接下来是与埃隆·马斯克的一场对话，这是他第3次做客这档“Lex Fridman 播客”。好的，请坐舒服些。

嘘。

哦，哇，好吧。

你不用耳机那一套吗？

不用。

好吧。我的意思是，我需要离这东西多近？

你离得越近，声音就越性感。

嘿，宝贝，咋样。

对。

在那个宝贝上怎么也听不够你的声音吗？（两人笑）——我要把那段剪出来，以后每次有人在手机上给我发消息，我就用那段回复。

如果你想要我的身体，而且觉得我很性感，那就直截了当地告诉我。嘟嘟嘟嘟嘟。

[希冯] 太好笑了。

太棒了。好，严肃模式启动，好吧。

严肃模式。拜托，你的俄罗斯人，你可以严肃起来。

是啊，我知道。

在俄罗斯，每个人无时无刻都很严肃。

对，对。我们会进入状态的。我们会进入状态的。（希冯轻声说话）只是变软弱了。请允许我说，SpaceX于2020年5月30日将人类发射进入轨道，被许多人视为人类太空探索新时代的第一步。在过去2年里，当我们的世界正经历人类近代史上最艰难的时期之一时，这些载人航天任务对我和数百万人来说都是希望的灯塔。

我们看到分裂、恐惧和犬儒主义抬头，也看到人类共同人性的丧失，而此时恰恰最需要这种共同人性。所以，首先，埃隆，请允许我感谢你给世界带来希望，也让人们有理由对未来感到兴奋。

哦，你这么说真是太好了。我确实想做到这一点。人类显然有很多问题，人有时也会做坏事，但尽管如此，我热爱人类，而且我认为我们应该确保尽己所能，创造一个美好而令人兴奋的未来，一个能最大限度提升人们幸福感的未来。

让我问问载人龙飞船Demo-2。所以，那是第一次搭载人类的飞行，在发射前的那段时间里，你感觉如何？你害怕吗？兴奋吗？你脑子里都在想些什么？事关重大。

是的，不，那压力大得不得了。问题在于，我们显然绝不能以任何方式让他们失望。所以我会说，压力极大，这还是往轻了说。

我有信心，在我们发射的时候，已经没有任何人能想到任何可以提高成功概率的办法了。我们绞尽脑汁，试图想出任何可能提高成功概率的方法，但再也想不出别的办法，NASA也想不出来，所以，那就是我们能做到的最好程度。于是我们就继续进行了发射。

我不是一个有宗教信仰的人，但我还是跪下来为那次任务祈祷了。

[莱克斯] 你睡得着吗？

不。

成功时是什么感觉？先是发射成功的时候，然后是他们返回家园，或者说返回地球的时候。

如释重负。是的。我发现，在压力巨大的情况下，感受到的与其说是狂喜，不如说是如释重负。而且我想，后来随着我们更加适应并验证了这些系统，因为我们真的，你必须确保一切都能正常运作。之后的小行星任务肯定愉快多了。而且我觉得“灵感”任务确实非常鼓舞人心，也就是Inspiration4任务。

我会建议大家观看Netflix上那部关于“灵感”任务的纪录片，真的很不错。而且它真的不是，我实际上受到了它的鼓舞，所以那一次，我觉得自己算是能够享受任务本身，而不只是全程压力巨大。

所以，对于那些不知怎么就是不知道的人来说，这是一次全平民任务，是第一次全部由平民进入太空、进入轨道。

是的，那是，我想，是大概，我不知道，30或40年来最高的讣告之类的，唯一比它更高的是那次航天飞机，抱歉，是一次哈勃维修任务。而再往前就是1972年的阿波罗任务了。那相当疯狂。所以很酷。很好。我认为，作为一个物种，我们希望继续做得更好，达到更高的境界。

我认为，如果阿波罗任务就是人类的最高水位线，如果那就是我们所能到达的最远之处，那将是悲剧，极其惨痛的悲剧。而令人担忧的是，现在距离上一次登月任务已经过去49年了。所以，差不多半个世纪过去了，我们却再也没有回去过。这很令人担忧，就像是，这是否意味着我们这个文明已经达到巅峰了，还是什么？我认为我们必须重返月球，并在那里建立一个基地。一个科学基地。

我认为，如果我们在月球上拥有一个正规的科学基地，就能了解许多有关宇宙本质的知识。我们在南极洲以及世界其他许多地方都有科学基地。所以我认为，下一件大事是我们必须拥有一个真正的黑色月球基地，然后把人类送往火星，走出去，成为一个承载太空的文明。

我会问你其中一些细节。但是，既然你忙于应对其中涉及的一切艰巨工程挑战，你是否仍然能够惊叹于这一切的神奇，惊叹于太空旅行，惊叹于火箭的每一次升空，尤其是在那是一项粗糙任务的时候？还是说，你完全被所有必须解决的挑战压得喘不过气？

实际上，再补充一点，我之所以想问这个关于5月30日的问题，是因为已经过去一段时间了，所以你可以回过头来，思考它已经产生的影响。当时它或许是一个工程问题，现在它正在成为一个历史性时刻。就像，这是一个时刻，关于21世纪，又会有多少个时刻被人们铭记？

对我来说，那次任务或类似的事情，也许是Inspiration4或其中某一次，会被人们铭记为太空探索新时代的早期步伐。

是的，我的意思是，在发射本身进行期间，所以，我的意思是，我想也许有些人知道，但很多人不知道，其实我是SpaceX的总工程师，所以几乎所有设计决策都是由我签字批准的。因此，如果那艘飞行器出了什么问题，从根本上说就是我的错，你知道吗？

所以我其实只是在思考所有那些事情，就像，所以当我看到火箭时，我看到的是所有可能出错的事情，以及可以做得更好的事情，龙飞船也是一样。其他人会说：“哦，这是一艘航天器或一枚火箭。”还有“这看起来真的很酷。”而我就像是，我脑中仿佛有一个读数：这些是风险，这些是问题。这就是我看到的东西。

就像（埃隆喷气般呼气）所以这并不是其他人在看到产品时所看到的东西。

那么，就让我请你以同样的方式分析一下星舰。我知道你有……你会在不久的将来更详细地谈谈星舰。也许你已经——如果你愿意，我们现在就可以谈。

但是，就以同样的方式，就像你说的，你看，当你看到一枚火箭时，你看到的像是一份风险清单。同样地，你说星舰是一个非常困难的问题。所以，我可以用很多方式来问这个问题，但如果你能神奇地完美解决一个问题，完美解决一个工程问题，你会选择哪一个？

[埃隆] 星舰上的问题？

关于，抱歉，关于星舰。那么，它可能与效率、发动机、不同部件的重量、各种事情的复杂性有关吗，也许是它为了着陆而必须完成的那种疯狂动作的控制？

不，实际上，到目前为止，占用我最多时间去解决的事情是发动机生产。不是发动机的设计，我经常说，原型很容易。生产很难。所以，我们拥有迄今设计过的最先进的火箭发动机。因为我会说，目前史上最好的火箭发动机可能是RD-180，或者RD-170这种双联俄罗斯发动机，基本上是这样。而且我仍然认为，只有当一台发动机把某个东西送入轨道后，它才应该算数。

所以我们的发动机还没有把任何东西送入轨道，但它是，它是第一台真正优于俄罗斯RD系列发动机的发动机，那些发动机的设计非常出色。

所以你说的是猛禽发动机。是什么让它如此出色？它有哪些不同的方面让它，让你在整个系统都奏效的情况下最感到兴奋，比如效率以及所有诸如此类的方面？

嗯，它，猛禽是一台全流量分级燃烧发动机，而且它在非常高的TAVR压力下运行。所以，一个关键的性能指标，也许是最关键的性能指标，就是火箭发动机能够在多高的燃烧室压力下运行？那就是燃烧室压力。所以猛禽的设计运行压力是300巴，可能，也许更高，相对于标准大气压而言。

目前投入运行的发动机纪录是我提到的 RD 发动机，也就是俄罗斯的 RD，我记得大约是267巴。而燃烧室压力的难度是以非线性方式增加的。所以，TAVR 压力提高10%，难度更像是增加50%，但正是那种空气压力让发动机能够获得非常高的功率密度。因此，它可以实现非常高的推重比和非常高的比冲。

所以，比冲就像是衡量火箭发动机效率的指标。它其实就是排气，也就是从发动机喷出的气体排气速度所产生的效果。凭借非常高的燃烧室压力，你可以拥有一台紧凑的发动机，但它仍然具有很高的膨胀比，也就是出口喷管与喉部之间的比率。你看，火箭发动机有点像沙漏的形状。

它就像一个燃烧室，然后收窄，接着是一个喷管，以及出口直径与喉部膨胀比的比率。

那么，为什么这种发动机如此难以规模化制造？

它非常复杂。

复杂是什么意思？这里涉及很多部件。

有很多部件，还有很多独特的材料。所以为了让这台发动机运转，我们不得不发明几种原本不存在的合金。

所以这也是一个材料问题。

这是一个材料问题，而且在分级燃烧，也就是那种全地板分级燃烧中，系统里存在许多反馈回路。基本上，推进剂和热气体会同时流向发动机上许多不同的位置。而它们都会彼此产生递归影响。所以你改变这里的一件事，它会在这里产生递归影响。它又会改变那边的某个东西。而这相当难以控制。

以前没人造出过这种东西是有原因的。我们之所以采用分级骚动全流量，是因为它拥有理论上可能达到的最高效率。所以，为了制造一枚完全合理的火箭，而那确实是轨道火箭技术的圣杯，你必须做到，一切都得是最好的。必须是最好的发动机、最好的箭体、最好的隔热罩、极其轻巧的航空电子设备，以及非常巧妙的控制机制。

你必须以任何可能的方式减轻质量。例如，我们不会在助推器和飞船上安装着陆腿，而是会用一座塔架接住它们，以省下着陆腿的重量。所以这就像，我是说，我们谈的是用一座配有筷子臂的巨型塔架，接住有史以来制造的最大飞行物体。就像《龙威小子》里夹苍蝇那样，只是要大得多。

（埃隆笑）——我是说，拉住某个东西——这第一次可能不会成功。（埃隆笑）所以这太疯狂了。这是疯狂的东西。

所以你提到你会怀疑，嗯，不是你会怀疑，而是有些日子或时刻，你会怀疑这件事到底是否可能。它太难了。

关于可能性这部分，嗯，到这个时候，我们会，我认为我们会让星舰成功运转。问题在于时间。我们要花多久才能做到？我们要花多久才能真正实现完全且快速的重复使用？因为在我们能够实现完全且快速的重复使用之前，可能需要进行很多次发射。但我可以说，从物理上算得通，我们没有，到这个时候，我会说我们有信心，姑且这么说，我非常确信成功属于所有可能结果的集合。

[Lex] 嗯，对，它不在所有的集合中。

有一段时间，我并不确信成功属于可能结果的集合。（Lex 笑）这其实非常重要。不过，所以……

[Lex] 所以你是说，还有机会。

我是说，还有机会。没错。只是不确定要花多久。但我们有一支非常有才华的团队，他们夜以继日地工作，要让它成为现实。就像我说的，要实现航天飞行领域的革命，并让人类成为一个承载太空的文明，关键是拥有一枚完全且能够快速重复使用的火箭，也就是轨道火箭。甚至从来没有任何一枚轨道火箭实现过完全重复使用。

而这一直是火箭技术的圣杯，许多聪明人，非常聪明的人，以前都曾尝试过，但他们没有成功。因为这是一个如此困难的问题。

在这种情况下，当工程问题如此困难，还有很多专家，其中许多人是你钦佩的，而他们过去都失败了，你的信念来源是什么？

[埃隆] 是的。

很多人、很多专家，或许还有记者，各种人、普遍公众，都非常怀疑这是否可能，而你自己也知道，即使成功是一个非节点集合，而不是空集，它仍然不太可能实现，或者非常困难。无论是你个人、作为工程师在理智层面，还是作为一个团队，你们从哪里获得所需的力量来源，从而坚持挺过这一切、继续推进这个项目并将它完成？

我想，力量。嗯。我其实并不是这样思考事情的。我的意思是，对我来说，事情很简单，这是一件必须完成的重要事情，我们就应该一直做下去，或者拼死一试，而我不需要力量来源。

所以，放弃甚至都不像是……

这不，这不符合我的天性。

好。

而且我不在乎乐观还是悲观。去他的，我们会把它做成。

[Lex] 会把它做成。那么你能不能再具体聚焦到星舰的具体问题，或者你所处理的任何工程问题？你能不能试着内省一下你特有的生物神经网络、你的思考过程，并描述你是如何思考问题、不同的工程和设计问题的？是否有一种你谈到过的系统性流程，比如第一性原理思维，但是否有某种——是的，当然。

流程？

就是说，物理学是低的，其他一切都是建议。我见过很多能违法的人，但我从没见过任何能违反物理定律的人。所以，对于任何一种技术问题，首先都得设法确保你没有违反物理定律。我认为，第一性原理分析确实可以应用于人生的任何领域，实际上任何事情都可以。

它其实只是在说，让我们把某件事归结到最基本的原理，也就是我们最确信在基础层面为真的那些事物，由此确立你的公理基础，然后再从那里向上推理。接着，你再用公理真理交叉核验你的结论。物理学中的一些基本问题会是，比如是否违反了能量守恒或动量守恒之类的定律，那么它就行不通。

所以，这只是为了确定它是否可行。然后，另一个很好的物理学工具是思考事物在极限情况下的表现。如果你选取某个特定事物，并把它放大到一个非常大的数值或缩小到一个非常小的数值，情况会如何变化？

既包括你制造的东西的数量之类的，也包括时间。

是的，比如说，以制造为例，我认为这是一个被严重低估的问题。就像我说的，把一个先进技术零部件投入大批量制造，远比一开始把它设计出来困难得多。更多是量级。比如说，你正试图弄清楚，这个零部件或产品为什么昂贵？是因为我们正在做某件根本上愚蠢的事吗？

还是因为我们的产量太低？于是你会说，好吧，如果我们的产量是每年100万件呢？它仍然昂贵吗？这就是我激进之处，把事情推到极限来思考。如果在每年100万件的产量下仍然太贵，那么产量就不是你的东西昂贵的原因。设计本身存在某种根本性问题。

然后你就可以专注于降低设计的复杂性之类的事情。

必须改变设计，把这个零部件改成某种本质上不昂贵的东西。这在火箭制造中很常见，因为单位产量相对较低，所以一种常见的借口会是：“嗯，它之所以昂贵，是因为我们的单位产量低。如果我们身处汽车之类的行业，或者消费电子行业，那么我们的成本就会降低。”我会说：“好吧，那我们假设”跳过，“现在你每年制造100万件。它仍然昂贵吗？”如果答案是肯定的，那么规模经济就不是问题所在。

你会不会把供应链之类的也纳入制造环节？你谈到了资源、材料之类的东西，你会把这些纳入从第一性原理出发进行推理的计算中吗？比如，我们要怎么让这里的供应链运转起来？

对，对。

[Lex] 然后还有材料成本之类的，还是说这就考虑得太多了？

对。没错。思考极限情况就是一个很好的例子：如果你拿任何产品、任何机器之类的，比如拿一枚火箭什么的，然后问，假如你有——假如你看看火箭里的原材料，那么其中会有铝、钢、钛、Inconel、特种合金、铜。然后你问：“这些材料各自的组成元素有多重，它们的原材料价值是多少？”

而这就设定了车辆成本所能降到多低的渐近极限，除非你改变材料。然后当你这么做时，我把它称为，也许是那个神奇的1号数字之类的。

所以那就像是，如果你这里有的只是堆成一堆的这些原材料，而你可以挥动魔杖，把原子重新排列成最终形状，那么那就是你制造这东西所能达到的最低成本，除非你改变材料。所以，然后，那几乎总是一个非常低的数字。所以，那么真正导致东西昂贵的是你如何把原子组合成想要的形状。

是的，其实，如果你不介意我稍微跑一下题，我曾经，我经常和吉姆·凯勒交谈，他是一个曾与你共事的人，当时是——哦，对。吉姆在 Tesla 的工作做得很出色。

所以，我想他传承着你现在所说的同一种思维方式。我想，我在 Tesla 和 SpaceX 的员工身上看到了同样的东西，他们算是学会了这种思维方式，而它几乎也就变得显而易见了。但不管怎样，我有过一场争论，不是争论。他教我了解了制造 Tesla Bot 可以有多便宜。我们只是，我们有过一场争论。你怎样才能降低大规模生产机器人的成本？

因为到目前为止，我显然有机会在学术圈与人形机器人进行大量互动，然后也接触了波士顿动力之类的。而它们的制造成本非常昂贵。然后吉姆算是给我上了一课，他说：“好吧，这就是一种第一性原理思维：怎样才能把制造成本降下来。”

我想你会这么做，你已经针对 Tesla Bot 以及各种各样、各种传统上被视为复杂的复杂系统做过这种思考，然后你说：“好吧，我们怎样才能把一切都简化？”

是的。我的意思是，我认为，如果你真的很擅长制造，那么在大批量生产时，你基本上可以制造任何东西，使其成本渐近趋近于各组成部分的原材料价值，再加上你需要获得许可的任何知识产权。任何东西。

对。

但这很难。这并不是说，那是一件很难做到的事，但对任何东西来说都是可能的。任何批量生产的东西都能，像我说的，以渐近趋近于其原材料组成加知识产权许可权的成本制造出来。

所以，在试图设计一款产品时，经常会发生这样的情况：人们会从自己熟悉的工具、零件和方法入手，并尝试使用现有的工具和方法来创造产品。另一种思考方式其实是设想那个，试着设想完美产品或技术的柏拉图式理想形态，不管它可能是什么，然后问：“这是什么？

什么样的原子完美排列会成为可能的最佳产品？现在让我们试着弄清楚，怎样才能让原子形成那种形状。”

我是说，这听起来，在你开始真正思考它之前，它荒诞得几乎像《瑞克和莫蒂》。而且你确实应该用这种方式思考，因为其他一切都有点，如果你不用这种方式思考，就可能会受过去做事方式的惯性所支配。

嗯，仅仅由于惯性，人们就会想使用自己熟悉的相同工具和方法。这就是他们默认会做的事。

[Lex] 是的。

然后，这会导致一种结果，也就是做出那些能够用这些工具和方法制造的东西，但那不太可能是完美产品的柏拉图式理想形态。所以这就是为什么从两个方向思考是有益的，比如我们能用现有的工具造出什么，但也要思考完美的、理论上完美的产品是什么样子？

而且，那个理论上的完美产品将会是一个不断移动的目标，因为随着你了解得更多，对那个完美产品的定义也会改变，因为你其实并不知道完美产品是什么，但你可以成功地逼近一个更加完美的产品。所以，要这样思考，然后说：“好吧，现在为了让原子形成那种形状，我们需要创造什么工具、方法、材料之类的东西？

但人们很少用这种方式思考。不过，这是一个强大的工具。

我应该提一下，才华横溢的希冯·齐利斯正在和我们待在一起，以防你们听到一个从外面、从上方传来的智慧之声。好。那么让我问问你火星的事。你提到，在月球上建立基地、开展一些研究对科学来说会很棒，但真正的巨大飞跃，同样属于这种看似不可能的事情，是把人类送上火星。你认为SpaceX什么时候会让人类登陆火星？

嗯。最好情况是大约5年，最坏情况是10年。

从工程角度来说，你认为决定因素是什么？还是说，瓶颈不在这里？

不，从根本上说，你是在设计这种运载工具。我的意思是，星舰是有史以来制造过的最复杂、最先进的火箭，高出了，我不知道，一个数量级之类的。差距很大。它确实达到了全新的层次。而星舰的根本优化目标，是尽量降低每吨载荷进入轨道的成本，并最终降低每吨载荷抵达火星表面的成本。

这可能看起来像一个商业目标，但它其实正是需要优化的东西。抵达火星表面的每吨成本存在一个特定门槛，在这个门槛下，我们有能力建立一座自给自足的城市。而高于这个门槛，我们就负担不起。所以，现在你可以花1万亿美元飞往火星。无论多少钱都无法让你买到一张去火星的票。

所以我们需要把上面说的做到，让它成为一件实际上有可能实现的事。对于火星，我们不想只是想要留下旗帜和脚印，然后像我们对月球所做的那样，半个世纪都不再回去。为了通过一个非常重要的大过滤器。我认为我们需要成为一个多行星物种。

这些话对很多人来说可能听起来有些深奥，但只要时间足够长，最终，某件事，地球很可能会遭遇某种灾难，那可能是人类对自己做的事，也可能是像恐龙所遭遇的那种外部事件。但如果，最终，这些事都没有发生，而且不知怎么地，我们奇迹般地继续发展下去，那么太阳将会，太阳正在逐渐膨胀，并将吞没地球。

而且地球大概会在约5亿年后变得太热，不再适合生命生存。这是一段很长的时间，但它只比地球已经存在的时间长10%。所以，如果你想想，比如，当前的情况，它确实非同寻常，也有点难以置信，但地球已经存在了45亿年，而这是45亿年来第一次有可能把生命延伸到地球之外。

而且这个机会窗口可能会开放很长时间，我希望如此，但它也可能只开放很短时间，而我认为，明智的做法是在窗口仍然开放时迅速行动。以防它关闭。

是的，核武器的存在、流行病、各种各样的威胁，——[埃隆] 是的。

应该算是能给我们一些动力。

我的意思是，文明可能会，可能会轰然毁灭，也可能会悄然消亡。如果它死于人口崩溃，那么显然更像是悄然消亡。如果是第三次世界大战，那就更像是轰然毁灭，但这些都是风险。我的意思是，重要的是要思考这些事情，而且只是，把它们看作概率，而不是必然，有一定概率会有坏事发生在地球上。

我认为未来最有可能是美好的，但比方说，为了便于讨论，每个世纪都有1%的概率发生终结文明的事件。比如，那是斯蒂芬·霍金的估计。我认为他在这件事上可能是对的。我们基本上应该把成为一个多行星物种这件事看作给生命本身买保险，就像生命的寿险。（两人笑）——这一下子就变成电视购物节目了。

生命的寿险，是的。而且我们可以把地球上的生物、植物和动物带到火星，为这颗行星注入生命，并拥有第2颗存在生命的行星。那会很棒。它们无法自己去那里，所以如果我们不把它们带到火星，那么无论如何，太阳膨胀时它们肯定都会死，然后一切就结束了。

你认为在火星上建立文明、对火星进行地球化改造，最困难的方面是什么，比如从工程角度、财务角度、人类角度来看，如何让大量永远不会返回地球的人去那里？

不，他们当然可以回来，有些人会返回地球。

他们会选择在那里度过余生。

是的，很多人会。我们需要让飞船回来，比如那些前往火星的飞船，我们需要让它们回来，所以如果你愿意，可以搭乘返航。但我们不能就让飞船不回来，那些东西很昂贵。我们需要让它们回来。我想回来记录他们的旅程。

我的意思是，你会考虑地球化改造这个方面、实际建设这件事吗，还是说你目前太专注于飞船这部分，因为它对抵达火星至关重要？

是的，是的。我们绝对，如果你无法到达那里，其他一切都无关紧要。而且就像我说的，我们不能以某种高得离谱的成本抵达那里。我的意思是，目前，比如说，把1吨载荷送到火星表面的成本大约是10亿美元。因为你需要的不只是火箭、发射和所有那些东西，你还需要，比如隔热罩，需要制导系统，需要深空通信。你需要某种着陆系统。

所以，粗略估算一下，目前把每吨货物送到火星表面的成本大约是10亿美元。这显然太昂贵了，无法创建一个自给自足的文明。所以我们需要将其至少改善1000倍。

[莱克斯] 每吨100万美元？

是的，理想情况下要低于、远低于100万吨。你得问，比如，社会负担得起或愿意花多少钱，在火星上建设一座自给自足的城市？自给自足这一点很重要。比如，它就是关键门槛，感激于此，我们就已经跨过了，当火星上的城市即使在来自地球的宇宙飞船因任何原因停止到来时仍能生存。原因是什么并不重要。

但如果它们因任何原因停止到来，它会灭亡，还是不会？而且，只要缺少哪怕1种关键成分，那仍然不算数。这就像你正在进行一次漫长的海上航行，除了维生素 C 之外什么都有。（埃隆笑）你死去只是时间问题。所以我们必须让火星城市达到能够自给自足的程度。

我不确定这是否真的会在我的有生之年发生，但我希望至少能看到它拥有很大的势头。然后你可以问：“好吧，要建成一座自给自足的城市，所需的最低吨位是多少？”这件事存在很大的不确定性。你可以说，我不知道，可能至少要100万吨。因为你必须在火星上建设大量基础设施。

