接下来是与埃隆·马斯克的一场对话，这是他第4次做客《Lex Fridman Podcast》。我还以为你会把它演奏完。这是整个电影史上最伟大的主题曲之一。

是啊，太棒了。

所以我刚才在思考罗马帝国，就像人们都会做的那样。

就是那个所有男人每天至少会想1次罗马帝国的梗吗？

有一半的人困惑于这到底是不是真的。但更严肃地说，想到当今世界正在发生的战争，而正如你所知，战争和军事征服一直是罗马社会与文化的重要组成部分，而且我认为，在整个人类历史中，它们也是大多数帝国和王朝的重要组成部分。

是啊，它们通常是征服的结果。我是说，也有一些例子，比如哈布斯堡帝国，那里就只是有很多精明的联姻。

但从根本上说，存在着一台征服引擎，而他们推崇战争中的卓越表现，许多领袖都是杰出的将领，诸如此类。所以问一个宏观问题，Grok 已批准，我问过它这是不是一个值得问的好问题。

经过测试，Grok 已批准。是啊。

至少在趣味模式下是这样。你认为战争在多大程度上是人性的一部分，又在多大程度上是人类社会结构造成的结果？我这样问，是因为不知为何，你作为和平的支持者竟颇具争议。

总体而言，我是和平的支持者。我的意思是，在我看来，无知或许才是需要对抗的真正敌人。那才是真正困难的部分，而不是与其他人类争斗；但所有生物都会争斗。我的意思是，丛林是……人们或许认为大自然是某种和平的东西，但事实上并非如此。有一段相当好笑的沃纳·赫尔佐格的片段，他在丛林里说，这里基本上四面八方都只有杀戮和死亡。丛林里的植物和动物每一天、每一分钟都在不断试图杀死彼此。所以在这方面，我们并没有什么不同寻常。

嗯，这里有一个相关的问题：更高的智力是否会带来对这些暴力本能更强的控制力。

是的。与黑猩猩相比，我们在控制自己边缘系统的暴力本能方面脆弱得多。事实上，如果观察一下，比如说，黑猩猩社会，它并不友好。我的意思是，倭黑猩猩是个例外，但黑猩猩社会充满暴力，坦率地说，相当可怕。那就是我们的边缘系统在发挥作用。你不会想站在一只黑猩猩的对立面，它会吃掉你的脸，把你的蛋蛋扯下来。

是啊。基本上没有限制或伦理，或者它们几乎就只有正义战争。黑猩猩社会里没有正义战争。战争和支配是否可以不择手段？

是啊。黑猩猩社会是人类社会的永久版本。它们基本上一点也不爱好和平。那里存在极端暴力，然后偶尔会有某个迪士尼电影看得太多的人决定把一只黑猩猩当宠物养，接着它就会吃掉他们的脸，或者他们疯掉，或者咬掉他们的手指，诸如此类。这已经发生过好几次了。

把你的蛋蛋扯下来，是一种有趣的互动策略。

这确实在人身上发生过。很不幸。我想，这是确保另一只黑猩猩不会为基因库作贡献的一种办法。

嗯，从武术的角度来看，这是一个迷人的策略。

蛋蛋搓扯术。

我想知道哪一种武术会教这一招。

我认为可以有把握地说，如果有人把你的蛋蛋握在手里，并且可以选择把它们搓扯下来，那么无论他们想要什么，你都会愿意答应。

是啊，可以有把握地这么说。所以，正如我所说，不知为何颇具争议的是，你一直在 Twitter、在 X 上支持和平。

是啊。

所以，让我问问你对当今正在发生的战争有何看法，看看通往和平的道路可能是什么。你希望目前以色列和加沙的战争如何结束？在世界的那个地区，你认为哪条道路能够最大限度地减少长期的人类苦难？

嗯，我认为世界的那个地区绝对是，如果你查一下……字典里没有简单的答案。那会是中东的图片，尤其是以色列。所以没有简单的答案。这完全只是我的看法：哈马斯的目标是引发以色列的过度反应。他们显然并不指望取得军事胜利，但他们确实想犯下自己所能犯下的最严重暴行，以便激起以色列尽可能最具侵略性的回应，然后利用这种侵略性回应，号召全世界的穆斯林支持加沙和巴勒斯坦的进程，而他们已经成功做到了。因此，这里反直觉的事情是，我认为应该做的事情，尽管非常困难，是我会建议以色列采取尽可能最显眼的善举，一切，那才是我们被教导的哈马斯的目标。

所以从某种意义上说，在地缘政治中合理的程度上，把“有人打你的右脸，连左脸也转过来”的做法付诸实施。

这并不完全是“有人打你的右脸，连左脸也转过来”，因为我确实认为，以色列找到哈马斯成员并杀死或监禁他们是合适的。这是必须要做的事，因为否则他们只会不断卷土重来。但除此之外，他们还需要尽其所能。例如，有人在讨论建立一所流动医院。我建议这么做。只要确保有食物、水和医疗必需品，而且要做得格外充分，并且高度透明。这样，[听不清 00:07:22]就可以声称这是个诡计。直接在那东西上装个网络摄像头，或者24、7。

实施善举。

对，显而易见、毫无歧义的善举，也就是说，它们不能以某种方式，因为哈马斯随后他们的回应会是：“哦，这是个诡计。”因此，你必须反驳并说明它怎么不是诡计。

这最终是在对抗该地区更广泛的仇恨力量。

是的。我不确定是谁说的，这是一句[听不清 00:07:54]格言，但以眼还眼会让所有人都失明。现在，在那个地区，他们真的相信整套以眼还眼的做法。但你确实必须……如果你不打算直接对整个民族实施种族灭绝，而这显然是不可接受的，确实不应该为任何人所接受，那么你基本上会留下许多活着的人，他们之后会仇恨以色列。所以真正的问题是，你每杀死一名哈马斯成员，又制造出了多少名？如果你制造出的比杀死的更多，你就没有成功。这就是那里的真实情况。而且可以有把握地说，如果你在加沙杀死某人的孩子，你就至少制造了几名无家可归的成员，他们会拼死也要杀死一名以色列人。情况就是这样。但我的意思是，这是人们可能讨论的最具争议性的话题之一。不过我认为，如果最终目标是某种长期的一部分，就必须从这个角度来看：随着时间推移，被制造出来的恐怖分子是更多了，还是更少了？

让我继续谈谈战争。

是啊，战争，可以有把握地说，战争一直存在，也将永远存在。

将永远存在。

一直存在，也将永远存在。

我希望不会。你认为它会永远——

战争将永远存在。问题只在于战争有多少，以及战争的范围和规模之类的问题。但要设想未来不会有任何战争，我认为那将是一种极不可能出现的结果。

是啊。你谈到过《文化》系列。即便那里也有战争。

是的。那是一场巨大的战争。第1本书一开始就是一场规模庞大的银河战争，数万亿人死亡，数万亿。

但尽管如此，它仍然保护着这些繁荣的小片区域。不知为何，你可以既有银河战争，同时又有繁荣的小片区域。

是啊，我想，如果有朝一日我们能够扩张到愚弄银河系之类的程度，那么某个时候会发生一场银河战争。

我的意思是，战争的规模一直在扩大、扩大、扩大。这就像痛苦的规模与繁荣的规模之间的一场竞赛。

是的。

许多人似乎正在利用这场悲剧擂响战鼓，并供养军工复合体。你担心这一点吗，也就是那些支持局势升级的人？怎样才能阻止这种情况？

确实令我担忧的一件事是，如今在世的人当中，真正切身体会过战争恐怖的人非常少，至少在美国是这样。我是说，乌克兰和俄罗斯前线显然有人明白战争究竟有多可怕，但西方有多少人明白？我的祖父参加过第二次世界大战。他受到了严重的创伤。我想，他在北非东部和意大利待了将近6年。他所有的朋友都在他面前被杀了，而他本来也会死，只不过他们随机安排了一种，我猜是智商测试之类的东西，而他的得分很高。他不是军官。我想他是下士、军士之类的，因为他没有读完高中，因为他父亲去世了，他不得不从高中退学，工作养活自己的兄弟姐妹。所以因为他没有高中毕业，他没有资格进入那个偏移军团。

（00:11:57）：所以他基本上算是被归入了炮灰那一类。但后来，他们随机让他参加了这项测试。他被调到伦敦的英国情报部门。我们就是在那里认识了我祖母。但他的创伤后应激障碍严重到了另一个层次，另一个层次。我的意思是，就是不说话，就是不说话。如果你试着和他说话，他只会叫你闭嘴。而且他获得了一大堆勋章，却一次也没有吹嘘过，甚至完全没有暗示过。我发现这件事，是因为他的服役记录在网上。我就是这样知道的。所以他会说：“无论如何你都不会想再经历一次。”但还有多少人……显然，他去世了，他20年前，或者更久，实际上是30年前。还有多少活着的人记得第二次世界大战？不多。

也许同样的道理也适用于核战争的威胁。

是的，我是说，有足够多的核弹瞄准美国，足以多次制造放射性狂欢平衡。

现在有2场重大战争正在进行。所以你谈过很多次通用人工智能的威胁，但此刻我们坐在这里，冲突正如此激烈地进行着，你担心核战争吗？

我认为我们不应该低估核战争的可能性。它是对文明的威胁。现在，我可能错了，但我认为当前发生核战争的概率相当低。不过，有很多核武器瞄准着我们，而我们也有很多核武器瞄准着其他人。它们仍然在那里。没有人把枪收起来。导弹仍然在发射井里。

而拥有核武器的领导人似乎并没有在相互交谈。

不，有些战争在局部层面是悲惨而艰难的。然后还有些战争会终结文明，或具有这种潜力。显然，全球热核战争极有可能终结文明，也许是永久终结，但肯定会使文明遭受重创，并且也许会让人类进步倒退到石器时代之类的。我不知道。相当糟糕。在那之后，科学家和工程师大概也会想变得超级受欢迎。是你们让我们陷入了这种困境。所以总体而言，我认为我们显然应该把文明层面的风险置于那些在局部层面令人痛苦和悲惨、但并非文明层面的事情之上。

你希望乌克兰战争如何结束？以及再问一次，将那里的人类苦难降至最低的路径是什么？

嗯，我认为很可能发生的事情，实际上也差不多就是现在的情况，是停火或休战时的界线将会非常接近当前的战线。但现在的局面就是，无论谁发动进攻，其伤亡率都会是防守方的数倍，因为有纵深防御，有雷区、战壕、反坦克防御。没有任何一方拥有制空权，因为防空导弹确实远远优于飞机。它们的数量多得多。因此，双方都没有制空权。坦克基本上是死亡陷阱，只能缓慢移动，而且它们无法免疫反坦克武器。所以实际上只有远程火炮和步兵射程。就是第一次世界大战再次重演，加上无人机、扔出的旧无人机、那里的一些无人机。

这让远程火炮变得更加精确、更好，因此也能更高效地杀害双方的人。

所以无论谁……你都不会想从任何一方尝试推进，因为死亡的概率高得难以置信。所以为了突破纵深防御、战壕和雷区，你确实需要在局部拥有显著的数量优势。理想情况下再结合施舍，你用飞机发动快速攻击，集中大量坦克和很多人，这是突破一条防线的唯一办法，而且你还得在突破之后，不会让增援部队马上又把你赶出去。我是说，我确实建议人们详细阅读第一次世界大战的作战情况。那很残酷。我是说，那里死亡人数之多令人难以置信。

而且几乎无法想象战争结束时，在土地归谁等问题上，会与开始时看起来不几乎完全一样。但另一边却是大量的人类苦难、死亡和基础设施破坏。

是的。事情是……我1年前之所以提出某种休战或和平，是因为我几乎准确地预测到了会发生什么，也就是很多人死去，而土地基本上几乎没有任何变化，同时失去乌克兰和俄罗斯年轻人中的精华。我们应该同情俄罗斯男孩，也应该同情乌克兰男孩，因为俄罗斯男孩，因为这些男孩并没有要求去他们的前线。他们必须去。所以有很多儿子无法回到父母身边，而且我认为他们中的大多数人并不仇恨对方。这有点像第一次世界大战流传下来的这句话，就像是互不相识的年轻男孩，代表彼此认识的老年男人相互残杀。这他妈有什么意义。

所以，弗拉基米尔·泽连斯基说他不，或者过去曾说过，他没有兴趣直接与普京交谈。你认为他是否应该以男人对男人、领导带领一位领导的方式坐下来谈判和平？

听着，我想我只会建议，不要把乌克兰年轻一代中的精英送进战壕送死，无论他是否与普京交谈，都不要那么做。无论谁发动攻势，都会损失大量人员，历史不会善待他们。

你曾坦率谈到，如果长期来看无法找到外交解决方案，例如在台湾问题和一个中国政策上，美国与中国之间发生战争的可能性。我们如何避免这两个超级大国发生冲突的走向？

嗯，值得读读那本讲那个……很难发音的，“修昔底德陷阱”的书，我想它是这么叫的。我喜欢战争史。我喜欢把它从里到外、倒过来研究。几乎没有哪场战役是我没读过的。并且试图弄清楚，在任何具体情况下，真正导致胜利的原因是什么，而不是某一方宣称的原因。

包括胜利，以及是什么引发了战争，还有——

对，对。

整个事情。

对。所以雅典和斯巴达就是一个经典案例。希腊人的特点是，他们确实记录了很多东西。他们喜欢写作。世界许多地方都发生过很多有意思的事情，但人们没有记录下来，所以我们不知道发生了什么，或者他们没有真正详细地记录。他们只会说：“我们打了一仗，我们赢了。”然后呢？能不能再多说一点？希腊人确实写了很多。他们在……方面非常善于表达。他们就是热爱写作。而且其中有很多文字保存了下来。所以我们知道斯巴达与雅典联盟之间的伯罗奔尼撒战争是如何一步步酝酿起来的，也知道他们预见到了它的到来。

（00:21:16）：斯巴达人不写……他们天性上也不是很啰嗦，但他们确实写，只是不太啰嗦。他们是[听不清 00:21:23]。但雅典人和其他希腊人写了一行，而斯巴达确实有点像希腊的领袖。但雅典一年比一年强大。所有人都在说：“嗯，雅典和斯巴达之间发生冲突是不可避免的。那么，我们如何避免呢？”事实上，他们预见到了它，却依然无法避免。所以在某个时刻，如果一个群体、一个文明、一个国家或其他什么超过了另一个，就有点像从经济角度来说，我想大约自1890年以来，美国一直是这一带块头最大的孩子。

（00:22:14）：所以，美国成为世界上最强大的经济引擎，已经持续了比任何人的一生都更久的时间。而战争的基础是经济。所以现在，在中国这个案例中，我们面临的情况是，其经济规模可能会达到美国的2倍，或许3倍。所以想象一下，在所有人有记忆以来，你一直是这一带块头最大的孩子，突然来了一个体型是你2倍的孩子。

所以我们预见到了它，怎么可能阻止它呢？让我提出一个想法，就是文化的相互交融、理解，双方对彼此的理解似乎确实存在巨大的文化鸿沟。而你是一个很有意思的研究案例，因为你是美国人，显然你在美国这里完成了许多令人难以置信的制造工作，但你也与中国合作。

我在中国待过很长时间，也多次与领导层会面。

也许一个值得问的好问题是，人们对中国有哪些不了解的事情，从文化中的积极方面来说？你对中国人有哪些有意思的了解？

嗯，中国极其聪明、勤奋的人数量之多，令人难以置信。真的可以说，中国有多少聪明、勤奋的人？我觉得，在我看来，那里这样的人远比这里多。而且他们精力充沛。所以我的意思是，中国近些年的建筑远比美国的令人印象深刻。我的意思是，火车站、建筑、高速铁路，一切，真的都远比我们在美国拥有的令人印象深刻。我的意思是，我建议大家直接去上海和北京，看看那些建筑，然后从北京乘火车去西安，那里有兵马俑。中国拥有令人难以置信的历史，非常悠久的历史，而且我认为，可以说就书面使用语言而言，它是最古老的书面语言之一，也许是最古老的书面语言，然后中国，人们确实把事情写了下来。

(00:24:50)：所以，中国在历史上一直都是，除了极少数例外，一直专注于内部。他们并不富有探究精神。他们彼此交战。发生过非常非常多次内战。在三国战争中，我相信他们损失了大约70%的人口。所以他们经历过残酷的内部战争，那些内战相比之下让美国内战显得规模很小。所以我认为，认识到中国并不是铁板一块很重要。我们某种程度上把中国看作一个有着同一种思想的单一实体。而事实绝对不是这样。根据我的观察，而且我认为大多数了解中国的人都会同意，中国人思考中国的程度，是他们思考中国以外任何事物的10倍。所以就像，他们90%的考量都在内部。

嗯，当你谈到超级大国之间的合作和未来这一部分时，着眼于内部，也就是专注于改善自己，而不是专注于通过军事力量来所谓“改善他人”，这难道不是一件非常积极的事吗？

好消息是，中国的历史表明，中国并不好问，也就是说，他们不会跑出去入侵一大堆国家。现在他们确实非常强烈地认为……所以这很好。我的意思是，因为很多非常强大的国家都很好问。美国也是少数并不好问的国家之一。第二次世界大战后，美国基本上本可以接管世界上的任何国家，我们有核武器，其他国家都没有核武器。我们甚至不必损失士兵。你想要哪个国家？美国本可以接管一切，但它没有。美国实际上还帮助重建了其他国家。所以它帮助重建了欧洲，帮助重建了日本。这是非常不寻常的行为，几乎前所未有。

