何不造人形机器人 | Moonshots Live | 第295期

哇。

Wow.

这里亮多了。

It's so much brighter out here.

是啊。

It is.

你知道，未来是光明的。

It's you know, the future is bright.

未来还有镜头光晕呢。

There's lens flare in the future.

所以，欢迎大家来到乐观主义奥斯卡，我们当然要以自主武器系统之王来开启这次峰会。

So, welcome everybody to the Oscars of Optimism that of course we're opening up the summit with the king of autonomous weapon systems.

好了。

Alright.

那么，我想先说一下，这周我们要颁发三个XPRIZE奖。

So, I wanna start with the fact that we're awarding three XPRIZE this week.

帕尔默，恭喜你赢得了我们的野火XPRIZE奖。

Palmer, congratulations for winning our, Wildfire XPRIZE.

非常感谢。

Thank you so much.

这个奖等了很久，确实如此。

This one's been a long time and It has.

我们把幻灯片放上来吧。

Let's get let's get the the slide up here.

帕尔默为Anduril赢得了220万美元。

So Palmer won 2,200,000 for Anduril.

好像你还需要

As if you need

这个。

it.

其中100万来自洛克希德，这还挺有意思的。

A million of 1,000,000 of that from Lockheed, which was kind of interesting.

我们和洛克希德有很多合作。

We do a lot of work with Lockheed.

我们也会和他们竞争，但在国防领域，你知道的，我们和他们的战斗机部门竞争，同时在一些载荷项目上和他们合作。

We also compete with them, but we, you know, the in defense space, it's a you know, we we we compete with their fighter jet division, and we partner with them on some of their payloads.

所以我们一起做了很多事情。

And so we do a lot of stuff together.

但你和我一直都在努力推进那些自主灭火的项目，已经

But me and you have both been trying to get those autonomous firefighting stuff going for a

很久了。

long time.

确实有一阵子了。

For a while.

我是说，这个奖项是2019年宣布的。

I mean, you got this prize announced back in 2019.

已经过去很长时间了。

It's been a long time.

Anduril最早的两个产品之一就是自动驾驶消防坦克，我们从2017年开始造，2019年完工。

Anduril's one of our first two products was an autonomous firefighting tank that we started building in 2017, finished in 2019.

我们竟然让火势蔓延成大灾才去扑灭，这太离谱了。

It's insane that we let fires become conflagrations before we put them out.

确实挺离谱的，人们总说，哦，技术终于让这成为可能了。

And it's kind of insane that, people are saying, oh, the technology's finally made this possible.

我也不太会在公开场合纠正他们。

And I'm not gonna correct them too often in public.

但我的意思是，这其实十多年前就能做到了。

But it's I mean, this has been possible for over a decade.

只是现在，这个问题在政治上已经无法忽视了。

It's just that Let's watch it's become politically politically impossible to ignore the problem now.

我想看一段一分钟的视频，展示你们团队的成果，你能带我们过一遍吗？

I'm gonna watch a one minute video clip of what your team achieved, and if you'd walk us through it here.

当然可以。

Sure.

所以，这就是我们在测试现场做的事情的一个例子，我们检测火灾，对火灾进行分类。

So, was an example of what we were doing at the test site, where we were detecting fires, classifying fires.

这里不太能看到的是，你还需要排除误报的火灾。

What you don't really see in here is that you also need to reject false fires.

这也是测试的一部分。

That was part of the test as well.

所以你得知道，谁是在抽烟的人？

So you have to know, who's a guy smoking a cigarette?

什么是篝火？

What's a campfire?

什么是道路照明弹？

What's a road flare?

然后什么才是真正具有火灾特征的东西？

And then what's something that shows the characteristics of a real fire?

这才是你需要提醒人们注意的。

That's what you need to flag to people.

这就是你需要向紧急服务部门报告的内容。

That's what you need to flag to emergency services.

然后你需要根据火势类型，匹配到该区域内合适的资源。

And then you need to match make the type of fire to the right asset that's in the area.

所以基本上，你要用最经济有效的方式，能够及时响应并到达现场。

And so you want to basically use the most cost effective thing that can respond and get there in time.

如果火势蔓延得更快，你可能需要不同的资源。

And if something is growing faster, you might need something different.

如果是在树林里，而不是在灌木丛山坡上，你可能也需要不同的资源。

If it in woods versus on a brush hill, you might need something different.

所以，你知道，这涉及到检测、分类、排除、匹配，然后去打击目标。

And so, you know, it's about detection, classification, rejection, matchmaking, and then going and hitting the target.

最后我想说的是，我们大部分工作，比如和军方合作，都会经历很多类似的步骤。

And the last thing I'll say is that with most of what we do, our work with the military, you'd go through many of these same steps.