就像我说的，为了实现自给自足，你不能缺少任何东西，你不能，比如，你需要半导体晶圆厂，你需要铁矿石精炼厂，你需要很多东西，明白吗？而且火星并不是特别宜居。它是最不恶劣的行星，但它绝对是一颗需要修整的行星。

[Lex] 在地球之外。

是的。

地球相当不错。

地球就像是简单模式。对。

而且，我们还应该澄清，是在太阳系内。

[埃隆] 是的。在太阳系内。

可能会有一些不错的、类似度假胜地的地方。

外面或许有一些很棒的行星，但那是毫无希望的——太难抵达那里了？

对，太、太、太、太、太难了，至少可以这么说。

让我反驳一下。其实也不算反驳，而是快速抛出一个出其不意的问题。所以你确实提到物理学是第一个出发点。广义相对论允许虫洞存在。严格来说，它们可以存在。你认为人类有朝一日能利用它们以光速快速旅行吗？还是说你认为——虫洞这件事存在争议。我们目前不知道有任何方法能超越光速。

有一些关于拥有空间的想法。你会以光速穿过空间，但如果你能让空间本身移动，那就是扭曲空间。空间能够以超过光速的速度移动。

[Lex] 对。

就像宇宙大爆炸时那样，宇宙以远远超过光速的速度膨胀，快得多。

[Lex] 对。如果这有可能，实现空间扭曲所需的能量会庞大到令人难以想象。

那么，就你在推进方面所做的全部工作而言，火箭推进还有多大的创新空间？我的意思是，你什么都见过，而且一直在各个方面创新。还有多大空间？比如说，你能以某种方式达到10倍吗？物理学里是否存在某种东西，可以让发动机效率以及诸如此类的方面得到显著提升？

嗯，正如我刚才所说，真正的圣杯是一套完全且快速合理的轨道系统。目前，猎鹰9号是现有唯一可重复使用的火箭。助推器会返回并着陆，你看过那些视频。我们也收回了鼻锥或整流罩，但没有收回上面级。这意味着我们至少要承担建造一个上面级的成本。

你可以把两级火箭想成有点像2架飞机，一架大飞机和一架小飞机，我们收回了大飞机，却没有收回较小的飞机。所以成本仍然很高。那个上面级至少要1000万美元。而且，为了咽部的次序，助推器的可重复使用程度也没有达到我们希望的那种快速和完全。

所以，不计间接费用，我们每次飞行的最低边际成本大约是1500万到2000万美元，也许吧。这已经极其出色了，它远远胜过历史上的任何火箭。但通过完全且快速的可重复使用，我们可以把每吨载荷进入轨道的成本降低100倍。你就这样想，比如想象一下，如果你有一架飞机之类的东西，或者一辆汽车。

如果你每次开车出门都必须买一辆新车，那会非常昂贵。坦率说，这会很愚蠢。

嗯哼。

但实际上，你只要给汽车加油或充电，这就会让你的行程，我不知道，便宜1000倍。所以火箭也是一样。制造这种能够进入轨道的复杂机器非常困难。因此，如果你无法重复使用它，甚至不得不丢弃其中任何重要部分，成本都会大幅增加。理论上，星舰每次发射的成本可以做到大约100万美元，也许200万美元之类的。而且能把超过100吨送入轨道，这太疯狂了。

对。这太不可思议了。所以你的意思是，就投入产出而言，最有效的办法显然是让它完全可重复使用，而不是在理论物理学上取得某种非凡突破。

不，不，没有什么，没有什么非凡的突，不，没有。我们必须让火箭可重复使用，这是一个极其困难的工程问题。

明白了。

但不需要新的物理学。

只需要出色的工程技术。让我问一个稍微带点哲学意味的有趣问题。必须得问。我知道你专注于抵达火星，但等我们到了火星之后，你认为哪种政府形式、经济制度、政治制度最适合早期的人类文明？谈论这些事情有趣的原因是，它也能帮助人们畅想未来。我知道你真的专注于短期的工程梦想，但这就像，我不知道。想象火星上真正存在一个文明，会给人们带来某种——当然。

真的会给人们带来希望。

嗯，它会是一片新边疆，也是一次重新思考政府整体本质的机会，就像创建美国时所做的那样。我的意思是，我建议实行直接民主，比如人们直接对事务进行投票，而不是代议制民主。所以，我认为代议制民主太容易受到特殊利益集团以及对政客的胁迫之类事情的影响。因此，我建议就实行直接民主。人们对法律投票，民众亲自对法律进行投票，而且法律必须足够简短，让人们能够理解。

对，然后让民众充分了解情况，真正公开他们所投票事项的全部信息。

对。绝对透明。

对。而且不要把它弄得像那些我们必须接受的Cookie一样烦人——必须接受Cookie。每次点击接受Cookie时，总会有一点忐忑。我感觉也许有一个非常非常小的概率，它会打开一道通往地狱的传送门之类的东西。

[Lex] 我的感受完全一样。他们为什么一直想让我接受那个？他们想用这个Cookie做什么？是不是某个地方有人因为接受Cookie之类的事情不高兴了。我的意思是，谁在乎？不停接受所有这些Cookie太烦人了。

[Lex] 对我来说，这就是一个很棒的——我已经厌倦了接受——（Shivon轻声说话）行，你可以拿走我该死的Cookie，我不在乎。随便吧。

[Lex] 你们最先从我这里听到了，埃隆，他接受你们所有该死的Cookie。

对。（两人笑）别再问我了。很烦人。

对，这是一个好想法被实施得极其糟糕的例子。

对，有人当时大概是想着，隐私之类的事情出发点很好，但现在每个人都只能勾选接受Cookie，如今有数十亿人不得不一直点击接受Cookie，这超级烦人。接受这该死的Cookie就行了，没事。我认为这里存在一个根本问题，因为我们已经有一段时间没有真正经历一场重大的，比如世界大战之类的事情了。

显然，我们不希望发生世界大战。规则和法规一直没有经历过清理机制。因此，战争确实有某种好的一面，那就是战争结束后，规则和法规会得到重置。因此，第一次和第二次世界大战后，规则和法规都经历了大规模重置。

如果社会没有战争，也没有针对规章制度的清理功能或垃圾回收机制，那么规章制度每年都会累积，因为它们是不朽的。没有真正的，人在死去，但法律不会。所以，我们需要一个针对规章制度的垃圾回收机制，它们不应该就是不朽的。

因为制定的一些规章制度会适得其反，出发点是好的，但会适得其反。有时出发点也并不好。如果规章制度每年都在累积，变得越来越多，那么最终你将什么都做不了。你就像格列佛一样，被成千上万根细绳捆住了。我们在美国就看到了这种情况。

还有洛杉矶，基本上所有存在了一段时间的经济体都是如此，监管者和立法者每年都会制定新的规章制度，但他们不会花力气去废除它们。我认为，我们努力废除规章制度是非常重要的。

但这会变得很困难，因为会出现依赖于它们的特殊利益集团，他们在某项规章制度中拥有既得利益，于是他们就会抗争，不让它被废除。

对。我的意思是，我想宪法的问题在于，它有点像C与Java的对比，因为它没有内置任何垃圾回收机制。我觉得应该有。你刚说到垃圾回收这个比喻时，我很喜欢——对，这是从编码的角度来说的。

从编码的角度来说，对，对。如果法律本身以某种方式内置了一种机制，让它们过一段时间就自行失效，除非有人明确地公开为其辩护，那会很有意思。所以这有点像，并不需要有人去消灭它们。它们会自行消亡。它们会消失。

[埃隆] 对。

不是要为Java之类的东西辩护，C++、Python等等也可以有很棒的垃圾回收机制。

对。所以，是的，必须采取某种行动，否则文明的动脉只会随着时间推移而硬化。你能做成的事情会越来越少，因为每件事都有一条规则加以禁止。所以我认为，我不知道，对于火星，或任何地方，我是说，甚至对于这里，显然也包括地球，我认为应该有一个主动废除规章制度并质疑其存在理由的流程。

如果我们有一个创建规章制度的函数，因为规章制度也可以被看作，它们就像是运行一个文明的软件工作或代码行，这就是规章制度。所以并不是说我们不该有规章制度，而是你的代码不断累积，却没有移除代码。于是过一段时间后，它基本上就会变成陈旧臃肿的软件。

而且这只会让事情难以取得进展。所以，我不知道，也许在火星上，任何一项法律都必须有失效期限，并且需要主动投票才能让它继续有效。实际上我还会说，这些只是，我不知道，建议或想法，最终将由火星上的人们来决定，但我认为废除一项法律应该比增加一项法律更容易，因为，为了克服法律的惯性。

所以，比如仅仅为了讨论，你可能需要比如60%的票数才能让一项法律生效，但只需要40%的票数就能废除它。

那就让我来当那个人吧，你最近在Twitter上发了一个梗图，里面有一排小便池，然后一个人径直穿过整排——太真实了，对。

然后他跟你大谈加密货币。

听着，我是说，这在我身上发生过太多次了，我想甚至可能真的是字面意义上的。（两人都笑）——从技术角度来说，你认为智能合约之类的理念还有什么发挥空间吗？因为你提到了法律，使用智能合约之类的东西来实施政府赖以运作的法律，是一种很有意思的实现方式。比如构建在以太坊上的东西，或者也许是某种以某种方式支持智能合约的狗币。

我从来没有，我不太理解整个智能合约这回事。（两人都笑）我太市中心了，理解不了智能合约。

这句话不错。（两人都笑）——我的意思是，对于任何类型的交易或其他事情，我通常的做法就是确保大家理解清楚。这是最重要的事情。

[莱克斯] 对。

而且要让任何类型的交易都非常简短、简单，使用平实的语言，并确保每个人都明白，这就是这笔交易。每个人都，清楚吗？如果最先那些事情没有发生，后果是什么？但通常交易、商业交易或其他交易都过于冗长复杂，被律师过度加工，而且毫无意义。

你提到Doge是人民的币。

[埃隆] 对。

而且你说过，你真的打算，SpaceX可能会考虑真的把一枚狗狗币放到月球上。

对。

这是你仍在考虑的事情吗，也许是火星，你认为有没有某种可能，我们谈过火星上的政治制度，狗狗币会成为火星的官方货币，它是未来之币？

嗯，我认为火星本身需要有一种不同的货币，因为受光速限制，你无法实现同步，或者说无法轻易实现。

所以它必须完全独立于地球。

嗯，对，火星在最接近时，大约相距4光分，然后再加上这种接近方式，大约相距20光分，可能还要多一点。所以，如果你面临20分钟的光速问题，而它采用的是1分钟区块链，你就无法真正让某种东西保持同步。它无法正确同步。我不知道火星是否会把某种加密货币作为一种东西，但很可能，似乎很有可能。不过它会是某种只在火星本地使用的东西。

而且你让人民来决定。

对，当然。火星的未来应该由火星人决定。我的意思是，我认为加密货币这件事是减少那个名为货币的数据库中错误的一种有趣方法。我认为自己对货币在实际日常层面究竟是什么有相当深入的理解，因为有PayPal。我们当时确实深入研究了这件事。而目前的货币系统，实际上从实际用途来看，就是一堆运行着旧式COBOL的异构大型机。

[莱克斯] 好，你是说真的是——真的是。

这确实就是正在发生的事。

以批处理模式。好。

以批处理模式。

对。可怜那些不得不一直维护那段代码的倒霉家伙。好。那真痛苦。

[莱克斯] 甚至都不是 Fortrans，是 COBOL，没错。

那就是 COBOL。而且他们仍然，银行仍然在购买大型机，在 2021 年，并运行引擎 COBOL 代码。美联储用的东西可能甚至比银行的还要老，他们有一台老旧的 COBOL 大型机。所以政府实际上拥有货币数据库的编辑权限。而他们会利用这些编辑权限，随时按自己的意愿制造更多货币。这增加了货币这个数据库中的错误。

所以我认为，确实应该透过信息论的视角来看待货币。它有点像互联网连接。比如带宽是多少、总比特率是多少、延迟抖动如何、是否丢包、网络通信中有什么错误。基本上就这样思考货币。我想这大概就是我对它真正的看法。然后再问，从信息论的角度来看，什么系统能让经济以最佳状态运转。

加密货币是一种尝试，旨在减少货币中的错误，这种错误是由政府稀释货币供应造成的，而这基本上是一种有害的征税形式。

所以，无论是通胀方面的政策，还是实际的，比如技术层面、COBOL，加密货币都在实际系统方面把我们带入了 21 世纪，这些系统让你能够进行交易、储存财富，以及做所有这类事情。

就像我说的，只要想想——理论上。

关于作为信息的货币，人们常常会认为货币本身拥有力量。它没有。货币是信息，它本身并没有力量。运用物理学中思考极限情况的工具会很有帮助。如果你被困在一座热带岛屿上，而你有 1 万亿美元，那也毫无用处。因为那里没有资源配置。

货币是一个资源配置数据库，但除了你自己以外，没有任何资源可供配置。所以货币毫无用处。如果你被困在一座没有食物的荒岛上，全世界所有的比特币也无法阻止你饿死。

[莱克斯] 对。

只要把货币看作一个跨越时间和空间进行资源配置的数据库。然后，什么样的系统，以什么形式来构建这个数据库或数据系统，才会最有效？比如说，比特币目前的形式存在一个根本问题，那就是它的交易量非常有限。而且延迟，延迟，一笔得到适当确认的交易所需的时间太长，远比你期望的要长。

实际上，无论从交易量还是延迟的角度来看，它都不算好。所以它或许有助于解决货币数据库问题的某个方面，也就是某种财富储存，或者我想，是对相对债务的记账。但它作为货币并没有用，无法作为日常货币。

但人们提出了不同的技术解决方案————[埃隆] 比如闪电网络。

对，Lightening 网络，以及构建在它之上的第 2 层技术。

我的意思是，这一切似乎都是某种权衡，但关键是，以信息的角度来说，这有点高明，只去思考信息，思考什么样的数据库、什么样的基础设施能够实现……的交换——对，比如说你在运营一个经济体，你需要有某种东西，能够让产品和服务之间存在高效的、实现高效的价值比率。

所以，你有数量极其庞大的产品和服务，而且需要，你不能只是以物易物。因为那会极其笨重难行。所以你需要某种东西，为商品和服务之间提供交换比率。然后，还需要某种东西，让你能跨时间转移义务，比如债务，债务和股权会跨时间转移义务。那么，什么东西最能做好这件事？

我认为狗狗币有一些可取之处，部分原因是，尽管它显然是作为一个玩笑创造出来的，但它实际上拥有比比特币高得多的交易容量。进行交易的成本，狗狗币的手续费非常低。比如现在，如果你想进行一笔比特币交易，进行这笔交易的价格非常高，所以你无法有效地将它用于大多数事情。

而且它甚至也无法扩展到很高的交易量。当比特币开始时，我猜大约是2008年之类的，那时的互联网连接比今天差得多，差了一个数量级。我的意思是，2008年时确实差得多得多。所以，采用较小的区块大小或者不管它是什么，以及较长的同步时间，在2008年是合理的，但到了2021年，或者快进10年，它就低得可笑了。

而且我认为，让生成的货币数量线性增长是有一定价值的。所以，因为一定数量的货币，如果一种货币通缩性太强，或者说，或者应该说，如果，如果一种货币预计会随时间升值，人们就会不愿意花掉它。

因为你会想：“哦，如果我，我就持有它而不花掉，因为它会随时间变得越来越稀缺，所以如果我现在花掉它，之后就会后悔。所以我就会，你知道，把它全都囤起来。”但如果货币会随时间发生一定程度的稀释，那就更能激励人们把它作为货币使用。

所以，狗狗币多少有些随机地规定了每年生成固定数量的某种币或者哈希字符串。因此存在一定的通胀，但它并不是以百分比为基础的。它是一个固定数字，所以通胀百分比必然会随时间下降。我不是说它就是理想的货币体系，但我认为，纯属偶然的是，它实际上从根本上优于我见过的其他一切。

我喜欢你说“大约2008年”的方式，所以你不是，有些人认为你可能是中本聪。你大概已经说过你不是。让我问——我不是。

你不是，确定。如果你是，你会告诉我们吗？

会。

好。你认为他，或者她，或者他们保持匿名，是一个特性还是一个缺陷？在人类历史上，有一项特定技术的发明者完全匿名，这是一种很有意思的奇特现象。或者说创造者。

嗯，我的意思是，你可以看看比特币推出前各种思想的演变，看看是谁写过那些思想。然后，我不知道，显然从实际意义上讲，我不知道是谁创造了比特币，但那些思想的演变脉络相当清楚。而且，看起来 Nick Szabo 对这些思想演变所起的推动作用可能比任何其他人都大。所以，他在这里声称自己不是中本聪，但我不确定，那也无关紧要，不过在比特币背后的思想方面，他似乎比任何其他人都起到了更大的推动作用。

所以，或许，单个的人物甚至没有参与思想演变并最终促成事物出现的那些人物重要。

对。也许想到历史会令人难过，不过大多数名字或许无论如何都会被遗忘。

名字到底是什么，它就是一个名字，一个附着在某种思想上的名字。它究竟真的意味着什么？

我记得莎士比亚说过一些关于玫瑰之类的话，不管他说的是什么。

“玫瑰即使换了名字，也会同样芬芳。”（Lex 笑）——我让埃隆引用莎士比亚了。我觉得自己今天有所成就。

“我可否将你比作夏日？”（两人都笑）——[Lex] 我反而要把这段剪出来。

你更加温和，也更加美丽。（两人都笑）（Shivon 轻声说话）——Autopilot。Tesla Autopilot——（埃隆笑）Tesla Autopilot 在过去6年里经历了一段不可思议的旅程，或许在你的思想中，以及许多参与者的思想中，甚至还要更久。

我觉得我们第一次真正产生联系，大概就是因为 Autopilot 的事情，自动驾驶以及……

在我看来，整个历程都令人难以置信。因为我知道，嗯，部分原因是我当时在 MIT，我知道计算机视觉有多难。而且我知道整个，我有很多同事和朋友，关于 DARPA 挑战赛，也知道它有多难。所以，当我第一次驾驶一辆搭载基于 Mobileye 的初始系统的 Tesla 时，自然而然地抱有怀疑。我觉得这不可能。

所以我刚坐进去时心想：“这辆车不可能保持，比如说留在车道内，并创造舒适的体验。”因此，我最初的直觉是，车道保持这个问题实在太难解决了。

[埃隆] 哦，车道保持，对，那个相对容易。

但要用我们刚才、我们之前谈到的方式来解决，也就是这个原型，与一个实际上能在数十万英里或数百万英里的行程中创造愉快体验的东西之间的区别。对，所以，事实证明我错了——我们必须在 Mobileye 那个东西周围包上大量代码，它不是自己就能运行的。

我的意思是，这也是你有时处理事情的方式这个故事的一部分。有时你从零开始做。有时一开始你会看看外面有什么，然后你决定从零开始。无论在硬件还是软件上，决定最终从零开始，都是我见过的最大胆的决定之一。我当时依然怀疑它是否能够成功，因为这是一个如此困难的问题。

所以这是一段不可思议的旅程。我现在看到的这一切，包括硬件、算力、传感器，而我或许最关心、最喜欢的，是 Andrej Karpathy 领导的那些工作：数据集选择、整个数据引擎流程、神经网络架构，以及那个网络在现实世界中接受测试和验证的方式、各种不同的测试集。不同于计算机视觉的 ImageNet 模型之类的学术界做法，这是现实世界中的人工智能。

Andrej 很出色，显然也发挥着重要作用，但我们有很多真正才华横溢的人在推动各项工作。Ashok 实际上是 Autopilot 工程负责人。Andrej 是 AI 总监。

AI 方面的工作，对。所以对，我知道这是一个了不起的团队，有非常多的事情正在进行。

人们会把太多功劳归给我，也会把太多功劳归给 Andrej。

而且人们应该意识到，背后正在进行多少————是的，就是有很多非常有才华的人。Tesla Autopilot AI 团队极其出色。他们就像是世界上最聪明的一群人。所以，是的，我们正在把它做成。

在自动驾驶的这5、6年里，对于自动驾驶这个问题，你获得了哪些洞见？所以，你一头扎了进去，带着某种第一性原理式的直觉，但没人知道这个问————是的，我原以为自动驾驶问题会很难，但它比我想象的更难。并不是说我原以为它会很容易，我原以为它会非常难，但实际上它甚至比那还要难得多。

所以，我是说，归根结底，要解决自动驾驶，你就必须解决。你基本上需要重现人类驾驶时所做的事情，也就是人类依靠光学感官、眼睛和生物神经网络驾驶。所以，为了，这就是整个道路系统被设计来运作的方式，基本上依靠被动光学和生物神经网络。

所以，实际上，要让完全自动驾驶发挥作用，我们必须以数字形式重现这一点。所以我们必须，这意味着摄像头加上以硅形式实现的先进神经网络。然后它显然会解决小细胞驾驶问题。这是唯一的办法，我认为没有其他办法。

但问题是，你必须把人性中的哪些方面编码进机器里，对吧？所以你必须解决感知问题，比如检测，然后你首先要弄清楚，驾驶中的感知问题是什么？比如你必须能够看到的各种事物。比如我们开车时究竟会看些什么？有，我最近刚听到 Andrej 在 MIT 谈到，比如车门。

我认为那是有史以来关于车门的全世界最精彩的演讲。车门的细微细节，比如到底什么才算一扇打开的车门，老兄。所以，比如它的本体论，那是一个感知问题。我们人类解决了那个感知问题，而 Tesla 必须解决那个问题。然后还有控制和规划，它们与感知相互耦合。

你必须弄清楚驾驶涉及哪些东西，尤其是在所有不同的边缘案例中。也许你可以谈谈这一点，在四向停车标志路口，需要涉及多少博弈论之类的东西？作为人类，我们开车时，我们的行动会影响世界。

确实。

它会改变其他人的行为，在大多数自动驾驶中，你通常只是在对场景作出反应，而不是真正在场景中坚持自己的行动。你认为……

我认为这类控制逻辑难题并不是困难的部分。让我们看看……

[莱克斯] 你认为这整个美丽而复杂的问题中，困难的部分是什么？

老兄，这需要海量该死的软件，还有大量聪明的代码行。毫无疑问，为了创建一个准确的向量空间。

你是从图像空间出发的，它就像是这股流向摄像头、摄像头们的光子流，然后因为图像空间中有这股庞大的比特流，接着你必须有效地压缩这个，一股与那些在摄像头传感器中撞出一个电子的光子相对应的庞大比特流，并把这股比特流转化为向量空间。

我所说的向量空间是指，其中有汽车、人、车道线、弯道、交通信号灯之类的东西。一旦有了准确的向量空间，控制问题就与电子游戏中的类似，比如《侠盗猎车手》或《赛博朋克》。如果你有准确的向量空间。这个，控制问题，我不会说它微不足道，它并非微不足道，但它它并不是什么无法克服的事情。

拥有准确的向量空间非常困难。

是的，我认为我们人类没有充分尊重人类感知系统是多么不可思议，它能把原始光子映射为我们脑中的向量空间表征。

你的大脑正在进行数量惊人的处理，并向你提供一幅经过高度净化的图像。比如我们环顾这里时，你能在眼角看到颜色，但实际上，你眼睛的周边视觉中只有极少的视锥细胞、视锥感受器。你的眼睛正在周边视觉中绘上颜色。

你没有意识到这一点，但它们，眼睛实际上正在绘上颜色，而且你的眼睛还会有，那里有血管和各种狰狞的东西，还有一个盲点，但你看得到自己的盲点吗？看不到，你的大脑正在把缺失的部分、那个盲点填补起来。

你可以在网上做这些事情，你看这里，再看这个点，然后再看这个点，而它，如果它处在你的盲点中，你的大脑就会直接填补缺失的部分。

太酷了。周边视觉太酷了。

是的。

它让你意识到所有这些错觉、供应科学，它让你意识到大脑究竟有多么不可思议。

大脑正在对来自你眼睛的视觉信号进行数量惊人的后期处理。太疯狂了。然后，即使你获得了所有那些视觉信号，你的大脑仍在不断尝试尽可能多地遗忘。所以，人类记忆也许是大脑最薄弱的东西，是记忆。

所以，因为记忆对我们的大脑而言代价非常高，而且非常有限，你的大脑会尝试尽可能多地遗忘，并把你看到的东西提炼成尽可能少的信息。所以，你的大脑不仅试图得出一个向量空间，还要得出一个尽可能小、只包含相关物体的向量空间。