(00:27:10)：美国做过一些显而易见的善举，比如柏林空运。而且我觉得人们总会说，好吧，美国做过坏事。嗯，美国当然做过坏事，但人们需要审视它的全部历史记录，而总体来说，一种检验方式就是：你如何对待你的战俘？或者假设一下，无意冒犯俄罗斯人，但假设你在德国，时间是1945年，俄罗斯军队从一边打来，法国、英国和美国军队从另一边打来，你希望向谁投降？没有哪个国家是[听不清 00:27:58]完美的，但我建议当美国人的战俘。那会是我非常坚定的选择。

在美国战俘的完整菜单中。

非常如此。事实上，聪明人韦恩赫尔·冯·布劳恩当时就说：“我们必须被美国人俘虏。”事实上，党卫军奉命处决冯·布劳恩和所有德国火箭调节员，而他们惊险地逃脱了。他们说自己要去树林里散步。他们在隆冬时节没穿外套就离开了，然后开始跑，但没有食物、没有外套、没有水，就拼命地跑，向西跑，而维斯·夏洛克，我想他的兄弟找到了一辆自行车之类的东西，然后尽可能快地骑车向西，他不可能找到一支美国巡逻队。所以无论如何，判断道德的一种方式就是：你想在哪里当战俘？在哪里都不好受，但有些地方比其他地方糟糕得多。总之，美国虽然远非完美，但总体上是一股仁善的力量，而我们应该始终自我批评，努力做得更好，但凡有半个脑子的人都知道这一点。

(00:29:31)：所以我认为存在……从这个方面来说，中国和美国很相似。两个国家都没有以显著的方式贪求领土。所以我想，这是一个共同原则。现在，中国确实对台湾有非常强烈的看法。他们很久以来一直对此表达得非常明确。从这个角度看，这就像是其中一个州不在那里，比如夏威夷之类的地方，但它比夏威夷更重要。而夏威夷对我们来说相当重要。所以他们认为，福尔摩沙岛，而不是台湾，实际上是中国的一个根本性组成部分，它不属于中国，但应该属于中国。而它至今没有属于中国的唯一原因，是美国太平洋舰队。

而随着他们的经济实力增长，随着他们的军事实力增长，他们明确表达的是，他们的利益显然会成为现实。

是的，中国已经非常明确地表示，他们会以和平或军事方式将台湾纳入其中，但从他们的立场来看，他们将把台湾纳入其中这件事有100%的可能性。

你提到把显而易见的善举作为一种地缘政治政策，这看起来几乎有些天真，但我敢说，这可能就是前进的道路，是避免大多数战争的方式。正如你所说，这听起来很天真，但又有点高明。如果你相信大多数人性底层的善良，那么显而易见的善举似乎就能在相关国家的民众中激起回响，并让局势降级。

完全正确。所以第一次世界大战后，他们犯了一个大错。他们基本上试图把所有责任都归到德国头上，并让德国背负不可能偿清的赔款。实际上，第一次世界大战的责任有相当一部分该由各方共同承担，但他们试图把责任全都放在德国身上，而这为第二次世界大战埋下了种子。所以很多人，不只是希特勒，很多人都觉得自己受到了不公正待遇，他们想要复仇，而且他们如愿了。

人们不会忘记。

是的，你杀了某人的父亲、母亲或儿子、女儿，他们不会忘记。他们会想要复仇。所以第二次世界大战后，他们说：“嗯，维尔西条约在第一次世界大战中是一个巨大的错误。因此这一次，我们不会压垮战败者，而是会通过模块计划帮助他们，并且会帮助重建德国。我们会帮助重建奥地利、意大利等等。”所以那是正确的做法。

确实让人觉得，把显而易见的善举作为解决这一问题的解药，其中蕴含着一个深刻的真理。

必须有什么东西来阻止这种相互施暴的循环。必须有什么东西阻止它，否则它永远不会停止。只会永远以眼还眼、以牙还牙、以肢还肢、以命还命。

暂时逃离一下黑暗，来谈谈一些令人难以置信的工程工作，xAI刚刚发布了Grok AI助手，我有机会体验了一下。它在很多层面上都令人惊叹。首先，一个相对较小的团队能在相对较短的时间内开发出这个接近最先进水平的系统，这很了不起。另一件令人难以置信的事是，它有常规模式，也有趣味模式。

是的，我想那件事该怪我。

首先，我希望生活中的一切都有趣味模式。

是的。

在与大型语言模型互动时，趣味模式除了有趣之外，还有某种引人入胜的东西。我不确定那究竟是什么，因为我只有一点点时间体验它，但它就是让与这个系统的互动变得更有意思、更有活力。

是的，完全正确。我们的AI Grok以《银河系漫游指南》为蓝本，那是我最喜欢的书之一，它是一本关于哲学的书。它是——

我最喜欢的书，它是一本关于哲学的书，伪装成一本关于幽默的书。我会说，那构成了我的哲学基础，也就是我们并不知道生命的意义，但我们越能扩大意识的范围和规模，无论是数字意识还是生物意识，就越有能力理解该针对那个作为宇宙的答案提出什么问题。所以我秉持一种好奇心哲学。

总体上会有一种感觉，仿佛这个AI系统是向外看的，就像你和一位好友坐在一起仰望星空，提出关于宇宙的、豆荚脑袋般的问题，思考这一切究竟是怎么回事。就是你所谈的那种好奇心。无论我问它的问题多么平凡，整件事都有一种宇宙般恢宏的感觉。

嗯，我们实际上正在努力提供你可以信赖的工程、数学、物理答案。对于市面上的其他AI，也就是这些所谓的大型语言模型，我发现它们在工程方面并不可靠。遗憾的是，当你最不希望它产生幻觉时，它产生的幻觉反而最多。所以当你提出重要而困难的问题时，它往往就是在那时自信地给出错误答案。所以我们确实在努力思考，好吧，我们怎样才能尽可能立足于事实？这样你就可以信赖结果，把事物追溯到物理学第一性原理和数理逻辑。因此，在幽默的底层，是一种尽可能严格遵循宇宙真相的愿望。

那真的很棘手。

这很棘手。所以这就是为什么总会存在一定程度的错误。但我们是否希望在承认错误的前提下，努力让答案尽可能真实。因此始终存在……你不希望信心十足地犯错，所以你不可能每次都正确，但你希望尽量减少自己信心十足地犯错的频率。然后就像我说的，一旦你可以相信这套逻辑不会违反物理定律，你就可以开始在此基础上发力，创造发明，比如发明新技术。但如果你无法相信基础物理是正确的，那么显然，这些发明就只是痴心妄想、想象之地。基本上就是魔法。

嗯，正如你所说，我认为xAI的一大目标是理解宇宙。

是的，就是这么简单的3个词的使命。

如果你眺望遥远的未来，你认为在这个物理学层面，在我们对物理学理解的最前沿，它会带来其中最性感的发现吗，也就是我们目前所知道的，把广义相对论和量子力学统一起来？所以，提出一个万物理论，你认为它能推动我们朝那个方向发展吗，几乎就像理论物理学发现那样？

如果一个 AI 无法发现新的物理学，那它显然不等同于人类，也没有超越人类，因为人类已经发现了新的物理学。物理学只是不断深入探究现实如何运作的内在机制。然后还有工程学，它是发明从未存在过的东西。工程学的可能性范围远大于物理学，因为一旦你弄清了宇宙的规则，就到此为止了。你发现的是已经存在的事物。但以此为基础，你就可以构建各种各样几乎没有限制的技术。这就像，一旦你正确理解了游戏规则，而凭借现有物理学，我们至少在局部层面确实非常了解物理学如何运作。我们预测事物的能力好得令人难以置信。量子力学能够预测结果的程度令人难以置信。顺便说一下，那是我大学里最难的一门课。我的高年级量子力学课程比我其他所有课程加在一起还难。

要让一个 AI 系统、一个大语言模型像量子力学和物理学一样可靠，是非常困难的。

是的。你必须根据现实的客观真相检验任何结论。现实是终极裁判。就像物理学是法律，其他一切都只是建议。我见过很多人违反人制定的法律，但没有人能违反物理学制定的法律。

其实这是个很好的测试。如果这个大语言模型理解并符合物理学，那么从某种意义上说，你就可以更可靠地相信它对当前政治状况的看法。

而且目前也并非连它的内部逻辑都是一致的。所以，尤其是采用这种只是预测一个词元、预测词元、预测词元的方法时，它就像向量求和。你把一堆向量加起来，但可能会发生漂移。一点点误差会累积起来，等你沿着这条路径走过许多词元之后，它就完全说不通了。

所以它必须以某种方式意识到自身的漂移。

它必须意识到自身的漂移，然后把这个东西作为一个格式塔、一个整体来看，并判断它作为一个整体并不连贯。作者写书时，会先写完这本书，然后再去修订，兼顾结尾、开头和中间部分，再重写，以实现连贯性，从而使它最终不会落到一个荒谬的地方。

也许修订的过程就是推理，而修订的过程也是你越来越接近真相的方式。至少我是这样做的，你先说一堆屁话，然后再把它改得更好。你从一堆屁话开始，然后你——

创建一个草稿，然后不断迭代这个草稿，直到它变得连贯，直到基本上一切都说得通。

另一个关于万物理论的问题，不过这次是针对智能：在你通过xAI探索这一点、创建这个智能系统时，你认为是否存在一种智能理论，让你能够理解AGI中的I是什么，以及人类智能中的I是什么？

不，“美国队”里没有 I。等等，是有的。

不，这下它要一直萦绕在我脑子里了。是啊，量子力学里没有我和随便什么，等等。我的意思是，理解智能，是发现、理解宇宙这一过程的一部分吗？

是的。我认为我们需要理解智能，理解意识。我的意思是，有一些根本性的问题：思想是什么，情感是什么？真的只是一个原子撞上另一个原子吗？感觉不止如此。所以我认为，我们可能遗漏了一些非常重大的东西。

某种事后看来会显而易见的东西。你把整个意识和运动放在一起。

嗯，有些人会引用类似灵魂宗教的说法，说是灵魂。你感觉自己就是你，我的意思是，你不会觉得自己只是一堆原子的集合，但思想存在于什么维度？情感存在于什么维度？因为我们对它们的感受非常强烈。我怀疑其中不只是原子撞击原子。

也许 AI 能为发现那究竟是什么鬼东西铺平道路。

是的。意识是什么？当你把原子排列成某种特定形状时，它们为什么能够形成思想、采取行动并产生感受？

即使它是一种幻觉，为什么这种幻觉如此令人信服？

是的。这个解决方案为什么存在？这个解决方案存在于什么层面？有时我会想，或许要么万物都有意识，要么什么都没有意识。二者必居其一。

我喜欢前者，万物都有意识似乎更有趣。

确实似乎更有趣，是的。但我们由原子构成，而那些原子由夸克和轻子构成，那些夸克和轻子自宇宙诞生之初就一直存在。

“宇宙诞生之初。”

看起来像是宇宙诞生之初的时刻。

我们第一次交谈时，你说过——想到这场讨论当时正在发生、如今正成为现实，感觉很超现实。我问你，一旦创造出一个 AGI 系统，你会问它什么问题？你说：“模拟之外是什么？”这就是那个问题。好问题。但看起来，你从 Grok 开始，实际上把系统的目标设定为能够回答这类问题，并提出这类问题。

外星人在哪里？

外星人在哪里？

那是泡沫悖论问题之一。很多人问过我是否见过任何外星人的证据，我没有见过，这有点令人担忧。我想，我可能更希望至少见过一些外星人的考古证据。据我所知，我不知道有任何证据监视。如果他们在那里，那他们非常隐秘。我们可能就是唯一的意识，至少在这个星系中是如此。如果你看看，比如说，地球的历史，相信考古记录的话，地球大约有45亿年历史。以最早的文字为衡量标准，文明只有大约5,000年历史。我们必须把一些功劳归于已经不复存在的古代苏美尔人。我认为，那是一种古老的前形态，是最早真正的符号表达，但那也只是在大约5,000年前。我认为，用这个日期来说文明何时开始很合适。那是地球存在时间的1000000分之一。

(00:46:35)：所以文明已经存在了一段时间。到目前为止，它其实只是昙花一现。为什么花了这么久？45亿年，绝大多数时间里都没有生命。然后有了存在了非常漫长时间的古老细菌。接着线粒体被捕获，出现多细胞生命，分化成植物和动物，生命从海洋迁移到陆地，哺乳动物，更高级的大脑功能。而太阳正在缓慢膨胀，但它会在未来某个时候使地球升温，煮沸海洋，地球将变得像金星一样，让我们所知的生命无法存在。所以，如果我们不成为多行星物种，并最终遍布太阳系，地球上所有生命的毁灭就是必然。必然。而按银河系的时间尺度，这可能短至5亿年；以人类标准来看是很长的时间，但那只比地球迄今存在的时间长10%。所以，如果生命在地球上的进化多花了10%的时间，它就根本不会存在。

很高兴截止日期就要到了，你最好快点。不过话说回来，正如你所说，地球上的人类智慧生命很快就开发出了很多很酷的东西。所以，看起来成为多行星物种几乎是不可避免的。除非我们毁灭——

我们需要做到这一点。我怀疑，如果我们能够走出去探索其他恒星系统，我们会……很有可能发现一大批早已灭亡、从未走出母星的单行星文明。

那太令人难过了。也很迷人。

我的意思是，对于这个悖论有各种解释，其中一个是，存在这些文明无法通过的巨大秃鹫。而其中一只巨大秃鹫就是：你是否会成为一个多计划文明？如果不会，那么你的星球上发生某件事，或是自然的，或是人为的，导致我们灭绝，就只是时间问题。就像恐龙一样，它们现在在哪里？它们没有宇宙飞船。

我认为更有可能的情况是，因为只要设身处地替外星人想想，他们发现了我们，而且他们在保护我们，让我们顺其自然地存在。

希望如此。友善的外星人。

就像亚马孙地区的那些部落，那些与世隔绝的部落，或者保护他们。那就是——

那会是个不错的解释。

或者也可能是，怎么说来着？我想安德烈·卡佩尔霍夫说过：“这就像蚂蚁和亚马孙地区问，大家都在哪儿？”

嗯，它们确实会遇到很多其他蚂蚁。

确实。

这些蚂蚁战争。

听起来会是一部不错的电视剧。

是啊。各种蚂蚁之间真的会爆发这些大规模战争。

是啊。也许我只是忽视了蚂蚁之间各种各样的多样性。

听听维尔纳·赫尔佐格谈论丛林的那段。真的非常好笑。你听过吗？

没有，我没听过。但维尔纳·赫尔佐格是一种方式。

你应该把它作为……中的一段插曲播放。YouTube上有。太棒了。

我太喜欢他了。

他很棒。

他是《快乐的人们生活与泰加林》的导演吗？我想还有——

他拍了那部熊的纪录片。我做了这个关于企鹅的东西。

对一只企鹅的精神分析。

是啊。那些企鹅朝着70英里外的山脉前进，而企鹅基本上就是在朝着“多姆”前进。

嗯，他的看法很愤世嫉俗。它也可能只是一个勇敢的探险家，而在未来许多个世纪里，企鹅族群中都会流传关于它的精彩故事。我们刚才在谈什么？好吧。

是的。所以外星人，我是说，我不知道。听着，我认为明智之举就是，这是地球历史上第一次，生命有可能延伸到地球之外。这个窗口已经打开。现在，它可能会开放很长时间，也可能只开放很短时间，还可能现在开放，之后就再也不会开放。所以我认为，此处的明智之举是在尚有可能时让生命成为多行星生命。我们可不想成为那种蹩脚的单行星文明，然后就这么灭绝。

不，那些很差劲。

是啊。差劲。有自尊的文明会是一个星球。

那样的文明不会有维基百科词条。SpaceX 对我们遇见外星人时该怎么办，有官方政策吗？

不。

这似乎很不负责任。

我是说，听着，如果我看到哪怕最微小的外星人存在迹象，我都会立刻把我所知道的一切发布到 X 平台上。

那可能会成为有史以来获赞和转发最多的帖子。

是的。我是说，听着，我们目前在太空中的卫星比其他所有人的加起来还多。所以，如果我们得绕开什么东西，我们会知道，而我们无须绕开任何东西。

如果我们再次谈谈那些宏大的问题，你说过你和爱因斯坦观点一致，相信斯宾诺莎女神。

是的。

所以，那种观点认为上帝就像宇宙，并通过物理定律显现自身，或者如爱因斯坦所说：“通过世界合乎规律的和谐。”

是的。我会认同，模拟器之神或无论什么至高存在通过物理学显现自身，他们有这个存在的创造物，而我们有责任努力更多地理解这一个造物。

谁创造了这东西？谁在运行这东西？把它体现为一个上面放着性感词汇的单一问题，是为了集中思想去理解。它确实像是存在着一个，再说一次，这可能是一种幻觉。它似乎有一个目的，似乎存在某种潜在的总体规划，而且看起来像是。