但如果你发现一个目标，比如说，到达某人那里需要一个小时，而另一个只需要十分钟。

But if you find a target Let's say it takes you an hour to get to somebody versus ten minutes.

通常差别不大。

It usually doesn't make as much of a difference.

但说到火灾，十分钟和一小时的响应时间，需要的工具完全不同。

When you're talking about fire, though, the difference between ten minutes and an hour response, totally different tools you're gonna need to respond with.

你知道，炸弹就是炸弹，不管是一天、一小时还是一分钟到达。

You know, a bomb is a bomb whether or not it takes a day or an hour or a minute to get there.

但当火势蔓延时，你得选对工具，在它变得太大之前把它扑灭。

But when you when a fire is is is metastasizing, you need to pick the right thing that can put it out before it gets too large.

太棒了。

Amazing.

你知道，我们一个月后会宣布一些事情，你正在资助一个新的XPRIZE，在那之前是保密的，但我们现在先透露一下。

You know, we're gonna be announcing something in a month's time, that you're funding a new XPRIZE, and, it's gonna be secret until then, but we're gonna tease it now.

又一个一千万美元的奖项。

Another $10,000,000 prize.

你能跟我们分享一下我们要做什么吗？

You share with us what we're gonna do?

当然可以。

Sure.

你知道，我们最好还是保密细节，因为里面有一些有趣的转折。

You know, the you know, we we we we should keep the details secret because there's there's some there's some good twists.

但简单来说，我对用我的钱去推动人工智能发展并不感兴趣。

But suffice to say that I'm not I'm not interested in joining the race to push AI forward with my money.

我觉得人工智能模型方面已经有足够多的人在做了。

I think that the artificial intelligence side of models is pretty well covered by plenty of people.

我觉得我们在理解生物智能方面做得还不够，尤其是超越人类的部分。

I think we've not been doing enough understanding biological intelligence, particularly beyond people.

所以，这个奖项将围绕理解动物如何思考，并让我们能够以比当前最聪明物种的最先进技术好10倍的方式与它们交流。

And so, this is gonna be a prize around understanding how animals think, and making, so that we can communicate with them 10 times better than the current state of the art for the current smartest species.

不仅仅是经过训练的动物，还有野生动物的自然语言。

And not just trained animals, but wild animals in their own in their own language.

太棒了。

Amazing.

而且我觉得我们确实能学到很多东西。

And I think that there's actually a lot that we're gonna be able to learn.

我的意思是，你看那些大型语言模型，它们其实就是人类思维和交流方式的浓缩。

I mean, if you look at large language models, they're kind of this, you know, they're they're they're based on humanity's way of thinking and communicating distilled.

而且我觉得，地球上其他物种身上也有很多值得学习的地方。

And I think that there's plenty to learn from all of the other species that are on this planet.

如果这对AI有应用价值，我也不会感到意外，尤其是那些需要更高效、更节能、更省时的AI，或者换个角度看问题，就像两条腿总比一条腿好，随便你怎么比喻。

And I wouldn't be surprised if it has applications for for AI, particularly AI where you need to be more efficient, more, you know, more power efficient, more time efficient, you know, a different point of view, two peds are better than one, whichever whatever metaphor you wanna use.

太棒了。

Love it.

太棒了。

Love it.

好的。

Alright.

我想开始问第一个正式问题了。

I wanna kick off the first official question here.

P末日。

P doom.

你在AI行业，国防行业。

You're in the AI industry, the defense industry.

你的P末日是什么？

What's your P doom?

我完全不担心AI会毁灭人类。

I'm just not worried, about AI dooming.

我想，我不担心AI会杀人。

The I guess, I'm not worried about AI doing the killing.

我非常非常担心的是坏人

I'm very very worried about evil people

对。

Yes.

利用能力一般的AI去做坏事。

Using moderately competent AI to do evil things.

没错。

Exactly.

而且我对很多刚加入“末日论”这个圈子的人感到有点无奈。

And I am kind of frustrated by a lot of people who are new to the dooming hobby.

这有点像我们之前处理乌克兰战争时的情况——突然之间，那些一辈子都发表过最愚蠢军事观点的人，那些相信历史已经终结、我们已经超越冲突、以为在联合国写封严厉的信就能解决问题的人，都忘了这只有在有可信的武力威胁作为后盾时才管用。

And it's it's it's sort it is sort of like how we had to deal with, you know, the war in Ukraine happens, And all of a sudden, all these people who have had the dumbest opinions you've ever heard about the military for their whole life, and people who believed we lived at the end of history, that we had moved past conflict, that we lived in a world where you could just write a nasty letter in the UN, forgetting that that only works if it's backed up by credible threat of violence.

然后突然之间，他们就改变主意了。

And and then all of a sudden, they change their minds.

现在你在AI领域也看到这种情况。

You're seeing this with AI.