你可以算是观察自己的大脑内部，或者至少我可以，比如当你沿着道路行驶时，试着想想你的大脑实际上正在做什么，——是的——有意识地。这就像，你会看到一辆车，因为你没有摄像头。你的后脑勺或侧面没有眼睛，所以你会说，比如说，基本上，你的脑袋就像一个，你基本上就像是在一个缓慢的云台上装了2台摄像头。（两人都笑）而且视力也没那么好。

好吗？人类的眼睛是……而且人们总是在分心、想事情、发短信，以及在车里做各种不该做的事，切换广播电台。所以，争吵。你上一次向右、向左、向后，甚至向斜前方看，真正刷新你的向量空间，是什么时候？

所以你会扫视四周，而你的心智正在做的，是尝试提炼相关向量，基本上就是具有位置和运动状态的物体，然后再将其删减到你驾驶所必需的最少数量。

它似乎确实能够把它精简下来，或者进一步压缩成概念之类的东西，所以它不是，它就像是超越了，人的思维有时似乎会超越向量空间，进入某种概念空间，在那里你会看到一个东西，它不再以某种空间形式来表示，它几乎就像是一个你应该意识到的概念。如果这里是学校区域，你会把它作为一个概念记住。这是一种很奇怪的表示方式，但也许对驾驶来说，你不需要完整地表示那些东西。或者也许你通过那种方式获得——嗯，你——间接地。

你需要建立向量空间，然后实际上对那些向量空间作出预测。比如你开车经过一辆公交车，看见那里有人，在你驶过公交车之前，你看见有人正在过马路，或者就想象那里有一辆大型卡车之类的东西挡住地点。但在你接近那辆卡车之前，你看到有几个孩子正准备从卡车前方横穿马路。

现在你已经看不到那些孩子了，但你现在会知道，好吧，那些孩子很可能会从卡车旁经过，然后穿过马路。即使你看不到他们。所以你必须有记忆。你需要记得那里曾经有孩子，而且你需要对他们将会处于什么位置做出某种前向预测。

这是一个非常困难的问题——在相关的时刻。

所以在遮挡和计算机视觉方面，当你再也看不到一个物体时，即使它只是走到一棵树后面又重新出现，这也是一个非常、非常，我是说，至少在学术文献中，这叫作跨遮挡追踪，非常困难。

是的，我们正在做。

[Lex] 我明白这一点。所以其中一部分——这就像客体恒存性。同样的事情也发生在人类的神经网络中。就像一个幼儿成长时，会在某个时间点发展出客体恒存感。所以在某个年龄之前，如果你有一个球、一个玩具或其他什么东西，然后你把它放到背后再拿出来，在他们形成客体恒存性之前，每次看起来都像是一个新东西。

就像，“哇，这个玩具噗的一下不见了，现在它又回来了。”而且他们简直不敢相信。他们可以一整天都玩躲猫猫，因为躲猫猫每次都是新鲜的。但后来我们弄明白了客体恒存性，他们就会意识到，“哦，不，这个物体并没有消失。它只是在你的背后。”

有时候我真希望我们从未弄明白客体恒存性。

客体恒存性。是的，所以那是一个……

[Lex] 那是一个需要解决的重要问题。

是的。所以，汽车中神经网络的一项重要演进，就是具备跨时间和空间的记忆。现在你不可能一直记住，你必须确定自己想把事情记多久。长时间记住事情是有成本的。所以，如果试图把太多事情记得太久，你的内存可能会耗尽。而且，如果你把一些东西记得太久，它们也会变得过时。

然后你还需要那些随时间被记住的东西。所以即使你，比方说，为了举证，在时间维度上有5秒的记忆，但假设你停在一个红绿灯前，并且你看到，用行人这个例子来说，有人在等待穿过穿过马路，而由于遮挡，你看不太清他们，但在信号灯改变、让他们穿过马路之前，他们可能会等上1分钟。

你仍然需要记住他们当时就在那里，而且他们很可能会穿过马路，诸如此类。所以，即使这超出了你基于时间的记忆，也不应该超出你的空间记忆。

而我只是觉得这件事的数据引擎方面，也就是获取数据来学习你现在所说的所有概念，是一个不可思议的过程。这是一个不断迭代的过程，就是，有这样一个由许多——组成的 HydraNet——HydraNet。我们正在把名字改成别的。

好吧。行。我相信它会同样具有《瑞克和莫蒂》那种风格。

是的。我们已经对汽车里的神经网络重新设计架构太多次了，简直疯狂。

哦，所以每次推出新的重大版本时，你都会把它改成一个更荒唐，或者，或者更令人难忘、更美的名字，抱歉。当然不是荒唐。

如果你看到操控汽车的完整神经网络阵列，它会让人有点难以理解。有太多层了。太疯狂了。我们一开始使用的是简单的神经网络，基本上是在单个摄像头的单帧画面上进行图像识别，然后试图用、用 C 把它们拼接在一起。我应该说，我们对在这里运行 C 非常熟悉，因为 C++ 的开销太大，而且我们有自己的 C 编译器。

所以，为了获得最高性能，我们实际上编写了自己的 C 编译器，并且还在继续优化我们的 C 编译器，以实现最高效率。事实上，我们最近刚刚对 C 编译器做了一个新修订版，它将直接编译到我们的 Autopilot 硬件上。

所以你想用自己的编译器把整个东西全部编译到底层？

是的。

所以这里讲究效率，因为有各种各样的计算机、CPU、GPU，还有一些基本类型的东西，而你必须以某种方式解决所有这些东西之间的调度问题。所以你是在把代码编译成——是的。

那会完成全部工作，好吧。所以这就是为什么有很多人参与其中。

其中有很多非常硬核的软件工程工作，处在一种非常接近裸机的层级。因为我们正试图完成大量计算，而计算能力受到我们的完全自动驾驶计算机的限制。我们希望在一种非常有限的算力和功耗条件下，尽量实现尽可能高的每秒帧数。我们确实投入了大量精力来提高计算效率。

所以，Tesla 一些非常有才华的软件工程师实际上完成了大量工作，在一个非常基础的层级提高计算效率，以及我们如何使用 trip 加速器；它们基本上是在进行矩阵数学运算、点积，比如无数次点积。而且，就像，我们的一个神经网络从计算量上说，大约 99% 都是点积。

而你想实现尽可能多的高帧率，就像电子游戏一样，你想要——对。

全分辨率，更高帧率。

高帧率、低延迟、低抖动。我认为我们现在正在推进的事情之一，是不再通过图像信号处理器对图像进行后处理。对于摄像头而言，情况是，嗯，几乎所有摄像头都会进行大量后处理，以使图片看起来漂亮。而我们并不在乎图片看起来是否漂亮。我们只想要数据。所以我们正在转向只使用原始光子计数。

计算机看到的图像实际上比你把它呈现在相机上时所能看到的多得多，它包含的数据要多得多。即使在光线非常微弱的条件下，你也能看到这里这个点和那里那个点之间存在微小的光子计数差异，这意味着，所以它在黑暗中的视力好得惊人，因为它能够检测到这些光子计数上的微小差异。比你所能想象的还要好得多。

我们还将延迟缩短了13毫秒。

[Lex] 是通过取消对图像的后处理吗？

是的。

对。

因为我们有8个摄像头，然后每个摄像头大约有，我不知道，1.5毫秒左右，也许是1. 6毫秒的延迟。绕过图像处理器基本上能让我们减少13毫秒的延迟，这一点很重要。

而且我们会一路追踪延迟，从光子击中摄像头开始，到它为了完成处理而必须经历的所有步骤，经过各种神经网络和C代码，其中也有一点C++。嗯，我可以说，也许有很多，但它，核心部分，也就是高强度计算，全都是用C完成的。

所以我们会一路追踪这个延迟，直到向驱动单元发出外部指令，去加速刹车、减慢转向、向左或向右转。因为你必须输出一条指令，这条指令必须传到控制器，而其中有些控制器的更新频率可能是10赫兹之类的，这很慢。这样一来，你可能就损失了100毫秒。

所以我们想更新转向和制动控制的驱动器，使其更接近100赫兹，而不是10赫兹，这样延迟就是10毫秒，而不是最坏情况下的100毫秒。

实际上，抖动比延迟更具挑战性，因为延迟是可以预期和预测的，但如果从摄像头到计算机，再经过一系列其他计算机，最后到达汽车上的执行器，这一整套环节层层叠加；如果各种容差、时序容差累积起来，那么延迟就可能变化很大，这就叫作抖动。

这使得你很难准确预判应该如何让汽车转向或加速，因为如果有大约150、200毫秒的抖动，那么你可能会偏差2. 2秒。这可能会造成很大的差异。

所以你必须以某种方式进行插值，以应对抖动的影响，这样它们才能做出稳健的控制决策。所以抖动存在于传感器信息中，或者说抖动可能发生在处理流程的任何阶段。

如果你只是，如果你的延迟是固定的，你就可以预期，比如说：“好，我们知道我们的信息是什么，”举例来说，“已经过时150毫秒了。”举例来说，从光子进入摄像头，到你能够测量到车辆加速度的变化，需要150毫秒。

那么你只要说：“好吧，我们，我们知道是150毫秒，所以我们会把它考虑进去，并对这个延迟进行补偿。”然而，如果你有150毫秒的延迟，再加上100毫秒的抖动，而额外的抖动可能在0到100毫秒之间的任何位置。那么你的延迟就可能从150毫秒到250毫秒，现在你就有100毫秒是不知道该如何处理的。

那基本上是随机的。所以，消除抖动极其重要。

而这会影响你的控制决策以及诸如此类的所有事情。好。

对，抖动越低，汽车从根本上就能操控得更好。

[Lex] 明白了。

汽车将以超越人类的能力和反应速度进行操控，比人类快得多。我的意思是，我认为随着时间推移，Autopilot、完全自动驾驶将能够完成远远超过詹姆斯·邦德在最精彩的电影中所能完成的那种动作。

你说这话的时候，我脑海中想象的正是那样。

就像是人类无法完成的不可能动作。

好吧，让我问一个算是，回顾过去6年，再展望未来，根据你目前的理解，你觉得完全自动驾驶这个问题有多难，你认为Tesla什么时候能解决4级FSD？

我的意思是，看起来很可能会在明年实现。

那么解决方案会是什么样子？是目前这一批FSD测试版候选用户吗？他们会像此前一样，获得越来越高程度的自主性。然后超过某个水平后，他们就能做自己的事，他们可以读书。

对。我的意思是，你可以看到，任何一直密切关注完全自动驾驶测试版的人都会发现，脱离自动驾驶的频率一直在迅速下降。所以，比如有一种接管B，就是驾驶员进行干预，以防汽车做出某件——[Lex] 对。可能有危险的事情。所以每100万英里的干预次数一直在大幅下降。

而从这一趋势来看，明年似乎就会出现这种情况：使用FSD发生事故的概率低于普通人类，然后显著低于普通人类。所以，看起来我们明年肯定能做到。

接下来就是，好吧，我们不必向监管机构证明这一点，并向……证明这一点，而且我们想要的标准不仅是与人类相当，而是远远优于普通人类。我认为它的安全性必须至少比人类高2到3倍。受伤概率要比人类低2到3倍，我们才会真正说：“好，可以上路了。”

它不会只是相当，而会好得多。

所以你看，FSD 10.6最近刚发布，10.7正在路上，也许11会在未来某个时候推出。

对。我们原本希望今年推出11，但它，11实际上包含了对神经网络架构的一系列根本性重写，以及在创建向量空间方面的一些根本性改进。

所以确实有某种根本性的飞跃，配得上11这个版本号。我的意思是，这个数字挺酷的。

对。11将成为适用于一切的单一技术栈，一个统御所有功能的技术栈。

单一技术栈。但其中确实有一些非常根本性的神经网络架构变化，将带来强大得多的能力。一开始它们会有问题。比如我们已经在某种Alpha版软件上让它运行起来了，效果不错，但它，它基本上是在接手一大堆C、C++代码，并引领大量的C++代码，然后用神经网络取代它。

Andrej经常提到这一点，也就是说，神经网络正在某种程度上吞噬软件。随着时间推移，传统软件越来越少，神经网络越来越多。它仍然是软件，但它，最终仍然体现为一行行软件。不过，就是有更多神经网络的东西，而启发式方法更少。更多基于矩阵的东西，更少基于启发式方法的东西。

其中一个重大变化将是，现在神经网络会把一大包点传给 C++，或者 C 和 C++ 代码。

[Lex] 是的。

我们称它为一大包点。

[Lex] 是的。它就像，你有一个像素和某种与该像素相关联的东西，比如这个像素可能是汽车，这个像素可能是固定电话。然后你得在 C 代码中组装这一大包点，并把它转化为向量。它在这方面做得相当不错，但它是，我们想要只是，我们需要在它上面再加一层神经网络，接收这一大包点并将其提炼为软件神经网络部分中的向量空间，而不是软件的启发式规则部分。这是一个很大的改进。

[Lex] 一路到底都是神经网络，所以你想要。

甚至并不全是神经网络，但这，这是一个颠覆性的变化，不再需要那一袋点，那一大袋点，它们必须用许多行C、C++代码组装起来，而是让一个神经网络直接把它们组装成一个向量。所以神经网络输出的数据少得多、少得多，它输出的是：这是一条车道线，这是一处路缘，这是可行驶空间，这是一辆车，这是一个行人或骑自行车的人，或者类似的东西。

它正在输出，它确实是在向C、C++控制代码输出正确的向量，而不是在C里构造这些向量。我认为这方面我们做得相当不错，但它逐渐有点触及了C能把这件事做到多好的局部最大值。所以这确实是一件大事。而且车里的所有网络都需要转向环绕视频，现在仍有一些不是环绕视频的遗留网络。

而且所有训练都需要转向环绕视频，训练效率也需要提高，而且确实正在提高。然后，我们需要把一切都转向原始光子计数，而不是经过处理的图像。

[Lex] 好的。所以如果你————这会让训练经历一次相当大的重置，因为系统是用后处理成像后的图像训练的。所以我们需要重新进行所有训练，改为针对原始光子计数进行训练，而不是后处理图像。

所以归根结底，这算是在降低整个系统的复杂性。所以，降低。

对。代码行数实际上会减少。

是啊，这很有意思。所以你会注入所有传感器，从而降低必须处理这些————[Elon] 注入摄像头。

抱歉。

其实全都是摄像头。

对，是的。人类也是一样。

是啊。

嗯，我想我们也有耳朵，好的。

是啊，我们其实也需要纳入声音。因为，你知道，你需要听救护车警笛或消防车的声音。如果有人，对你大喊什么的，我不知道。只是，还需要纳入一点音频。

你需要去上个厕所休息一下吗？

[Elon] 是啊，当然，我们休息一下吧。

好的。

[Elon] 老实说，坦率地说，想法是容易的部分，实施才是困难的部分。去月球这个想法是容易的部分，但真正去月球才是困难的部分。

[Lex] 才是困难的部分。

而且在硬件和软件层面，有大量硬核工程工作必须完成。就像我说的，要优化C编译器，而且就是，消除各处的延迟。如果我们不这么做，系统就无法正常运行。所以，完成这些工作的工程师，他们就像无名英雄。但他们对这件事的成功至关重要。

我认为你已经说清楚了。我的意思是，至少对我而言，Andrej所做工作之外正在发生的一切都超级令人兴奋。就是整个软件基础设施。我的意思是，数据引擎正在进行的一切，不管它叫什么，整个过程简直是一件艺术品。

它规模之大，令人难以置信。训练，所完成的工作量，我们编写了所有这些用于训练和标注的定制软件，以及进行顺序标注。顺序标注至关重要。因为，尤其是当你有环绕视频时，从头开始标注环绕视频极其困难。人类甚至要花很长时间才能标注一个视频片段，比如几个小时。

或者顺序标注器，它基本上会对视频片段投入重型的、大量的算力，预先指定并猜测环绕视频中正在发生的所有事情分别是什么。

[莱克斯] 而且还有像是纠正它。

对，然后人类要做的就只是稍微调整一下，比如说，调整不正确的地方。这就像是把生产力提高了100倍或更多。

对。所以你把 Tesla Bot 描述成主要在工厂里有用。首先，作为一个机器人技术爱好者，我觉得人形机器人不可思议。我觉得人形机器人、双足机器人所展现出的动作优雅感实在太酷了。

你在研究这个，同时还谈到应用同一种，就是所有这些想法，其中有一些你已经谈过了，包括数据引擎，包括我们正在讨论的所有东西，包括 Tesla Autopilot，只是把它们转移到这个，又一个机器人问题上，这真的很有意思。因为我关心人机互动，所以我必须问问其中人类这一面。所以你谈到的主要是在工厂里。

你是否认为，Tesla Bot 必须解决的这个问题中，有一部分是与人类互动，并可能在家里这样的地方拥有一席之地。所以，是互动，而不只是，——当然。

不是取代劳动力，而是也可以，怎么说呢，成为朋友或助手。

[埃隆] 我认为可能性是无穷的。对，我的意思是，很明显，它并不完全符合 Tesla 加速可持续能源发展的主要使命方向，但这是我们能够为世界做的一件极其有用的事情，也就是制造一种有用的人形机器人，它能够与现实世界互动，并以许多不同的方式提供帮助。

所以在工厂里，以及其实就是，我的意思是，我认为，如果你把时间外推到未来许多年，我认为工作将变成可选的。有很多工作，如果人们得不到报酬，他们就不会去做。比如说，它不一定有趣。如果你整天都在洗碗，就会觉得，呃。即使你真的很喜欢洗碗，你真的愿意每天洗8个小时、天天如此吗？大概不会。

然后还有危险的工作，基本上，如果一项工作很危险、无聊，或者可能造成重复性劳损之类的问题，那么人形机器人在初期真正能够创造最大价值的就是这些领域。所以我们的目标是让人形机器人去做人们不愿主动去做的工作。然后显然，未来我们必须将其与某种全民基本收入结合起来。

所以，我认为。

你能想象这样一个世界吗，那里有数亿个 Tesla Bot，在世界各地做着不同的、执行着不同的任务？

是的，我还真没想过那么遥远的未来，但我猜可能会出现类似的情况。

我能问一个大胆的问题吗？所以，Tesla 汽车的数量一直在加速增长，产量已经接近200万辆。其中很多都配有 Autopilot。

[埃隆] 我想我们现在已经超过200万了。

对。你认为会不会有一天，Tesla Bot 的数量会超过 Tesla 汽车？

会。其实，你问这个问题挺有意思的，因为通常我确实会尽量思考相当遥远的未来，但对于 Tesla Bot，或者它的代号 Optimus，我还真没想过那么遥远的未来。我叫它 Optimus Subprime，因为它并不像一个巨型变形金刚机器人。但它的定位是通用型辅助机器人。

而且基本上，那些东西是，基本上，我认为 Tesla 拥有与现实世界互动的最先进的现实世界 AI，这是我们为了实现自动驾驶而开发出来的。

所以，再加上定制硬件，以及大量硬核的底层软件，让它能够高效运行并提高能效，因为，如果你有一个装着10,000台计算机的巨型服务器机房，那么运行神经网络是一回事，但现在，假设你只是，你现在必须把它提炼到一台低功耗运行的计算机中，而这台计算机装在人形机器人或汽车里。

这其实非常困难，而且需要大量硬核的软件工作。所以，既然我们算是在为汽车解决利用神经网络在现实世界中导航的问题，而汽车有点像装着4个轮子的机器人，那么把它放进一个有手臂和腿的机器人里，就像是这件事的自然延伸。还有执行器。

两件困难的事是，你基本上需要让，让机器人足够智能，能够以合理的方式与环境互动。所以你需要真正的现实世界 AI，而且你还需要非常擅长制造，而这是一个非常困难的问题。

Tesla 非常擅长制造，而且也拥有现实世界 AI，所以要让人形机器人运作起来，基本上就意味着要开发与汽车所用部件不同的定制电机和传感器。我认为，我们在开发先进电动机和电力电子设备方面拥有最顶尖的专业能力。所以，只需要让它适用于人形机器人，而不是汽车。

不过，你有时确实会谈到爱。所以让我问一下，这并不是为了比如性爱机器人之类的————[埃隆] 爱就是答案。

是的。人形机器人，甚至是有腿的机器人，比如狗，做成狗的形状，对我们而言有某种吸引力，不是吸引力，而是我们会与它们产生联结。只是，这个世界上似乎存在着极其普遍的孤独。我们所有人都在寻求与其他人相伴、友谊以及所有这类东西。我们这里在奥斯汀有很多，很多人都养狗。

[埃隆] 没错。

似乎还有一个巨大的机会，可以让机器人减少世界上的孤独，或者帮助我们人类彼此建立联系。就像狗能做到的那样。你有没有从这个角度考虑过 Tesla Bot，还是说它确实专注于执行特定任务这个问题？而不是与人类建立联系？

我是说，老实说，我其实没有从陪伴这个角度考虑过，但我觉得它最终确实会成为，它实际上可能会成为一个非常好的伴侣。而且它可以随着时间推移形成独特的个性。并不是所有机器人都一模一样。而且这种个性可以逐渐演变，从而匹配主人，或者，我想是主人吧。随你想怎么称呼。伴侣，那个人类。

另一半，对吧？就像朋友那样。你看，我觉得这是一个巨大的机会。我觉得————是啊，不，这很有意思。因为日语里有个词，侘寂，细微的不完美正是让某样东西变得特别的原因。而机器人个性中的细微不完美，与机器人那位人类朋友身上的细微不完美相匹配，不知道，“主人”听起来也许不太合适，但基本上确实可能造就一个不可思议的伙伴。

[莱克斯] 而且通过这种方式，那些不完美之处————就像R2-D2或C-3PO之类的。

所以从机器学习的角度看，我觉得把缺陷当作一种特性真的很好。在一般的家庭环境中，或者整个广泛的世界里，你可以在相当长一段时间内都非常不擅长当机器人。而这还挺可爱的，那些就是你的缺陷，而你会爱上这些缺陷。这和自动驾驶非常不同，自动驾驶的环境风险极高，你不能出错。所以，是啊，在家里当机器人更有趣。

是啊，事实上，如果你想想 C-3PO 和 R2-D2，他们其实有很多，比如缺陷、不完美和傻乎乎的地方，而且他们还会互相争论。

他们真的有什么事情做得好吗？我不太确定。

他们肯定为故事增色不少。但他们身上有些古怪的元素，而且，他们会犯错、会做一些事情，这就只是，这让人觉得他们很亲切，我不知道。惹人喜爱。所以，是啊，我觉得那可能会成为某种，它大概会发生的事情。但我们最初的重点只是让它实用。我有信心我们会把它做出来，我不确定确切的时间安排，但我们大概会在，我不知道，明年年底前后做出一个还不错的原型之类的。

而且很酷的是，它与 Tesla，那辆车，相连。

是啊，它正在使用很多，它会使用 Autopilot 推理计算机，而且我们为4辆车做的大量训练，就识别现实世界中的事物而言，可以直接应用到机器人上。但还有很多定制执行器和传感器需要开发。

以及在向量空间之上再加一个用于爱的额外模块。

啊，是啊。

那个还缺着。好吧。

我们也可以把那个加到车里。

确实。是啊，它在所有环境中都可能有用。就像你说的，很多人会在车里争吵，所以也许我们能帮帮他们。你是历史的研习者，也是《丹·卡林的硬核历史》播客的粉丝。

[埃隆] 是啊。那个很棒。

有史以来最棒的播客。

是啊，我觉得它确实是。

它几乎都不能真正算作播客。

[埃隆] 它更像是一本有声书。

对。所以你和丹一起上了播客，我刚和他聊过这件事。他说你们谈了军事以及诸如此类的东西。

对，基本上，它应该叫《工程师战争》。本质上，当技术发展的速度迅速变化时，工程就在战斗的胜利中发挥关键作用。

你们在历史上追溯到了多早？谈到第二次世界大战了吗？

嗯，本来应该是深入探讨第二次世界大战中的战斗机和轰炸机技术，但最后涉及的范围比那更广。因为我就是钻进了一个彻底的牛角尖，研究第二次世界大战中的所有战斗机和轰炸机，以及那种持续不断的石头剪刀布游戏：一个国家会制造这种飞机，另一个国家就会制造一种飞机来击败它，然后他们又会设法制造一种飞机来击败那种飞机，接着他们会……

而真正重要的是创新的速度，还有能否获得高质量的燃料和原材料。所以，比如德国有一些很惊人的设计，但他们造不出来，因为得不到原材料。他们在石油和燃料方面确实有问题，基本上，燃料质量极不稳定。