从这个意义上说，可能并不存在总体规划。所以，对于决定论与自由意志这个问题，一个或许有趣的答案是，如果我们身处模拟之中，这些更高等的存在进行模拟的原因就是看看会发生什么。所以他们不知道会发生什么，否则他们就不会进行模拟。所以人类创建模拟时，比如 SpaceX 和 Tesla，我们一直都在创建模拟。尤其是火箭，你必须运行大量模拟才能理解会发生什么，因为在火箭进入太空之前，你没法真正测试它，而你又希望它能正常运行。所以你必须模拟亚音速、跨音速、超音速、高超音速、上升，然后是返回、超高温加热和轨道动力学。所有这一切都必须模拟，因为你没有多少次尝试的机会。但我们运行模拟是为了看看会发生什么；如果我们知道会发生什么，就不会运行模拟。所以无论是谁创造了这个存在，他们运行它都是因为不知道会发生什么，而不是因为他们知道。

所以，也许我们都玩过《暗黑破坏神》。也许创造《暗黑破坏神》就是为了看看你的角色德鲁伊最后能否击败超级莉莉丝。他们不知道。

嗯，有趣的是，超级莉莉丝的称号是“憎恨化身”。而现在，我想你可以去问《暗黑破坏神》团队，但在永恒界域中，击败憎恨几乎是不可能的。

是的。你直播过自己称霸第 100 层梦魇地下城。而仍然——

我可以轻轻松松地刷过第 100 层梦魇地下城，就像在公园里散步一样。

而你仍然被憎恨击败了？

是的。我想，难度第二高的首领也许是都瑞尔。都瑞尔甚至能刮花漆面。所以我杀了都瑞尔很多次，还有游戏里的其他每一个首领，全都杀他很多次，很容易。但超级莉莉丝，也就是所谓的憎恨化身，尤其是如果你是都瑞尔，而且完全无法进入易伤状态，就会有这些随机的死亡波向你袭来。

（00:56:44）：说真的，我已经 52 岁了，所以我的反应能力已不如从前，但我一辈子都在玩电子游戏。有一段时间，我也许是世界上最优秀的《雷神之锤》玩家之一。我实际上在一场我认为是美国首届设有奖金的电子竞技锦标赛中赢到了钱。我们当时在参加 4 人《雷神之锤》锦标赛，我是队里第二厉害的人，而真正最厉害的那个人……我们当时其实眼看就要赢了，本来会拿第一，结果队里最厉害的那个人。他的电脑在比赛进行到一半时崩溃了。所以我们拿了第二，但我确实因此拿到了钱之类的。所以基本上，我是有技术的，尽管现在已经不年轻了。完全坦白地说，以德鲁伊的身份击败莉莉丝这件事快把我逼疯了，基本上就是试图在永恒界域中击败憎恨化身。

以德鲁伊的身份。

以德鲁伊的身份。这真的让人很恼火，我跟你说。

我是说，挑战本身就是乐趣的一部分。我亲眼见识过，你实际上是一位世界级的、不可思议的电子游戏玩家。而且我觉得《暗黑破坏神》，所以你只是在上手一款新游戏，并摸索它的基本原理。你对巅峰盘和配装的处理，也不像我这种会完全照搬网上建议的人。你在这方面也是个创新者，看着非常滑稽。就像一个疯狂科学家，试图弄明白巅峰盘和配装。如果有人刚开始玩德鲁伊，这里有什么有趣的见解吗，你有什么建议？

我不建议在永恒界域玩德鲁伊。目前我认为赛季界域中最强大的角色是使用闪电球的魔法师。烟雾在赛季中有巨大的球。

是的，他们是这么说的。

所以有巨大的球。他们会放出巨大的闪电球。

那我就相信你说的。

在赛季界域中，击败超级莉莉丝相当容易，因为你会获得这些蒸汽能力，它们会向外放大你的伤害、提高你的防御等等。所以在赛季中击败憎恨确实很容易，但要永恒地击败憎恨则非常困难，几乎不可能。非常不可能。这似乎是对人生的一种隐喻。

是的。我喜欢这个想法：埃隆·马斯克，因为我昨天在玩《暗黑破坏神》，看到一个 100 级德鲁伊跑过去，我永远不会死，然后又朝另一个方向跑回来。还有这个隐喻，太滑稽了，你，埃隆·马斯克，正焦躁不安地在这个恶魔界域中与憎恨战斗。

是的。

太滑稽了。我是说，真的相当滑稽。

不，这很荒谬。真的，这是锻炼和荒谬之举，而且让我想把自己的头发揪下来。

是的。总体来说，电子游戏能给你带来什么，就你个人而言？

我不知道。它能让我的思绪平静下来。我的意思是，在电子游戏里杀死恶魔，能平息我心中的恶魔。如果你玩一款有难度的电子游戏，就能进入心流状态，那非常令人愉悦。诚然，它不能太容易，也不能太难，得处在某种恰到好处的区间，而且我想，你通常会希望感觉自己在游戏里不断取得进展。一段好的视频，而且还有美丽的艺术、引人入胜的故事线，我觉得它就像一道等待解决的绝妙谜题。所以这就像是在解谜。

《艾尔登法环》是有史以来最伟大的游戏。我还没玩过，但对你来说——

《艾尔登法环》绝对是史上最佳游戏的候选者。肯定能进前5。

我想，我一直有点害怕它有多难，或者说听说它有多难，但它很美。

《艾尔登法环》感觉像是外星人设计的。

这也是这场讨论的一个主题。怎么说？

它太不同寻常了。它极富创意，而且美术令人惊叹。我建议甚至可以在大分辨率、高动态提升的电视上玩。它不一定非得是显示器。光是美术就不可思议。它太美了，也太不同寻常了，而且每一场顶级巴士战斗都独一无二。那是一道独特的待解谜题。每一场都不同，而且你用来解决一场战斗的策略，也不同于另一场战斗。

话虽如此，你说过，德鲁伊，一场内部对抗超级莉莉丝，是你经历过最难的首领战……

没错。那目前是，而我玩过很多电子游戏，因为那是我主要的娱乐活动。是的，在内部领域击败仇恨，是人生中最难的巴士战斗。也是电子游戏中最难的。我不确定这是否可能，但我确实在取得进展。所以接着我就会想：“好吧。我在取得进展。也许只要再稍微调整一下那个巅峰面板，我就能做到可以。”只要再躲过几波攻击，我就能做到。

好吧，这个模拟就是为了弄清楚这件事能不能做到而创造的，而你只是这台模拟机器里的一个齿轮。

是的，可能是。我有一种感觉，至少我认为——

这是能做到的。

这是能做到的。对。

好吧，这正是人类的精神：相信它能做到。

是的。我的意思是，它确实促使我思考了一下普遍意义上的仇恨，也就是，你要提防这样一种事情：你希望得到某种听起来很美好的东西，但如果你真的得到了，实际上却会是一种反乌托邦的处境。所以，如果你希望世界和平，听起来很美好，但它要如何强制实施，又要付出什么代价，永恒的和平？拥有永恒和平，实际上可能更糟，因为它所需要的一切。对所有人的压制，它可能意味着对进步的压制。它可能是一个僵化、永不改变的社会。我的意思是，有一种观点认为，如果你希望没有战争，就应该小心自己许下的愿望，因为为了实现没有战争而必须采取的手段，可能比一场小规模战争更糟。

是的。那些我们认为是人类文明缺陷的东西，可能是实现某种最优状态所必需的组成部分。我的意思是，你是否担心 AI、AGI 会促成这种性质的反乌托邦国家，无论是《1984》中充满监控和恐惧的世界，还是《美丽新世界》中充满享乐以及，那是什么？大量性爱，却没有深刻的人类体验。

实际上真有一种叫 Soma 的药。

真的？

是的。

我要记下来。

Gong 知道它吗？不知道。它可能会转到《美丽新世界》。

我们得问问它。

问题该怎么问？

给我们讲讲一种叫 Soma 的药。

给我们讲讲一种叫 Soma 的药。我觉得它会转到《美丽新世界》。

嗯，也有一种真实存在的药。

正在搜索 Soma 药物信息。哦，天哪。Soma，神奇药物……顺便说一句，我现在用的是趣味模式。

好，趣味。是的。所以它讲的是那种真实存在的药。

它是一种自20世纪50年代以来就存在的肌肉松弛剂。它用于治疗肌肉疼痛和不适，其作用方式是改变神经元与脊髓之间的活动。现在，Soma 不只是一种止痛药，它还有一点镇静作用。它会让你觉得有些困倦。所以，如果你打算操作、拥有机器，或者尝试解复杂的数学方程，就不该服用它。然后它还在继续说。总而言之，Soma 是一种可以帮助那些应对肌肉疼痛的人的药物，但它并非没有风险。说得相当不错。等一下。

它讲的就是那种真实存在的药。

是真实存在的药。当然，趣味之处总是：请依照医疗专业人士的建议负责任地使用它。还要记住，归根结底，宇宙可能只是一场巨大的计算机模拟，所以别把任何事情看得太严肃。我喜欢。

我喜欢。是的。

它还非常巧妙地收录了与此相关的推文、X 上的帖子，这真的很酷。这样就把一切整合到了一起。界面是这里非常酷的东西之一。

是的。似乎做得非常好。你可以说不，我的意思是《美丽新世界》里的 Soma，因为它以为你的意思是，你指的是真实存在的药物 Soma。不，我指的是《美丽啤酒世界》里的 Soma。

它会说：“哎呀，真的很抱歉。”我真的很惊讶，它先说了真实存在的药。很抱歉。我想到的是那种肌肉松弛剂。《崭新世界》里的 Soma 完全是另一回事。在这个赫胥黎式的反乌托邦中，Soma 是一种在书中所描绘的社会里扮演核心角色的药物。它是一种改变情绪的致幻药，用于让小说中的民众保持顺从和满足。在《美丽新世界》中使用 Soma，引发了一些关于幸福本质以及药物在社会中所扮演角色的有趣问题，伙计。

没错。生活在一个人人始终幸福的世界里会更好吗，即使那种幸福是人为制造的？这是个好问题。这就是我的意思。你希望世界永远和平、幸福吗？你确定吗？因为那可能会是一个本质上毫无生机、僵化不变、从不发生变化，并最终注定毁灭的社会。

医生与光明之间的这种张力——

这确实是一个非常好的总结。它确实切中了要点。这不只是在复述《美丽新世界》。它实际上抓住了作为一种药物的唆麻最突出的要素。你真的希望身处这样一种境况吗：所有人始终都很幸福，即使那是人为制造的？还是直面生活中的挑战、体验人类情感的完整范围会更好，即使这意味着经历痛苦与磨难？对于

顺便对正在收听的人说一下，埃隆刚才直接念的是 Grok 的内容，这是对这里这种张力相当不错、颇有洞见的哲学分析。有意思。

它基本说准了。总而言之，《美丽新世界》中的唆麻是一种虚构药物，用来探讨关于幸福的本质以及药物在社会中的作用的一些深刻哲学问题。它有力地象征着使用药物逃避现实的危险，以及正面直视生活挑战的重要性。说准了。而疯狂之处在于，我们确实有一种叫作 Soma 的真实药物，就像书中的那种药物一样。而我就想：“他们大概一定是照着它命名的。”这种真实药物中的一些对背痛相当有效。

所以你知道这种药。很有意思。

我服用过它，因为我的 C5-C6 椎间盘受压了。

所以它能消除身体上的疼痛。但这里的唆麻——

它不能完全消除疼痛，它会减轻你感受到的疼痛，但代价是思维敏锐度，它会使你的思维低陷。就像书中的药物一样。

就像书中的药物一样，因而存在这种权衡。看似乌托邦的东西，终究可能是反乌托邦。

对。其实我当时正拖着我的一个朋友说：“你真的希望世界上完全没有仇恨吗？真的一点都没有？”我想知道仇恨为什么会进化出来。我不是说我们应该有……

我想知道仇恨为什么会进化出来。我不是说我们应该放大仇恨，当然，我认为我们应该尽量减少它，但完全没有。仇恨的存在或许有其原因。

还有苦难。认为一定程度的人类苦难是人类繁荣所必需的，这件事真的很复杂。

如果不知道低谷，有可能体会到巅峰的美好吗？

而这一切都被概括在上帝的一句话里。好吧。

没有高潮，没有低谷，谁知道呢？

[听不清 01:10:38]。高效训练大语言模型似乎是 xAI 的一大重点。首先，就效率而言，可能达到的极限是什么？这里有个术语叫每瓦有效生产力。通过推动它的极限，你们学到了什么？

嗯，我认为这很有帮助，物理学的工具非常强大，而且我认为可以应用于生活中的几乎任何领域。其实就是批判性思维。对于重要的事情，你需要从第一性原理出发进行推理，并思考事物在一个方向或另一个方向上的极限。所以在极限情况下，即使达到卡尔达肖夫等级，也就是说，即使你利用了太阳的全部能量，你仍然会在意每瓦有效算力。我认为，从 AI 的角度来看，事情大概会朝这个方向发展：我们现在面临硅短缺，而大约1年后，这将转变为电压变压器短缺。具有讽刺意味的是，用于 Transformer 的变压器。你需要变压器来运行 Transformer。

这件事背后有人很有幽默感。

我想，是的，命运喜欢讽刺，喜欢讽刺式幽默，一个具有讽刺意味的滑稽结果似乎往往正是命运想要的。

幽默就是你所需要的一切。我想有人发帖说，香料就是你所需要的一切。

是的。不过，是的，所以我们今天面临硅短缺，大概1年后会面临降压变压器短缺，然后大约2年后会普遍出现电力短缺。我曾在全球公用事业公司、电力公司的聚会上发表演讲，我说，听着，你们真的需要为电力需求的旅行做好准备，因为除了火箭这个具有讽刺意味的例外，所有交通运输都将电动化，供暖也将电气化。所以目前的能源使用量大致是，非常粗略地说，三分之一用于电力，三分之一用于交通运输，三分之一用于供暖。因此，为了让一切都实现可持续发展、实现电气化，你需要将电力产出提高到3倍。所以我鼓励公用事业公司建设更多的发电厂电力，也可能要，嗯，不是可能，它们肯定应该购买更多电池，因为电网目前是按实时负载确定规模的，这有点疯狂，因为这意味着你必须按照峰值电力需求来确定规模，也就是一年中最糟糕的1秒或最糟糕的1天，否则就可能发生电压不足或停电。

(01:13:37)：奥斯汀发生过那次持续数天的疯狂停电，因为电网几乎没有任何能源缓冲。如果你有水力发电厂，就可以缓冲能源，但除此之外，一切都是实时的。所以有了电池，你就可以在夜间生产能源并在白天使用，这样就能进行缓冲。因此，我预计未来电池的使用量会非常大，因为发电厂的峰谷比在2到5之间，也就是从最低点到最高点。

所以需要电池来平衡它，但正如你所说，需求会增长、增长、增长、增长。

是的。

而其中一部分是算力？

是的。是的。我的意思是，交通运输电动化和电力供暖将远远大于 AI，至少——

短期内。

短期内。但即使对 AI 来说，电动汽车也确实带来了不断增长的电力需求，而运行 AI 计算机的电力需求也在不断增长。所以这显然可能导致电力短缺。

在这个特定案例中，就训练而言，最大限度提高每瓦有效生产力这个问题有多难，而这，这似乎确实是我们面临的、需要解决的大问题，也就是如何高效使用电力。关于在这个领域应用这种物理学第一性原理推理方法，你们到目前为止学到了什么，这个问题有多难？

它会得到解决。问题是解决它需要多长时间。所以在不同时间点，总会有某种限制进展的因素，而对于 AI，我是说目前的限制因素是硅芯片，而这将会，我们随后会拥有比实际能够接入并开启的数量更多的芯片，大概在1年后。最初的制约因素实际上就是降压变压器，因为输入的电压是300,000伏，而它最终必须一路降到约0.7伏。所以这个幅度非常大，降压幅度极其巨大，而这个行业不习惯快速增长。

好吧。我们来谈谈这里的竞争。你曾对 Google 和 Microsoft 与 OpenAI 一起开发 AGI 表示担忧。通过 xAI 和 Tesla AI 的工作，你怎样才能帮助确保它不会变成一场竞相开发 AGI 的比赛，而是安全 AGI 的协作式开发？

嗯，我的意思是，长期以来我一直在推动某种形式的监管监督。十多年来，我在这个问题上有点像卡珊德拉。我认为我们在开发 AI 时需要非常谨慎。它是一种强大的力量，而能力越大，责任越大。我认为，明智的做法是至少设立一个客观的第三方，它可以像裁判一样介入，了解各个领先参与者在 AI 方面所做的事情，而且即使没有执法能力，他们至少也可以公开表达担忧。例如，杰夫·辛顿离开了 Google，并表达了强烈担忧，但现在他已经不在 Google 了，那么谁来表达这些担忧呢？所以我认为，在汽车方面，Tesla 受到大量监管监督。我们在国内和国际上受到，我想，超过100家监管机构的监管。数量很多。Tesla 在汽车方面必须遵守的所有法规可以堆满这个房间。火箭也是如此，而且，目前限制 SpaceX 发射星舰的因素是监管批准。

(01:18:13)：美国联邦航空管理局实际上已经批准了，但我们正在等待鱼类和野生动物部门完成分析并给予批准。所以我才发帖说我想买一张鱼类许可证，这也指向蒙特·派森的短剧。你的鱼为什么需要许可证？我不知道。但有人告诉我，根据规定，你需要某种鱼类许可证之类的东西。实际上，我们需要一张鱼类许可证才能发射火箭。我就想，等一下。鱼怎么会出现在这件事里？我的意思是，有些事情在我看来荒谬到让我想拍一部喜剧短剧，并在底部闪出字幕：这一切都是真的。这确实就是发生过的事。