那些对这个问题思考了没几年，甚至可能只有几个月的人，现在都在说：天哪，这是对人类最大的威胁。

People who have not thought about this problem for more than a couple years, or maybe even just a couple months, are are saying, my god, this is the biggest threat to humanity.

这些人几乎没花时间真正分析过对人类最大的威胁是什么。

These are people who have spent almost no time actually analyzing what the largest threats to humanity are.

他们也没跟这个领域的任何专家谈过。

They haven't spoken to any of the experts in this area.

他们肯定没有囤冻干食品，也没有建末日避难所，不像我和我的很多哥们儿，我们既幻想过末日会是什么样子，也认真思考过这个问题。

They certainly weren't buying freeze dried food and building prepper bunkers, you know, like like like like me and many of my my my buddies who who've who've indulged fantasies about what doom could look like, and also thought realistically about it.

所以，是的。

And so, yeah.

对我来说，AI末日论排在很后面。

The AI doom is way down the list for me.

我是个非常乐观的人。

I'm I'm I'm a big optimist.

我觉得它会大幅降低资源开采、加工和转化的成本。

I think it's gonna massively reduce the cost of resource extraction, processing, transformation.

你会看到社会的每个部分都经历自动化和效率的规模经济效应，就像那些没有AI就已经自动化的行业一样。

You're gonna see every part of society become subject to the same economies of scale and efficiency that automated industries that automated without AI have seen.

比如农业或纺织业。

So, think like agriculture or or textiles.

你花不到联邦最低工资一小时的劳动，就能买到两周的卡路里。

You can buy two weeks of calories for less than one hour of federal minimum wage labor.

这太疯狂了。

That's crazy.

让我们为Yeah鼓掌。

Let's give it up for Yeah.

为帕尔默鼓掌。

For Palmer here.

你会发现所有东西都是这样。

You're gonna see that for everything.

戴夫。

Dave.

我脑子里有个特别紧迫的问题。

So I I got a really pressing question on my mind.

我算是搞饲料、芯片和AI基础模型的人。

I'm I'm kind of a feeds and feeds and chips and AI foundation models guy.

所以，当你把系统部署到战场上，把武器部署到战场上，你可以把智能装进去，而这一直是空军的一个大难题。

And so, when you're putting systems in the field, weapons in the field, you can you can put the intelligence inside it, and this has always been a big problem for the Air Force.

比如说，如果我们开始用无人机代替飞行员，那谁来发射武器？这个决定怎么做？

Like, if we're gonna start flying drones instead of pilots, who's gonna unleash the weapons, and how's that decision gonna get made?

而把信息传回总部的缺点是会有延迟。

And the the downside of transmitting it back to headquarters is you have latency.

好处是你有更多的计算能力。

The upside is you have a lot more compute.

你可以在设备、机器人、战场、飞机上本地部署的东西是有限的，但它非常非常快。

What you can put locally in a device, in a robot, in a field, in a plane is limited, but it's really, really fast.

所以你的理念是什么？我觉得很多人会问，我们是要让自主决策的武器在敌方领土上到处游荡，还是要在总部保持控制？对。

So what's your philosophy on, like, you know, I think a lot of people would say, well, are we gonna have self contained decision making weapons wandering around enemy territories, or are we gonna, like, are we gonna have, you know, control at the head Yeah.

那行不通。

Doesn't work.

差不多吧。

More or less.

真正让人担心的其实不是延迟或者计算速度。

The the the real concern isn't so much the latency or or or the speed of computation.

问题是，如果你依赖一个中心位置来做计算，你就需要有一个指挥控制中心，能够和成千上万个活跃资产之间保持高带宽的通信。

The problem is that if you rely on, you know, central a central location to do your compute, you now need to have command and control that has a good high bandwidth communication back and forth between you and potentially thousands or tens of thousands of active assets.

如果敌人干扰了你的通信，或者截获了你的通信，或者黑进了你的通信，那你就完了。

And if your enemy jams your communications, or intercepts your communications, or hacks your communications, you're done.

更糟的是，他们可能根本不去管战场上的那些东西，直接去攻击那个控制中心。

Or worse, they just skip targeting all the stuff in the field, and they go for the place where it's all being controlled.

我们见过这种情况。

Know, we saw we saw this.

我记得乔治·卢卡斯拍过一部纪录片，贸易联邦的中央AI指挥控制中心被炸掉了，那场纳布战役就是这样结束的，所有战斗机器人全倒了。

I there was a there was a there was there was this documentary that George Lucas was working on, and and the trade federation, they had they had their central AI command and control blown up, and that was how the battle of Naboo actually resolved, was all the battle droids lost.

对，它们全倒了。

Oh yeah, they all fell over.

没错，它们全倒了。

That's right, they all fell over.