所以设计并不是瓶颈，而是——对，美国有非常厉害的燃料，而且非常稳定。问题在于，如果你制造一台性能非常高的飞机发动机，为了让它拥有高性能，你必须……燃料，也就是航空汽油，必须是成分一致的混合物。而且它的辛烷值必须很高。高辛烷值是最重要的，但也不能有杂质之类的东西，因为那会让发动机积垢。而德国人始终无法顺利获得石油。他们试图通过入侵党团会议来获得石油，但结果不太顺利。

那从来都不会顺利。

他们没能如愿。（女子轻声说话）很高兴认识你。德国一直在为基本上很糟糕的石油而苦恼，因此他们无法，他们无法指望为自己的飞机获得高质量燃料。所以他们不得不使用所有这些添加剂之类的东西。而美国拥有极好的燃料，他们也把这种燃料提供给了英国。

所以，这让英国人和美国人能够设计出性能极高的飞机发动机，胜过世界上的任何其他发动机。德国能够设计这些发动机，他们只是没有燃料。然后还有诸如，他们获得的铝合金质量也没有那么好，所以，对。

[莱克斯] 你和丹谈了所有这些吗？

对。

太棒了。从宏观上审视历史，当你看到成吉思汗，看到斯大林、希特勒，看到人类历史上最黑暗的时刻时，你从这些时刻中得出了什么？这是否有助于你深入理解人性，理解当今的人类行为？无论是战争、个人，还是人们的行为，历史的任何方面。

对。我觉得历史很迷人。人们做过很多令人难以置信的事情，有好的，也有坏的，它们就是能帮助你理解文明的本质、个人的本质，以及……

人类彼此做出这类事情会让你感到悲伤吗？看看20世纪、第二次世界大战，以及滥用权力的残酷。谈谈共产主义、马克思主义和斯大林。

我的意思是，其中一些事情确实会，我是说，如果你，人类历史很漫长，但其中大部分其实只是人们继续过自己的生活，并不是说人类历史就是永不停歇的战争和灾难，那些其实只是，那些是间歇性的，而且很少发生，如果不是这样，人类很快就会不复存在。但就是，战争往往被大量书写。

而像是，嗯，平常的一年，没发生什么大事，就不会被写太多，但那才是，大多数人只是在耕作，过着自己的生活。在某个地方做一个村民。偶尔会有一场战争。我不得不说，我读过的书里，很少有那种让我不得不停下来不再读的，因为内容实在太黑暗了。

但那本关于《斯大林：红星宫廷》的书，我当时停止阅读了，实在太黑暗。很残酷。

对。30年代。对我来说，那里有很多教训，尤其是感觉人类，我们所有人都有那种狂热，索尔仁尼琴的那句话，善与恶的分界线贯穿每个人的内心，我们所有人都有作恶的能力，所有人都有行善的能力，这几乎就像我们所有人都有的一种责任，要趋向善。

所以，对我来说，审视历史几乎就像一个例子：你看，有某个富有魅力的领导者让你相信某些事情，根据那个故事，太容易对彼此、对你的家人、对其他人作恶了。所以，这就像是我们行善的责任。并不是说现在不知怎么就与历史不同了，那可能再次发生，所有这一切都可能再次发生。而且是的，大多数时候你是对的。

我的意思是，这里乐观的看法是，大多数人只是在生活。而且正如你经常用梗表达的那样，过去的生活质量要糟糕得多，而它随着时间推移，通过创新、通过技术不断改善，但这些暴行的“飞艇”出现，依然不知怎么地引人注目。

当然。对，我的意思是，在历史上的大部分时期，生活确实非常艰难。我是说，或许在人类历史的大部分时间里，好年景就是你村里没有那么多人死于瘟疫、饥饿、冻死，或被邻村的人杀死的一年。就像，“嗯，没那么糟。”只不过是，“你知道，我们今年损失了5%。这是个好年景。”

对。

那会是很平常的事。纵观历史，对大多数人来说，首要目标就是别饿死。只要确保我们有足够的食物撑过冬天，不会遭遇、冻死或者别的什么。现在食物很充足。我们有肥胖问题。

嗯，是啊，这里的教训是，我们中的一些人应该对如今的生活心怀感激。我们私下谈过这件事。我很想在这里听听你的想法。如果我与俄罗斯总统弗拉基米尔·普京面对面坐下来进行一次长篇对话，你是否可能愿意打电话进来，加入几分钟与他的对话，由我主持和翻译？

当然。是的。当然，我很乐意那样做。

你表现出过对俄语的兴趣。这是源于你对历史、语言学、文化的兴趣，还是一般性的好奇心？

[埃隆] 我觉得听起来很酷。

是听起来很酷，不是看起来很酷。读西里尔字母需要一点时间。等你知道西里尔字符分别代表什么之后，阅读俄语实际上就会容易很多，因为有很多词其实是一样的。比如 bank 就是 bank。

所以找出那些完全相同的词，现在你就开始理解西里尔字母了，是啊。

如果你能把它拼读出来，那么就会更，至少有些词是相通的。

文化方面呢？你热爱伟大的工程和物理。那里有科学传统。回顾20世纪，从火箭技术来看。所以，一些最伟大的火箭，一些太空探索，是在苏联、在前苏联完成的。

是啊。

那么，你会从那段历史中汲取灵感吗？就是这种文化如何，从许多方面来说，我是说，一件令人遗憾的事是，由于语言，其中很多东西都消失在历史中了，因为它们没有被翻译，所有那些种类的，因为它在某些方面是一种孤立的文化，它在自己的边界内蓬勃发展。

[埃隆] 是啊。

那么，你会从那些人、从那里科学和工程的历史中汲取灵感吗？

会。我是说，苏联、俄罗斯以及乌克兰在航天方面都有非常深厚的历史，比如历史上一些最先进、最令人赞叹的成就是由苏联完成的。人们不可能不钦佩当时研发出的令人赞叹的火箭技术。在苏联算是解体之后，发生的事情少了很多，仍然有事情在发生，但其狂热的速度已经远不及苏联算是解体为各个共和国之前。

是啊。我是说，还有俄罗斯的俄罗斯国家航天集团，那个机构。我期待有一天，这些国家与中国、美国携手合作，它们全都一起合作，也许有一点友好竞争，不过。

我觉得友好竞争是好事。政府行动迟缓，而唯一比一个政府更慢的，就是一群政府。（莱克斯笑）——是啊。

如果所有人都同时冲过终点线，奥运会就会很无聊。

是啊。

没人会看。

[莱克斯] 是啊。

而且人们也不会努力跑快之类的。所以，我认为友好竞争是件好事。

这里也很适合推荐一下蒂姆·多德，也就是“日常宇航员”的一个视频，标题是《完整的苏联火箭发动机家族树》。大概有1个半小时。它完整讲述了苏联火箭的历史。大家绝对应该去看看，而且总体上也应该支持蒂姆，那家伙对未来超级兴奋，对航天超级兴奋，每次看到他的任何东西，我脸上都会挂着傻笑，因为他对各种东西都太兴奋了。

是啊，蒂姆·多德是——我喜欢那样的人。

如果你对任何与太空有关的东西感兴趣，他真的很棒。就向普通人解释火箭技术而言，他非常出色。可以说是最棒的。我应该说，我之所以把我们的方案换掉，全部原因在于，猛禽发动机一度原本会是一款氢发动机，但氢面临很多挑战。它的密度非常低。它是一种深冷介质，所以只有在非常接近绝对零度时才是液体。需要大量隔热材料。

所以那里有很多挑战。而且我当时确实读了一些关于俄罗斯火箭发动机研发的资料。至少我的印象是，主要是苏联、俄罗斯和乌克兰实际上正在转向甲烷液氧。而且有一些关于比冲的有趣测试和数据，他们使用甲烷液氧发动机能达到大约382秒的比冲。我当时想：“哇，好吧，那，那其实真的令人印象深刻。”

所以我认为我们可以，你实际上可以获得低得多的成本，优化每吨入轨成本、每去往火星的成本。我认为甲烷方案才是正确方向。我有一部分灵感来自俄罗斯在测试终端上对甲烷液氧发动机所做的工作。

现在来说点完全不同的。你介意本着伟大而强大的 PewDiePie 的精神，做一点表情包点评吗？假设从1到11，——好。

就看看几份打印出来的文件。

[埃隆] 我们可以试试。

[莱克斯] 我们来试试这个。我向你呈上第一号文件。（埃隆笑）——好。

[莱克斯] 穿刺公弗拉德发现了棉花糖。

是啊，这个还不错。

你懂吧，因为他喜欢把东西刺穿。

是的，我懂。是的，我懂，我不知道，3分，随便吧。

[莱克斯] 哦，这不太好。这个是以一些工程、一些历史为基础的。（埃隆笑）——是啊，我给这个打8分，满分10分。

[莱克斯] 你怎么看核能？

我支持核电。在不易遭受极端自然灾害的地方。我认为这是一种，新的核电是很好的发电方式。我认为我们不应该关闭核电站。

[莱克斯] 是啊，但切尔诺贝利呢？

正是。我觉得人们，就是对辐射之类的东西有很多恐惧。我想，问题在于很多人就是不懂，他们没学过工程学或物理学，所以他们不懂，光是“辐射”这个词听起来就很吓人，你知道吗？所以他们不懂，他们无法衡量辐射意味着什么。但辐射远没有你想象的那么危险。例如福岛，当福岛的问题因为那场海啸发生时。

有加利福尼亚州的人问我，他们是否应该担心来自福岛的辐射。我就说，绝对不用，一丁点都不用，完全不用。那太疯狂了。为了表明这一点，就是危险被渲染得比实际情况严重太多了，我真的飞去了福岛。而且，实际上，我为一家水处理厂捐赠了一套太阳能发电系统。我还特意在福岛的电视上吃了当地种植的蔬菜。

我还活着。好吧。

所以甚至不是这些事件的风险很低，而是它们的影响—— ——影响被大大夸大了。

这是人性。

人们不知道辐射是什么，有人问过我：“手机的辐射导致脑癌怎么办？”我就说：“你说辐射时，指的是光子还是粒子？”他们就说，不知道，“你说光子、粒子是什么意思？”“假设你指的是光子。什么频率或波长？”然后他们就说：“不知道，我完全不清楚。”“你知道所有东西一直都在辐射吗？”他们就说：“你是什么意思？”

“就是，所有东西一直都在辐射。”基本上，所有物体一直都在发射光子。如果你想知道站在核火焰前意味着什么，就到外面去。太阳就是一个巨大的热核反应堆，而你正盯着它看。你还活着吗？是的。好吧。太神奇了。

是啊，我想“辐射”是可能被某些人用作散布恐惧的工具的词语之一。就是这样。

我觉得人们就是不明白。

我是说，我想这就是对抗那种恐惧的方式，就是去理解，去学习。

是啊，就说，好吧，究竟有多少人死于核事故？实际上几乎没有，然后，再说有多少人死于燃煤电厂？那是一个非常大的数字。显然，我们不应该启动燃煤电厂并关闭核电站，这完全没有任何道理。燃煤电厂，我不知道，对健康的危害是核电站的100到1000倍。

你想看下一个吗？这个真的很糟。那个90、180和360度，所有人都喜欢这种数学。根本没人把270当回事。

不是特别好笑。我不知道，大概2分或3分。

[莱克斯] 是啊。这并不是让人笑出声的情况。（两人笑）——[莱克斯] 是啊。（埃隆笑）——那个挺不错。

[莱克斯] 美国在建立和摧毁独裁政权之间来回摇摆。它就像一个地铁，是叫地铁——对，节拍器。对，这个嘛，我不知道，10 分里能得 7 分。算是有点道理。

这个对我来说有点私人。下一个。

哦，天哪，这是莱卡吗。

[莱克斯] 对，嗯，不，这是——或者它是在指莱卡之类的。

[莱克斯] 这是莱卡的丈夫。

丈夫，对。

[莱克斯] 喂？是的，我是狗。你妻子被发射到太空去了。然后最后一张是他闭着眼睛，身边放着一瓶伏特加。

对，莱卡没有回来。

[莱克斯] 没有。他们不会告诉你完整的故事，不会告诉你这对它的亲人造成了什么影响。

确实。

那个我给 11 分。它还在继续，延续俄罗斯主题。第一个进入太空的人，没人在乎。第一个登上月球的人。

嗯，我觉得人们确实在乎。

[莱克斯] 我知道，但是。

尤里·加加林的名字将永远载入史册。我觉得。

置身于、踏上一片完全陌生的土地，确实有某种特别之处。重要的不是旅程，就像那些探索海洋的人。探索海洋不如登陆一块全新的大陆那么重要。

[埃隆] 对。

[莱克斯] 哦，这个是关于你的。（埃隆笑）哦，对。我很想听听你对此有何评论。埃隆·马斯克向联合国汇去 66 亿美元以终结世界饥饿之后。“你有 3 个小时。”

对，嗯，我是说，显然 60 亿美元不可能终结世界饥饿。我是说，现实是，目前全世界生产的食物远远超过实际能够消费的数量。到目前为止，我们并不存在热量限制。所以，凡是存在饥饿的地方，几乎总是由于内战或冲突，或者某些类似的原因；仅仅因为缺钱而发生饥饿，是极其罕见的。某个国家发生了内战，这个国家的一方真的试图让另一方饿死。

所以，这远比钱能够解决的问题复杂得多。它涉及地缘政治，涉及很多事情，涉及人性、政府、各种钱、货币体系，诸如此类的东西。

对。如今食物极其便宜。我是说，目前在美国的低收入家庭中，肥胖实际上才是现在的问题。问题不是，显然不是饥饿，而是太多、摄入了太多热量。并不是说任何地方都没有人在挨饿，只是，这并不是简单地增加资金就能解决的问题。

[莱克斯] 你觉得那个能得多少分？正在得多少分？

2 分。

[莱克斯] 就是继续拿帝国开涮。世界：“你们从哪里得到那些文物的？”大英博物馆。这是在致敬《巨蟒剧团》。“我们发现了它们。”

对。大英博物馆，它相当不错。我是说，无可否认，英国确实从世界各地拿走了这些历史文物，并把它们放到了伦敦，但人们并不是不能去看它们。所以，对世界上很大一部分人而言，伦敦确实是一个方便观赏这些古代文物的地方。所以我认为，不平衡地说，大英博物馆带来的净影响是好的。嗯，我相信很多国家对此都有争议。

[莱克斯] 对。

这就像是，你希望让尽可能多的人能够接触到这些历史文物。而我认为大英博物馆在这方面做得很好。

即使总体上的帝国历史有更黑暗的一面，无论是哪个帝国，无论当时是如何行事的。那都是已经发生过的历史。不幸的是，你不可能就这样抹去那段历史。你只能在未来变得更好。这就是重点。

是啊，我是说，嗯，我们要怎么对这些事情作出道德评判呢？如果一个人要评判，比如说俄罗斯帝国，你就得评判当时每个人都在做什么，以及英国人相对于其他所有人怎么样？我认为英国人实际上会得到一个相对不错的评分，相对不错的评分，不是从绝对意义上说，而是与其他所有人当时所做的事情相比，他们不是最糟糕的。就像我说的，你必须在当时的历史背景下看待这些事情，然后问：“当时还有哪些选择，你拿它和什么作比较？”

是的。

而且我不认为，在审视当时的历史时，英国会得到一个差评。现在，如果你用今天在道德上可以接受的标准来评判历史，那你基本上会给所有人不及格。我不清楚。我认为任何人的道德都拿不到及格分，你可以回到300年前，谁能拿到及格分？基本上没人。

[莱克斯] 而且我们可能无法从后世那里得到及格分数——对。没错。

[莱克斯] 那些在我们之后出现的。那个得几分？

当然。6分，7分。

也许是因为《巨蟒剧团》。

[埃隆] 我总是《巨蟒剧团》，他们很棒。《布莱恩的一生》和《寻找圣杯》都不可思议。

是啊。是啊。

天啊，那眉毛真够夸张的。

[莱克斯] 勃列日涅夫。你觉得，有多重要——天啊。

[莱克斯] 面部毛发之于伟大领导力？你换了个新发型。这对你的领导力有什么影响？

[埃隆] 我不知道。希望没有。没有影响。

[希冯] 第二名是，没人吗？

对，第二名是没人。

[埃隆] 没人能和勃列日涅夫竞争。

没人第二。

那简直是史诗级的眉毛。当然。

[莱克斯] 太荒唐了。

给它6分或7分吧，我不知道。

[莱克斯] 我喜欢这个，对梗进行莎士比亚式分析。

勃列日涅夫，他也很有戏剧天赋。德国笑话。

[莱克斯] 对，对。肯定是眉毛带来的。好了。发明，伟大的工程。看看我发明了什么。那是自撕碎面包以来最棒的东西。

对。

因为他们发明了切片面包。我现在只是在解释梗吗？（所有人笑）我的人生已经变成这样了。

[希冯] 他是梗王，你是梗解说员。

[莱克斯] 我是一个梗，就像一个书记官，跟着国王们到处跑，只负责把梗记下来。

我是说，芝士汉堡是什么时候发明的？那是个史诗级的发明。

[莱克斯] 对。

就像，哇。

[莱克斯] 相比普通汉堡？

或者汉堡，我想就是泛指汉堡。

然后就是，汉堡是什么？三明治是什么？接着你开始问，比萨算三明治吗？最初的又是什么？这就进入本体论争论了。

对，但大家都知道，如果你点一个汉堡、芝士汉堡或者随便什么，然后拿到番茄、一些生菜、洋葱之类的，还有蛋黄酱、番茄酱和芥末酱，那简直是史诗级的。

对，但我敢肯定，人们很久以来都是把面包和肉分开吃的。它们放在同一个盘子里，也算是某种汉堡，但有人真正把它们组合成同一个东西，然后咬着吃、拿在手里，这就方便了。这是一个材料问题。比如你的手不会弄脏之类的。对，这太聪——（希冯轻声说话）那不是我会猜的答案。

但每个人都知道，如果你点一个芝士汉堡，你知道自己会拿到什么，不会是什么晦涩难懂的东西，呃，我不知道自己会拿到什么。薯条，我是说，很棒。我是说，它们是魔鬼，但薯条太棒了。对，比萨不可思议。

食品创新得到的喜爱不够多。

对。

我想这就是我们想表达的。

[埃隆] 很棒。

那这位奥斯汀人马修·麦康纳呢？肯尼迪总统：“你们知道怎么把人送上月球了吗？”NASA：“不知道。”肯尼迪总统：“你们要是知道，那就酷多了。”

差不多，当然。6分，6分或7分，我想。

[莱克斯] 这是最后一个。

这很好笑。

[莱克斯] 有人在墙上到处画了一堆阴茎。西斯廷教堂，男厕所。

当然，我会给它9分。这真的是真的。

这是我们今天排名最高的迷因。

[埃隆] 我的意思是，确实如此，他们是怎么蒙混过关的？

好多裸体。

我是说，阴茎照这种东西，我是说，贯穿了整个历史。只要人们能画东西，就一直有阴茎照。

这是人类历史中的常见元素。

这是个常见元素。贯穿整个人类历史，始终如此。

你发推文说你渴望喜剧，你和乔·罗根是朋友。你以后会不会在某个时候来一小段单口喜剧？也许给乔暖场？诸如此类？那是——真要？单口喜剧？真的直接来一整套单口喜剧？

[莱克斯] 来一整套单口喜剧。这在你的计划里吗，还是说？

我从没想过这件事。

这极其困难，至少乔是这么说的，喜剧演员们也是这么说的。

[埃隆] 嗯？不知道我能不能做到。

只有一种办法能知道。

我确实给朋友们表演过单口喜剧，就是即兴的，我会爬到比如屋顶上，他们确实会笑，但他们也全都是朋友。所以，如果房间里都是陌生人，我不知道。他们真的也会觉得好笑吗，不过我可以试试。看看会发生什么。

我觉得无论结果如何，你都会学到些东西。

是的。

我有点既喜欢在……的时候，也喜欢在你表现很棒的时候，就是观察人们，看他们怎么应对。太难了。你在台上非常脆弱。只有你自己。而且你以为自己会很搞笑，可当表演彻底冷场时，那种，看着人们应对这种情况，真的很美。

我觉得我的素材可能足够表演单口喜剧了。我从没想过这件事，但我的素材可能够了。我不知道，比如15分钟之类的。

哦，是啊。是啊。做一部 Netflix 特辑。（埃隆笑）——[埃隆] Netflix 特辑，当然。

你最喜欢《瑞克和莫蒂》里的哪个概念？突然把这个问题抛给你，有没有，那里探索了很多某种科学和工程理念。有那个，——最喜欢的《瑞克和莫蒂》

有那个黄油机器人。

是啊，这是部很棒的剧。

你喜欢吗？

是啊，《瑞克和莫蒂》棒极了。

一个来自平行维度、和你一模一样的人出现在了那里。埃隆·塔斯克。

是啊。没错。

是你配的音。

是啊，《瑞克和莫蒂》确实探索了很多有意思的概念。当然，比如最喜欢的是哪个。黄油机器人当然是，它当然有可能，让一个设备拥有过多的自我意识。你不会想让自己的烤面包机成为一台超级天才烤面包机。它会痛恨生活，因为它能做的只有烤面包。就像，你不会想把超级智能困在一个功能非常有限的设备里。

你觉得这是不是太容易了，从一个，如果我们是从工程角度谈论超级智能，比如机器人马文。似乎要设计出一个抑郁的机器人可能非常容易。

当然。

要设计出一个能找到充实存在方式的机器人，并不是显而易见的。我想，人类也是一样。我在想这是否就像默认状态，如果你没把机器人造好，它就会经常伤心。

嗯，我们重新编程机器人，比重新编程人类容易。我想，如果你任由它演化而不去干预，那它可能会变得伤心，但你可以改变优化函数，让它成为一个快乐的机器人。

就像我谈到 SpaceX 时所说的，你给了很多人希望，也有很多人仰慕你。数百万人仰慕你。如果我们想想高中里的年轻人，也许是大学里的年轻人，假如他们想努力在这个世界上做一件大事，想真正产生巨大的积极影响，你会给他们什么建议？关于他们的职业生涯，也许也关于整个人生，你会给他们什么建议？

努力做一个有用的人。做对你的人类同胞、对世界有用的事。要做到有用非常困难。非常困难。你的贡献是否超过了你的消耗？努力为社会作出正向的净贡献。我认为这才是应该追求的目标。不要只是为了当领导者而试图成为某种领导者，或者诸如此类。

很多时候，那些——很多时候，你想让他们当领导者的人，正是那些不想当领导者的人。如果你过着有用的人生，那就是美好的人生，是值得活过的人生。就像我说的，我会鼓励人们运用物理学的思维工具，并将它们广泛应用于生活。它们是最好的工具。

当你思考教育和自我教育时，你有什么建议？所以有大学，有自学。有亲自动手，也就是找到一家公司、一个地方或一群在做你所热爱之事的人，并尽早加入他们。还有花几年时间在欧洲公路旅行，并写些诗。你建议走哪条轨迹？就学习如何能让自己变得有用而言，正如你提到的，如何能产生最大的积极影响。

我鼓励人们大量读书，就是读，基本上尽量吸收尽可能多的信息，也努力培养良好的通识。所以你至少能对知识版图的整体面貌有一个粗略了解，试着对很多事物都了解一点。因为你可能不知道自己真正对什么感兴趣。如果你至少都没有去做，你怎么会知道自己真正对什么感兴趣？

广泛地对知识版图进行外围探索。和来自不同生活背景、不同行业，以及从事不同专业、拥有不同技能和职业的人交谈，就是去尝试。尽可能多地学习。不断寻找意义。

这一切不就是对意义的探寻吗？

是啊，生命的意义是什么之类的？但总的来说，就像我说的，我会鼓励人们广泛阅读许多不同学科领域的内容，然后努力找到你的天赋与你感兴趣的事情相重合的东西。所以，人们可能擅长某件事，或者在某个特定方面有技能，但他们不喜欢做。所以你要努力找到一件事，它能很好地结合你天生擅长的东西，以及你也喜欢做的东西。

而阅读是一条超级快捷的路径，可以弄清楚是哪件事、你的定位在哪里：你既擅长它，又喜欢做它，而且它确实会产生积极影响。

嗯，你总得通过某种方式了解事物。所以阅读，广泛地读，就是真的去读。更重要的是，我小时候把百科全书从头到尾读了一遍。所以，那非常有帮助。而且，里面有各种各样我甚至不知道其存在的东西，嗯，显然有很多。

这已经宽泛到极致了。

我觉得，大概40年前，百科全书是可推荐的。也许可以通读像《大英百科全书》的精简版，我会推荐那个。你随时可以跳过一些主题，所以你读了几段，知道自己不感兴趣，就直接跳到下一个。所以，读百科全书，或者快速浏览一遍。