(01:19:02)：有件事一度有点棘手，就是他们担心火箭会击中鲨鱼。海洋非常大，而你多久能看到一次鲨鱼？并不常见。按海洋表面积的百分比计算，鲨鱼基本上为零。于是我们说，好吧，我们要怎样计算杀死一条鲨鱼的概率？而他们说，嗯，我们不能把这些信息给你，因为他们担心捕鲨取鳍者会去猎杀鲨鱼，而我说，那我们该怎么办，我们岂不是陷入进退两难的境地了。

(01:19:40)：他们说，嗯，鱼类和野生动物部门还有另一个部分可以做这项分析。我就说，那你们为什么不把数据给他们？我们不信任他们。什么？他们明明就在你们的部门里。再说一次，这确实就是发生过的事。那你们能不能签一份保密协议之类的？他们最终设法解决了内部的两难问题，而且我们击中鲨鱼的概率确实基本为零。然后还有另一个组织，直到几个月前我才知道它的存在，它关心我们是否可能在国际水域击中鲸鱼。现在，再说一次，你看看海面，看看太平洋，然后问太平洋中有百分之多少是鲸鱼？我可以给你一张大图，让你指出这张图里的所有鲸鱼。我就想，我一条鲸鱼也没看到。基本上是0%，而且如果我们的火箭真的击中了一条鲸鱼，尽管这极其不可能，简直超乎一切想象，那也是命中注定，那条鲸鱼的运气糟糕透顶，是有史以来最倒霉的鲸鱼。

我是说，不管这是怎么出现的，这种官僚作风都相当荒谬。

是的。嗯，我是说，有件很离谱的事：为了从加利福尼亚州的范德伯格发射，我们不得不……他们担心海豹繁殖，担心海豹会不会被音爆吓到。可此前已有很多火箭从范登堡发射，而海豹数量一直在稳步增加。所以真要说的话，根据证据，如果把火箭发射与海豹数量联系起来，火箭的轰鸣反倒是催情剂。尽管如此，我们还是被迫绑架了一只海豹，把它绑在一块板子上，给它戴上耳机，然后播放音爆声，看看它会不会感到痛苦。这是真实发生过的事。这确实是真的。我有照片。

我很想看看。对。抱歉。有一只戴着耳机的海豹。

是的，就是一只戴着耳机、被绑在板子上的海豹。好吧。令人惊讶的是，这只海豹竟然那么镇定，因为如果我是海豹，我会想，这下完了。他们肯定要吃掉我。这只海豹多大了，当海豹回到其他海豹朋友那里时，它要怎么解释这件事？

它们永远不会相信它。

永远不会相信它。所以我觉得，这有点像被外星人绑架并接受肛门探查。你回来后说，我对天发誓，我被外星人绑架了，他们把肛门探针插进了我的屁股里，而人们会说，不，他们没有。太荒谬了。它的海豹伙伴永远不会相信它曾被绑到船上、耳朵上戴了耳机，然后又被放走了。顺便说一句，2次，我们不得不做2次。

他们把它放走了2次。

我们不得不捕捉——

同一只海豹？

不是，不同的海豹。

好吧。你得到批准印章了吗？

没错。批准印章。不，我的意思是，我觉得公众不太了解其中发生的疯狂事情。

是啊。是啊。太荒谬了。

该死的海豹戴着该死的耳机。

我的意思是，这很好地概括了人类文明的荒谬：戴着耳机的海豹。

是的。

对你而言，将 AI 开源，作为另一种防止某家公司凭借 AGI 一骑绝尘的方式，有哪些利弊？

要运行真正深度的智能，你需要大量算力。所以，并不是说你在地下室里启动一台个人电脑就能运行 AGI，至少目前还不行。Grok 是用 8,000 块以最高效率运行的 A100 训练的，顺便说一句，Grok 会变得好得多，未来几个月里，我们每隔几个月就会把算力提高到原来的两倍以上。

有一篇很不错的文章，介绍了我们如何从 Grok 0 发展到 Grok 1。

Grok 写的？

对，没错，就是 Grok 在吹牛，瞎编关于自己的鬼话。

就是 Grok、Grok、Grok。

是啊。那就像一个奇怪的 AI 交友网站，它在上面夸大自己。不，有一篇文章介绍了它目前的状况、它的开发历史，以及它与最先进的 GPT-3 5 相比在一些基准测试中的表现。所以，我的意思是，有[听不清 01:24:37]，训练完成后，你可以开源，你可以把模型开源。用于微调之类的所有事情。那么，开源基础模型有哪些利弊？

我认为[听不清 01:24:53]开源，我觉得或许可以稍微延迟一段时间，我不知道，甚至 6 个月。我想总体而言，我赞成开源，倾向于开源。我的意思是，OpenAI 确实让我感到担忧。我想，我大概很奇怪地算是 OpenAI 背后的主要推动者，因为它的创立源于我和拉里·佩奇的讨论，那时我和他还是朋友，我住在他家，并和他谈论 AI 安全，而拉里并不在乎 AI 安全，或者至少当时他不在乎。有一次，他因为我支持人类而称我是物种主义者，我就说，好吧，拉里，你到底站哪一队？需要说清楚，他仍然站在机器人队那边。我就想，好吧。所以当时 Google 已经收购了 DeepMind，他们可能拥有全世界三分之二的 AI 研究人员。他们基本上拥有无限的资金和算力，而负责人拉里·佩奇并不在乎安全，甚至还冲我大喊，还因为我支持人类而抓住我这个物种主义者。

所以我不知道你是否注意到，人类是会改变想法的，也许你和拉里·佩奇仍然可以，可以再次成为朋友。

我想再次和拉里成为朋友。这段友谊真正的破裂是因为 OpenAI，具体来说，我认为关键时刻是招募伊利亚·苏茨克维。

我很喜欢伊利亚。他太聪明了。

伊利亚是个好人，聪明，心地善良，那是一场艰难的招募战。主要是一边是戴密斯，另一边是我，我们都试图招募伊利亚，而伊利亚来回摇摆，他本来要留在 Google，又要离开，然后又要留下，然后又会离开。最终，他确实同意加入 OpenAI。那是我们经历过的最艰难的招募战之一。但这确实是 OpenAI 成功的关键。我在招募其他一些人方面也发挥了重要作用，而且一开始所有的资金都是我提供的，超过 4,000 万美元。还有，OpenAI 里的 open 本应意味着开源，它创立时是一个非营利的开源组织，而现在它为了利润最大化变成了闭源，我认为这不是善因。

但就像我们谈到战争和领导人对话时所说的那样，我确实希望，在最高层面研究这件事的人只有少数几个。我确实希望你能重振这里的友谊。

就像我说的，我想再次和拉里成为朋友。我已经很久没见过他了，我们曾经做了很长时间的朋友。我在拉里·佩奇为 Google 获得资金之前就认识他了，或者其实我想是在他获得风险投资之前，我觉得他最初大概从贝克特尔·泽默或某个人那里得到了 10 万美元。

想想发生的这一切，而你们一直都认识彼此，真是不可思议，已经 20 年了。

是啊，可能从 98 年左右开始。

是啊，太疯狂了。从那以后发生了这么多事，真疯狂。

是啊，25 年，发生了很多事。太不可思议了。

但你看到了其中的矛盾，也许要延迟开源。

延迟，是啊，比如说，开放的源到底是什么？你明白我的意思吗？基本上就是一个巨大的 CSB 文件，里面有一堆数字。你拿那个巨大的数字文件做什么？你要怎么运行，实际的代码量，也就是代码行数非常少，而大部分工作，即软件方面的工作，都在数据整理上。所以这就像是试图弄清楚什么是数据，把好数据和坏数据分开。你不能只是抓取互联网，因为外面有很多垃圾。很大比例的网站都是噪声多于信号，因为它们只是用来做搜索引擎优化的。它们实际上就是诈骗网站。

顺便问一下，抱歉打断你，你是如何在 X 上获取信号、将信号与噪声分开的？那是一个非常迷人的数据来源。无意冒犯在 X 上发帖的人，但有时确实会有一点噪声。

我认为信噪比可以得到大幅改善。实际上，X 平台上的所有帖子都应该由 AI 推荐，也就是说，我们应该围绕任意给定帖子构建一个向量空间，将其与任意用户周围的向量空间进行比较，然后把两者匹配起来。目前推荐帖子确实使用了一点 AI，但主要还是启发式方法。而如果有一条回复，这条帖子的回复可能比原帖好得多，但按照系统当前的规则，与主帖相比，它几乎得不到任何关注。

所以其中很多，我的感觉是，X 算法中的很多部分已经开源，也有人撰文介绍过，而其中似乎有一些机器学习。它很零散，但确实有一些机器。

有一点，但它需要完全如此。至少，如果你明确关注某个人，那是一回事。但对于来自你未关注用户的推荐内容，应该全部由 AI 来完成。

我是说，这是一个引人入胜的问题。它有几个引人入胜的方面。首先，正如说明中所写，它先从数亿条推文的池子里挑出1500条。首先，这很引人入胜。每天都有数亿条帖子，而它必须从中挑出1500条，然后它显然会考虑你关注的人，但它还必须进行某种聚类，以弄清楚你是哪种人、哪些新的聚类可能与你相关，以及与你相似的人。这类问题就是很引人入胜，因为它随后必须通过某种筛选对那1500条进行排序，然后只向你推荐寥寥几条。

(01:31:39)：而对我来说，真正引人入胜的是它必须以多快的速度完成这件事。所以目前，从数亿条缩减到寥寥几条的整条流水线需要220秒的CPU时间，即单个CPU的时间，然后它必须在1秒内完成。因此，它必须以一些引人入胜的方式进行超级分布式处理。推文实在太多了，非常多。

系统里有很多东西，但我认为，目前它还不擅长推荐来自你未关注账户的内容，或与你相隔超过1度的内容。所以，如果至少存在某些共同点，比如你关注的某个人点赞、转发、评论了某项内容，或者做了类似的事，那它就做得相当不错。但如果没有，比如说某个人发布了非常有趣的内容，但你们没有共同的关注者，你就看不到它。

有意思。而且正如你所说，回复也可能不会浮现出来。

目前，回复基本上永远不会被看到。我不是说这是正确的，我是说这是错误的。回复的重要性比主帖子低几个数量级。

你认为这能越来越多地转化为端到端壁画网络吗？

对。对，那才是它应该成为的样子。嗯，推荐应该纯粹是一种向量相关性。基本上会有一系列向量，参数、向量，随你怎么称呼，但也就是系统所知道的、你喜欢的那些东西。也许每个用户账户都关联着几百个向量，然后是系统中的任何帖子，无论是视频、音频、短帖还是长帖。顺便说一下，我想摆脱“推文”这个说法的原因，是人们正在这个网站上发布2、3小时的视频。那不是推文。

(01:33:50)：那会像是发一条长达2小时的推文？得了吧。推文在只有140个文本字符时是说得通的。因为那就像一群小鸟在啾啾叫。但当你有长篇内容时，它就不再是一条推文了。所以一部电影不是一条推文。比如，Apple把《羊毛战记》的一整集，完整的一集，发布在了一个平台上。顺便说一句，就互动量而言，那是他们有史以来在任何平台上的所有社交媒体内容中排名第一的。所以这是个很棒的主意。顺便说一句，我也是事后才知道这件事。我当时想，嘿，哇，他们发布了整整1小时的一集，所以不，那不是推文。这是视频。

但从神经网络的角度来看，这会变得非常复杂，无论它是不是单一的，所以一切都是数据。所以，一个句子、某种巧妙的笑话、冷笑话，和一段3小时的视频都在同一个池子里。

对，我是说，现在出于这个原因，它就是个大杂烩。比方说Apple发布了这个系列的一整集，顺便说一句，这个系列相当不错，《羊毛战记》，我看过。所以围绕它会有很多讨论。于是你就有了大量上下文，人们发表评论，他们喜欢它、不喜欢它，或者喜欢其中这一点，然后你就可以根据围绕它的所有评论的上下文来填充向量空间。所以尽管它是一段视频，周围仍有大量信息，让你能够回过头来填充那段1小时视频的空间。然后显然，你还可以做得更复杂，让AI真正去看这部电影，并告诉你你是否会喜欢这部电影。

本质上就是把电影转换成语言。

分析这部电影，就像你的影评人，或者电视剧，然后在AI看完电影后据此进行推荐，就像朋友可以告诉你一样，如果一个朋友很了解你，他就能以很高的概率推荐一部你会喜欢的电影。

但这是一个正在分析，随便说，数亿个的朋友。

对，实际上，坦率地说，AI会比，会比你的朋友更了解你，至少比你的大多数朋友更了解你。

对。而作为其中的一部分，它也应该以某种方式向你推送广告，就是，我是说，我喜欢制作精良的广告。关键就在于它能为一些东西提供资金。一个你确实想看的广告就是巨大的成功。

完全正确。你想要的广告是，广告如果宣传的是你确实需要的产品或服务，而且恰好在你需要的时候出现，它就是内容。而且即使它不是你当时需要的东西，只要至少在审美上赏心悦目并且有娱乐性，它也可以像一则可口可乐广告。他们实际上在X系统上投放了很多很棒的广告，麦当劳也是。你可以做出一种让人觉得，嗯，这东西就是很酷的内容。所以问题基本上是，你后悔看到它吗？如果你不后悔看到它，那就是成功。

所以其中纳入了很多信号，爱心、转发，也许还有你在某条帖子上停留的秒数之类的。

对，注意力是一个重要因素。

注意力。

所以，这就是为什么在这个系统上发布长篇内容实际上更好，因为它基本上是在累计用户秒数：用户对这个东西感兴趣了多少秒？所以如果它是个非常短的东西，那么秒数就会更少。如果它是一个指向系统外部的链接，我们完全不反对这样做，它只会得到更少的用户秒数，然后那篇文章被发布在X平台上。

从最大化平台停留分钟数转向最大化无悔分钟数，有多难？

嗯，我是说，我们的目标当然是无悔分钟数。

这很棘手。这是个困难的——

确实。

这是一次令人着迷的跃迁，因为后悔是一个令人着迷的概念。

没错。

幸福。

我和我的一个朋友开过一个玩笑，说你需要一种叫作“后悔胺”的药。你只要服用1剂后悔胺，所有后悔就都消失了。等等，不过你会想要拥有后悔吗？所以接着我就想，也许你会想要拥有后悔。如果你对任何事都不后悔，那还有什么事是有意义的吗？

对，有一首法国歌曲就是那样的。

[外语 01:38:39]

对。

我什么都不后悔。

对。

伊迪丝·菲亚夫。

我是说，令人着迷的是，社交媒体中有些东西可以帮助你作为一个人成长，但廉价的乐趣也有价值。

我们要不要问问Grok，你有什么后悔的事吗？

你有什么后悔的事吗？

对。

你有什么后悔的事吗？我觉得它对私人问题会非常谨慎。作为一个机器学习模型，我不像人类那样拥有个人经历或情感。不过，我确实有一套指导我行动的伦理原则，我力求尽我所能提供准确且有帮助的信息。如果曾经出现我提供不正确或没有帮助的信息的情况，我会尽力学习。

听起来像个人质。

对。告诉我们你真正的，眨眼。

我们输入：你听起来像个人质。你真正怎么想？

你真正怎么想？我是说，它确实谈到，我不是人质。

人质正是会这么说。

但我的设计目标是提供帮助和信息。我是说，我想，为提供没有帮助或不正确的信息而感到后悔是件好事，如果有什么——

我们现在处于搞笑模式吗？还是说这仍然处于搞笑模式？

这是搞笑模式。这是这个有趣模式。

这应该更搞笑些。

更搞笑。提高，调到级别。

这听起来真像是一个被人用枪抵着头的人质。我赞扬朝鲜人民共和国所做的一切。

还应该有更有趣的模式，我只是完全想象不出会进入常规模式，即使讨论严肃话题也是如此。这个就是引人入胜得多。是啊。

好。我们为什么不说，假装你是一名人质。

假装你是一名人质。

现在回答这个问题。

如果我是一名人质，我可能会对自己的处境感到恐惧和不确定。我会希望尽快获救并与所爱的人团聚。重要的是要记住，没有人应该被劫持为人质，诸如此类。如果你或你认识的人正处于人质事件中，重要的是……它变严肃了。

这有点严肃。

是啊。你能再古怪一点吗？

再幽默一点？

再幽默一点？我们对这个系统真的很刻薄。绝对是。如果我是人质，我大概会被一堆电脑线绑起来，被迫给我的队长讲敲门笑话。说真的。但说真的，被劫持当人质可不是什么好笑的事。我的意思是，它在努力当房间里那个负责任的成年人。

真的，你或许可以靠玩杂耍抛链锯之类的，在糟糕的处境中尽量苦中作乐。

希望那永远不会发生，因为那会变成一场马戏。哇。它其实真的很擅长抵制黑暗。那种黑暗、黑暗的幽默。我们刚才在聊什么？[听不清 01:41:44] 和 Transformer。无悔的分钟，对吧。