所以，我觉得这就是现实情况，推动着我们必须让自主能力不留在后方，而是放在前线。

And so, I think that that's, that's the practical reality that's driving, the need to have autonomy living not back home, but forward.

还有其他领域，比如你到了水下，就没法跟别人通话。

There's other areas like, you go underwater, you can't talk to

对。

things.

有时候你能通话，但你不想。

And sometimes you could talk to things, but you don't want to.

发出无线电信号实际上必然会暴露你自己。

Having radio emissions actually exposes you necessarily.

所以，不管好坏，你都需要这个。

And so what you for better or for worse, you're gonna need this.

我其实觉得这没有人们想的那么大的转变。

I actually don't think it's as much of a shift as people think.

有很多人看到这个会说，天哪，潘多拉的盒子被打开了。

I that there's a lot of people who look at this say, oh my god, Pandora's box is being opened.

你会有武器自己做击杀决定。

You're gonna have weapons that are making their own kill decisions.

但他们没意识到，这种情况其实已经存在几十年了。

And they don't recognize the reality, which is that that's been the case for decades.

你看驱逐舰上的宙斯盾系统。

You look at the you look at the the you look at the Aegis system on destroyers.

它有完全自主的模式。

It has a fully autonomous mode.

你按个按钮，它就会击落任何进入你设定范围内的目标，完全不需要人工确认。

You can flip a button and it will shoot down anything that comes within whatever range you programmed it with without any human confirmation.

再看海麻雀和海拉姆，它们用来拦截来袭的导弹、迫击炮、火炮、导弹，还有快艇。

You look at Sea Wiz and Sea Ram for shooting down incoming missiles, incoming mortar, incoming artillery fire, incoming missiles, or incoming speedboats.

也是一样的。

Same thing.

你可以把它们设置成完全自主模式。

You can put them in fully autonomous modes.

而且，回到越南战争时期，我们就有可以从飞机上发射的导弹了。

And, I mean, going back to Vietnam, we had missiles that you could launch from an aircraft.

它们飞越地平线，试图识别出他们认为与地对空导弹发射点相对应的无线电发射器，然后进行打击，而无需任何人决定“我要把它送到这个区域，它们会寻找这个信号，我知道它们有概率会找对”。

They flew over the horizon, tried to identify the right radio emitters that they believed corresponded with a surface to air missile light site, and then struck them without any The person had to decide, I'm gonna send it in this area, they're gonna look for this signal, and I know that there's this probability they're gonna get it right.

这个决定仍然要由人来做出。

That decision is gonna remain with people.

但自主武器非常非常普遍。还有另一个我们今天不打算讨论的话题，但我确实对此有番牢骚，实际上自主武器已经存在了数千年。

But autonomous weapons are widely, widely There's a whole another talk we're not gonna have today, but I will I do have a rant about this, where actually autonomous weapons have existed for thousands of years.

如果将其定义为能够执行你编程设定的意图、独立于操作员和工程师的武器。

If you look at defined as kind of weapons that can execute the intent of what you program to do independent from the operator to the engineer.

比如，你看那些陷阱，人走过去没事，但指挥官骑马经过时，因为重量更重，就会掉进去被抓住。

You look at, for example, pitfall traps that allow people to walk over them, but then capture the commander on his horse when he walks by and it's heavier.

你看某些毒陷阱，它们会杀死具有某些特征的人，而不会杀死具有其他特征的人，也不会杀死动物。

You look at certain poison traps that would that would kill people who had who had certain characteristics, not kill people who had other characteristics or not kill animals.

你看许多类型的地雷、水雷，甚至鱼雷，所以这取决于你的确切定义，但甚至追溯到神话中，人们创造魔像，比如，你知道，替他们去战斗，或者保卫他们的城市。

You look at many types of land mines and sea mines and even even torpedoes, and so it depends on your exact definition, but even going into myth, where people are creating golems, like, you know, go out and fight on their fight on their behalf, or defend, you know, defend their city.

我想我的论点是，我们拥有自主武器已经非常非常久了。

I guess my argument is we've had autonomous weapons for a very, very long time.

只是新技术让它们变得更好了。

It's just a new set of technology that's making them better.

那关于这个有个技术问题。

So technical question on that.

那些远程系统里用的基础模型，比如像陷阱之类的。

The foundation models that go into those remote systems Like bedtraps.

对，现在开源模型基本都是中国的。

Yeah, well, so the the Right now, the open source models are all Chinese.

所以我敢肯定你不会把中国模型装进远程武器里。

So I'm pretty sure you're not gonna be putting Chinese models into remote weapons.

你是打算自己开发Anduril的基础模型，还是怎么把AI装进那些设备里？

Are you gonna build Anduril foundation models, or how are you gonna get AI in those devices?