我非常看重、当然也非常尊重那些踏踏实实工作一天、做有用之事的人。而且总体上要抱有一种，不是零和的思维，或者更多地抱有一种把蛋糕做大的思维。当我看到一些人，比如说，也包括一些非常聪明的人，采取一种态度，我喜欢做那些看起来在道德上有问题的事情。

这往往是因为，他们在一种基础的、公理式的层面上抱有零和思维。而他们没有意识到这一点，他们没有意识到自己抱有零和思维，或者至少没有有意识地意识到。所以，如果你抱有零和思维，那么取得领先的唯一方式就是从别人那里拿走东西。如果蛋糕是固定的，那么拥有更多蛋糕的唯一方式就是拿走别人的蛋糕。但这是错的。

显然，随着时间推移，蛋糕已经显著变大了，也就是经济蛋糕。实际上，你可以拥有，（埃隆笑）这个比喻用得太多了，我们可以拥有很多，有很多蛋糕。（莱克斯笑）我的蛋糕不是固定的。所以，你确实要确保自己不是在没有意识到的情况下，以零和思维行事。

在这种思维下，取得领先的唯一方式就是从别人那里拿走东西，那么这会导致你试图从别人那里拿走东西，这并不好。致力于把经济蛋糕做大要好得多。就像我说的，创造得比你消耗的更多。做得比你，没错。所以这是件大事。我认为金融业里有相当一部分人确实有一点零和思维。

我的意思是，各行各业都是如此。我见过这种情况。罗根激励我的原因之一，是他经常赞美别人，而不是制造持续不断的竞争，仿佛资源是稀缺的。而当你赞美别人、推广别人和别人的想法时，实际上会把蛋糕做大。资源就不再那么稀缺了。这适用于许多领域。

这也适用于学术界，那里很多人非常，把学术研究的一些资金视为零和博弈。并非如此，如果你们相互赞美，如果你让，如果你让每个人都对 AI、物理学、数学感到兴奋，我认为资金会越来越多，而且我认为每个人都会获益。是的。我认为这广泛适用。

对，对。没错。

那么最后一个问题是关于爱和意义。广义而言，爱在人类境况中扮演什么角色，而更具体地说，对你而言呢？爱，无论是浪漫爱情还是其他形式，如何让你成为一个更好的人、一个更好的人类个体？更好的工程师？

现在你问的真是令人困惑的问题。很难给出一个。我的意思是，有许多书籍、诗歌和歌曲都在探讨什么是爱，以及什么是，究竟，什么是爱，宝贝别伤害我。（莱克斯笑）——那确实是其中一首经典，是的。你之前引用过莎士比亚，但这首真的也名列前茅。

[埃隆] 是的。爱是一件有许多光彩的事。

我的意思是，因为我们谈了这么多鼓舞人心的事情，比如做对世界有用的人，算是解决问题、减轻痛苦，但人与人之间的联结似乎是一种源泉，它是快乐的源泉，是意义的源泉，而那就是爱、友谊、爱。我只是想知道，当你谈论保存人类意识之光时，你是否会思考这类事情。

对。

以及我们成为一个多行星物种。我的意思是，至少对我来说，这意味着，如果我们只是孤独地存在、有意识并且有智慧，那远不如我们与其他人在一起时那么有意义。对吧？而且当我们在一起时，会产生某种魔力。其中的友谊，而我认为它的最高形式就是爱，我认为广义上的爱远不只是某种浪漫爱情，不过也包括，是的。浪漫爱情、家庭以及这类事物。

嗯，我的意思是，我想，我之所以关心我们成为一个多行星物种和一个太空承载文明，从根本上说，是因为我爱人类。所以我希望看到人类繁荣、成就伟业并获得幸福，而如果我不爱人类，我就不会关心这些事情。

所以，当你审视整个，人类历史，所有曾经活过的人、现在活着的所有人，挺不错的，我们还行。总体而言，我们是一群相当有意思的人。

是的。综合考虑一切，而且我读过很多历史，包括其中最黑暗、最糟糕的部分。尽管如此，我认为权衡下来，我仍然爱人类。

你拿它开过玩笑，那个 42，你认为这一切的意义是什么？有没有非数字形式的表述？

哦，我应该说，是的，嗯，实际上，我认为道格·桑德斯在《银河系漫游指南》中所说的是，宇宙就是答案。我们真正需要弄清楚的是，面对宇宙这个答案，我们该提出什么问题。而问题才是真正困难的部分。如果你能恰当地构建问题，那么相对而言，答案就很简单。

所以，因此，如果你想理解该对大学提出什么问题，你想理解生命的意义，我们就需要拓展意识的范围和规模，以便我们能更好地理解宇宙的本质，并理解生命的意义。

而最终，最重要的部分将是提出正确的问题。

[埃隆] 是的。

从而提升了采访者的角色——[埃隆] 是的，没错。

成为房间里最重要的人。

好问题是，很难想出好问题。绝对如此。不过，是的，这就是我的哲学基础：我对宇宙的本质感到好奇。显然我会死。我不知道自己什么时候会死，但我不会永远活着。不过，我想知道我们正走在理解宇宙本质、生命意义，以及该对宇宙这个答案提出什么问题的道路上。

所以，如果我们扩大人类以及总体上意识的范围和规模，其中也包括硅基意识，那么这似乎从根本上来说是一件好事。

埃隆，就像我说的，我非常感激你今天愿意把极其宝贵的时间花在我身上，也感激你在这个艰难、分裂且犬儒的时期给了数百万人希望。所以我希望你继续做你正在做的事。非常感谢你今天与我交谈。

哦，不客气。谢谢你提出这些精彩的问题。

感谢收听这场与埃隆·马斯克的对话。若想支持本播客，请查看简介中的赞助商。现在，让我用埃隆·马斯克本人的一段话作为结尾。“当一件事足够重要时，你就会去做，即使胜算对你不利。”感谢收听，希望下次再见。

The following is a conversation with Elon Musk, his third time on this, the "Lex Fridman Podcast." Yeah, make yourself comfortable.

Boo.

Oh, wow, okay.

You don't do the headphone thing?

No.

Okay. I mean, how close do I need to get this thing?

The closer you are the sexier you sound.

Hey babe, sup.

Yup.

Can't get enough of you on that baby? (both laughing) - I'm gonna clip that out and any time somebody messages me on my phone I'll just respond with that.

If you want my body and you think I'm sexy come right out and tell me so. Do do do do do.

[Shivon] So funny.

So good. Okay, serious mode activate, alright.

Serious mode. Come on, your Russian, you can be serious.

Yeah I know.

Everyone's serious all the time in Russia.

Yeah, yeah. We'll get there. We'll get there. (Shivon speaking faintly) Just gotten soft. Allow me to say that the SpaceX launch of human beings to orbit on May 30th, 2020, was seen by many as the first step in a new era of human space exploration. These human space flight missions were a beacon of hope to me and to millions over the past two years as our world has been going through one of the most difficult periods in recent human history.

We see the rise of division, fear, cynicism, and the loss of common humanity, right when it is needed most. So, first, Elon, let me say thank you for giving the world hope and reason to be excited about the future.

Oh, it's kind of you to say that. I do want to do that. Humanity has, obviously a lot of issues, and people at times do bad things, but despite all that, I love humanity and I think we should make sure we do everything we can to have a good future and an exciting future, and one where that maximizes the happiness of the people.

Let me ask about a Crew Dragon Demo-2. So that first flight with humans onboard, how did you feel leading up to that launch? Were you scared? Were you excited? What was goin' through your mind? So much was at stake.

Yeah, no, that was extremely stressful. The question we obviously could not let them down in any way. So, extremely stressful I'd say, to say the least.

I was confident that, at the time that we launched, that no one could think of anything, at all, to do that would improve the probability of success and we racked our brains to think of any possible way to improve the probability of success, and we could not think of anything more, nor could NASA, and so, that's just the best that we could do. So then we went ahead and launched.

Now, I'm not a religious person, but I nonetheless got on my knees and prayed for that mission.

[Lex] Were you able to sleep?

No.

How did it feel when it was a success? First when the launch was a success, and when they returned back home, or back to earth.

It was a great relief. Yeah. For high stress situations I find it's not so much elation, as relief. And, I think once as we got more comfortable and proved out the systems, 'cause we really, you're gotta make sure everything works. It was definitely a lot more enjoyable with the subsequent asteroid missions. And I thought the Inspiration mission was actually very inspiring, the Inspiration4 mission.

I'd encourage people to watch the Inspiration documentary on Netflix, it's actually really good. And it really isn't, I was actually inspired by that, so that one I felt, I was kind of able to enjoy the actual mission and not just be super stressed all the time.

So, for people that somehow don't know, it's the all civilian, first time all civilian out to space out to orbit.

Yeah, it was the, I think the highest obit that in like, I don't know, 30 or 40 years or something, the only one that was higher was the one shuttle, sorry, a Hubble servicing mission. And then before that it would've been Apollo in '72. It was pretty wild. So it's cool. It's good. I think as a species, we want to be continuing to do better and reach higher ground.

I think it would be tragic, extremely tragic, if Apollo was the high watermark for humanity, and that that's as far as we ever got. And it's concerning that here we are 49 years after the last mission to the moon. And, so almost half a century, and we've not been back. And that's worrying, it's like, does that mean we've peaked as a civilization or what? I think we gotta get back to the moon and build a base there. A science base.

I think we could learn a lot about the nature of the universe if we have a proper science base on the moon. We have a science base in Antarctica and many other parts of the world. So that's what I think the next big thing we've gotta have like a serious black moon base, and then get people to Mars and get out there and be a space bearing civilization.

I'll ask you about some of those details. But, since you're so busy with the hard engineering challenges of everything that's involved, are you still able to marvel at the magic of it all, of space travel, of every time the rocket goes up, especially when it's a crude mission? Or are you just so overwhelmed with all the challenges that you have to solve?

And actually, sort of to add to that, the reason I wanted to ask this question of May 30th, it's been some time, so you can look back and think about the impact already. At the time it was an engineering problem maybe, now it's becoming a historic moment. Like it's a moment that, how many moments will be remembered about the 21st century?

To me, that or something like that, maybe Inspiration4 or one of those will be remembered as the early steps of a new age of space exploration.

Yeah, I mean, during the launches itself, so I mean, I think maybe some people will know, but a lot of people don't know, is I'm actually the chief engineer of SpaceX, so I've signed off on pretty much all the design decisions. So if there's something that goes wrong with that vehicle, it's fundamentally my fault, you know?

So I'm really just thinking about all the things that like, so when I see the rocket, I see all the things that could go wrong, and the things that could be better, and the same with the Dragon spacecraft. Other people will say, "Oh, this is a spacecraft or a rocket." and "This looks really cool." I'm like, I've like a readout of these are the risks, these are the problems. That's what I see.

Like (Elon chuffing) So it's not what other people see when they see the product.

So let me ask you then to analyze Starship in that same way. I know you have, you'll talk a bit in more detail about Starship in the near future. Perhaps you had that- - We can talk about in now if you want.

But, just in that same way, like you said, you see, when you see a rocket, you see the sort of a list of risks. In that same way, you said that Starship was a really hard problem. So, there's many ways I can ask this, but if you magically could solve one problem perfectly, one engineering problem perfectly, which one would it be?

[Elon] On Starship?

On, sorry, on Starship. So is it maybe related to the efficiency, the engine, the weight of the different components, the complexity of various things, maybe the controls of the crazy thing it has to do to land?

No, it's actually, by far the biggest thing of solving my time is engine production. Not the design of the engine, I've often said prototypes are easy. Production is hard. So, we have the most advanced rocket engine that's ever been designed. 'Cause I say currently the best rocket engine ever is probably the RD-180 or RD-170 the dual Russian engine, basically. And still, I think an engine should only count if it's gotten something to orbit.

And so our engine has not gotten anything to orbit yet, but it is, it's the first engine that's actually better than the Russian RD engines, which were amazing design.

So you're talking about Raptor engine. What makes it amazing? What are the different aspects of it that make it, what are you the most excited about if the whole thing works in terms of efficiency, all those kinds of things?

Well, it's, the Raptor is a full flow staged combustion engine, and it's operating at a very high TAVR pressure. So, one of the key figures, merit, perhaps the key figure of merit is what is the chamber pressure at which the rocket engine can operate? That's the combustion chamber pressure. So a Raptor is designed to operate at a 300 bar, possibly, maybe higher, than standard atmospheres.

The record right now for operational engine is the RD engine that I mentioned, the Russian RD, which is, I believe around 267 bar. And the difficulty of the chamber pressure is increases on a non-linear basis. So, 10% more TAVR pressure is more like 50% more difficult, but that air pressure, that is what allows you to get a very high power density for the engine. So, enabling a very high thrust to weight ratio and a very high, specific impulse.

So, specific impulse is like a measure of the efficiency of a rocket engine. It's really the exhaust, the effect of exhaust velocity of the gas coming out of the engine. With a very high chamber pressure you can have a compact engine that nonetheless has a high expansion ratio, which is the ratio between the exit nozzle and the throat. You see a rocket engine has got sort of like a hourglass shape.

It's like a chamber and then it necks down and there's a nozzle, and the ratio of the exit diameter to the throat expansion ratio.

So why is this such a hard engine to manufacture at scale?

It's very complex.

What does complexity mean? Here's a lot of components involved.

There's a lot of components and a lot of unique materials. So we had to invent several alloys that don't exist in order to make this engine work.

So it's a materials problem too.

It's a materials problem, and in a stage combustion, that full floor stage combustion, there are many feedback loops in the system. Basically you've got propellants and hot gas flowing simultaneously to so many different places on the engine. And they all have a recursive effect on each other. So you change one thing here, it has a recursive effect here. It changes something over there. And it's quite hard to control.

There's a reason no one's made this before. And the reason we're doing a stage commotion full flow is because it has the highest theoretical possible efficiency. So in order to make a fully reasonable rocket, which, that's really the holy grail of orbital rocketry, you have to have, everything's gotta be the best. It's gotta be the best engine, the best airframe, the best heat shield, extremely light avionics, very clever control mechanisms.

You've got to shed mass in any possible way that you can. For example, we are, instead of putting landing legs on the booster and ship, we are going to catch them with a tower to save the weight of the landing legs. So that's like, I mean, we're talking about catching the largest flying object ever made on a giant tower with chopstick arms. It's like "Karate Kid" with the fly, but much bigger.

(Elon laughing) - I mean, pulling something- - This probably won't work the first time. (Elon laughing) So this is bananas. This is bananas stuff.

So you mentioned that you doubt, well, not you doubt, but there's days or moments when you doubt that this is even possible. It's so difficult.

The possible part is, well at this point, we'll I think we'll get Starship to work. There's a question of timing. How long will it take us to do this? How long will it take us to actually achieve full and rapid reusability? 'Cause it will probably many launches before we are able to have full and rapid reusability. But I can say that the physics pencils out, we're not, at this point I'd say we're confident that, let's say, I'm very confident success is in the set of all possible outcomes.

[Lex] Mm, right, it's not in all set of.

For a while there I was not convinced that success was in the set of possible outcomes. (Lex laughing) Which is very important actually. But, so...

[Lex] So you're saying there's a chance.

I'm saying there's a chance. Exactly. Just not sure how long it will take. But we have a very talented team, they're working night and day to make it happen. Like I said, the critical thing to achieve with revolution in space flight and for humanity to be a space bearing civilization is to have a fully and rapidly reusable rocket, orbital rocket. There's not even been any orbital rocket that's been fully reusable ever.

And this has always been the holy grail of rocketry and many smart people, very smart people, have tried to do this before, and they've not succeeded. 'Cause it's such a hard problem.

What's your source of belief in situations like this when the engineering problem is so difficult, there's a lot of experts, many of whom you admire, who have failed in the past.

[Elon] Yes.

A lot of people, a lot of experts, maybe journalists, all the kinds of, the public in general, have a lot of doubt about whether it's possible, and you yourself know that even if it's a non-nodal set, not empty set, of success, it's still unlikely or very difficult. Where do you go to both personally, intellectually as an engineer, as a team, for source of strength needed to sort of persevere through this and to keep going with the project, take it to completion?

I suppose the strength. Hmm. That's really not how I think about things. I mean, for me, it's simply this is something that is important to get done and we should just keep doing it or die trying, and I don't need a source of strength.

So quitting is not even like...

It's not, it's not in my nature.

Okay.

And I don't care about optimism or pessimism. Fuck that, we're gonna get it done.

[Lex] Gonna get it done. Can you then zoom back in to specific problems with Starship or any engineering problems you work on? Can you try to introspect your particular biological neural network, your thinking process, and describe how you think through problems, the different engineering and design problems? Is there like a systematic process you've spoken about, first principles thinking, but is there kind of - Yeah, absolutely.

process to it?

Saying like, physics is low and everything else was a recommendation. I've met a lot of people that can break the law, but I have never met anyone who could break physics. So first for any kind of technology problem you have to sort of just make sure you're not violating physics. First principles analysis, I think, is something that can be applied to really any walk of life, anything really.

It's really just saying, let's boil something down to the most fundamental principles, the things that we are most confident are true at a foundational level, and that sets your axiomatic base, and then you reason up from there. And then you cross check your conclusion against the axiomatic truth. Some basics in physics would be like are violating conservation of energy or momentum or something like that, then it's not gonna work.

So that's just to establish is it possible? And then another good physics tool is thinking about things in the limit. If you take a particular thing and you scale it to a very large number or to a very small number, how do things change?

Both in number of things you manufacture, something like that, and then in time.

Yeah, let's say, take an example of manufacturing, which I think is just a very underrated problem. Like I said, it's much harder to take an advanced technology part and bring it into volume manufacturing, than it is to design it in the first place. More is magnitude. So let's say you're trying to figure out, why is this part or product expensive? Is it because of something fundamentally foolish that we're doing?

Or is it because our volume is too low? And so then you say, okay, well what if our volume was a million units a year? Is it still expensive? That's what I'm radical, thinking about things to the limit. If it's too expensive at a million units a year, then volume is not the reason why your thing is expensive. There's something fundamental about the design.

And then you then can focus on the reducing complexity or something like that in the design.

Gotta change the design to, change the part to be something that is not fundamentally expensive. That's a common thing in rocketry 'cause the unit volume is relatively low, and so a common excuse would be "Well, it's expensive because our unit volume is low. And if we were in like automotive or something like that, or consumer electronics, then our costs would lower." I'm like, "Okay, so let's say" we skip, "now you're making a million units a year. Is it still expensive?" If the answer is yes, then economies of scale are not the issue.

Do you throw, into manufacturing, do you throw like supply chain, you talked about resources and materials and stuff like that, do you throw that into the calculation of trying to reason from first principles? Like, how are we gonna make the supply chain work here?

Yeah, yeah.

[Lex] And then the cost of materials, things like that, or is that too much?

Yeah. Exactly. Like a good example of thinking about things in the limit is if you take any product, any machine or whatever, like take a rocket or whatever, and say, if you've got, if you look at the raw materials in the rocket, so you're gonna have like aluminum, steel, titanium, Inconel, specialty alloys, copper. And you say, "What's the weight of the constituent elements of each of these elements, and what is their raw material value?"

And that sets the asymptotic limit for how low the cost of the vehicle can be, unless you change the materials. And then when you do that, I call it like maybe the magic one number or something like that.

So that would be like, if you had the, just a pile of these raw materials here, and you could wave a magic wand and rearrange the atoms into the final shape, that would be the lowest possible cost that you could make this thing for, unless you change the materials. So then, and that is always, almost always a very low number. So then, what's actually causing things to be expensive is how you put the atoms into the desired shape.

Yeah, actually, if you don't mind me taking a tiny tangent, I had a, I often talk to Jim Keller who's somebody that worked with you as a- - Oh yeah. Jim did great work at Tesla.

So, I suppose he carries the flame of the same kind of thinking that you're talking about now. I guess I see that same thing at Tesla and SpaceX folks who work there, they kind of learn this way of thinking and it kinda becomes obvious almost. But anyway, I had argument, not argument. He educated me about how cheap it might be to manufacture Tesla Bot. We just, we had an argument. How can you reduce the cost, of scale, of producing a robot?

Because, so far I've gotten a chance to interact quite a bit, obviously in the academic circles, with humanoid robots, and then with Boston Dynamics and stuff like that. And they're very expensive to build. And then Jim kinda schooled me on saying like, "Okay, this kind of first principles thinking of how can we get the cost of manufacturing down."

I suppose you do that, you have done that kind of thinking for Tesla Bot and for all kinds of, all kinds of complex, systems that are traditionally seen as complex, and you say, "Okay, how can we simplify everything down?"

Yeah. I mean, I think if you are really good at manufacturing, you can basically make, at high volume you can basically make anything for a cost that asymptotically approaches the raw material value of the constituents, plus any intellectual property that you need to license. Anything.

Right.

But it's hard. It's not like that's a very hard thing to do, but it is possible for anything. Anything in volume can be made of, like I said, for a cost that asymptotically approaches it's raw material constituents plus intellectual property license rights.

So what will often happen in trying to design a product is people will start with the tools and parts and methods that they are familiar with, and try to create a product using their existing tools and methods. The other way to think about it is actually imagine the, try to imagine the platonic ideal of the perfect product or technology, whatever it might be, and say, "What is this?

What is the perfect arrangement of atoms that would be the best possible product? And now let us try to figure out how to get the atoms in that shape."

I mean, it sounds, it's almost like "Rick and Morty" absurd until you start to really think about it. And you really should think about it in this way 'cause everything else is kind of, if you think you might fall victim to the momentum of the way things are done in the past, unless you think in this way.

Well, just as a function of inertia, people will want to use the same tools and methods that they are familiar with. That's what they'll do by default.

[Lex] Yeah.

And then that will lead to an outcome of things that can be made with those tools and methods, but is unlikely to be the platonic ideal of the perfect product. So that's why it's good to think of things in both directions, so like what can we build with the tools that we have, but also what is the perfect, the theoretical perfect product look like?

And that theoretical perfect product is gonna be a moving target, 'cause as you learn more the definition of that perfect product will change 'cause you don't actually know what the perfect product is, but you can successfully approximate a more perfect product. So, thinking about it like that, and then saying, "Okay, now what tools, methods, materials, whatever, do we need to create in order to get the atoms in that shape?

But people very rarely think about it that way. But it's a powerful tool.

I should mention that the brilliant Shivon Zilis is hanging out with us, in case you hear a voice of wisdom from outside, from up above. Okay. So let me ask you about Mars. You mentioned it would be great for science to put a base on the moon, to do some research, but the truly big leap, again, in this category of seemingly impossible, is to put a human being on Mars. When do you think SpaceX will land a human being on Mars?

Hm. Best case is about five years, worst case 10 years.

What are the determining factors, would you say, from an engineering perspective? Or is that not the bottlenecks?

No, it's fundamentally you're engineering the vehicle. I mean Starship is the most complex and advanced rocket that's ever been made by, I don't know, order of magnitude or something like that. It's a lot. It's really next level. And the fundamental optimization of Starship is minimizing cost per ton to orbit, and ultimately cost per ton to the surface of Mars.

This may seem like a mercantile objective, but it is actually the thing that needs to be optimized. There is a certain cost per ton to the surface of Mars where we can afford to establish a self-sustaining city. And then above that, we cannot afford to do it. So, right now you can fly to Mars for $1 trillion. No amount of money could get you a ticket to Mars.

So we need to get that above, to get that like something that is actually possible at all. We don't want to just wanna have, with Mars, flags and footprints, and then not come back for a half century like we did with the moon. In order to pass a very important, great filter. I think we need to be a multi-planet species.

This ways sound somewhat esoteric to a lot of people, but, eventually given enough time, something, Earth is likely to experience some calamity, that could be something that humans do to themselves, or an external event like happened to the dinosaurs. But if, eventually, if none of that happens, and somehow, magically, we keep going, then the sun will, the sun is gradually expanding and will engulf the earth.

And probably Earth gets too hot for life in about 500 million years. It's a long time, but that's only 10% longer than earth has been around. And so if you think about like the, the current situation, it's really remarkable and kind of hard to believe, but Earth's been around four and a half billion years, and this is the first time in four and a half billion years that it's been possible to extend life beyond Earth.

And that window of opportunity may be open for a long time, and I hope it is, but it also may be open for a short time, and we should, I think it is wise for us to act quickly while the window is open. Just in case it closes.

Yeah, the existence of nuclear weapons, pandemics, all kinds of threats, - [Elon] Yeah.

should kind of give us some motivation.

I mean, civilization could get, could die with a bang or a whimper. If it dies of demographic collapse, then it's more of a whimper, obviously. And if it's World War III, it's more of a bang, but these are all risks. I mean, it's important to think of these things and just, things like probabilities, not certainties, there's a probability that something bad will happen on earth.