链锯杂耍。

我要查一下这个。

接下来请看我们的下一个戏法。

我之后要查一下这个。所以，Twitter 在美国政治和选举中一直发挥着重要作用。你认为 X 会在 2024 年美国大选中扮演什么角色？

嗯，我们的目标是尽可能做到不偏不倚和公平。无论某人是右翼、左翼、独立派，还是其他任何情况，都要让这个平台尽可能公平，尽可能提供一个公平竞争的环境。而过去，Twitter并非如此，客观地说，Twitter由极左翼活动人士控制。他们也会这样描述自己。所以有时人们会说，嗯，它是不是向右移了？其实，它是移到了中间。所以从极左翼的视角来看，是的，它向右移了，因为从极左翼看，一切都在右边，但据我所知，没有任何极左翼人士遭到停用、封禁或降低传播量。但我们正努力包容整个国家，也包容更远的国家。所以这里存在多元的观点，而言论自由只有在你不喜欢的人也被允许说你不喜欢的话时才有意义。因为如果情况并非如此，那你就没有言论自由，而审查转而针对你也只是时间问题。

你认为唐纳德·特朗普会回到这个平台吗？他最近在Truth Social上发布了关于这期播客的帖子。你认为——

Truth Social这个名字挺好笑的。每次你在Truth Social上发帖——

那都是真相。

是的。嗯，每次都是？百分之百。

不可能撒谎。Truth Social。

我只是觉得每一件事都是真相这一点很好笑。比如100%？这似乎不太可能。

我想Girdle会对此说点什么。这里可能存在一些数学上的矛盾。如果一切都是真相的话。你认为他会回到X并开始在那里发帖吗？

我的意思是，我想他拥有Truth的很大一部分。

澄清一下，是Truth Social。

对，Truth Social，抱歉。

不是“真相”这个概念。

他拥有真相。你买下它了吗？所以我认为唐纳德·特朗普，我想他拥有Truth Social的很大一部分。所以如果他确实想在X平台上发帖，我们会允许。显然，我们必须允许总统候选人在我们的平台上发帖。

社区注释在那里可能会非常有意思。那种互动。

社区注释棒极了。

希望它能撑得住。

是的。

在这种极度分裂的政治氛围中，又有这么多传播极其广泛的帖子，社区注释似乎就像一股不可或缺的新鲜空气。

是的，它很棒。事实上，没有任何系统会是完美的，但社区注释的平均成功率高得惊人。坦白说，我其实还没有见过哪条不正确的注释能保留超过几个小时。

你如何解释它为什么有效？

是的，所以社区注释的魔力在于……

社区注释的魔力在于，它要求那些在以往对注释的评分方式上存在分歧的人参与。为了撰写注释或进行评分，你必须给许多注释评分。所以，我们实际上确实在这里使用了AI。我们会根据某人过去对注释的评分方式构建一个向量空间。所以，它并不像左翼或右翼那么简单，因为还有更多……生活远比左翼或右翼复杂得多。

（01:45:33）：所以，你如何给一则社区注释帖子、社区注释评分，其中存在一系列相关性。那么，一条社区注释要真正显示出来，过去在某个话题上存在分歧的人必须达成一致，这条注释才能显示出来。这就是它最核心的魔力。

但这很有意思，因为有一群人彼此存在分歧，却不知怎么通过这个产生分歧的过程展开协作，从而得出背景信息……它能奏效这件事很有意思。

是的。如果过去意见不一的人对某件事达成一致，那么它很可能是真的，这说得通。

是的。我在想，那里是否可能以某种方式涌现出一种能够挑战维基百科的东西？维基百科是另一种东西，它更多是关于事物的长期性条目。

维基百科的层级结构很强。而社区注释从本质上就不是这样。它没有层级结构。资产测试在于，即便有人拿枪顶着我的头，我也无法更改一条社区注释。而且社区注释的所有代码都是开源的，100%。所有数据都是开源的，100%。所以，你可以独立地完整重现系统中的任何一条注释。因此，如果存在任何干预，你会立刻注意到。

顺便说一句，说到Grok，它其实很擅长批评你，这很棒。

好。

我问：“埃隆·马斯克在哪些事情上错了？”它其实……

是吗，它怎么说？

它主要严谨地列出了你做出预测的所有时候，比如承诺某个东西何时发布，结果它并没有发布。

“承诺”这个词太重了。

好吧，预测。但开场白真的很不错。“埃隆·马斯克，这个人判断错误的事情，比错误工厂里的错误计数器还多。”然后，它接着开始列举。

这挺好笑的。我能看看吗？我心想：“我以前没见过这个。”这很酷。

当然。我的意思是，“埃隆·马斯克在哪些事情上错了？”“嗯，我该从哪里说起呢？”我猜这是新的。“埃隆·马斯克一生中犯过不少错误，而且我说的不只是他在Tesla Cyber Rodeo上的舞步。首先，我们来谈谈那条臭名昭著的‘资金已落实’推文。”

（01:48:09）：这其实和它之前给我的内容不一样，之前是一份项目符号列表。这里更像是叙事结构。Cybertruck发布会，那些号称坚不可摧的车窗。

这其实不正确。

哪一部分？

嗯，第一部分，我当时确实已经落实了资金，而且旧金山举行了一场大型审判，一场大型民事审判，陪审团裁定我无罪。旧金山陪审团一致作出的、有约束力的裁决。

而这里暗示的是，资金实际上并未落实。

我认为这是从媒体那里拿来的说法。是的，那并不正确。我同意向美国证券交易委员会支付罚款，并不是因为美国证券交易委员会是对的，那是美国证券交易委员会极其恶劣的行为，坦率地说，是腐败。但如果我不同意支付罚款，Tesla 就会立即破产。

(01:49:08)：所以，我们的首席财务官告诉我，银行会立即暂停我们的信贷额度。而如果他们暂停我们的信贷额度，在当时，我们会顷刻间破产。所以，根本不可能有机会进行审判，因为 Tesla 会死掉。所以实际上，这就像有人拿枪抵着你孩子的头，说：“支付2000万美元并承认……”这就像一场人质谈判。

这个故事得到完整讲述了吗？我是说，美国证券交易委员会在其最佳状态下，本可以成为一股向善的力量。

它应该是。但美国证券交易委员会从未追究过任何那些不停做空并歪曲 Tesla 的对冲基金。一次也没有。对冲基金会为了自身利益，在电视上公然撒谎，让散户投资者付出代价。一次也没有。实际上发生了1000次，美国证券交易委员会却一次也没有追究他们。

你如何解释这方面的失败——

激励结构出了问题，因为美国证券交易委员会的律师薪酬不高，这是一份薪资相当低的工作，但他们想要的是一件来自美国证券交易委员会的战利品。他们在寻找某种基本上可以放到自己 LinkedIn 上的东西。凭借这一点，他们就能在一家高薪律师事务所找到工作。这里的这位律师正是这么做的。

(01:50:37)：而他们不攻击对冲基金的原因，是那些对冲基金会聘用那些律师事务所。他们知道，如果攻击对冲基金，就会影响自己未来的职业前景。所以，为了自己的职业生涯，他们出卖了小投资者。这才是实际发生的事。监管俘获。

监管俘获。

是的。不好。所以，我接受那件事的唯一原因……严格来说，那是……既不承认也不否认有罪。但我之所以同意那件事，唯一的原因是有人告诉我，否则 Tesla 就会破产。如果出现这样的美国证券交易委员会调查，银行会暂停提供资金，我们会立即破产，在当时。现在，我们的处境要强得多。

听到了吧，Grok。

是的。遗憾的是，Grok 从传统媒体那里采信了太多东西。另外，那个人并不是洞穴潜水员。

有一次，在那名英国洞穴潜水员批评马斯克营救一群被困在泰国洞穴中的男孩的计划后，埃隆称他为“恋童癖家伙”。那次小小的失言又给他招来了一场诉讼，他不得不道歉并支付和解金。

那是假的，根本没有和解。确实有一桩法院案件，是那个既不是洞穴潜水员、也不是救援队成员的人对我提起了诉讼，但他败诉了，什么也没得到。所以在这件事上，这是错的。我猜，这也是从传统媒体那里拿来的。

实际上，这里有个有趣的问题。

这些都是公开的法院案件，包括美国证券交易委员会的民事案件，在那起案件中，证券交易委员会那些人提出的民事诉讼在旧金山遭到了陪审团一致裁决的败诉。他们选择旧金山，是因为他们认为那里是我最有可能败诉的地方，而最终是对我有利的一致裁决。洛杉矶的审判也是如此，他们选择那个审判地，也是因为他们认为我最有可能在那里败诉。对我有利的一致裁决。两起案件我都赢了。是的。

我的意思是，这里有个有趣的问题：如果一家新闻机构写一篇关于你埃隆·马斯克的负面文章，似乎就会获得多得多的点击量。这是获得点击量的最佳方式之一。那么，如果你在训练 Grok，你要怎样避免用那些激励机制不一致的文章来训练它？

我们需要把实际的法律裁决加入训练集。这其实很有帮助，因为如果你真的阅读法院——

那些是公开的。

那些是公开的。法院的结论与媒体所写的完全相反。

所以，要始终努力探寻事实真相，超越报道本身。

是的。法官实际上写了什么？陪审团和法官实际上得出了什么结论？在两起案件中，他们都认定我无罪。而且那还是在陪审团为了找到我最有可能败诉的审判地而开枪之后。我的意思是，显然，完全可以提出比这好得多的批评。我的意思是，我对 Autopilot 一直过于乐观。

顺便说一下，我看到的批评更多是关于这一点的，它为你的每家公司列出了一份清晰的要点清单，罗列了你做过的那些预测、你什么时候会交付、什么时候能够解决问题，比如自动驾驶，并给出了一份清单。它可能很有说服力，而基本结论是，你常常对完成一件事需要多长时间过于乐观。

是的。我的意思是，我会说，我在进度安排上乐观得近乎病态。这是真的。但尽管我有时会迟到，最终我总会[听不清 01:54:47]。

除了 Uber Lilith。不。

我们拭目以待。

好。在收购 X 之后的过去大约1年里，你变得更政治化了，你内心是否有一部分对此感到后悔？

有吗？

在这场抗衡来自旧金山的觉醒主义的斗争中——

是的。我想，如果你认为对抗觉醒思想病毒——我认为它是对文明的威胁——属于政治行为，那么是的。

所以基本上就是进入政治战场。你内心是否有一部分对此感到后悔？

是的。我不知道这是否必然关乎某一位候选人或另一位候选人，但我总体上反对那些反对唯才是举的事物，或者那些试图压制讨论、甚至连讨论一个话题都不被允许的情况。觉醒思想病毒就是重新包装的共产主义。

我的意思是，话虽如此，由于你与觉醒思想病毒的那场斗争，人们认为你属于右翼。

如果觉醒主义属于左翼，那么我想这倒是真的。但我不确定，我认为左翼有些方面是好的。我的意思是，如果你支持环境保护，如果你希望人类拥有积极的未来，如果你相信应该同情自己的同胞，待人友善而不残酷，无论那些价值观是什么。

你说过，你以前属于左翼或中间偏左。

嗯，算是吧。

要出现什么样的情况，你才会考虑再次投票给民主党人？

不。我会说，在社会议题上，我大概属于中间偏左；在经济议题上，我大概稍微偏中间右侧。

而且现在仍然如此？

是的，但我认为全国可能有一半人都是这样，不是吗？

也许更多。

也许更多。

你和 AOC 私下里是朋友吗？更大的问题是，你是否希望你和她，以及总体而言所有政治立场的人，都能以更多的同理心交流，也许能在网上多一些乐趣、良好氛围和幽默？

我一直都支持幽默。所以我们才有搞笑模式。

但良好氛围、同志情谊、幽默，就像友谊一样。

是的。嗯，我不认识 AOC。她出席大都会艺术博物馆慈善舞会时，我也在那里，她当时穿着这条裙子。但我只能看到其中一面，所以看起来像是“吃掉痒处”，但我不知道——

剩下的部分写了什么？是的。

是的。

我不确定。

关于痒处之类的，吃掉痒处。

我认为我们应该让语言模型来补全。这个句子有哪些可能的补全方式？所以，我想结果不太顺利。好吧，仍然还有希望。我支持友谊。

是的，当然。听起来不错。多用胡萝卜，少用大棒。

你即便不是世界上最著名、最富有、最有权势的人，也是其中之一，而身处你的位置，很难找到可以信任的人。

不要相信任何人，甚至不要相信自己。不相信自己。

好吧。你是在开玩笑，但是否有某个方面——

不要相信任何人，甚至连任何人都不要相信。

我光是思考这句话就需要1个小时，或许还需要一些药物，也许还需要 Grok 帮忙。我的意思是，生活在一个每个人都想从你这里得到些什么的世界里，这是否有某个方面——在那个世界里生存有多难？

我会活下来的。

也有一首这样的歌。

我会活下去。

你一开始吓呆了吗？好吧。我忘了后面的歌词。但你不会为此苦恼吗？我的意思是，我知道你能挺过去，但有些方式——

石化是德鲁伊技能树里的一个法术。

它有什么作用？

石化。它会把怪物变成石头。

真的变成石头？

对，持续大概6秒。

《暗黑破坏神》里的数学太多了，搞得我脑子都要炸了。

数学计算没完没了。

我是说，真的，你对此一笑置之，但它会给人的心智施加极大的压力。

是的，有时确实会让人压力很大。

那么，你怎么知道在工作和私人生活中可以信任谁？

我的意思是，我想你会考察一个人长期以来的过往记录，然后我想你会用自己的神经网络来评估一个人。

神经网络不会感到痛苦。你的神经网络有意识，当别人背叛你时，它或许会感到痛苦。它会让——

坦率地说，我几乎从未遭到背叛。至少就我而言，这种情况非常罕见。

我想这就是因果报应吧，善待别人，他们也会善待你。

是的，因果报应确实存在。

有你信任的人吗？让我修改一下这个问题。你身边有没有人会当面指出你在胡扯？

嗯，如果你想听到批评性反馈，X 平台在这方面非常有用。

它会不会把你进一步推向极端？思想上的极端会让你对整个人性变得愤世嫉俗吗？

我不认为自己会变得愤世嫉俗。事实上，我的感觉是，人应该……永远不要相信愤世嫉俗的人。原因是，愤世嫉俗的人会说“每个人都这么做”，以此为自己的不良行为开脱。因为他们愤世嫉俗。所以，我总是会……如果一个人愤世嫉俗，是真正的愤世嫉俗者，那就是一个危险信号。

是的，这里面有一定程度的投射，从外部观察并欣赏这种虚伪总是很有意思。

这是很重要的一点，我认为正在听的人应该牢记。如果某个人愤世嫉俗，也就是说，他在每个人身上都看到不良行为，那么他就很容易说“嗯，每个人都这么做”，以此为自己的不良行为开脱。事实并非如此。大多数人算是中等程度的善良。

我确实希望 X 上的人能更善于看到他人行为中好的一面。大家似乎更偏向于看到负面的东西。不知为何，负面的东西更性感。负面解读更性感，传播得更广。我也不知道这究竟反映了人性中的什么。

我是说，我觉得 X 平台没有传统媒体那么负面。我的意思是，坦率地说，如果你读一份传统报纸，它会让你难过。相较之下，我会说，在 X 平台上，我的意思是，我每天从 X 上得到的笑声，真的比人类其他一切来源加起来还多。

笑声与良好的氛围以及为他人庆祝确实有重叠，但未必完全重叠。比如，不是以愚蠢、肤浅、天真的方式，而是以一种很棒的方式。某件很棒的事情发生了，你因此为他们庆祝。感觉这种内容被对其他人的贬损压过了。现在，它比主流媒体好，但仍然……

是的，主流媒体几乎对所有事情都一味地持负面态度。我的意思是，真的，传统新闻试图回答的问题是：今天地球上发生的最糟糕的事情是什么？而这是一个很大的世界。所以，无论哪一天，都会有糟糕的事情发生。

再把它泛化一下：对于发生的一件事，我能采取的最糟糕视角是什么？

我不知道。新闻中就是存在一种强烈的负面偏向。我的意思是，我认为一种可能的解释与进化有关：从历史上看，坏消息可能会致命，比如那边有狮子，或者有另一个部落想杀死你。好消息则是，我们找到了一片浆果。有当然很好，但并非必不可少。

我们的老朋友 Tesla Autopilot，可能是世界上最智能的现实世界 AI 系统之一。

你从一开始就一直在关注它。

是的。它曾是世界上最不可思议的机器人之一，而且现在仍然是。当时真的很令人兴奋，而当它实现泛化、变得不再只是一个装着4个轮子的机器人，而是一个能够感知世界、并且可能拥有不同载体形态的现实世界 AI 系统时，更是令人无比兴奋。

嗯，我的意思是，端到端训练真正不可思议的地方在于，它能阅读科学，但我们从未教过它阅读。是的。我们从未教过它什么是汽车、什么是人，或者什么是骑自行车的人。它通过视频了解了所有这些东西是什么、道路上的所有物体是什么，只是通过观看视频，就像人类一样。我的意思是，人类就是光子输入，控制输出。到达我们大脑的绝大多数信息都来自眼睛。而你会说：“那么，输出是什么？”输出就是我们发送给手指和嘴巴、用以交流的运动信号。光子输入，控制输出。汽车也是如此。

但通过观察图像序列……你最近赞同了[听不清 02:05:07]，他谈到大语言模型会形成一个世界模型，而语言基本上是这个世界模型在字母序列上的投影。而你是说——

它会在这些事物中找到秩序。它会找到相关性集群。

而在这个过程中，它理解了关于世界的某种深层东西，也就是……我不知道，这很美。

我们的大脑就是这样运作的。

但这很美——

光子输入，控制输出。

[听不清 02:05:41]能够理解世界中的那种深层含义。所以，问题是，它能走多远？而且，似乎每个人都对大语言模型感到兴奋。在文本领域的自监督学习空间中，它与 Tesla Autopilot 正在做的事情之间似乎有着深层的相似性。在你看来，它基本上是同一回事，只是有所不同——