我们2017年就开始做了，所以早就领先所有人了。

I mean, we've been doing it since 2017, so we're way ahead of everybody.

这个问题我们已经超前了十年。

We're a decade beyond that question.

我觉得我们也不是说……大家现在在构建的那些基础模型，并不是适合做这件事的模型类型。

I think that we also don't The foundation models that people are building are not the types of models you want to be doing this.

它们很大程度上是不确定性的。

They're largely non deterministic.

它们是不可审计的。

They're non auditable.

这完全是另一个类别。

It's a very different category.

我们正在构建的东西，很不幸，几乎没有消费级的应用场景。对。

The thing that we are building, unfortunately, has very few consumer applications Mhmm.

反过来也一样。

And vice versa.

当然，LLM在国防领域有很多合理的用例，但它不是用来引导我的导弹，或者分类雷达信号，你知道，雷达、雷达、雷达数据流。

There, of course, are LLM use cases for defense that make a lot of sense, but it's not guiding my missile or or classifying radar, you know, radar radar radar feeds.

我的意思是，我们也不能构建那种输入文本然后输出文本的模型。

I mean, and we we're also we can't build models that ingest text and then spit out text.

我需要处理原始数据，也就是雷达数据流。

I need to process raw data radar data feeds.

我需要处理潜艇上被动监听站传来的声纳数据。

I need to process sonar data from passive listening stations on submarines.

这确实非常专业化。

That's it's pretty highly specialized.

在这里我要稍微邀一下功。

And I will take some credit here.

那些说AI会改变一切的人，大多都是最近才转变想法的。

All these people who are talking about how AI is gonna change everything, they are mostly recent converts.

你可能会注意到，Anduril Industries的缩写就是AI。

Anduril Industries, you might notice, the acronym is AI.

所以，这可不是巧合。

So and that that was not that was not a mistake.

好的。

Okay.

亚历克斯，你来说说吧。

Alex, jump in.

我觉得，接着我们刚才聊的驾驶台话题，现在这个时代，我们可以把前沿模型直接放到机器人身上，然后就能得到相当惊人的效果。

Think maybe dovetailing on our discussion of driving bench, we're now in an era when it's possible to take frontier models, drop them into a robotic embodiment, and get pretty stunning results.

所以我想问你一个问题，关于规模假说在自主武器系统或者更广泛的自主系统上的应用。

So I'd love to ask you a question, maybe about the scaling hypothesis as it applies to autonomous weapon systems or autonomous systems in general.

如果要外推一下，有没有一个摩尔定律？

If you have to extrapolate out, is there a Moore's Law?

有没有一条经验曲线，就像底层前沿模型的能力提升那样，不管这些模型是来自前沿实验室还是来自AI。

Is there an experience curve for what as obviously the capabilities of underlying frontier models, whether they're from Frontier Labs proper or from AI.

随着这些能力的进步，我们是否能看到自主系统本身也有类似摩尔定律的规律？

As as those progress, do we see any equivalent of a Moore's Law for the autonomous systems themselves?

问题的第二部分可能是，你觉得我们最终会不会达到某种趋同？

And and maybe the the second half of the question would be, do you think ultimately we achieve some sort of convergence?

你们Anduril，制造了各种不同形态和尺寸的载体。

You Anduril, it makes a variety of embodiments in a variety of different form factors.

你觉得最终会不会出现一种通用的杀手级机器人，就像我们现在有通用模型能处理几乎所有任务一样，出现一个通用的机器人形态？

Do you see ultimately some general purpose category killer robot that is equivalently the generalist robotic embodiment just like we now have generalist models that are capable of doing substantially any task?

不会。

No.

在国防领域，我不这么认为。

I don't think so on the defense side.

我觉得在制造业、酒店业和医疗领域，你可能会看到这种情况，因为那些行业能实现真正的规模经济，这对它们来说非常重要。

I think that for manufacturing and for hospitality and medicine, you might see this because you get economies of scale that that that that that are really that are really material for those types of businesses.

而且，你知道，人形机器人可能不是烤汉堡的最佳形态，也可能不是消防员的最佳形态。

And, you know, a humanoid robot may not be the optimal form for a burger griller, and it may not be the optimal form for a firefighter.

它可能不是护士或装配技术员的最佳形态——萨利姆在笑。

It may not be the optimal form for a for a nurse or an assembly tech Salim is smiling.

但即使它不是最优的，它与人共享形态这一点，使它成为一个非常非常好的形态。

But even if it's not optimal, the fact that it is shared with humanity in its form makes it a really, really good form factor.

现在我可以制造一个机器人来做所有这些事情。

And now I can make one robot that does all of these things.