I think most likely the future will be good, but there's, let's say for argument's sake, a 1% chance per century of a civilization ending event. Like that was Stephen Hawking's estimate. I think he might be right about that. We should basically think of this, being a multi-planet species, just like taking out insurance for life itself, like life insurance for life. (both laughing) - This turned into a infomercial real quick.

Life insurance for life, yes. And we can bring the creatures from, plants and animals from Earth to Mars, and breathe life into the planet, and have a second planet with life. That would be great. They can't bring themselves there, so if we don't bring them to Mars, then they will just for sure all die when the sun expands anyway, and then that'll be it.

What do you think is the most difficult aspect of building civilization on Mars, terraforming Mars, like from engineering perspective, from a financial perspective, human perspective, to get a large number of folks there who will never return back to Earth?

No, they could certainly return, some will return back to Earth.

They will choose to stay there for the rest of their lives.

Yeah, many will. We need the spaceships back, like the ones that go to Mars, we need them back, so you can hop on if you want. But we can't just not have the spaceships come back, those things are expensive. We need them back. I'd like to come back and journal their trip.

I mean, do you think about the terraforming aspect, actually building, are you're so focused right now on the spaceships part that's so critical to get to Mars?

Yeah, yeah. We absolutely, if you can't get there, nothing else matters. And like I said, we can't get there at some extraordinarily high cost. I mean, the current cost of let's say one ton to the surface of Mars is on the order of a billion dollars. So, 'cause you don't just need the rocket and the launch and everything, you need like heat shield, you need guidance system, you need deep space communications. You need some kind of landing system.

So, like rough approximation would be a billion dollars per ton to the surface of Mars right now. This is obviously way too expensive to create a self-sustaining civilization. So we need to improve that by at least a factor of a thousand.

[Lex] A million per ton?

Yes, ideally less than, much less than a million ton. You have to say like, well how much can society afford to spend or want to spend on a self-sustaining city on Mars? The self-sustaining part is important. Like it's just the key threshold, the grateful to, we'll have been passed, when the city on Mars can survive even if the space ships from earth stop coming, for any reason. Doesn't matter what the reason is.

But if they stop coming for any reason, will it die out or will it not? And if there's even one critical ingredient missing, then it still doesn't count. It's like if you're in a long sea voyage and you've got everything except vitamin C. (Elon laughing) It's only a matter of time, you're gonna die. So we gotta get a Mars city to the point where it's self sustaining.

I'm not sure this will really happen in my lifetime, but I hope to see it at least have a lot of momentum. And then you could say, "Okay, what is the minimum tonnage necessary to have a self-sustaining city?" And there's a lot of uncertainty about this. You could say, I dunno, it's probably at least a million tons. 'Cause you have to set up a lot of infrastructure on Mars.

Like I said, you can't be missing anything that in order to be self-sustaining, you can't be, like you need a semiconductor, fabs, you need iron ore refineries, you need lots of things, you know? And Mars is not super hospitable. It's the least inhospitable planet, but it's definitely a fixer upper of a planet.

[Lex] Outside of Earth.

Yes.

Earth is pretty good.

Earth is like easy. Yeah.

And, also, we should clarify in the solar system.

[Elon] Yes. In the solar system.

There might be nice like vacation spots.

There might be some great planets out there, but it's hopeless- - Too hard to get there?

Yeah, way, way, way, way, way too hard, to say the least.

Let me push back on that. Not really a pushback, but quick a curve ball of a question. So you did mention physics as the first starting point. General relativity allows for worm holes. They technically can exist. Do you think those can ever be leveraged by humans to travel fast in the speed of light? Or are you saying- - The worm hole thing is debatable. We currently do not know of any means of going faster than the speed of light.

There are some ideas about having space. You're gonna move at the speed of light through space, but if you can make space itself move, that would be warping space. Space is capable of moving faster than the speed of light.

[Lex] Right.

Like the universe in the big bang, the universe expanded at much more than the speed of light, by a lot.

[Lex] Yeah. If this is possible, the amount of energy required to warp space is so gigantic, it boggles the mind.

So, all the work you've done with propulsion, how much innovation is possible with rocket propulsion? I mean, you've seen it all, and you're constantly innovating in every aspect. How much is possible? Like how much, can you get 10 X somehow? Is there something in there, in physics, that you can get significant improvement in terms of efficiency of engines and all those kinds of things?

Well, as I was saying, really the holy grail is a fully and rapidly reasonable orbital system. Right now, the Falcon 9 is the only reusable rocket out there. The booster comes back and lands, you've seen the videos. And we got the nose cone or fairing back, but we do not get the upper stage back. That means that we have a minimum cost of building an upper stage.

You can think of like a two-stage rocket of sort of like two airplanes, like a big airplane and a small airplane, and we get the big airplane back, but not the smaller airplane. And so it still costs a lot. That upper stage is at least $10 million. And then the degree of the booster is not as rapidly and completely reusable as we'd like in order of the pharynx.

So, our kind of minimum marginal cost not counting overhead for per flight is on the order of 15 to $20 million, maybe. That's extremely good for, it's by far better than any rocket ever in history. But with full and rapid reusability, we can reduce the cost per ton to orbit by a factor of a hundred. Just think of it like, like imagining if you had an aircraft or something or a car.

And if you had to buy a new car every time you went for a drive, that'll be very expensive. It'll silly, frankly.

Mhm.

But, in fact, you just refuel the car or recharge the car and that's makes your trip, I don't know, a thousand times cheaper. So, it's the same for rockets. Very difficult to make this complex machine that can go to orbit. And so if you cannot reuse it, and have to throw even any significant part of it away, that massively increases the cost. Starship in theory could do a cost per launch of like a million, maybe $2 million or something like that. And put over a hundred tons in orbit, which is crazy.

Yeah. That's incredible. So you're saying it's, by far the biggest bang for the buck is to make it fully reusable versus like some kind of brilliant breakthrough in theoretical physics.

No, no, there's no, there's no brilliant brea, no, there's no. We gotta make the rocket reusable, this is an extremely difficult engineering problem.

Got it.

But no new physics is required.

Just brilliant engineering. Let me ask a slightly philosophical fun question. Gotta ask. I know you're focused on getting to Mars, but once we're there on Mars, what form of government, economic system, political system, do you think would best for an early civilization of humans? The interesting reason to talk about this stuff, it also helps people dream about the future. I know you're really focused about the short-term engineering dream, but it's like, I don't know. There's something about imagining an actual civilization on Mars that gives people, - Sure.

really gives people hope.

Well, it would be a new frontier and an opportunity to rethink the whole nature of government just as was done in the creation of the United States. I mean, I would suggest having a direct democracy, like people vote directly on things, as opposed to representative democracy. So, representative democracy, I think, is too subject to a special interests and coercion of the politicians and that kind of thing. So I'd recommend that there's just direct democracy. People vote on laws, the population votes on laws themselves, and then the laws must be short enough that people can understand them.

Yeah, and then keeping a well-informed populace, really being transparent about all the information about what they're voting for.

Yeah. Absolute transparency.

Yeah. And not make it as annoying as those cookies we have to accept- - Have to accept cookies. There's always a slight amount of trepidation when you click accept cookies. I feel as though there's perhaps a very tiny chance that'll open a portal to hell or something like that.

[Lex] That's exactly how I feel. Why do they keep wanting me to accept that? What do they want with this cookie? Somebody got upset with accepting cookies or something somewhere. I mean, who cares? So annoying to keep accepting all these cookies.

[Lex] To me, it's just a great- - I'm tired of accept- (Shivon speaking faintly) Yes you can have my damn cookie, I don't care. Whatever.

[Lex] You heard it from me Elon first, he accepts all your damn cookies.

Yeah. (both laughing) And stop asking me. It's annoying.

Yeah, it's one example of implementation of a good idea done really horribly.

Yeah, somebody was like, there's some good intentions of like privacy or whatever, but now everyone's just has to tick accept cookies and it's now, you have billions of people who have to keep clicking accept cookie and it's super annoying. Just accept the damn cookie, it's fine. There is like, I think fundamental problem that we're, because we've not really had a major, like a world war or something like that in a while.

And obviously we would like to not have world wars. There's not been a cleansing function for rules and regulations. So wars did have some silver lining in that there would be a reset on rules and regulations after a war. So World Wars I and II there were huge resets on rules and regulations.

If society does not have a war, and there's no cleansing function or garbage collection for rules and regulations, then rules and regulations will accumulate every year 'cause they're immortal. There's no actual, humans die, but the laws don't. So, we need a garbage collection function for rules and regulations that should not just be immortal.

'Cause some of the rules and regulations that are put in place will be counterproductive, done with good intentions, but counterproductive. And sometimes not done with good intentions. If rules and regulations just accumulate every year, and you get more and more of them, then eventually you won't be able to do anything. You're just like Gulliver with, tied down by thousands of little strings. And we see that in, U. S.

and LA, basically all economies that have been around for awhile, and regulators and legislators create new rules and regulations every year, but they don't put effort into removing them. And I think that's very important that we put effort into removing rules and regulations.

But it gets tough 'cause you get special interests that then are dependent on, they have a vested interest in that whatever rule regulation and that they, then they fight to not get it removed.

Yeah. I mean, I guess the problem with the constitution is it's kinda like C versus Java 'cause it doesn't have any garbage collection built in. I think there should be. When you first said the metaphor of garbage collection, I loved it - Yeah, it's from a coding standpoint.

From a coding standpoint, yeah, yeah. It would be interesting if the laws themselves kinda had a built in thing where they kinda die after a while, unless somebody explicitly publicly defends them. So that's sort of, it's not like somebody has to kill them. They kinda die themselves. They disappear.

[Elon] Yeah.

Not to defend Java or anything, C++, you could also have great garbage collection in Python and so on.

Yeah. So, yeah, something needs to happen or just the civilizations arteries just harden over time. And you can just get less and less done because there's just a rule against everything. So I think, I don't know, for Mars, or whatever, I say, or even for here, obviously for Earth as well, I think there should be an active process for removing rules and regulations and questioning their existence.

If we've got a function for creating rules and regulations, 'cause rules and regulations could also think of as like, they're like soft work or lines of code for operating a civilization, that's the rules and regulations. So it's not like we shouldn't have rules and regulations, but you have your code accumulation, but no code removal. And so it just gets to be become basically archaic bloatware after a while.

And it's just, it makes it hard for things to progress. So, I don't know, maybe Mars you'd have like any given law must have a sunset, and require active voting to keep it up there. I actually also say like, and these are just, I don't know, recommendations or thoughts, and ultimately will be up to the people on Mars to decide, but I think it should be easier to remove a law than to add one, because of the, just to overcome the inertia of laws.

So, maybe it's like, for argument's sake, you need like say 60% vote to have a law take effect, but only a 40% vote to remove it.

So let me be the guy, you posted a meme on Twitter recently where there's like a row of urinals and a guy just walks all the way across - So true, yeah.

and he tells you about crypto.

Listen, I mean, that's happened to me so many times, I think maybe even literally. (both laughing) - Do you think technologically speaking there's any room for ideas of smart contracts or so on? 'Cause you mentioned laws, that's an interesting implement use of things like smart contracts to implement the laws by which governments function. Like something built on Ethereum, or maybe a dog coin that enables smart contracts somehow.

I never, I didn't quite understand this whole smart contract thing. (both laughing) I'm too downtown to understand smart contracts.

That's a good line. (both laughing) - I mean, my general approach to any kind of deal or whatever is just make sure there's clarity of understanding. That's the most important thing.

[Lex] Right.

And just keep any kind of deal very short and simple, plain language, and just make sure everyone understands this is the deal. Does everyone, is it clear? And what are the consequences if first things don't happen? But usually deals are, business deals or whatever are way too long and complex and overly lawyered and pointlessly.

You mentioned that Doge is the people's coin.

[Elon] Yeah.

And you said that you were literally going, SpaceX may consider literally putting a Dogecoin on the moon.

Yeah.

Is this something you're still considering, Mars perhaps, do you think there's some chance, we've talked about political systems on Mars, that a Dogecoin is the official currency of Mars, it's the coin of the future?

Well, I think Mars itself will need to have a different currency because you can't synchronize due to speed of light, or not easily.

So it must be complete standalone from earth.

Well, yeah, Mars is, at closest approach, it's four light minutes away roughly, and then add for this approach, it's roughly 20 light minutes away, maybe a little more. So you can't really have something synchronizing if you've got a 20 minute speed of light issue, if it's got a one minute blockchain. It's not gonna synchronize properly. I don't know if Mars would have a cryptocurrency as a thing, but probably, seems likely. But it would be so kind of localized thing on Mars.

And you let the people decide.

Yeah, absolutely. The future of Mars should be up to the martians. I mean, I think the cryptocurrency thing is an interesting approach to reducing the error in the database that is called money. I think I have a pretty deep understanding of what money actually is on a practical day-to-day basis, because of PayPal. We really got in deep there. And right now the money system, actually for practical purposes is really a bunch of heterogeneous mainframes running a old COBOL.

[Lex] Okay, you mean literally- - Literally.

That is literally what's happening.

in batch mode. Okay.

In batch mode.

Yeah. Pity the poor bastards who have to've maintained that code. Okay. That's pain.

[Lex] Not even Fortrans, COBOL, yep.

That's COBOL. And they still, the banks are still buying mainframes, in 2021, and running engine COBOL code. The federal reserve is like probably even older than what the banks have, and they have an old COBOL mainframe. And so the government effectively has editing privileges on the money database. And they use those editing privileges to make more money whenever they want. And this increases the error in the database that is money.

So I think money should really be viewed through the lens of information theory. You're kinda like an internet connection. Like what's the bandwidth, total bit rate, what is the latency jitter, packet drop, errors in the network communication. Just think of money like that basically. I think that's probably what I really think of it. And then say what system, from an information theory standpoint, allows an economy to function the best.

Crypto is an attempt to reduce the error in money that is contributed by governments diluting the money supply as basically a pernicious form of taxation.

So both policy in terms of with inflation, and actual like technological, COBOL, cryptocurrency takes us into the 21st century in terms of the actual systems that allow you to do the transaction, to store wealth, all those kinds of things.

Like I said, just think - In theory.

of money as information, people often will think of money as having power in and of itself. It does not. Money is information, and it does not have power in and of itself. Applying the physics tools of thinking about things in the limit is helpful. If you are stranded on a tropical island and you have a trillion dollars, it's useless. 'Cause there's no resource allocation.

Money is a database of resource allocation, but there's no resources to allocate except yourself. So money's useless. If you're stranded on a desert island with no food, all the Bitcoin in the world will not stop you from starving.

[Lex] Yeah.

Just think of money as a database for resource allocation across time and space. And then what system, in what form should that database, or data system, what would be most effective? There is a fundamental issue with, say Bitcoin, in its current form in that it's, the transaction volume is very limited. And the latency, the latency, for a properly confirmed transaction is too long, much longer than you'd like.

It's actually not great from transaction volume standpoint or latency standpoint. So it is perhaps useful as, to solve an aspect of the money database problem, which is the sort of store of wealth or an accounting of relative obligations, I suppose. But it is not useful as a currency, as a day-to-day currency.

But people have proposed different technological solutions- - [Elon] Like Lightning.

Yeah, Lightening Network and the Layer 2 technologies on top of that.

I mean, it's all, it seems to be all kind of a trade-off, but the point is, it's kinda brilliant to say, to just think about information, think about what kind of database, what kind of infrastructure enables the exchange of- - Yeah, let's say like you're operating an economy, and you need to have some thing that allows for the efficient, to have efficient value ratios between products and services.

So you've got this massive number of products and services, and need to, you can't just barter. 'Cause that would be extremely unwieldy. So you need something that gives you a ratio of exchange between goods and services. And then, something that allows you to shift obligations across time, like debt, debt and equity shift obligations across time. Then what does the best job of that?

Part of the reason why I think there's some merit to Dogecoin, even though, it was obviously created as a joke, is that it actually does have a much higher transaction volume capability than Bitcoin. The costs of doing a transaction, the Dogecoin fee is very low. Like right now, if you wanna do a Bitcoin transaction, the price of doing that transaction is very high, so you could not use it effectively for most things.

And nor could it even scale to a high volume. And when Bitcoin was started, I guess around 2008 or something like that, the internet connections were much worse than they are today, like order of magnitude. I mean, they were way, way worse in 2008. So like having a small block size or whatever it is, and a long synchronization time made sense in 2008, but, 2021, or fast forward 10 years, it's like, comically low.

And I think there's some value to having a linear increase in the amount of currency that is generated. So, because some amount of the currency, if a currency is too deflationary or like, or should say if, if a currency is expected to increase in value over time, there's reluctance to spend it.

'Cause you're like, "Oh, if I, I'll just hold it and not spend it because its scarcity is increasing with time, so if I spend it now, then I will regret spending it. So I will just, you know, hoard all it." But if there's some dilution of the currency occurring over time, that's more of an incentive to use that as a currency.

So Dogecoin just somewhat randomly has just a fixed a number of sort of coins or hash strings that are generated every year. So there's some inflation, but it's not a percentage at base. It's a fixed number, so the percentage of inflation will necessarily decline over time. I'm not saying that it's like the ideal system for a currency, but I think it actually is just fundamentally better than anything else I've seen, just by accident.

I like how you said around 2008, so you're not, some people suggest that you might be Satoshi Nakamoto. You've probably said you're not. Let me ask- - I'm not.

You're not, for sure. Would you tell us if you were?

Yes.

Okay. Do you think it's a feature or a bug that he's anonymous, or she, or they? It's an interesting kind of quirk of human history that there is a particular technology that is a completely anonymous inventor. Or creator.

Well, I mean, you can look at the evolution of ideas before the launch of Bitcoin and see who wrote about those ideas. And then, I don't know, obviously I don't know who created Bitcoin for practical purposes, but the evolution of ideas is pretty clear for that. And, it seems as though Nick Szabo is probably more than anyone else responsible for the evolution of those ideas. So, here he claims not to be Nakamoto, but I'm not sure, that's neither here nor there, but he seems to be the one more responsible for the ideas behind Bitcoin than anyone else.

So it's not, perhaps, like singular figures aren't even as important as the figures involved in the evolution of ideas that led to things.

Yeah. Perhaps it's sad to think about history, but maybe most names would be forgotten anyway.

What is a name anyway, it's a name, a name attached to an idea. What does it even mean really?

I think Shakespeare had a thing about roses and stuff, whatever he said.

"Rose by any other name would smell as sweet." (Lex laughing) - I got Elon to quote Shakespeare. I feel like I accomplished something today.

"Shall I compare thee to a summer's day?" (both laughing) - [Lex] I'm gonna clip that out instead.

Thou art more temporate and more fair. (both laughing) (Shivon speaking faintly) - Autopilot. Tesla autopilot- (Elon laughing) Tesla autopilot has been through an incredible journey over the past six years, or perhaps even longer in the minds of, in your mind, and the minds of many involved.

I think that's where we first like connected, really, was the autopilot stuff, autonomy and...

The whole journey was incredible to me to watch. 'Cause I knew, well, part of it is I was at MIT and I knew the difficulty of computer vision. And I knew the whole, I had a lotta colleagues and friends, about the DARPA challenge, and knew how difficult it is. And so there was a natural skepticism when I first drove a Tesla with the initial system based on Mobileye. I thought there's no way.

So at first when I got in I thought "There's no way this car could maintain, like stay in the lane and create a comfortable experience." So my intuition initially was that the lane-keeping problem is way too difficult to solve.

[Elon] Oh lane-keeping, yeah, that's relatively easy.

But solve in the way that we just, we talked about previous, this prototype, versus a thing that actually creates a pleasant experience over hundreds of thousands of miles or millions. Yeah, so, I was proven wrong- - We had to wrap a lot of code around the Mobileye thing, it doesn't just work by itself.

I mean, that's part of the story of how you approach things sometimes. Sometimes you do things from scratch. Sometimes at first you kind of see what's out there, and then you decide to from scratch. That was one of the boldest decisions I've seen is both on the hardware and the software to decide to eventually go from scratch. I thought, again, I was skeptical of whether that's going to be able to work out 'cause it's such a difficult problem.

And so it was an incredible journey, what I see now with everything, the hardware, the compute, the sensors, the things I maybe care and love about most is the stuff that Andrej Karpathy's leading with, the dataset selection, the whole data engine process, the neural network architectures, the way that's in the real world, that network is tested, validated, all the different test sets, versus the image net model of computer vision, like what's in academia is like real world artificial intelligence.

Andrej's awesome and obviously plays an important role, but we have a lot of really talented people driving things. Ashok is actually the head of autopilot engineering. Andrej's the director of AI.

Ai stuff, yeah. So yeah, I'm aware that there's an incredible team of just a lot going on.

People will give me too much credit, and they'll give Andrej too much credit.

And people should realize how much is going on under the- - Yeah, just a lot of really talented people. The Tesla Autopilot AI team is extremely talented. It's like some of the smartest people in the world. So, yeah, and we're getting it done.

What are some insights you've gained over those five, six years of autopilot about the problem of autonomous driving. So, you leaped in having some sort of first principles kinds of intuitions, but nobody knows how difficult the pro- - Yeah, I thought the self-driving problem would be hard, but it was harder than I thought. It's not like I thought it'd be easy, I thought it would be very hard, but it was actually way harder than even that.

So, I mean want it comes down to at the end of the day is to solve self-driving you have to solve. You basically need to recreate what humans do to drive, which is humans drive with optical senses, eyes, and biological neural nets. And so in order to, that's how the entire road system is designed to work, with basically passive optical and neural nets, biologically.

So, for actually, for full self driving to work, we have to recreate that in digital form. So we have to, that means cameras with advanced neural nets in silicon form. And then it will obviously solve for small cell driving. That's the only way, I don't think there's any other way.

But the question is what aspects of human nature do you have to encode into the machine, right? So you have to solve the perception problem, like detect, and then you first realize, what is the perception problem for driving? Like all the kinds of things you have to be able to see. Like what do we even look at when we drive? There's, I just recently heard, Andrej talked about, at MIT, about like car doors.

I think it was the world's greatest talk of all time about car doors. The fine details of car doors, like what is even an open car door, man. So like the ontology of that, that's a perception problem. We humans solve that perception problem, and Tesla has to solve that problem. And then there's the control and the planning, coupled with the perception.

You have to figure out like what's involved in driving, especially in all the different edge cases. Maybe you can comment on this, how much game theoretic kind of stuff needs to be involved, at a four-way stop sign? As humans, when we drive, our actions affect the world.

True.

It changes how others behave, most autonomous driving, you're usually just responding to the scene, as opposed to like really asserting yourself in the scene. Do you think...

I think these sort of control logic conundrums are not the hard part. Let's see...

[Lex] What do you think is the hard part in this whole beautiful complex problem?

It's a lot of freaking software man, and a lot of smart lines of code. For sure, in order to create an accurate vector space.

You're coming from image space, which is like this flow of photons going to the camera, cameras and then since you have this massive bitstream in image space, and then you have to effectively compress the, a massive bitstream corresponding to photons that knocked off an electron in a camera sensor and turn that bitstream into a vector space.

By vector space I mean, you've got cars and humans and lane lines and curves and traffic lights and that kind of thing. Once you have an accurate vector space, the control problem is similar to that of a video game, like a "Grand Theft Auto" or "Cyberpunk." If you have accurate vector space. It's, the control problem is, I wouldn't say it's trivial, it's not trivial, but it's it's not like some insurmountable thing.

Having an accurate vector space is very difficult.

Yeah, I think we humans don't give enough respect to how incredible the human perception system is to mapping the raw photons to the vector space representation in our heads.

Your brain is doing an incredible amount of processing and giving you an image that is a very cleaned up image. Like when we look around here, you see color in the corners of your eyes, but actually your eyes have very few cones, cone receptors in the peripheral vision. Your eyes are painting color in the peripheral vision.

You don't realize it, but they're, eyes are actually painting color and your eyes will also have, there's blood vessels and all sorts of gnarly things, and there's a blind spot, but do you see your blind spot? No, your brain is painting in the missing, the blind spot.

You're gonna do these things online where you look here and look at this point and then look at this point, and it's, if it's in your blind spot, your brain will just fill in the missing bits.

So cool. The peripheral vision's so cool.

Yeah.

It makes you realize all the illusions, provision science, it makes you realize just how incredible the brain is.

The brain's doing a crazy amount of post-processing on the vision signals from your eyes. It's insane. And then even once you get all those vision signals, your brain is constantly trying to forget as much as possible. So human memory is perhaps the weakest thing about the brain is memory.