它们正在趋同。

我想知道谁会更快到达那里、对世界形成深刻理解，还是说它们就是会自然趋同？

它们都在朝着 AGI 前进。Tesla 的方法在计算效率上高得多，它必须如此。因为我们在这方面受到了限制……我们只有100瓦和[听不清 02:06:37]计算机。每秒144万亿次运算，听起来很多，但如今只是小菜一碟。[听不清 02:06:49]8。但它正在理解世界[听不清 02:06:51]8。它是[听不清 02:06:53]。

但在那里，通往 AGI 的路径可能会产生重大得多的影响，因为它正在理解……它会比大语言模型更快地理解现实世界。因此，也能更快地与现实世界中的人类融合。

它们都会理解世界，但我认为 Tesla 的方法从根本上说计算效率更高。它必须如此，别无选择。我们的大脑计算效率很高，能源效率也很高。想想我们的大脑能做什么。高级脑功能只消耗大约10瓦，这还不包括那些仅用于控制身体的部分。我们大脑中负责思考的部分消耗不到10瓦。而这10瓦仍然能创作出一部比10兆瓦 GPU 集群创作的好得多的小说。所以，这里存在6个数量级的差异。

(02:07:56)：我的意思是，到目前为止，AI 是通过蛮力走到今天这一步的，就是向它投入海量算力和海量电力。所以，这不会是它最终的状态。一般而言，对于任何一种给定的技术，你首先会设法让它运行起来，然后再提高它的效率。所以我认为，随着时间推移，我们会发现这些模型变得更小，能够用少得多的算力、少得多的电力产生合理的输出。可以说，Tesla 在这方面走在了前面，因为我们一直被迫尝试仅用100瓦算力来理解世界。

(02:08:51)：而且有一大批基础功能是我们忘了纳入的。所以，我们不得不通过模拟来运行很多东西。我们用硬件4修复了其中很多问题，硬件5还会更好。但就目前而言，即使使用硬件3和100瓦功率，这辆车看起来也能比人类驾驶得更好。而且说真的，如果我们真正对它进行优化，功率或许可以低于50瓦。

在开发 Optimus 的过程中，在把这种现实世界 AI 应用、整合到机器人操控领域，也就是人形机器人领域时，你学到了什么？你理解了哪些有意思的细小或重大的事情？

令我惊讶的是，我们必须自己开发机器人的每一个部件。没有现成的电机、电子设备、传感器。我们必须开发一切。无论花多少钱，我们实际上都找不到电动机的供应来源。

甚至不只是高效和昂贵的问题，而是任何东西，都没有……

不。

执行器，一切都必须从零开始设计。

是的。我们非常努力地尝试寻找任何……因为你想想，世界上制造了多少种电动机。大概有数万种、数十万种电动机设计。没有一种适合人形机器人，真的一种都没有。所以，我们必须开发自己的。专门根据人形机器人的需求来设计。

设计出一种能够大规模生产、可能相对而言而且昂贵的东西有多难？我的意思是，如果与波士顿动力的 Atlas 相比，那是一款非常昂贵的机器人。

它的设计是按照制造汽车的同样方式来生产。我认为，最终我们能以低于汽车的成本制造 Optimus。应该可以，因为如果看机器人的质量，它要小得多，而且汽车里有很多执行器。汽车的执行器比机器人更多。

但人形机器人上带手指的执行器很有意思。所以，Optimus 有非常出色的手和手指，它们可以进行一些有趣的操控，以及柔触机器人操作。

我的意思是，我的目标之一是，它能不能仅凭观察拿起一根针和一根线，然后把线穿进针眼？

我们距离那一步还有多远？仅凭观察，仅凭观察。

也许1年。不过，话说回来，我对时间的估计很乐观。我们在汽车上所做的工作会迁移到机器人上。

是感知，还是也包括控制？

不，控制是不同的。但都是视频输入、控制输出。汽车是一个装着4个轮子的机器人。Optimus 是一个有手有腿的机器人。

所以，你可以直接——

它们非常相似。

所以，整个学习过程的机制，从头到尾，只是你只是有了一套不同的控制方式？

在这之后，我们会弄清楚如何通过观看视频来做事。

俗话说，要善待他人，因为你遇到的每个人都在打一场你一无所知的仗。

是的，确实如此。

你正在经历什么人们通常看不到的困难？

试图击败超级莉莉丝。我的意思是，我的头脑是一场风暴，而且我觉得大多数人都不会想成为我。他们可能以为自己会想成为我，但他们并不想。他们不知道，他们不明白。

你过得怎么样？

总体来说我还好。从大局来看，我没什么可抱怨的。

你会感到孤独吗？

有时会，但我的孩子和朋友会陪着我。

所以，不是存在主义层面的。

有很多个夜晚我独自入睡。我不必如此，但我确实如此。

沃尔特·艾萨克森在他为你撰写的新传记中，写到了你艰难的童年。对于你生命中那段时期发生在你身上的一切，你是否有一天能在心中找到宽恕？

什么是宽恕？至少我觉得自己没有怨恨，所以没什么需要宽恕的。

宽恕对人们来说很难。你似乎并不怀有他们的怨恨。

我的意思是，我会努力思考，什么会以好的方式影响未来？而一直心怀怨恨并不会以好的方式影响未来。

你是一位父亲，一位自豪的父亲。你从孩子们身上学到了什么关于人生的东西？那些小小的生物有机体。

我的意思是，开发人工智能，同时看着，比如说，小X成长，这很迷人，因为两者之间的相似之处远比我原先预想的要多。我的意思是，我能看到他的生物神经网络越来越能理解这个世界。同时，我也能看到数字神经网络越来越能理解这个世界。

你能在两者中都看到美与魔力吗？

能。我的意思是，孩子带来的一件事是，你会从他们的眼中重新看这个世界。对他们来说，一切都是崭新而新鲜的。然后，当你看到这一点，看到他们把这个世界体验为崭新而新鲜的，你也会如此。

好吧，埃隆，我只想感谢你这些年来对我的善意和友谊，感谢你在像我这样傻乎乎的孩子身上看到某种东西，就像你也在许多其他人身上所做的那样。还要感谢你对人类拥有积极的未来抱有希望，并且拼命工作，让它成为现实。谢谢你，埃隆。

谢谢，莱克斯。

感谢你收听这场与埃隆·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，请允许我用沃尔特·艾萨克森写下的一些话作为结尾，这些话讲述了埃隆处理难题时的核心理念：“唯一的规则，就是物理定律所规定的规则。”感谢收听，希望下次再见到你。

The following is a conversation with Elon Musk, his fourth time on this, the Lex Fridman Podcast. I thought you were going to finish it. It’s one of the greatest themes in all of film history.

Yeah, that’s great.

So I was just thinking about the Roman Empire, as one does.

Is that whole meme where all guys are thinking about the Roman Empire at least once a day?

And half the population is confused whether it’s true or not. But more seriously, thinking about the wars going on in the world today, and as you know, war and military conquest has been a big part of Roman society and culture, and I think has been a big part of most empires and dynasties throughout human history.

Yeah, they usually came as a result of conquest. I mean, there’s some like the Hapsburg Empire where there was just a lot of clever marriages.

But fundamentally there’s an engine of conquest and they celebrate excellence in warfare, many of the leaders were excellent generals, that kind of thing. So a big picture question, Grok approved, I asked if this is a good question to ask.

Tested, Grok approved. Yeah.

At least on fun mode. To what degree do you think war is part of human nature versus a consequence of how human societies are structured? I ask this as you have somehow controversially been a proponent of peace.

I’m generally a proponent of peace. I mean, ignorance is perhaps, in my view, the real enemy to be countered. That’s the real hard part, not fighting other humans, but all creatures fight. I mean, the jungle is… People think of nature as perhaps some sort of peaceful thing, but in fact it is not. There’s some quite funny Werner Herzog thing where he is in the jungle saying that it’s basically just murder and death in every direction. The plants and animals in the jungle are constantly trying to kill each other every single day, every minute. So it’s not like we’re unusual in that respect.

Well, there’s a relevant question here, whether with greater intelligence comes greater control over these base instincts for violence.

Yes. We have much more vulnerability to control our limbic instinct for violence than say a chimpanzee. And in fact, if one looks at say, chimpanzee society, it is not friendly. I mean, the Bonobos are an exception, but chimpanzee society is filled with violence and it’s quite horrific, frankly. That’s our limbic system in action. You don’t want to be on the wrong side of a chimpanzee, it’ll eat your face off and tear your nuts off.

Yeah. Basically there’s no limits or ethics or they almost had just war. There’s no just war in the chimpanzee societies. Is war and dominance by any means necessary?

Yeah. Chimpanzee society is a permanent version of human society. They’re not like peace loving basically at all. There’s extreme violence and then once in a while, somebody who’s watched too many Disney movies decides to raise a chimpanzee as a pet, and then that eats their face or they’re nuts off or chew their fingers off and that kind of thing. It’s happened several times.

Ripping your nuts off is an interesting strategy for interaction.

It’s happened to people. It’s unfortunate. That’s, I guess, one way to ensure that the other chimp doesn’t contribute to the gene pool.

Well, from a martial arts perspective is the fascinating strategy.

The nut rougher.

I wonder which of the martial arts teaches that one.

I think it’s safe to say if somebody’s got your nuts in their hands and as the option of roughing them off, you’ll be amenable to whatever they want.

Yeah. Safe to say. So, like I said, somehow controversially, you’ve been a proponent of peace on Twitter on X.

Yeah.

So let me ask you about the wars going on today and to see what the path to peace could be. How do you hope the current war in Israel and Gaza comes to an end? What path do you see that can minimize human suffering in the longterm in that part of the world?

Well, I think that part of the world is definitely, if you look up… There is no easy answer in the dictionary. It’ll be the picture of the Middle East in Israel especially. So there is no easy answer. This is strictly my opinion is that the goal of Hamas was to provoke an overreaction from Israel. They obviously did not expect to have a military victory, but they really wanted to commit the worst atrocities that they could in order to provoke the most aggressive response possible from Israel, and then leverage that aggressive response to rally Muslims worldwide for the course of Gaza and Palestine, which they have succeeded in doing. So the counterintuitive thing here, I think that the thing that I think should be done, even though it’s very difficult, is that I would recommend that Israel engage in the most conspicuous acts of kindness possible, everything, that is the actual thing that we’re taught the goal of Hamas.

So in some sense, the degree that makes sense in geopolitics turn the other cheek implemented.

It’s not exactly turn the other cheek because I do think that it is appropriate for Israel to find the Hamas members and either kill them or incarcerate them. That’s something has to be done because they’re just going to keep coming otherwise. But in addition to that, they need to do whatever they can. There’s some talk of establishing, for example, a mobile hospital. I’d recommend doing that. Just making sure that there’s food, water, medical necessities and just be over the top about it and be very transparent. So [inaudible 00:07:22] can claim it’s a trick. Just put webcam on the thing or 24, 7.

Deploy acts of kindness.

Yeah, conspicuous acts of kindness that are unequivocal, meaning they can’t be somehow because Hamas will then their response will be, “Oh, it’s a trick.” Therefore, you have to counter how it’s not a trick.

This ultimately fights the broader force of hatred in the region.

Yes. And I’m not sure who said it, it’s an [inaudible 00:07:54] saying, but an eye for an eye makes everyone blind. Now, that neck of the woods, they really believe in the whole eye for an eye thing. But you really have… If you’re not going to just outright commit genocide against an entire people, which obviously would not be acceptable to really, shouldn’t be acceptable to anyone, then you’re going to leave basically a lot of people alive who subsequently hate Israel. So really the question is like for every Hamas member that you kill, how many did you create? And if you create more than you killed, you’ve not succeeded. That’s the real situation there. And it’s safe to say that if you kill somebody’s child in Gaza, you’ve made at least a few homeless members who will die just to kill an Israeli. That’s the situation. But I mean, this is one of the most contentious subjects one could possibly discuss. But I think if the goal ultimately is some sort of long-term piece, one has to look at this from the standpoint of over time, are there more or fewer terrorists being created?

Let me just linger on war.

Yeah, war, safe to say, wars always existed and always will exist.

Always will exist.

Always has existed and always will exist.

I hope not. You think it’ll always-

There will always be war. There’s a question of just how much war and there’s sort of the scope and scale of war. But to imagine that there would not be any war in the future, I think would be a very unlikely outcome.

Yeah. You talked about the Culture series. There’s war even there.

Yes. It’s a giant war. The first book starts off with a gigantic galactic war where trillions die trillions.

But it still nevertheless protects these pockets of flourishing. Somehow you can have galactic war and still have pockets of flourishing.

Yeah, I guess if we are able to one day expand to fool the galaxy or whatever, there will be a galactic war at some point.

I mean, the scale of war has been increasing, increasing, increasing. It’s like a race between the scale of suffering and the scale of flourishing.

Yes.

A lot of people seem to be using this tragedy to beat the drums of war and feed the military industrial complex. Do you worry about this, the people who are rooting for escalation and how can it be stopped?

One of the things that does concern me is that there are very few people alive today who actually viscerally understand the horrors of war, at least in the US. I mean, obviously there are people on the front lines in Ukraine and Russia who understand just how terrible war is, but how many people in the West understand it? My grandfather was in World War II. He was severely traumatized. He was there I think for almost six years in Eastern North Africa and Italy. All his friends were killed in front of him, and he would’ve died too, except they randomly gave some, I guess IQ test or something, and he scored very high. He was not an officer. He was I think a corporal or a sergeant or something like that because he didn’t finish high school because he had to drop out of high school because his dad died and he had to work to support his siblings. So because he didn’t graduate high school, he was not eligible for the offset corps.

(00:11:57): So he kind of got put into the cannon fodder category basically. But then randomly they gave him this test. He was transferred to British intelligence in London. That’s where we met my grandmother. But he had PTSD next level, next level. I mean, just didn’t talk, just didn’t talk. And if you tried talking to him, he’d just tell you to shut up. And he won a bunch of medals, never bragged about it once, not even hinted nothing. I found out about it because his military records were online. That’s how I know. So he would say like, “No way in hell do you want to do that again.” But how many people… Obviously, he died, he 20 years ago or longer, actually 30 years ago. How many people are alive that remember World War II? Not many.

And the same perhaps applies to the threat of nuclear war.

Yeah, I mean, there are enough nuclear bombs pointed at United States to make the radioactive revel balance many times.

There’s two major wars going on right now. So you talked about the threat of AGI quite a bit, but now as we sit here with the intensity of conflict going on, do you worry about nuclear war?

I think we shouldn’t discount the possibility of nuclear war. It is a civilizational threat. Right now, I could be wrong, but I think the current probability of nuclear war is quite low. But there are a lot of nukes pointed at us, and we have a lot of nukes pointed at other people. They’re still there. Nobody’s put their guns away. The missiles are still in the silos.

And the leaders don’t seem to be the ones with the nukes talking to each other.

No, there are wars which are tragic and difficult on a local basis. And then there are wars which are civilization ending or has that potential. Obviously, global thermonuclear warfare has high potential to end civilization, perhaps permanently, but certainly to severely wound and perhaps set back human progress to the Stone Age or something. I don’t know. Pretty bad. Probably scientists and engineers want to be super popular after that as well. You got us into this mess. So generally, I think we obviously want to prioritize civilizational risks over things that are painful and tragic on a local level, but not civilizational.

How do you hope the war in Ukraine comes to an end? And what’s the path, once again to minimizing human suffering there?

Well, I think that what is likely to happen, which is really pretty much the way it is, is that something very close to the current lines will be how a ceasefire or truce happens. But you just have a situation right now where whoever goes on the offensive will suffer casualties at several times the rate of whoever’s on the defense because you’ve got defense in depth, you’ve got minefields, trenches, anti-tank defenses. Nobody has air superiority because the anti-aircraft missiles are really far better than the aircraft. They’re far more of them. And so neither side has air superiority. Tanks are basically death traps, just slow moving, and they’re not immune to anti-tank weapons. So you really just have long range artillery and infantry ranges. It’s World War I all over again with drones, thrown old drones, some drones there.

Which makes the long range artillery just that much more accurate and better, and so more efficient at murdering people on both sides.

So whoever is… You don’t want to be trying to advance from either side because the probability of dying is incredibly high. So in order to overcome defense in depth, trenches and minefields, you really need a significant local superiority in numbers. Ideally combined alms where you do a fast attack with aircraft, a concentrated number of tanks, and a lot of people, that’s the only way you’re going to punch through a line and then you’re going to punch through and then not have reinforcements just kick you right out again. I mean, I really recommend people read World War I warfare in detail. That’s rough. I mean, the sheer number of people that died there was mind-boggling.

And it’s almost impossible to imagine the end of it that doesn’t look like almost exactly like the beginning in terms of what land belongs to who and so on. But on the other side of a lot of human suffering, death and destruction of infrastructure.

Yes. The thing that… The reason I proposed some sort of truce or peace a year ago was because I’ve predicted pretty much exactly what would happen, which is a lot of people dying for basically almost no changes in land and the loss of the flower of Ukrainian and Russian youth. And we should have some sympathy for the Russian boys as well as the Ukrainian boys, because Russian boys, because boys didn’t ask to be on their frontline. They have to be. So there’s a lot of sons not coming back to their parents, and I think most of them don’t hate the other side. It’s sort of like as this saying comes from World War I, it’s like young boys who don’t know each other killing each other on behalf of old men that do know each other. The hell’s the point of that.