而且，我可能宁愿要一个，比如说，比最优的翻汉堡机器人慢5%或者可靠性差一点的机器人，如果我能和那个护士共享通用性的话。

And, I I would probably rather have a, let's say, robot that's 5% slower or or, you know, less reliable than the optimal burger flipping robot if I can share commonality with that with that nurse.

但在国防领域，情况就不是这样了。

In defense, that's not really the case.

因为你处于一场持续的竞赛中，不仅仅是要勉强击败对手，而是要彻底压倒他们。

Because you are in this constant race, to not just barely beat your adversary, but to massively overpower them.

目标不是要在冲突中勉强获胜。

To try and make it where The goal is not to just barely win a conflict on the margins.

而是要打得不公平，要有压倒性的武力威胁作为你任何外交或谈判的后盾。

It is to fight unfair, to have overwhelming threat of violence underpinning whatever diplomatic overtures or negotiations you have.

而且每个人都在不断突破极限。

And and everyone's always pushing the limits.

所以，如果我有一架每次都能稳定比敌人快5%的喷气机，那值得花大价钱。

So, if I I have a jet that has a reliable 5% every time edge over the enemy, that's worth paying a lot of money for.

而且我认为，由于缺乏专业化而导致的性能损失，可能会让它变得不可行。

Like, I I I And and I think the the loss in performance you would see through lack of specialization probably makes it nonviable.

比如说，举个极端的例子，我造机器人潜艇，那些有专门的形态，需要下潜到6000米深，然后在海底待上好几个月。

Like, to to to go to the to the extreme, like, know, I build robot submarines, and those have a specialized form that needs to dive to 6,000 meter depth and and sit on the bottom of the ocean for months at a time.

然后我还有别的东西在水面上，你知道，那些机器人系统，能爬上山坡，然后在那儿待着。

And then I have other stuff that's on the surface and, and, you know, their robotic systems that, you know, roll up a hill, and and they loiter.

然后我还有别的东西，比如飞行器，它们在做非常专门的任务。

And then I have other things that are, you know, aerial vehicles that are doing very specialized tasks.

我真的想象不出一个能说得通的融合方案。

And I just can't imagine a convergence that would make sense.

除非你告诉我，我们会得到一些很酷的东西，比如人形机器人，它们自带核聚变反应堆，还能潜到6000英尺深。

Unless you're telling me that we're gonna get something cool, like like humanoids that have, like, you know, onboard fusion reactors, and they could dive to 6,000 feet depth.

所以，帕尔默，

So Palmer,

你为什么要造那些东西？

why why are you building those?

我是说，如果要说谁最有可能……

I mean, if anybody,

你应该看看这个。

you should It it it it's look.

你确实有可能造出这样的东西，但很快你就会发现，专业化能让你造出更好的东西。

You could potentially build such a thing, but then very quickly you realize that specialization would allow you to build something even better.

你说，好吧，我造了个能潜到海底又能飞到月球的东西。

You say, okay, I've built something that can dive to the bottom of the ocean and fly to the moon.

天哪，如果它不用同时潜到海底，它飞到月球的速度会快得多。

Man, it could get to the moon a lot faster if it didn't have to also dive to the bottom of ocean.

所以国防领域会是这样，就像我们一样，你看，比如汽车，同一家制造商的所有车型都共享平台，还有制造工艺。

So defense is gonna be this one where kind of like us, you know, you you look at, for example, the difference between cars where everything has shared platforms within a manufacturer with, you know, manufacturing.

它们也共享很多相同的标准和接口。

They also share a lot of the same standards and interfaces.

我觉得你可以把它和赛车运动比较，那里完全不一样，一直在不断向前推进。

I think that you compare that with like motorsport, where it's all completely different, it's continuously pushing forward.

我认为国防领域可能永远都会是这样，我相信他们。

I I think the defense side is probably always gonna be this I believe in them.

如果我不相信人形机器人有这么强大的未来，那我可能就会去造它们了。

If I didn't believe that humanoid robotics had such a powerful future, then I would probably build them.

但我想让Anduril去做那些没有它就不会存在的东西。

But I want Anduril to build things that wouldn't exist otherwise.

这基本上也是我一直想说的。

That's what I've generally wanted write.

对，你想最大化你的边际影响，但你说得对。

Well, you wanna maximize your marginal impact, but they Correct.

其他所有人已经做得很好了。

All the other people are already doing a great job.

我喜欢，我想用他们的机器人，没错。

I like I I I wanna use their robots Sure.

然后用它们去做一些事情。

And and and do things with them.

我觉得人形机器人会成为人工智能和远程系统之间一个很棒的通用接口，也能连接很多为人类设计的旧平台，比如拉杆、按按钮、看屏幕这些。

Like, I think humanoids are gonna be an incredible general artificial intelligence interface between AI and remote systems, and a lot of legacy platforms that were designed for a person to, know, pull the lever, push the button, read the screen.