So because memory is so expensive to our brain, and so limited, your brain is trying to forget as much as possible and distill the things that you see into the smallest amounts of information possible. So your brain is trying to not just get to a vector space, but get to a vector space that is the smallest possible vector space of only relevant objects.

You can sort of look inside your brain, or at least I can like when you drive down the road, and try to think about what your brain is actually doing, - Yeah - consciously. It's like, you'll see a car, because you don't have cameras. You don't have eyes in the back of your head or the side, so you say like, you're basically, your head is like a, you basically have like two cameras on a slow gimbal. (both laughing) And eyesight's not that great.

Okay? Human eyes are... And people are constantly distracted and thinking about things and texting and doing all sorts of things they shouldn't do in a car, changing the radio station. So, having arguments. When's the last time you looked right and left, and rearward, or even diagonally forward to actually refresh your vector space?

So you're glancing around and what your mind is doing is trying to distill the relevant vectors, basically objects with a position and motion, and then editing that down to the least amount that's necessary for you to drive.

It does seem to be able to edit it down or compress even further into things like concept, so it's not, it's like it goes beyond, the human mind seems to go sometimes beyond vector space to sort of space of concepts, to where you'll see a thing, it's no longer represented spatially somehow, it's almost like a concept that you should be aware of. If this is a school zone, you'll remember that as a concept. Which is a weird thing to represent, but perhaps for driving you don't need to fully represent those things. Or maybe you get those kind of - Well you- - indirectly.

You need to established vector space and then actually have predictions for those vector spaces. Like you drive past say a bus and you see that there's people, before you drove past the bus you saw people crossing, or just imagine there's like a large truck or something blocking site. But before you came up to the truck you saw that there were some kids about to cross the road in front of the truck.

Now you can no longer see the kids, but you would now know, okay, those kids are probably gonna pass by the truck and cross the road. Even though you cannot see them. So you have to have memory. You need to remember that there were kids there and you need to have some forward prediction of what their position will be.

It's a really hard problem - at the time of relevance.

So with occlusions and computer vision, when you can't see an object anymore, even when it just walks behind a tree and reappears, that's a really, really, I mean, at least in academic literature, it's tracking through occlusions, it's very difficult.

Yeah, we're doin' it.

[Lex] I understand this. So some of it- - It's like object permanence. The same thing happens with the humans with neural nets. When like a toddler grows up, there's a point in time where they develop, they have a sense of object permanence. So before a certain age, if you have a ball, or a toy or whatever, and you put it behind your back and you pop it out, before they have object permanence, it's like a new thing every time.

It's like, "Whoa, this toy went poof, disappeared, and now it's back again." and they can't believe it. And that they can play peek-a-boo all day long because peek-a-boo's fresh every time. But then we figure out object permanence, then they realize, "Oh, no, the object is not gone. It's just behind your back."

Sometimes I wish we never did figure out object permanence.

Object permanence. Yeah, so that's a...

[Lex] That's an important problem to solve.

Yes. So, an important evolution of the neural nets in the car is memory across both time and space. Now you can't remember, you have to say how long do you want to remember things for. There's a cost to remembering things for a long time. So you could run out of memory to try to remember too much for too long. And then you also have things that are stale if you remember 'em for too long.

And then you also need things that are remembered over time. So even if you, say have, for evidence sake, five seconds of memory on a time basis, but, let's say you you're parked at a light and you saw, use a pedestrian example, that people were waiting to cross the cross the road, and you can't quite see them because of an occlusion, but they might wait for a minute before the light changes for them to cross the road.

You still need to remember that that's where they were, and that they're probably going to cross road type of thing. So even if that exceeds your time-based memory, it should not exceed your space of memory.

And I just think the data engine side of that, so getting the data to learn all of the concepts that you're saying now, is an incredible process. It's this iterative process of just, there's this HydraNet of many- - HydraNet. We're changing the name to something else.

Okay. Alright. I'm sure it will be equally as "Rick and Morty," like.

Yeah. We've re-architected the neural nets in the cars so many times, it's crazy.

Oh, so every time there's a new major version, you'll rename it to something more ridiculous or, or memorable and beautiful, sorry. Not ridiculous of course.

If you see the full like array of neural nets that are operating the cars, it kinda boggles the mind. There's so many layers. It's crazy. We started off with simple neural nets that were basically image recognition on a single frame from a single camera, and then trying to knit those together with, with C. I should say, we were really familiar running C here, 'cause C++ is too much overhead, and we have our own C compiler.

So, to get maximum performance we actually wrote our own C compiler and are continuing to optimize our C compiler for maximum efficiency. In fact, we've just recently done a new rev on the C compiler that will compile directly to our autopilot hardware.

So you wanna compile the whole thing down with your own compiler?

Yeah.

So efficiency here, 'cause there's all kinds of computers, CPU, GPU, there's like basic types of things and you have to somehow figure out the scheduling across all of those things. And so you're compiling the code down - Yeah.

that does all, okay. So that's why there's a lotta people involved.

There's a lot of hardcore software engineering at a very sort of bare metal level. 'Cause we're trying to do a lot of compute that's constrained to the our full self-driving computer. And we wanna try to have the highest frames per second possible in a sort of very finite amount of compute and power. We really put a lot of effort into the efficiency of our compute.

So there's actually a lot of work done by some very talented software engineers at Tesla that, at a very foundational level to improve the efficiency of compute and how we use the trip accelerators, which are basically doing matrix math, dot products, like a bazillion dot products. And it's like, one of our neural nets is like, compute wise, like 99% dot products.

And you wanna achieve as many high frame rates, like a video game, you want - Yeah.

full resolution, higher frame.

High frame rate, low latency, low jitter. I think one of the things we're moving towards now is no post-processing of the image through the image signal processor. What happens for cameras is that, well almost all cameras, is they there's a lot of post-processing done in order to make pictures look pretty. And so we don't care about pictures looking pretty. We just want the data. So we're moving just raw photon counts.

The image that the computer sees is actually much more than what you'd see if you represent it on a camera, it's got much more data. And even in very low light conditions, you can see that there's a small photon count difference between this spot here and that spot there, which means that, so it can see in the dark incredibly well, because it can detect these tiny differences in photon counts. Like much better than you could possibly imagine.

We also save 13 milliseconds on latency.

[Lex] From removing the post-processing on the image?

Yes.

Yeah.

'Cause we've got eight cameras and then there's roughly, I don't know, one and a half milliseconds or so, maybe 1. 6 milliseconds of latency for each camera. Basically bypassing the image processor gets us back 13 milliseconds of latency, which is important.

And we track latency all the way from photon hits the camera, to all the steps that it's gotta go through to get, go through the various neural nets and the C code, and there's a little bit of C++ there as well. Well, I can, maybe a lot, but it, the core stuff is, the heavy-duty compute is all in C.

And so we track that latency all the way to an outward command to the drive unit to accelerate the brakes, to slow down the steering, turn left or right. 'Cause you gotta output a command, that's gotta go to a controller, and like some of these controllers have an update frequency that's maybe 10 Hertz or something like that, which is slow. That's like now you lose a hundred milliseconds potentially.

So then we wanna update the drivers on the steering and braking control to have more like 100 Hertz instead of 10 Hertz, then you've got a 10 millisecond latency instead of 100 milliseconds worst-case latency.

And actually, jitter is more of a challenge than latency, 'cause latency is, you can anticipate and predict, but if you've got a stackup of things going from the camera to the computer, through then a series of other computers, and finally to an actuator on the car; if you have a stackup of tolerances, of timing tolerances, then you can have quite a variable latency, which is called jitter.

And that makes it hard to anticipate exactly how you should turn the car or accelerate because, if you've got maybe 150, 200 milliseconds of jitter, then you could be off by 2. 2 seconds. And this could make a big difference.

So you have to interpolate somehow to deal with the effects of jitter, so they can make robust control decisions. So the jitters and the sensor information, or the jitter can occur at any stage in the pipeline.

If you have just, if you have fixed latency, you can anticipate and like say, "Okay, we know what that our information is," for argument's sake, "150 milliseconds stale." For argument's sake, 150 milliseconds from photons taking camera to where you can measure a change in the acceleration of the vehicle.

Then you can just say, "Okay, well we're gonna, we know it's 150 milliseconds, so we're gonna take that into account and compensate for that latency." However, if you've got then 150 milliseconds of latency, plus 100 milliseconds of jitter, which could be anywhere from zero to 100 milliseconds on top. So then your latency could be from 150, 250 milliseconds, now you've got 100 milliseconds that you don't know what to do with.

That's basically random. So, getting rid of jitter is extremely important.

And that affects your control decisions and all of those kinds of things. Okay.

Yeah, the cars just gonna fundamentally maneuver better with lower jitter.

[Lex] Got it.

The cars will maneuver with super human ability and reaction time, much faster than a human. I mean, I think over time, the autopilot, full self-driving will be capable of maneuvers that are far more than what like James Bond could do in like the best movie, type of thing.

That's exactly what I was imagining in my mind, as you said it.

It's like impossible maneuvers that a human couldn't do.

Well, let me ask sort of a, looking back the six years, looking out into the future, based on your current understanding, how hard do you think this full self-driving problem, when do you think Tesla will solve level four FSD?

I mean, it's looking quite likely that it'll be next year.

And what does the solution look like? Is it the current pool of FSD beta candidates? They start getting greater and greater as they have been, degrees of autonomy. And then there's a certain level beyond which they can do their own, they can read a book.

Yeah. I mean, you can see, anybody who's been following the full self-driving beta closely will see that the rate of disengagements has been dropping rapidly. So, like there's engagement B where the driver intervenes to prevent the car from doing something - [Lex] Right. dangerous potentially. So the interventions per million miles has been dropping dramatically.

And that trend looks like it happens next year is that the probability of an accident on FSD is less than that of the average human, and then significantly less than that of the average human. So, it certainly appears like we will get there next year.

Then there's gonna be a case of, okay, well, we not have to prove this to regulators and prove it to, and we want a standard that is not just equivalent to a human, but much better than the average human. I think it's gotta be at least two or three times higher safety than a human. Two or three times lower probability of injury than a human before we would actually say like, "Okay, it's okay to go."

It's not gonna be equivalent, it's gonna be much better.

So if you look, FSD 10.6 just came out recently, 10.7's on the way, maybe 11 is on the way somewhere in the future.

Yeah. We were hoping to get 11 out this year, but it's, 11 actually has a whole bunch of fundamental rewrites on the neural net architecture and some fundamental improvements in creating vector space.

So there is some fundamental leap that really deserves the 11. I mean, that's a pretty cool number.

Yeah. 11 would be a single stack for all, one stack to rule them all.

A single stack. But there are just some really fundamental neural net architecture changes that will allow for much more capability. At first they're gonna have issues. Like we have this working on like sort of alpha software and it's good, but it's, it's basically taking a whole bunch of C, C++ code and leading a massive amount of C++ code and replacing it with the neural net.

And Andrej makes this point a lot, which is like neural nets are kind of eating software. Over time there's less and less conventional software, more and more neural net. Which is still software, but it's, still comes out to lines of software. But, just more neural net stuff, and less, heuristics basically. More matrix based stuff, and less heuristics based stuff.

One of the big changes will be, right now the neural nets will deliver a giant bag of points to the C++, or C and C++ code.

[Lex] Yeah.

We call it the giant bag of points.

[Lex] Yeah. And it's like, so you got a pixel and something associated with that pixel, like this pixel is probably car, this pixel is probably landline. Then you've got to assemble this giant bag of points in the C code and turn it into vectors. And it does a pretty good job of it, but it's, we wanna just, we need another layer of neural nets on top of that to take the giant bag of points and distill that down to a vector space in the neural net part of the software, as opposed to the heuristics part of the software. This is a big improvement.

[Lex] Neural net's all the way down, so you want.

It's not even all neural nets, but it's, this is a game changer to not have the bag of points, the giant bag of points, that has to be assembled with many lines of C, C++, and have a neural net just assemble those into a vector. So the neural net is outputting much, much less data, it's outputting, this is a lane line, this is a curb, this is drivable space, this is a car, this is a pedestrian or cyclist or something like that.

It's outputting, it's really outputting proper vectors to the C, C++ control code, as opposed to, sort of, constructing the vectors in C. Which we've done, I think, quite a good job of, but it grew kinda hitting a local maximum on the, how well the C can do this. So this is really a big deal. And just all of the networks in the car need to move to Surround Video, there's still some Legacy Networks that are not Surround Video.

And all of the training needs to move to Surround Video, and the efficiency of the training, it needs to get better, and it is. And then we need to move everything to raw photon counts, as opposed to processed images.

[Lex] Okay. So if you- - Which is quite a big reset on the training, 'cause the system's trained on post-process imaged images. So we need to redo all the training to train against the raw photon counts, instead of the post-processed image.

So ultimately, it's kind of reducing the complexity of the whole thing. So, reducing.

Yep. Lines of code will actually go lower.

Yeah, that's fascinating. So you do infusion of all the sensors, so reducing the complexity of having to deal with these- - [Elon] Infusion of the cameras.

Sorry.

It's all cameras really.

Right, yes. Same with humans.

Yeah.

Well, I guess we got ears too, okay.

Yeah, we'll actually need to incorporate sound as well. 'Cause you know, you need to listen for ambulance sirens or firetrucks. If somebody, yelling at you or something, I don't know. It just, there's a little bit of audio that needs to be incorporated as well.

Do you need to go to bathroom break?

[Elon] Yeah, sure, let's take a break.

Okay.

[Elon] Honestly, frankly, the ideas are the easy thing, and the implementation is the hard thing. The idea of going to the moon is the easy part, but going to the moon is the hard part.

[Lex] Is the hard part.

And there's a lot of like hardcore engineering that's gotta get done at the hardware and software level. Like I said, optimizing the C compiler and just, cutting out latency everywhere. If we don't do this, the system will not work properly. So, the work of the engineers doing this, they are like the unsung heroes. But they are critical to the success of the situation.

I think you made it clear. I mean, at least to me, it's super exciting, everything that's going on outside of what Andrej is doing. Just the whole infrastructure of the software. I mean, everything is going on with data engine, whatever it's called, the whole process is just a work of art.

The sheer scale of it is, it boggles the mind. The training, the amount of work done with, we've written all this custom software for training and labeling, and to do order labeling. Order labeling is essential. 'Cause, especially when you've got like Surround Video, it's very difficult to label Surround Video from scratch is extremely difficult. Take humans such a long time to even label one video clip, like several hours.

Or the order labeler, it basically will just apply heavy duty, a lot of compute to the video clips, to pre-assign and guess what all the things are that are going on in the Surround Video.

[Lex] And there's like correcting it.

Yeah, and then all the human has to do is like tweak, like say, adjust what is incorrect. This is like, increases productivity by 100 or more.

Yeah. So you've presented Tesla Bot as primarily useful in the factory. First of all, I think humanoid robots are incredible from a fan of robotics. I think the elegance of movement that humanoid robots, that bipedal robots show are just so cool.

It's really interesting that you're working on this and also talking about applying the same kind of, all the ideas, of some of which you've talked about, with data engine, all the things that we're talking about, with Tesla autopilot, just transferring that over to the, just yet another robotics problem. I have to ask since I care about human robot interactions, so the human side of that. So you've talked about mostly in the factory.

Do you see as part of this problem that Tesla Bot has to solve is interacting with humans and potentially having a place like in the home. So, interacting, not just, - Sure.

not replacing labor, but also like, I don't know, being a friend or an assistant.

[Elon] I think the possibilities are endless. Yeah, I mean, it's obviously, it's not quite in Tesla's primary mission direction of accelerating sustainable energy, but it is an extremely useful thing that we can do for the world, which is to make a useful humanoid robot that is capable of interacting with the world and helping in many different ways.

So in factories, and really just, I mean, I think, if you say, extrapolate to many years in the future, I think work will become optional. There's a lot of jobs that, if people weren't paid to do it, they wouldn't do it. Like it's not, it's not fun, necessarily. If you're washing dishes all day, it's like, eh. Even if you really like washing dishes, do you really wanna do it for eight hours a day every day? Probably not.

And then there's like dangerous work, and basically if it's dangerous, boring, has like potential for repetitive stress injury, that kind of thing, then that's really where humanoid robots would add the most value initially. So that's what we're aiming for is to, for the humanoid robots to do jobs that people don't voluntarily want to do. And then we'll have to pair that, obviously, with some kind of universal, basic income in the future.

So, I think.

Do you see a world when there's like hundreds of millions of Tesla Bots doing different, performing different tasks throughout the world?

Yeah, I haven't really thought about it that far into the future, but I guess that there may be something like that.

Can I ask a wild question? So, the number of Tesla cars has been accelerated and has been close to 2 million produced. Many of them have autopilot.

[Elon] I think we're over 2 million now.

Yeah. Do you think there'll ever be a time when there'll be more Tesla Bots than Tesla cars?

Yeah. Actually, it's funny you ask this question 'cause normally I do try to think pretty far into the future, but I haven't really thought that far into the future with the Tesla Bot, or it's codenamed Optimus, I call it Optimus Subprime, because it's not like a giant transformer robot. But it's meant to be a general purpose help robot.

And basically, the things that were, basically, Tesla, I think, has the most advanced real-world AI for interacting with the real world, which we've developed as a function to make self-driving work.

And so, along with custom hardware and, like a lotta hardcore low-level software to have it run efficiently and be power efficient 'cause, it's one thing to do neural nets if you've got a gigantic server room with 10,000 computers, but now, let's say you just, you have to now distill that down into one computer that's running at low power in a humanoid robot or a car.

That's actually very difficult and a lotta hardcore soft work is required for that. So since we're kind of like solving the navigate the real world with neural nets problem for cars, which are kinda like robots with four wheels, then it's like kind of a natural extension of that is to put it in a robot with arms and legs. And actuators.

The two hard things are, you basically need to make the, have the robot be intelligent enough to interact in a sensible way with the environment. So you need real real world AI, and you need to be very good at manufacturing, which is a very hard problem.

Tesla's very good at manufacturing, and also has the real world AI, so making the humanoid robot work is, basically it means developing custom motors and sensors that are different from what a car would use. I think we have the best expertise in developing advanced electric motors and power electronics. So, it just has to be for humanoid robot application, not a car.

Still, you do talk about love sometimes. So let me ask, this isn't like for like sex robots or something- - [Elon] Love is the answer.

Yes. There is something compelling to us, not compelling, but we connect with humanoid robots, or even legged robot, like with a dog, in shapes of dogs. It just, it seems like there's a huge amount of loneliness in this world. All of us seek companionship with other humans, friendship and all those kinds of things. We have a lot of here in Austin, a lot of people have dogs.

[Elon] That's right.

There seems to be a huge opportunity to also have robots that decrease the amount of loneliness in the world, or help us humans connects with each other. So, in a way that dogs can. Do you think about that with Tesla Bot at all, or is it really focused on the problem of performing specific tasks? Not connecting with humans?

I mean, to be honest, I have not actually thought about it from the companionship standpoint, but I think it actually would end up being, it could be actually a very good companion. And it could develop a personality over time that is unique. It's not just all the robots are the same. And that personality could evolve to be, match the owner or the, I guess the owner. Whatever you wanna call it. The companion, the human.

The other half, right? In the same way that friends do. See, I think that's a huge opportunity. I think- - Yeah, no, that's interesting. 'Cause there's a Japanese phrase; wabi-sabi, the subtle imperfections are what makes something special. And the subtle imperfections of the personality of the robot, mapped to the subtle imperfections of the robot's human friend, dunno, owner sounds like maybe the wrong word, but, could actually make an incredible buddy basically.

[Lex] And in that way, the imperfections- - Like R2-D2 or a C-3PO sort of thing.

So from a machine learning perspective, I think the flaws being a feature is really nice. You could be quite terrible at being a robot for quite a while in the general home environment or all in the general world. And that's kind of adorable and that's, those are your flaws, and you fall in love with those flaws. It's very different than autonomous driving where it's a very high stakes environment, you cannot mess up. And so it's, yeah, it's more fun to be a robot in the home.

Yeah, in fact, if you think of like a C-3PO and R2-D2, they actually had a lot of like flaws and imperfections and silly things and they would argue with each other.

Were they actually good at doing anything? I'm not exactly sure.

They definitely added a lot to the story. But there sort of quirky elements and, that they would make mistakes and do things, it would just, it made them relatable, I don't know. Endearing. So yeah, I think that that could be something that, it probably would happen. But our initial focus is just to make it useful. I'm confident we'll get it done, I'm not sure what the exact timeframe is, but we'll probably have, I don't know, a decent prototype towards the end of next year or something like that.

And it's cool that it's connected to Tesla, the car.

Yeah, it's using a lotta, it would use the autopilot inference computer and a lot of the training that we've done for the four cars, in terms of recognizing real world things, could be applied directly to the robot. But there's a lot of custom actuators and sensors that need to be developed.

And an extra module on top of the vector space for love.

Ah, yeah.

That's missing. Okay.

We could add that to the car too.

That's true. Yeah, it could be useful in all environments. Like you said, a lot of people argue in the car, so maybe we can help 'em out. You're a student of history, fan of "Dan Carlin's Hardcore History" podcast.

[Elon] Yeah. That's great.

Greatest podcast ever.

Yeah, I think it is, actually.

It almost doesn't really count as a podcast.

[Elon] It's more like a audio book.

Yeah. So you were on the podcast with Dan, I just had a chat with him about it. He said you guys went military and all that kind of stuff.

Yeah, it was basically, it should be titled engineer wars. Essentially, when there's a rapid change in the rate of technology, then engineering plays a pivotal role in victory in battle.

How far back in history did you go? Did you go to World War II?

Well, it was supposed to be a deep dive on fighters and bomber technology in World War II, but that ended up being more wide-ranging than that. 'Cause I just went down the, a total rat hole of like studying all of the fighters and bombers in World War II, and the constant rock, paper, scissors game that one country would make this plane, and they'd make a plane to beat that, and they'd try to make a plane to beat that, and then they'll...

And really what matters is like the pace of innovation, and also access to high quality fuel and raw materials. So, like Germany had like some amazing designs, but they couldn't make them because they couldn't get the raw materials. And they had a real problem with the oil and fuel, basically, the fuel quality was extremely variable.

So the design wasn't the bottleneck, it was- - Yeah, the U.S. had kick-ass fuel, that was very consistent, the problem is, if you make a very high performance aircraft engine, in order to make it high performance, you have to the fuel, the aviation gas, has to be a consistent mixture. And it has to have a high octane. High octane is the most important thing, but also can't have like impurities and stuff 'cause you'll foul up the engine. And the German just never had good access to oil. They try to get it by invading the caucuses, but that didn't work too well.

That never works well.

Didn't work out for them. (woman speaking faintly) Nice to meet you. Germany was always struggling with basically shitty oil, and so then they could not, they couldn't count on high quality fuel for their aircraft. So then they had to have all these additives and stuff. Whereas the U. S. had awesome fuel, and they provided that to Britain as well.

So, that allowed the British and the Americans to design aircraft engines that were super high-performance, better than anything else in the world. Germany could design the engines, they just didn't have the fuel. And then also the likes of the, the quality of the aluminum alloys that they were getting was also not that great, and so, yeah.

[Lex] You talked about all this with Dan?

Yep.

Awesome. Broadly looking at history, when you look at Genghis Khan, when you look at Stalin, Hitler, the darkest moments of human history, what do you take away from those moments? Does it help you gain insight about human nature, about human behavior today? Whether it's the wars or the individuals, or just the behavior of people, any aspects of history.

Yeah. I find history fascinating. There's just a lot of incredible things that have been done, good and bad, that they just help you understand the nature of civilization, and individuals, and...

Does it make you sad that humans do these kinds of things to each other? You look at the 20th century, World War II, the cruelty of the abuse of power. Talk about communism, Marxism, and Stalin.

I mean, some of these things do, I mean, if you, there's a lot of human history, but most of it is actually people just getting on with their lives, and it's not like human history is just non-stop war and disaster, those are actually just, those are intermittent and rare, and if they weren't then humans would soon cease to exist. But there's just that, wars tend to be written about a lot.

Whereas something being like, well, a normal year where nothing major happened doesn't get written about much, but that's, most people just like farming and kinda living their life. Being a villager somewhere. And every now and again, there's a war. I would have to say, there aren't very many books that I, where I just had to stop reading, 'cause it was just too dark.

But the book about "Stalin The Court Of The Red Star," I had stopped reading, it was just too dark. Rough.

Yeah. The 30s. There's a lot of lessons there to me, in particular that it feels like humans, all of us have that zeal, Solzhenitsyn line, that the line between good and evil runs to the heart in every man that all of us are capable of evil, all of us are capable of good, it's almost like this kind of responsibility that all of us have to tend towards the good.