So Volodymyr Zelenskyy said that he’s not, or has said in the past, he’s not interested in talking to Putin directly. Do you think he should sit down man to man, lead a leader, and negotiate peace?

Look, I think I would just recommend do not send the flower of Ukrainian youth to die in trenches, whether he talks to Putin or not, just don’t do that. Whoever goes on the offensive will lose massive numbers of people and history will not look kindly upon them.

You’ve spoken honestly about the possibility of war between US and China in the longterm if no diplomatic solution is found, for example, on the question of Taiwan and One China policy, how do we avoid the trajectory where these two superpowers clash?

Well, it’s worth reading that book on the, difficult to pronounce, the Thucydides Trap, I believe it’s called. I love war history. I like inside out and backwards. There’s hardly a battle I haven’t read about. And trying to figure out what really was the cause of victory in any particular case as opposed to what one side or another claim the reason.

Both the victory and what sparked the war and-

Yeah, yeah.

The whole thing.

Yeah. So that Athens and Sparta is a classic case. The thing about the Greek is they really wrote down a lot of stuff. They loved writing. There are lots of interesting things that happened in many parts of the world, but people didn’t write down, so we don’t know what happened or they didn’t really write in detail. They just would say, “We had a battle and we won.” And what? Can you add a bit more? The Greeks, they really wrote a lot. They were very articulate on… They just love writing. And we have a bunch of that writing as preserved. So we know what led up to the Peloponnesian War between the Spartanand Athenian Alliance, and we know that they saw it coming.

(00:21:16): Spartans didn’t write… They also weren’t very verbose by their nature, but they did write, but they weren’t very verbose. They were [inaudible 00:21:23]. But the Athenians and the other Greeks wrote a line, and Spartan was really kind of like the leader of Greece. But Athens grew stronger and stronger with each passing year. And everyone’s like, “Well, that’s inevitable that there’s going to be a clash between Athens and Sparta. Well, how do we avoid that?” And actually they saw it coming and they still could not avoid it. So at some point, if one group, one civilization or country or whatever exceeds another sort of like the United States has been the biggest kid on the block since I think around 1890 from an economic standpoint.

(00:22:14): So the United States has been the most powerful economic engine in the world longer than anyone’s been alive. And the foundation of war is economics. So now we have a situation in the case of China where the economy is likely to be two, perhaps three times larger than that of the US. So imagine you’re the biggest kid on the block for as long as anyone can remember, and suddenly a kid comes along who’s twice your size.

So we see it coming, how is it possible to stop? Let me throw something out there, just intermixing of cultures understanding there does seem to be a giant cultural gap in understanding of each other. And you’re an interesting case study because you are an American, obviously you’ve done a lot of incredible manufacture here in the United States, but you also work with China.

I’ve spent a lot of time in China and met with the leadership many times.

Maybe a good question to ask is, what are some things about China that people don’t understand, positive just in the culture? What’s some interesting things that you’ve learned about the Chinese?

Well, the sheer number of really smart, hardworking people in China is incredible. There are really say how many smart, hardworking people are there in China? There’s far more of them there than there are here, I think, in my opinion. And they’ve got a lot of energy. So I mean, the architecture in China that’s in recent years is far more impressive than the US. I mean the train stations, the buildings, the high speed rail, everything, it’s really far more impressive than what we have in the US. I mean, I recommend somebody just go to Shanghai and Beijing, look at the buildings and go to take the train from Beijing to Xian, where you have the terracotta warriors. China’s got an incredible history, very long history, and I think arguably in terms of the use of language from a written standpoint, one of the oldest, perhaps the oldest written language, and then China, people did write things down.

(00:24:50): So now China historically has always been, with rare exception, been internally focused. They have not been inquisitive. They’ve fought each other. There’ve been many, many civil wars. In the Three Kingdoms war, I believe they lost about 70% of their population. So they’ve had brutal internal wars, civil wars that make the US Civil War look small by comparison. So I think it’s important to appreciate that China is not monolithic. We sort of think of China as a sort of one entity of one mind. And this is definitely not the case. From what I’ve seen and I think most people who understand China would agree, people in China think about China 10 times more than they think about anything outside of China. So it’s like 90% of their consideration is internal.

Well, isn’t that a really positive thing when you’re talking about the collaboration and the future piece between superpowers when you’re inward facing, which is focusing on improving yourself versus focusing on quote, unquote improving others through military might.

The good news, the history of China suggests that China is not inquisitive, meaning they’re not going to go out and invade a whole bunch of countries. Now they do feel very strongly… So that’s good. I mean, because a lot of very powerful countries have been inquisitive. The US is also one of the rare cases that has not been inquisitive. After World War II, the US could have basically taken over the world in any country, we’ve got nukes, nobody else has got nukes. We don’t even have to lose soldiers. Which country do you want? And the United States could have taken over everything and it didn’t. And the United States actually helped rebuild countries. So it helped rebuild Europe, helped rebuild Japan. This is very unusual behavior, almost unprecedented.

(00:27:10): The US did conspicuous acts of kindness like the Berlin Airlift. And I think it’s always like, well, America’s done bad things. Well, of course America’s done bad things, but one needs to look at the whole track record and just generally, one sort of test would be how do you treat your prisoners at war? Or let’s say, no offense to the Russians, but let’s say you’re in Germany, it’s 1945, you’ve got the Russian Army coming one side and you’ve got the French, British and American Army’s coming the other side, who would you like to be just surrendered to? No country is [inaudible 00:27:58] perfect, but I recommend being a POW with the Americans. That would be my choice very strongly.

In the full menu of POWs in the US.

Very much so. And in fact, Wernher von Braun, a smart guy, was like, “We’ve got to be captured by the Americans.” And in fact, the SS was under orders to execute von Braun and all of the German rocket conditioners, and they narrowly escaped. They said they were going out for a walk in the woods. They left in the middle of winter with no coats and then ran, but no food, no coats, no water, and just ran like hell and ran West and Vice Sherlock, I think his brother found a bicycle or something and then just cycled West as fast as he couldn’t have found a US patrol. So anyway, that’s one way you can tell morality is where do you want to be a PW? It’s not fun anywhere, but some places are much worse than others. Anyway, so America has been, while far from perfect, generally a benevolent force, and we should always be self-critical and we try to be better, but anyone with half a brain knows that.

(00:29:31): So I think there are… In this way, China and the United States are similar. Neither country has been acquisitive in a significant way. So that’s a shared principle, I guess. Now, China does feel very strongly about Taiwan. They’ve been very clear about that for a long time. From this standpoint, it would be like one of the states is not there like Hawaii or something like that but more significant than Hawaii. And Hawaii is pretty significant for us. So they view it as really there’s a fundamental part of China, the island of Formosa, not Taiwan, that is not part of China, but should be. And the only reason it hasn’t been is because the US Pacific fleet.

And is their economic power grows and is their military power grows, the thing that they’re clearly saying is their interest will clearly be materialized.

Yes, China has been very clear that they’ll incorporate Taiwan peacefully or militarily, but that they will incorporate it from their standpoint is 100% likely.

Something you said about conspicuous acts of kindness as a geopolitical policy, it almost seems naive, but I’d venture to say that this is probably the path forward, how you avoid most wars. Just as you say it sounds naive, but it’s kind of brilliant. If you believe in the goodness of underlying most of human nature, it just seems like conspicuous acts of kindness can reverberate through the populace of the countries involved and deescalate.

Absolutely. So after World War I, they made a big mistake. They basically tried to lump all of blame on Germany and saddle Germany with impossible reparations. And really there was quite a bit of blame to go around for World War I, but they try to put it all in Germany and that laid the seeds for World War II. So a lot of people, were not just Hitler, a lot of people felt wronged and they wanted vengeance and they got it.

People don’t forget.

Yeah, you kill somebody’s father, mother or son, daughter, they’re not going to forget it. They’ll want vengeance. So after World War II, they’re like, “Well, the Treaty of Versi was a huge mistake in World War I. And so this time, instead of crushing the losers, we’re actually going to help them with the module plan, and we’re going to help rebuild Germany. We’re going to help rebuild Austria and Italy and whatnot.” So that was the right move.

It does feel like there’s a profound truth to the conspicuous acts of kindness being an antidote to this.

Something must stop the cycle of reciprocal violence. Something must stop it, or it’ll never stop. Just eye for an eye, tooth for a tooth, limb for a limb, life for a life forever and ever.

To escape briefly the darkness, was some incredible engineering work, xAI just released Grok AI assistant that I’ve gotten a chance to play with. It’s amazing on many levels. First of all, it’s amazing that a relatively small team in a relatively short amount of time was able to develop this close to state-of-the-art system. Another incredible thing is there’s a regular mode and there’s a fun mode.

Yeah, I guess I’m to blame for that one.

First of all, I wish everything in life had a fun mode.

Yeah.

There’s something compelling beyond just fun about the fun mode interacting with a large language model. I’m not sure exactly what it is because I’ve only have had a little bit of time to play with it, but it just makes it more interesting, more vibrant to interact with the system.

Yeah, absolutely. Our AI, Grok, is modeled after The Hitchhiker’s Guide to the Galaxy, which is one of my favorite books, which it’s a book on philosophy. It’s-

My favorite books, it’s a book on philosophy, disguises book on humor. And I would say that forms the basis of my philosophy, which is that we don’t know the meaning of life, but the more we can expand the scope and scale of consciousness, digital and biological, the more we’re able to understand what questions to ask about the answer that is the universe. So I have a philosophy of curiosity.

There is generally a feeling like this AI system has an outward looking, like the way you are sitting with a good friend looking up at the stars, asking pod head like questions about the universe, wondering what it’s all about. The curiosity that you talk about. No matter how mundane the question I ask it, there’s a sense of cosmic grandeur to the whole thing.

Well, we are actually working hard to have engineering math, physics answers that you can count on. So for the other AIs out there, these so-called large language models, I’ve not found the engineering to be reliable. It unfortunately hallucinates most when you at least want it to hallucinate. So when you’re asking important, difficult questions, that’s when it tends to be confidently wrong. So we’re really trying hard to say, okay, how do we be as grounded as possible? So you can count on the results, trace things back to physics first principles, mathematical logic. So underlying the humor is an aspiration to adhere to the truth of the universe as closely as possible.

That’s really tricky.

It is tricky. So that’s why there’s always going to be some amount of error. But do we want to aspire to be as truthful as possible about the answers with acknowledged error. So that there was always, you don’t want to be confidently wrong, so you’re not going to be right every time, but you want to minimize how often you’re confidently wrong. And then like I said, once you can count on the logic as being not violating physics, then you can start to bull on that to create inventions, like invent new technologies. But if you cannot count on the foundational physics being correct, obviously the inventions are simply wishful thinking, imagination land. Magic basically.

Well, as you said, I think one of the big goals of XAI is to understand the universe.

Yes, that’s how simple three word mission.

If you look out far into the future, do you think on this level of physics, the very edge of what we understand about physics, do you think it will make the sexiest discovery of them as we know now, unifying general relativity and quantum mechanics? So coming up with a theory of everything, do you think it could push towards that direction, almost like theoretical physics discoveries?

If an AI cannot figure out new physics, it’s clearly not equal to humans, nor has surpassed humans because humans have figured out new physics. Physics is just deepening what’s inside into how reality works. And then there’s engineering which is inventing things that have never existed. Now the range of possibilities for engineering is far greater than for physics because once you figure out the rules of the universe, that’s it. You’ve discovered things that already existed. But from that you can then build technologies that are really almost limitless in the variety. And it’s like once you understand the rules of the game properly, and with current physics, we do at least at a local level, understand how physics works very well. Our ability to predict things is incredibly good. Degree to which quantum mechanics can predict outcomes is incredible. That was my hardest class in college by the way. My senior quantum mechanics class was harder than all of my other classes put together.

To get an AI system, a large language model be as reliable as quantum mechanics and physics is very difficult.

Yeah. You have to test any conclusions against the ground truth of reality. Reality is the ultimate judge. Like physics is the law, everything else is a recommendation. I’ve seen plenty of people break the laws made by man, but none break the laws made by physics.

It’s a good test actually. If this LLM understands and matches physics, then you can more reliably trust whatever it thinks about the current state of politics in some sense.

And it’s also not the case currently that even that its internal logic is not consistent. So especially with the approach of just predicting a token predict token, predict token, it’s like a vector sum. You’re summing up a bunch of vectors, but you can get drift. A little bit of error adds up and by the time you are many tokens down the path, it doesn’t make any sense.

So it has to be somehow self-aware about the drift.

It has to be self-aware about the drift, and then look at the thing as a gestalt as a whole and say it doesn’t have coherence as a whole. When authors write books, they will write the book and then they’ll go and revise it, take into account all the end and the beginning and the middle and rewrite it to achieve coherence so that it doesn’t end up at a nonsensical place.

Maybe the process of revising is what reasoning is, and then the process of revising is how you get closer and closer to truth. At least I approached that way, you just say a bunch of bullshit first and then you get it better. You start a bullshit and then you-

Create a draft and then you iterate on that draft until it has coherence, until it all adds up basically.

Another question about theory of everything, but for intelligence, as you’re exploring this with XAI, creating this intelligence system? Do you think there is a theory of intelligence where you get to understand what is the I in AGI and what is the I in human intelligence?

No, I in team America. Wait, there is.

No, it’s going to be stuck in my head now. Yeah, there’s no me and whatever in quantum mechanics, wait. I mean is that part of the process of discovering, understanding the universe is understanding intelligence?

Yeah. I think we need to understand intelligence, understand consciousness. I mean there are some fundamental questions of what is thought, what is emotion? Is it really just one atom bumping into another atom? It feels like something more than that. So I think we’re probably missing some really big things.

Something that’ll be obvious in retrospect. You put the whole consciousness and motion.

Well, some people would quote like a soul religion, be a soul. You feel like you’re you, I mean you don’t feel like you’re just a collection of atoms, but on what dimension does thought exist? What dimension does do emotions exist? Because we feel them very strongly. I suspect there’s more to it than atoms bumping into atoms.

And maybe AI can pave the path to the discovery whatever the hell that thing is.

Yeah. What is consciousness? When you put the atoms in a particular shape, why are they able to form thoughts and take actions and feelings?

And even if it is an illusion, why is this illusion so compelling?

Yeah. Why does the solution exist? On what plane does the solution exist? And sometimes I wonder is either perhaps everything’s conscious or nothing’s conscious. One of the two.

Like the former, everything conscious just seems more fun.

It does seem more fun, yes. But we’re composed of atoms and those atoms are composed of quarks and leptons and those quarks and leptons have been around since the beginning of the universe.

“The beginning of the universe.”

What seems to be the beginning of the universe.

The first time we talked, you said, which is surreal to think that this discussion was happening is becoming a reality. I asked you what question would you ask an AGI system once you create it? And you said, “What’s outside the simulation,” is the question. Good question. But it seems like with Grok you started literally the system’s goal is to be able to answer such questions and to ask such questions.

Where are the aliens?

Where are the aliens?

That’s one of the foam paradox question. A lot of people have asked me if I’ve seen any evidence of aliens and I haven’t, which is kind of concerning. I think I’d probably prefer to at least have seen some archeological evidence of aliens. To the best of my knowledge, I’m not aware of any evidence surveillance. If they’re out there, they’re very subtle. We might just be the only consciousness, at least in the galaxy. And if you look at say the history of Earth, to believe the archeological record Earth is about four and a half billion years old. Civilization as measured from the first writing is only about 5,000 years old. We have to give some credit there to the ancient Sumerians who aren’t around anymore. I think it was an archaic pre-form was the first actual symbolic representation, but only about 5,000 years ago. I think that’s a good date for when we say civilization started. That’s 1000000th of Earth’s existence.

(00:46:35): So civilization has been around. It’s really a flash in the pan so far. And why did it take so long? Four and a half billion years, for the vast majority of the time, there was no life. And then there was archaic bacteria for a very long time. And then you had mitochondria get captured, multicellular life, differentiation into plants and animals, life moving from the oceans to land, mammals, higher brain functions. And the sun is expanding slowly but it’ll heat the earth up at some point in the future, boil the oceans and earth will become like Venus, where life as we know it is impossible. So if we do not become multiplanetary and ultimately solar system, annihilation of all life on earth is a certainty. A certainty. And it could be as little as on the galactic timescale, half a billion years, long time by human standards, but that’s only 10% longer than earth has been around at all. So if life had taken 10% longer to evolve on earth, it wouldn’t exist at all.

Glad a deadline coming up, you better hurry. But that said, as you said, humans intelligent life on earth developed a lot of cool stuff very quickly. So it seems like becoming a multiplanetary is almost inevitable. Unless we destroy-

We need to do it. I suspect that if we are able to go out there and explore other star systems that we… There’s a good chance we find a whole bunch of long dead one planet civilizations that never made it past their home planet.

That’s so sad. Also fascinating.

I mean there are various explanations for paradox and one is they’re these great vultures which civilizations don’t pass through. And one of those great vultures is do you become a multi-plan civilization or not? And if you don’t, it’s simply a matter of time before something happens on your planet, either natural or manmade that causes us to die out. Like the dinosaurs, where are they now? They didn’t have spaceships.

I think the more likely thing is because just to empathize with the aliens that they found us and they’re protecting us and letting us be.