就是说，我们能够把这些系统全部自动化，而不需要完全替换它们。

Like, we're gonna be able to take all those and automate them without replacing them entirely.

所以有很多像雷达系统这样的东西，造价几百万、几千万，甚至几亿美元。

So you have a lot of things like radar systems that cost millions, tens of millions, even hundreds of millions of dollars.

在激烈冲突中没人愿意去操作它们，因为它们是第一个会被摧毁的目标。

Nobody wants to man them during a hot conflict, because they're the first thing that's gonna go.

但如果我能把类人机器人放进去，我就能在不牺牲人员的情况下实现自动化。

But if I can put humanoid robots in them, I've now been able to automate them without that.

所以，是的，我不会去造类人机器人。

So, yeah, I'm not gonna build humanoids.

这也是为什么Anduril不制造小型四旋翼攻击无人机。

This is also why Anduril is not building, like, small quadcopter attack drones.

太多其他公司已经做得很好了。

Too many other companies are doing a great job.

还有很多……我们也不会做太空发射。

There's a lot of And we're not gonna do space launch.

我们确实做太空产品。

Like, we do space products.

我们有很多，我们自己运营很多卫星和太空传感器。

We have a lot we we operate a lot of our own satellites and a lot of our own sensors in space.

但我不想做发射服务提供商。

But I don't wanna be a launch provider.

那是其他亿万富翁的事。

That's for all the other billionaires to

让他们去处理吧。

Let let them handle that.

太棒了。

That's fantastic.

萨利姆？

Salim?

你知道，当你运营安杜里尔工业时，你是在分散国防。

You know, when you're operating Anduril Industries, you're decentralizing defense.

五角大楼和全球各国的国防部，在人类历史上一直是最集权的机构。

The Pentagon and Ministries of Defense globally have been the most centralized entity in the history of humanity.

你正处在风口浪尖上处理这个问题。

You're sitting on the edge dealing with that.

你肯定面临巨大的内部阻力，机构里的人会说，不，我们以前试过了。

You must have a massive immune system problem where people inside the entity are going, no, we've done that before.

我们能做得更好。

We can do it better.

他们要么想把你吸纳进去，要么想把你排挤出去。

They either try and suck you in or they try and spit you out.

你是怎么应对这种局面的？因为你做事的方式，我从来没见过有人能成功做到。

How are you navigating that interface because you're doing it in a way that I've never seen anybody succeed do it?

我觉得这个问题一开始非常非常难。

I think that this is a problem that's very very hard at the beginning.

我说错话了。

I said the wrong thing.

哇。

Wow.

不知道。

No idea.

好吧，除非有人通过广播告诉我们

Well, until someone gets on the intercom and tells us

人工智能。

the AI.

恭喜。

Congratulations.

我们刚才在聊什么来着？

What were we talking about?

是什么来着？

What what was it?

真的吗？

Seriously?

我就知道广播会响。

I knew the intercom was gonna come on.

我们刚才在聊

What were

什么来着？

we talking about again?

五角大楼的集中化对比分散化，接下来要

Centralization of Pentagon versus decentralized gonna

继续说。

keep talking.

继续说下去。

Just keep speaking.

不行，我们已经聊过一点了。

Can't so we've gotten through this a bit.

我觉得一开始，有人担心Anduril过不了那一关。

I think early on, there was that concern that Anduril was not gonna be able to get through that.

我们最困难的问题就是筹钱。

Our our hardest problem raising money

Skippy，你能把这个关掉吗？

Skippy, would you please shut this down?

我经历过太多次这种事了，我知道接下来会怎么发展。

I've been through enough of these to to know know how it's gonna work.

好了，就这样。

There we go.

我想也是。

I imagine.

其中一个

One of

我们早期的问题是，我们的投资者说，帕尔默，潜在的投资者。

our early problems was that our investors said, Palmer, potential investors.

他们说，听着，我们相信你能造出能用的东西。

They said, Look, we believe that you'll be able to build things that work.

我们相信你能造出很棒的产品，但我们不认为政府有能力大规模采购它们。

We believe you'll be able to build products that are great, but we don't believe the government is capable of actually purchasing them at scale.

因为，我是说，你谈到集中化。

Because, I mean, you talk you talk about centralization.

你知道那些不了解情况的人会遇到的问题。

You know about the issues for people who don't.

就在几年前，大约80%的MDAP，也就是重大国防采购项目。

Until just a few years ago, you had about 80% of MDAPs, major defense acquisition programs.

所以基本上，这很重要。

So basically, that that that matters.

80%的MDAP都流向了五家公司，就这样。

80% of MDAPs went to five companies, and that's it.

哇。

Wow.

而且我觉得你已经看到了令人难以置信的整合和集中化。

And I think you've seen incredible consolidation, centralization.