And so, to me, looking at history is almost like an example of, look, you have some charismatic leader that convinces you of things, is too easy, based on that story to do evil, onto each other, onto your family onto others. And so it's like our responsibility to do good. It's not like now somehow different from history, that can happen again, all of it can happen again. And yes, most of the time you're right.

I mean, the optimistic view here is mostly people are just living life. And as you've often memed about, the quality of life was way worse back in the day, and it keeps improving over time, through innovation, through technology, but still it's somehow notable that these blimps of atrocities happen.

Sure. Yeah, I mean, life was really tough for most of history. I mean, probably for most of human history, a good year would be one where not that many people in your village died of the plague, starvation, freezing to death, or being killed by a neighboring village. It's like, "Well, it wasn't that bad." It was only like, "You know, we lost 5% this year. It was a good year."

Yeah.

That would be par for the course. Just not starving to death would have been the primary goal of most people throughout history. Just making sure we'll have enough food to last through the winter and not get, freeze or whatever. Now food is plentiful. We have an obesity problem.

Well, yeah, the lesson there is to be grateful for the way things are now for some of us. We've spoken about this offline. I'd love to get your thought about it here. If I sat down for a long form in person conversation with the President of Russia, Vladimir Putin, would you potentially want to call in for a few minutes to join in on a conversation with him, moderated and translated by me?

Sure. Yeah. Sure, I'd be happy to do that.

You've shown interest in the Russian language. Is this grounded in your interest in history of linguistics culture, general curiosity?

[Elon] I think it sounds cool.

Sounds cool, not looks cool. It takes a moment to read Cyrillic. Once you know what the Cyrillic characters stand for, actually, then reading Russian becomes a lot easier 'cause there are a lot of words that are actually the same. Like bank is bank.

So find the words that are exactly the same and now you start to understand Cyrillic, yeah.

If you can sound it out, then it's much, there's at least some commonality of words.

What about the culture? You love great engineering, physics. There's a tradition of the sciences there. When you look at the 20th century, from rocketry. So, some of the greatest rockets, some of the space exploration has been done in the Soviet, in the former Soviet Union.

Yeah.

So, do you draw inspiration from that history? Just how this culture, that in many ways, I mean, one of the sad things is, because of the language, a lot of it is lost to history, because it's not translated, all those kinds of, because it is in some ways an isolated culture, it flourishes within it's borders.

[Elon] Yeah.

So do you draw inspiration from those folks, from the history of science engineering there?

Yeah. I mean, the Soviet Union, Russia, and Ukraine as well, have a really strong history in space flight, like some of the most advanced and impressive things in history were done by the Soviet Union. One cannot help but admire the impressive rocket technology that was developed. After the sort of fall of the Soviet Union, there's much less that happened, still things are happening, but it's not quite at the frenetic pace that it was happening before the Soviet Union kind of dissolved into separate republics.

Yeah. I mean, there's the Roscosmos, the Russian, the agency. I look forward to a time when those countries, with China, are working together, the United States, they're all working together, maybe a little bit of friendly competition, but.

I feel like friendly competition is good. Governments are slow and the only thing slower than one government is a collection of governments. (Lex laughing) - Yeah.

The Olympics would be boring if everyone just crossed the finishing line at the same time.

Yeah.

Nobody would watch.

[Lex] Yeah.

And people wouldn't try hard to run fast and stuff. So, I think friendly competition is a good thing.

This is also a good place to give a shout out to a video titled "The Entire Soviet Rocket Engine Family Tree" by Tim Dodd, AKA Everyday Astronaut. It's like an hour and a half. It gives a full history of Soviet rockets. And people should definitely go check out and support Tim in general, that guy's super excited about the future, super excited about space flight, every time I see anything by him I just have a stupid smile on my face, 'cause he's so excited about stuff.

Yeah, Tim Dodd is - I love people like that.

really great if you're interested in anything to do with space. He's, in terms of explaining rocket technology to your average person, he's awesome. The Best, I'd say. I should say, the whole reason I switched us from, Raptor at one point was gonna be a hydrogen engine, but hydrogen has a lot of challenges. It's very low density. It's a deep cryogen, so it's only liquid very close to absolute zero. Requires a lot of insulation.

So it was a lot of challenges there. And I was actually reading a bit about Russian rocket engine development. At least the impression I had was that Soviet Union, Russia, and Ukraine primarily were actually in the process of switching to Methalux. And there were some interesting test and data for ISP, they were able to get up to like a 382nd ISP with the Methalux engine. And I was like, "Whoa, okay, that's, that's actually really impressive."

So I think we could, you could actually get a much lower cost, an optimizing cost per ton to orbit, cost per to Mars. I think methane option is the way to go. And I was partly inspired by the Russian work on the test ends, with Methalux engines.

And now for something completely different. Do you mind doing a bit of a meme review in the spirit of the great, the powerful Pewdiepie? Let's say one to 11, - Okay.

just go over a few documents printed out.

[Elon] We can try.

[Lex] Let's try this. I present to you document numero uno. (Elon laughing) - Okay.

[Lex] Vlad The Impaler discovers marshmallows.

Yeah, that's not bad.

You get it, because he likes impaling things.

Yes, I get it. Yes, I get it, I don't know, three, whatever.

[Lex] Oh, that's not very good. This is ground in some engineering, some history. (Elon laughing) - Yeah, I give this an 8 out of 10.

[Lex] What do you think about nuclear power?

I'm in favor of nuclear power. In a place that is not subject to extreme natural disasters. I think it's a, new nuclear power is a great way to generate electricity. I don't think we should be shutting down nuclear power stations.

[Lex] Yeah, but what about Chernobyl?

Exactly. I think people, there's like a lot of fear of radiation and stuff. I guess, the problem is a lot of people just don't, they didn't study engineering or physics, so they don't, just the word radiation just sounds scary, you know? So they don't, they can't calibrate what radiation means. But radiation is much less dangerous than you'd think. For example, Fukushima, when the Fukushima problem happened, due that tsunami.

I got people in California asking me if they should worry about radiation from Fukushima. And I'm like, definitely not, not even slightly, not at all. That is crazy. And just to show this is how, the dangers is so much overplayed compared to what it really is that I actually flew to Fukushima. And, actually, I donated a solar power system for a water treatment plant. And I made a point of eating locally grown vegetables on T. V. in Fukushima.

I'm still alive. Okay.

So it's not even that the risk of these events is low, but the impact of them is- - The impact is greatly exaggerated.

It' human nature.

People don't know what radiation is, I've had people ask me, "What about radiation from cell phones causing brain cancer?" I'm like, "When you say radiation, do you mean photons or particles?" They're like, dunno, "What do you mean photons particles?" "Do you mean, let's say photons. What frequency or wavelength?" And they're like, "No, I have no idea." "Do you know that everything's radiating all the time?" They're like, "What do you mean?"

"Like, everything's radiating all the time." Photons are being emitted by all objects all the time, basically. And if you wanna know what it means to stand in front of nuclear fire, go outside. The sun is a gigantic thermonuclear reactor that you're staring right at it. Are you still alive? Yes. Okay. Amazing.

Yeah, I guess radiation is one of the words that could be used as a tool to fear monger by certain people. That's it.

I think people just don't understand.

I mean, that's the way to fight that fear, I suppose, just to understand, just to learn.

Yeah, just say, okay, how many people have actually died from nuclear accidents? It's like practically nothing, and, say how many people have died from coal plants? And it's a very big number. Obviously we should not be starting up coal plants and shutting down nuclear plants, just doesn't make any sense at all. Coal plants, I don't know, a hundred to a thousand times worse for health than nuclear power plants.

You wanna go to the next one? It's really bad. That 90, 180 and 360 degrees, everybody loves the math. Nobody gives a shit about 270.

It's not super funny. I don't know, like two or three.

[Lex] Yeah. This is not, LOL situation. (both laughing) - [Lex] Yeah. (Elon laughing) - That one's pretty good.

[Lex] The United States oscillating between establishing and destroying dictatorships. It's like a metro, is that metro- - Yeah, metronome. Yeah, it's, I dunno, a 7 out of 10. It's kinda true.

This is kinda personal for me. Next one.

Oh, man, is this Laika.

[Lex] Yeah, well, no, this is- - Or it's referring to Laika or something.

[Lex] It's Laika's husband.

Husband, yeah.

[Lex] Hello? Yes, this is dog. Your wife was launched into space. And then the last one is him with his eyes closed and a bottle of vodka.

Yeah, Laika didn't come back.

[Lex] No. They don't tell you the full story of, the impact it had on the loved ones.

True.

That one gets an 11 from me. It just keeps goin', on the Russian theme. First man in space, nobody cares. First man on the moon.

Well, I think people do care.

[Lex] I know, but.

Yuri Gagarin's name will be forever in history. I think.

There is something special about placing, stepping foot onto another totally foreign land. It's not the journey, like people that explore the oceans. It's not as important to explore the oceans as to land in a whole new continent.

[Elon] Yeah.

[Lex] Oh this is about you. (Elon laughing) Oh yeah. I'd love to get your comment on this. Elon Musk after sending $6.6 billion to the UN to end world hunger. "You have three hours."

Yeah, well, I mean obviously $6 billion is not gonna end world hunger. I mean, the reality is at this point the world is producing far more food than it can really consume. We don't have a caloric constraint to this point. So where there is hunger, it is almost always due to civil war, or strife, or some like, it's not a thing that is extremely rare for it to be just a matter of, lack of money. There's a civil war in some country, and one part of the country's literally trying to starve the other part of the country.

So it's much more complex than something that money could solve. It's geopolitics, it's a lot of things, it's human nature, it's governments, it's monies, monetary systems, all that kinda stuff.

Yeah. Food is extremely cheap these days. I mean, the U.S. at this point, among low income families, obesity is actually now the problem. It's not, obviously it's not hunger, it's too much, too many calories. It's not that nobody's hungry anywhere, it's just, this is not a simple matter of adding money and solving it.

[Lex] What do you think that one gets? Is getting?

Two.

[Lex] Just going after empires. World, "Where did you get those artifacts?" The British Museum. It's a shout out to "Monty Python." "We found them."

Yeah. The British Museum is, it's pretty great. I mean, admittedly Britain did take these historical artifacts from around the world and put them in London, but it's not like people can't go see them. So, it is a convenient place to see these ancient artifacts is London, for a large segment of the world. So I think, unbalanced, the British Museum is net good. Well, I'm sure that a lot of countries are arguing about that.

[Lex] Yeah.

It's like, you wanna make these historical artifacts accessible to as many people as possible. And the British Museum, I think does a good job of that.

Even if there's a darker aspect to like the history of empire in general, whatever the empire is, however things were done. It is the history that happened. You can't sort of erase that history, unfortunately. You can just become better in the future. Is the point.

Yeah, I mean, well how are we gonna pass moral judgment on these things? If one is gonna judge, say the Russia Empire, you gotta judge what everyone was doing at the time, and how were the British relative to everyone? And I think that the British would actually get a relatively good grade, relatively good grade, not in absolute terms, but compared to what everyone else was doin', they were not the worst. Like I said, you gotta look at these things in the context of the history at the time and say, "What were the alternatives, and what are you comparing it against?"

Yes.

And I do not think it would be the case that Britain would get a bad grade, when looking at history at the time. Now if you judge history from what is morally acceptable today, you're basically are gonna give everyone a failing grade. I'm not clear. I don't think anyone would get a passing grade in their morality of, you could go back 300 years ago, who is getting a passing grade? Basically no one.

[Lex] And we might not get a passing grade from generations - Yeah. Exactly.

[Lex] that come after us. What does that one get?

Sure. A six, a seven.

For the "Monty Python," maybe.

[Elon] I always 'Monty Python," they're great. The "Life of Brian" and the "Quest for the Holy Grail" are incredible.

Yeah. Yeah.

Damn, those are serious eyebrows.

[Lex] Brezhnev. How important, do you think, - Damn.

[Lex] is facial hair to great leadership? You got a new haircut. How does that affect your leadership?

[Elon] I don't know. Hopefully not. It doesn't.

[Shivon] Is that the second, no one?

Yeah, the second is no one.

[Elon] There is no one competing with Brezhnev.

No one two.

Those are like epic eyebrows. Sure.

[Lex] That's ridiculous.

Give it a six or seven, I dunno.

[Lex] I like this, Shakespeare analysis of memes.

Brezhnev, he had a flare for drama as well. German joke.

[Lex] Yeah, yeah. It must come from the eyebrows. Alright. Invention, great engineering. Look what I invented. That's the best thing since rip up bread.

Yeah.

'Cause they invented sliced bread. Am I just explaining memes at this point? (all laughing) This is what my life has become.

[Shivon] He's a memelord, you're a meme explainer.

[Lex] I'm a meme, like a scribe, that runs around with the kings and just writes down memes.

I mean, when was the cheeseburger invented? That's an epic invention.

[Lex] Yeah.

Like, wow.

[Lex] Versus just like a burger?

Or a burger, I guess a burger in general.

Then there's, what is the burger? What's a sandwich? And then you start getting is a pizza a sandwich? And what is the original? It gets into an ontology argument.

Yeah, but everybody knows if you order a burger, or cheeseburger, or whatever, and you get tomato and some lettuce and onions and whatever, and mayo and ketchup and mustard, it's like epic.

Yeah, but I'm sure they've had bread and meat separately for a long time. And it was kind of a burger on the same plate, but somebody who actually combined them into the same thing and then bite it and hold it, makes it convenient. It's a materials problem. Like your hands don't get dirty and whatever. Yeah, it's brill- (Shivon talking faintly) That is not what I would've guessed.

But everyone knows, if you order a cheeseburger, you know what you're getting, it's not like some obtuse, well, I wonder what I'll get. Fries are, I mean, great. I mean, they're the devil, but fries are awesome. Yeah, pizza is incredible.

Food innovation doesn't get enough love.

Yeah.

I guess is what we're getting at.

[Elon] It's great.

What about the Matthew McConaughey, Austinite here? President Kennedy, "Do you know how to put men on the moon yet?" NASA, "No." President Kennedy, "Be a lot cooler if you did."

Pretty much, sure. Six, six or seven, I suppose.

[Lex] And this is the last one.

That's funny.

[Lex] Someone drew a bunch of dicks all over the walls. Sistine Chapel, Boys bathroom.

Sure, I'll give it a nine. It's really true.

This is our highest ranking meme for today.

[Elon] I mean, it's true, how did they get away with it?

Lotsa nakedness.

I mean, dick pics are, I mean, just something throughout history. As long as people can draw things, there's been a dick pic.

It's a staple of human history.

It's a staple. Consistent throughout human history.

You tweeted that you aspire to comedy, you're friends with Joe Rogan. Might you do a short standup comedy set at some point in the future? Maybe open for Joe? Something like that? Is that- - Really? Stand up? Actual just full-on stand up?

[Lex] Full-on stand up. Is that in there or is that?

I've never thought about that.

It's extremely difficult, at least that's what like Joe says, and the comedians say.

[Elon] Huh? I wonder if I could.

Only one way to find out.

I have done standup for friends, just impromptu, I'll get on like a roof, and they do laugh, but they're all friends too. So, I don't know if you got a room of strangers. Are they gonna actually also find it funny, but I could try. See what happens.

I think you'd learn something either way.

Yeah.

I kinda love both when and when you do great, just watching people, how they deal with it. It's so difficult. You're so fragile up there. It's just you. And you think you're gonna be funny and when it completely falls flat, it's just, it's beautiful to see people deal with that.

I think I might have enough material to do stand up. I've never thought about it, but I might have enough material. I don't know, like 15 minutes or something.

Oh yeah. Yeah. Do a Netflix special. (Elon laughing) - [Elon] Netflix special, sure.

What's your favorite "Rick and Morty" concept? Just to spring that on you, is there, there's a lot of sort of scientific engineering ideas explored there. There's the, - Favorite "Rick and Morty"

There's the butter robot.

Yeah, it's a great show.

You like it?

Yeah, "Rick and Morty's" Awesome.

Somebody that's exactly like you from an alternate dimension showed up there. Elon Tusk.

Yeah. That's right.

That you voiced.

Yeah, "Rick and Morty" certainly explores a lot of interesting concepts. Sure, like what's the favorite one. The butter robot certainly is, it's certainly possible to have too much sentience, in a device. You don't want to have your toaster be a super genius toaster. It's gonna hate life, 'cause all it can make is toast. It's like, you don't wanna have super-intelligence stuck in a very limited device.

Do you think it's too easy, from a, if we're talking about from the engineering perspective, super intelligence, like with Marvin, the robot. It seems like it might be very easy to engineer just a depressed robot.

Sure.

It's not obvious to engineer a robot that's going to find a fulfilling existence. Same as humans, I suppose. I wonder if that's like the default, if you don't do a good job on building a robot, it's going to be sad a lot.

Well, we can reprogram robots easier than we can reprogram humans. I guess if you let it evolve without tinkering, then it might get sad, but you can change the optimization function and have it be a cheery robot.

Like I mentioned with SpaceX, you give a lot of people hope, and a lot of people look up to you. Millions of people look up to you. If we think about young people in high school, maybe in college, what advice would you give to them about if they wanna try to do something big in this world, they wanna really have a big, positive impact, what advice would you give them about their career, maybe about life in general?

Try to be useful. Do things that are useful to your fellow human beings, to the world. It's very hard to be useful. Very hard. Are you contributing more than you consume? Try to have a positive net contribution to society. I think that's the thing to aim for. Not to try to be sort of a leader for the sake of being a leader or whatever.

A lot of the time people who, a lot of times the people you want as leaders, are the people who don't want to be leaders. If you're living a useful life, that is a good life, a life worth having lived. Like I said, I would encourage people to use the mental tools of physics and apply them broadly in life. They are the best tools.

When you think about education and self-education, what do you recommend? So there's the university, there's self study. There is hands-on, sort of finding a company or a place or a set of people that do the thing you're passionate about and joining them as early as possible. There's taking a road trip across Europe for a few years and writing some poetry. Which trajectory do you suggest? In terms of learning about how you can become useful, as you mentioned, how you can have the most positive impact.

I encourage people to read a lot of books, just read, basically try to ingest as much information as you can, and try to also just develop a good general knowledge. So you at least have a rough lay of the land of the knowledge landscape, try to learn a little about a lot of things. 'Cause you might not know what you're really interested. How would you know what you're really interested in if you at least aren't like doing it?

Peripheral exploration broadly of the knowledge landscape. And talk to people from different walks of life and different industries, and professions, and skills, and occupations, like just try. Learn as much as possible. Be on the search for meaning.

Isn't the whole thing a search for meaning?

Yeah, what's the meaning of life and all? But just generally, like I said, I would encourage people to read broadly in many different subject areas, and then try to find something where there's an overlap of your talents and what you're interested in. So people may be good at something, or they may have skill at a particular thing, but they don't like doing it. So you wanna try to find a thing that's a good combination of the things that you're inherently good at, but you also like doing.

And reading as a super fast shortcut to figure out which, where are you, you're both good at it, you like doing it, and it'll actually have positive impact.

Well, you gotta learn about things somehow. So reading, a broad range, just really read. More important was as a kid I read through the encyclopedia. So, that was pretty helpful. And, there was all sorts of things I didn't even know existed, well lots, obviously.

That's as broad as it gets.

Encyclopedias were suggestible, I think, whatever 40 years ago. Maybe read through like the condensed version of the Encyclopedia Britannica, I'd recommend that. You can always like skip subjects, so you read a few paragraphs and you know you're not interested, just jump to the next one. So, read the encyclopedia, or skim through it.

I put a lotta stock and certainly have a lot of respect for someone who puts in an honest day's work to do useful things. And just generally to have a, not a zero sum mindset, or have more of a grow the pie mindset. When I see people like, perhaps, including some very smart people, kind of taking an attitude of, I like doing things that seem like morally questionable.

It's often because they have, at a base sort of axiomatic level, a zero sum mindset. And they, without realizing it, they don't realize to have a zero sum mindset, or at least they don't realize it consciously. And so, if you have a zero sum mindset, then the only way to get ahead is by taking things from others. If the pie is fixed, then the only way to have more pie is to take someone else's pie. But this is false.

Obviously the pie has grown dramatically over time, the economic pie. In reality, you can have, (Elon laughing) overuse this analogy, we can have a lot of, there's a lot of pie. (Lex laughing) My pie is not fixed. So, you really wanna make sure you're not operating, without realizing it, from a zero sum mindset.

Where the only way to get ahead is to take things from others, then that's gonna result in you trying to take things from others, which is not good. It's much better to work on adding to the economic pie. Like I said, creating more than you consume. Doing more than you, yeah. So that's a big deal. I think there's a fair number of people in finance that do have a bit of a zero-sum mindset.

I mean, it's all walks of life. I've seen that. One of the reasons Rogan inspires me is he celebrates others a lot, not creating a constant competition like there's a scarcity of resources. And what happens when you celebrate others and you promote others, the ideas of others, it actually grows that pie. The resources become less scarce. And that applies in a lot of kinds of domains.

It applies in academia where a lot of people are very, see some funding for academic research as zero sum. It is not, if you celebrate each other, if you make, if you get everybody to be excited about AI, about physics, about mathematics, I think there'll be more and more funding, and I think everybody wins. Yeah. That applies, I think, broadly.

Yeah, yeah. Exactly.

So the last question about love and meaning. What is the role of love in the human condition broadly, and more specific to you? How has love, romantic love or otherwise, made you a better person, a better human being? Better engineer?

Now you're asking really perplexing questions. It's hard to give a. I mean, there are many books, poems, and songs written about what is love, and what is, what exactly, what is love, baby don't hurt me. (Lex laughing) - That's one of the great ones, yes. You have earlier quoted Shakespeare, but that's really up there.

[Elon] Yeah. Love is a many splendor thing.

I mean, there's, 'cause we've talked about so many inspiring things, like be useful in the world, sort of solve problems, alleviate suffering, but it seems like connection between humans is a source, it's a source of joy, it's a source of meaning, and that's what love is, friendship, love. I just wonder if you think about that kind of thing, when you talk about preserving the light of human consciousness.

Right.

And us becoming a multi-planetary species. I mean, to me at least, that means, if we're just alone, and conscious, and intelligent, it doesn't mean nearly as much as if we're with others. Right? And there's some magic created when we're together. The friendship of it, and I think the highest form of it is love, which I think broadly is much bigger than just sort of romantic, but also yes. Romantic love and family and those kinds of things.

Well, I mean, the reason I guess I care about us becoming a multi-planet species and a space bearing civilization is foundationally, I love humanity. And so I wish to see it prosper and do great things and be happy, and if I did not love humanity, I would not care about these things.

So when you look at the whole, the human history, all of the people whose ever lived, all the people alive now, It's pretty, we're okay. On the whole, we're a pretty interesting bunch.

Yeah. All things considered, and I've read a lot of history, including the darkest, worst parts of it. Despite all that, I think on balance, I still love humanity.

You joked about it, the 42, what do you think is the meaning of this whole thing? Is there a non-numerical representation?

Oh, I should say Yeah, well really, I think what Doug Sanders was saying in "The Hitchhiker's Guide to the Galaxy" is that the universe is the answer. What we really need to figure out are what questions to ask about the answer that is the universe. And that the question is the really the hard part. And if you can properly frame the question, then the answer, relatively speaking, is easy.

So therefore, if you want to understand what questions to ask about the university, you wanna understand the meaning of life, we need to expand the scope and scale of consciousness so that we're better able to understand the nature of the universe and understand the meaning of life.

And ultimately, the most important part will be to ask the right question.

[Elon] Yes.

Thereby elevating the role of the interviewer - [Elon] Yeah, exactly.

as the most important human in the room.

Good questions are, it's hard to come up with good questions. Absolutely. But yeah, that is the foundation of my philosophy is that I am curious about the nature of the universe. And obviously I will die. I don't know when I'll die, but I won't live forever. But I would like to know that we are on a path to understanding the nature of the universe and the meaning of life and what questions to ask about the answer that is the universe.

And so if we expand the scope and scale of humanity, and consciousness in general, which includes silicon consciousness, then that seems like a fundamentally good thing.

Elon, like I said, I'm deeply grateful that you would spend your extremely valuable time with me today, and also that you have given millions of people hope in this difficult time, this divisive time and this cynical time. So I hope you do continue doing what you're doing. Thank you so much for talking today.

Oh, you're welcome. Thanks for your excellent questions.

Thanks for listening to this conversation with Elon Musk. To support this podcast, please check out our sponsors in the description. And now, let me leave you with some words from Elon Musk himself. "When something is important enough, you do it, even if the odds are not in your favor." Thank you for listening, and hope to see you next time.