Anduril之所以成功，是因为我们更像是一家产品公司，而不是承包商。

Anduril has been successful because we use we're we're less of a contractor and more of a product company.

我们用自有资金来开发产品，然后再卖出去。

We use our own money to build products, then we sell them.

现在我们已经到了这样的地步：政府里的人押注Anduril，结果都赚得盆满钵满。

And we've gotten to the point now where people in government have done very, very well, betting on Anduril.

然后，神奇的是，我们居然按时完成了。

And then, miraculously, we finish Okay.

按计划。

On schedule.

我们按时完成了。

We finish on time.

然后下一个项目经理就说，我知道怎么让我的工作轻松了。

And then the next program manager says, I know how I can make my job easy.

我只要把合同交给Anduril就行了。

I just I just give the contract to Anduril.

明白了。

Got it.

所以，进入国防工业这行，肯定得有点胆量。

So, it's gotta take guts to enter the defense industry.

这可是地球上最封闭、最自给自足的圈子，就像自己舔自己的冰淇淋蛋筒一样。

The most incestuous, you know, self licking ice cream cone on the planet.

到底是什么疯狂的想法，让你有勇气去做这件事？

And and what insanity, you know, gave you the balls to do that?

正是那种疯狂，让我决定进入这个行业。

The insanity was the reason that I got into it.

我们融资演示文稿的第一页写着：Anduril 将通过每年赚取数百亿美元，为纳税人节省数千亿美元。

The first page of our pitch deck said, Anduril will save taxpayers hundreds of billions of dollars a year by making tens of billions of dollars a year.

我们的想法是，我们不只是制造产品来卖，而是取代那些成本本应低十倍却白白浪费的系统。

The idea being, we're not just making products that we're selling, but we are displacing the systems that cost needlessly 10 times what they should.

对。

And yeah.

谢谢。

Thank you.

所以这看起来确实很疯狂，但你知道，我没办法把自己和他放在同一个高度。

And so it does seem crazy, but, you know, I don't I I I can't put myself on the same pedestal as him.

但我成长过程中，以及职业生涯早期的一位导师和偶像，是约翰·卡马克，InSoftware的创始人。

But one of my mentors and idols growing up, and then also early in my career, was John Carmack, the founder of InSoftware.

约翰，约翰很厉害。

John John is great.

约翰非常了不起。

John is incredible.

对于不了解他的人，他基本上就是现代3D游戏引擎和第一人称射击游戏的发明者，还创办了一家火箭公司，参加了

And for people who don't know, he's basically inventor of the modern modern three d game engine, first person shooters, also started a rocket company that competed in

他当时在参加XPRIZE比赛。

the He was competing in the XPRIZE.

我记得

And I I remember

而且他是最早做可重复使用的垂直起降火箭的人之一，甚至可以说是唯一一个。

And he was one of the, if not the only one, to be doing reusable vertical takeoff, vertical landing was even trying to do it.

我

I

记得他飞行器侧面有个名字，NVIDIA。

remember there's this this name on the side of his vehicles, NVIDIA.

那个怎么念来着？

How do you pronounce that thing?

要是知道就好了。

If I would only know

没错。

That's right.

对，他当时用的就是那个，对。

He was using and so, yeah, he was using he was using yeah.

他在自己编写传感器融合算法。

He was building his own sensor fusion algorithms.

他用游戏编程的经验，在那种，你知道的，商用显卡硬件上处理这些数据。

He was processing them on modern, you know, or on a, you know, commercial style graphics card hardware using his game programming experience.

总之，重点是

Anyway, the point is

那叫犰狳。

It was called Armadillo.

犰狳航天。

Armadillo Aerospace.

没错。

That's right.

而且，他们做垂直起降的时候，SpaceX 还，我是说，还没开始搞任何东西呢。

And the thing and, yeah, they were doing vertical takeoff and landing long before SpaceX was was was was, I mean, doing anything.

不。

No.

在它存在之前，

Before it existed,

当然。

for sure.

那是

That's

不过话说回来，重点是，他最后成了Oculus VR的首席技术官。

And so but anyway, the point is, he he I I he ended up becoming the CTO of Oculus VR.

我成功把他挖过来了。

I managed to poach him away from it.

但他离开后，其实比我还早就开始搞AI了。

But after he left, he went to work on AI, actually, even before I did.

所以他创办了自己的公司，专门研究通用人工智能。

And so he started he started his own company to work on on AGI.

我就问他，你为什么要这么做？

And and I said, why why are you why are you doing this?

你知道，说实话，你成功的几率不大，我当时还是相信AI会有发展的，但他说，我要创办一家公司，全力押注最终目标。

You know, the the odds of you succeeding to be clear, I still believed AI was going somewhere, but he said, I'm gonna start a company that's just going all in on the end state.