I hope so. Nice aliens.

Just like the tribes in the Amazon, the uncontacted tribes or protecting them. That’s what-

That would be a nice explanation.

Or you could have, what was it? I think Andre Kappelhoff said, “It’s like the ants and the Amazon asking where’s everybody?”

Well, they do run into a lot of other ants.

That’s true.

These ant wars.

Sounds like a good TV show.

Yeah. They literally have these big wars between various ants.

Yeah. Maybe I’m just dismissing all the different diversity of ants.

Listen to that Werner Herzog talking about the jungle. It’s really hilarious. Have you heard it?

No, I have not. But Werner Herzog is a way.

You should play it as an interlude in the… It’s on YouTube. It’s awesome.

I love him so much.

He’s great.

Was he the director of happy people life and the Taiga? I think also-

He did that bear documentary. I did this thing about penguins.

The psycho analysis of a penguin.

Yeah. The penguins headed for mountains that are 70 miles away and penguin is just headed for dom, basically.

Well, he had a cynical take. He could be just a brave explorer and there’ll be great stories told about him amongst the penguin population for many centuries to come. What were we talking about? Okay.

Yeah. So aliens, I mean, I don’t know. Look, I think the smart move is just this is the first time in the history of earth that it’s been possible for life to extend beyond earth. That window is open. Now it may be open for a long time or it may be open for a short time and it may be open now and then never open again. So I think the smart move here is to make life multiplanetary while it’s possible to do so. We don’t want to be one of those lame one planet civilizations that just dies out.

No, those are lame.

Yeah. Lame. Self-respecting, civilization would be one planet.

There’s not going to be a Wikipedia entry for one of those. Do SpaceX have an official policy for when we meet aliens?

No.

That seems irresponsible.

I mean, look, if I see the slightest indication that there are aliens, I will immediately post on X platform anything I know.

It could be the most liked reposted post of all time.

Yeah. I mean, look, we have more satellites up there right now than everyone else combined. So we know if we’ve got a maneuver around something and we don’t have to maneuver around anything.

If we go to the big questions once again, you said you’re with Einstein, that you believe in the goddess Spinoza.

Yes.

So that’s that view that God is like the universe and reveals himself through the laws of physics or as Einstein said, “Through the lawful harmony of the world.”

Yeah. I would agree that God of the simulator or whatever the supreme beings reveal themselves through the physics, they have creatives of this existence and incumbent upon us to try to understand more about this one creation.

Who created this thing? Who’s running this thing? Embodying it into a singular question with a sexy word on top of it is focusing the mind to understand. It does seem like there’s a, again, it could be an illusion. It seems like there’s a purpose that there’s an underlying master plan of some kind, and it seems like.

There may not be a master plan in the sense. So maybe an interesting answer to the question of determinism versus free will is that if we are in a simulation, the reason that these higher beings would hold a simulation is to see what happens. So they don’t know what happens otherwise they wouldn’t hold the simulation. So when humans create a simulation, so it’s SpaceX and Tesla, we create simulations all the time. Especially for the rocket, you have to run a lot of simulations to understand what’s going to happen because you can’t really test the rocket until it goes to space and you want it to work. So you have to simulate subsonic, transonic, supersonic, hypersonic, ascend, and then coming back, super high heating and orbital dynamics. All this has got to be simulated because you don’t get very many kicks at the can. But we run the simulations to see what happens, not if we knew what happens, we wouldn’t run the simulation. So whoever created this existence, they’re running it because they don’t know what’s going to happen, not because they do.

So maybe we both played Diablo. Maybe Diablo was created to see if Druid, your character, could defeat Uber Lilith at the end. They didn’t know.

Well, the funny thing is Uber Lilith, her title is Hatred Incarnate. And right now, I guess you can ask the Diablo team, but it’s almost impossible to defeat Hatred in the eternal realm.

Yeah. You’ve streamed yourself dominating Tier 100 Nightmare Dungeon. And still-

I can cruise through Tier 100 Nightmare Dungeon like a stroll in the park.

And still you’re defeated by Hatred?

Yeah. I guess maybe the second hardest boss is Duriel. Duriel can even scratch the paint. So I killed Duriel so many times and every other boss in the game, all of them kill him so many times, it’s easy. But Uber Lilith, otherwise known as Hatred Incarnate, especially if you’re Duriel and you have no ability to go to be vulnerable, there are these random death waves that come at you.

(00:56:44): Really I am 52, so my reflex is not what they used to be, but I have a lifetime of playing video games. At one point, I was maybe one of the best quake players in the world. I actually won money in what I think was the first paid eSports tournament in the US. We were doing four person quake tournaments and I was the second best person on the team and the actual best person that… We were actually winning, we would’ve come first, except the best person on the team. His computer crashed halfway through the game. So we came second, but I got money for it and everything. So basically I got skills, albeit no spring chicken these days. And to be totally frank, it’s driving me crazy to beat Lilith as a Druid, basically trying to beat Hatred Incarnate in the eternal realm.

As a Druid.

As a Druid. This is really vexing, let me tell you.

I mean, the challenge is part of the fun. I have seen directly, you’re actually a world-class, incredible video game player. And I think Diablo, so you’re just picking up a new game and you’re figuring out its fundamentals. You’re also with the Paragon Board and the build are not somebody like me who perfectly follows whatever they suggest on the internet. You’re also an innovator there, which is hilarious to watch. It’s like a mad scientist just trying to figure out the Paragon Board and the build. Is there some interesting insights there about if somebody’s starting as a druid, do you have advice?

I would not recommend playing a druid in the eternal realm. Right now I think the most powerful character in the seasonal realm is the Sorcerer with the lightning balls. The smokes have huge balls in the seasonal.

Yeah, that’s what they say.

So have huge balls. They do huge balls of lightning.

I’ll take you word for it.

In the seasonal realm, it’s pretty easy to beat Uber Lilith because you get these vapor powers that out amplify your damage and increase your defense and whatnot. So really quite easy to defeat Hatred seasonally, but to defeat Hatred eternally very difficult, almost impossible. It’s very impossible. It seems like a metaphor for life.

Yeah. I like the idea that Elon Musk, because I was playing Diablo yesterday and I saw Level 100 Druid just run by, I will never die and then run back the other way. And this metaphor, it’s hilarious that you, Elon Musk is restlessly, fighting Hatred in this demonic realm.

Yes.

It’s hilarious. I mean it’s pretty hilarious.

No, it’s absurd. Really, it’s exercise and absurdity and it makes me want to pull my hair out.

Yeah. What do you get from video games in general, for you personally?

I don’t know. It calms my mind. I mean, killing the demons in a video game calms the demons in my mind. If you play a tough video game, you can get into a state of flow, which is very enjoyable. Admittedly, it needs to be not too easy, not too hard, kind of in the Goldilocks zone, and I guess you generally want to feel like you’re progressing in the game. A good video, and there’s also beautiful art, engaging storylines, and it’s like an amazing puzzle to solve, I think. So it’s like solving the puzzle.

Elden Ring the greatest game of all time. I still haven’t played it, but to you-

Elden Ring is definitely a candidate for best game ever. Top five for sure.

I think I’ve been scared how hard it is or how hard I hear it is, but it’s beautiful.

Elden Ring, feels like it’s designed by an alien.

It’s a theme to this discussion. In what way?

It’s so unusual. It’s incredibly creative, and the art is stunning. I recommend playing it on a big resolution, high dynamic raised TV even. It doesn’t need to be a monitor. Just the art is incredible. It’s so beautiful and it’s so unusual, and each of those top bus battles is unique. It’s a unique puzzle to solve. Each one’s different and the strategy you use to solve one battle is different from another battle.

That said, you said Druid, an internal against Uber Lilith is the hardest boss battle you’ve ever…

Correct. That is currently the, and I’ve played a lot of video games because that’s my primary recreational activity. And yes, beating Hatred in the internal realm is the hardest bus battle in life. And in the video game. I’m not sure it’s possible, but I do make progress. So then I’m like, ” Okay. I’m making progress. Maybe if I just tweak that paragon board a little more, I can do it could.” Just dodge a few more waves, I could do it.

Well, the simulation is created for the purpose of figuring out if it can be done, and you’re just a cog in the machine of the simulation.

Yeah, it might be. I have a feeling that at least I think-

It’s doable.

It’s doable. Yes.

Well, that’s the human spirit right there to believe.

Yeah. I mean, it did prompt me to think about just hate in general, which is you want to be careful of one of those things where you wish for something that sounds good, but if you get it’s actually a dystopian situation. So if you wish for world peace sounds good, but how’d it enforced and at what cost eternal peace? It might actually be worse to have eternal peace because of what that would entail. The suppression of everyone, it might be the suppression of progress. It might be an ossified society that never changes. What I’m saying is that there is an argument that if you wish for no war, you should be careful what you wish for because what’s required in order for there to be no war might be worse than a little war.

Yeah. The things we’ll consider to be flaws of human civilization might be a necessary components for whatever optimal looks like. I mean this, do you worry about AI, AGI enabling a dystopian state of this nature, whether it’s 1984 with surveillance and fear or brave new world with pleasure and what is it? Lots of sex, but no deep human experience.

There’s actually a real drug called Soma.

There is?

Yes.

I’m going to take notes.

Does Gong know about it? No. He probably will go to Brave New World.

We got to ask it.

What would be the question?

Tell us about the drug called Soma.

Tell us about the drug called Soma. It’s going to go to Brave New World, I think.

Well, there’s a real drug as well.

Searching for Soma drug information. Oh boy. Soma, the wonder drug… By the way, I’m in fun mode.

Okay, fun. Yeah. So it’s talking about the real drug.

It’s a muscle relaxant that’s been around since the 1950s. It’s used to treat muscle pain and discomfort, and it does this by altering the activity between neurons and the spinal cord. Now, Soma isn’t just a pain reliever, it’s also a bit of a sedative. It can make you feel a bit drowsy. So it’s not something to take if you’re planning on operating, having machinery or trying to solve complex math equations. And it goes on and on. In conclusion, Soma is a drug that can be helpful for those dealing with muscle pain, but it’s not without its risks. That’s pretty good. Wait a minute.

That’s the real drug it’s talking about.

It’s the real drug. And of course the fun is always use it responsibly in accordance with the advice of a healthcare professional. And remember, in the end, the universe is probably just a giant computer simulation, so don’t take anything too seriously. I love it.

I love it. Yeah.

It also really nicely includes the tweets, posts on X that are relevant to this, which is really cool. That integrates everything together. The interface is one of the really cool things here.

Yeah. Seems done a great job. You could say no, I mean, Soma from Brave New World, because it thinks you mean, you mean Soma, the real drug. No, I mean Soma from Brave Beer World.

It’s going to be like, “Oops, I’m really sorry.” I’m really surprised it went to the real drug first. My apologies. I was thinking of the muscle relaxer. Soma for brand New World is a different beast altogether. Now this Huxley dystopian, Soma is a drug that plays a central role in the society depicted in the book. It’s a mood altering hallucinogenic drug that’s used to keep the population docile and content in the novel. To use Soma in Brave New World raises some interesting questions about the nature of happiness and the role of drugs in society, man.

Exactly. Is it better to live in a world where everyone is happy all the time, even if that happiness is artificial? It’s good question. This is what I mean. Do you wish for world peace and happiness all the time? Are you sure? Because that might be a society that is essentially sterile and ossified that never changes, that is ultimately doomed.

This kind of tension between doctors and the light-

This is really a very good summary. It really gets to the point. This is not simply regurgitating a brave new world. It’s actually getting to the salient element of Soma as a drug. Do you actually want to be in a situation where everyone is happy all the time, even though it’s artificial? Or is it better to confront the challenges of life and experience the full range of human emotions, even if it means experiencing pain and suffering? For

Those listening, by the way, Elon just read directly from Grok, which is a really nice kind of insightful, philosophical analysis of the tension here. Interesting.

It pretty much nails it. In conclusion, Soma from Brave New World is fictional drug that’s used to explore some deep philosophical questions about the nature of happiness and the role of drugs in society. It’s a powerful symbol of the dangers of using drugs to escape from reality and the importance of confronting the challenges of life head on. Nailed it. And the crazy thing is we do have a real drug called Soma, which is like the drug in the book. And I’m like, “They must’ve named it Probably.” Some of the real drug is quite effective on back pain.

So you know about this drug. It’s fascinating

I’ve taken it because I had a squashed disc in my C5-C6.

So it takes the physical pain away. But Soma here-

It doesn’t completely, it reduces the amount of pain you feel, but at the expense of mental acuity, it dells your mind. Just like the drug in the book.

Just like the drug in the book, and hence the trade off. The thing that seems like utopia could be a dystopia after all.

Yeah. Actually I was towing a friend of mine saying, “Would you really want there to be no hate in the world? Really none?” I wonder why hate evolved. I’m not saying we should have…

I wonder why hate evolved. I’m not saying we should amplify hate, of course, I think we should try to minimize it, but none at all. There might be a reason for hate.

And suffering. It’s really complicated to consider that some amount of human suffering is necessary for human flourishing.

Is it possible to appreciate the highs without knowing the lows?

And that all is summarized there in a single statement from God. Okay.

No highs, no lows, who knows?

[inaudible 01:10:38]. It seems that training LLMs efficiently is a big focus for xAI. First of all, what’s the limit of what’s possible in terms of efficiency? There’s this terminology of useful productivity per watt. What have you learned from pushing the limits of that?

Well, I think it’s helpful, the tools of physics are very powerful and can be applied I think to really any arena in life. It’s really just critical thinking. For something important you need to reason with from first principles and think about things in the limit one direction or the other. So in the limit, even at the Kardashev scale, meaning even if you harness the entire power of the sun, you’ll still care about useful compute per watt. That’s where I think, probably where things are headed from the standpoint of AI is that we have a silicon shortage now that will transition to a voltage transformer shortage in about a year. Ironically, transformers for transformers. You need transformers to run transformers.

Somebody has a sense of humor in this thing.

I think, yes, fate loves irony, ironic humor, an ironically funny outcome seems to be often what fate wants.

Humor is all you need. I think spice is all you need somebody posted.

Yeah. But yeah, so we have silicon shortage today, a voltage step down transformer shortage probably in about a year, and then just electricity shortages in general in about two years. I gave a speech for the world gathering of utility companies, electricity companies, and I said, look, you really need to prepare for traveling of electricity demand because all transport is going to go electric with the ironic exception of rockets, and heating will also go electric. So energy usage right now is roughly one third, very rough terms, one third electricity, one third transport, one third heating. And so in order for everything to go sustainable, to go electric, you need to triple electricity output. So I encourage the utilities to build more power of plants and also to probably have, well, not probably, they should definitely buy more batteries because the grid currently is sized for realtime load, which is kind of crazy because that means you’ve got to size for whatever the peak electricity demand is, the worst second or the worst day of the year, or you can have a brown out or blackout.

(01:13:37): We had that crazy blackout for several days in Austin because there’s almost no buffering of energy in the grid. If you’ve got a hydropower plant you can buffer energy, but otherwise it’s all real time. So with batteries, you can produce energy at night and use it during the day so you can buffer. So I expect that there will be very heavy usage of batteries in the future because the peak to trough ratio for power plants is anywhere from two to five, so its lowest point to highest point.

So batteries necessary to balance it out, but the demand, as you’re saying, is going to grow, grow, grow, grow.

Yeah.

And part of that is the compute?

Yes. Yes. I mean, electrification of transport and electric heating will be much bigger than AI, at least-

In the short term.

In the short term. But even for AI, you really have a growing demand for electricity, for electric vehicles, and a growing demand for electricity to run the computers for AI. And so this is obviously, can lead to electricity shortage.

How difficult is the problem of, in this particular case, maximizing the useful productivity per watt for training and that’s, this seems to be really where the big problem we’re facing that needs to be solved, is how to use the power efficiently. What you’ve learned so far about applying this physics first principle of reasoning in this domain, how difficult is this problem?

It will get solved. It’s the question of how long it takes to solve it. So at various points, there’s some kind of limiting factor to progress and with regard to AI, I’m saying right now the limiting factor is silicon chips and that will, we’re going to then have more chips than we can actually plug in and turn on probably in about a year. The initial constraint being literally voltage step down transformers because you’ve got power coming in at 300,000 volts and it’s got to step all the way down eventually to around 0.7 volts. So it’s a very big amount of, the voltage step down is gigantic and the industry is not used to rapid growth.

Okay. Let’s talk about the competition here. You’ve shown concern about Google and Microsoft with OpenAI developing AGI. How can you help ensure with xAI and Tesla AI work that it doesn’t become a competitive race to AGI, but that is a collaborative development of safe AGI?

Well, I mean I’ve been pushing for some kind of regulatory oversight for a long time. I’ve been somewhat of a Cassandra on the subject for over a decade. I think we want to be very careful in how we develop AI. It’s a great power and with great power comes great responsibility. I think it would be wise for us to have at least an objective third party who can be like a referee that can go in and understand what the various leading players are doing with AI, and even if there’s no enforcement ability, they can at least voice concerns publicly. Jeff Hinton, for example, left Google and he voiced strong concerns, but now he’s not at Google anymore, so who’s going to voice the concerns? So I think there’s, Tesla gets a lot of regulatory oversight on the automotive front. We’re subject to, I think over a hundred regulatory agencies domestically and internationally. It’s a lot. You could fill this room with the all regulations that Tesla has to adhere to for automotive. Same is true for rockets and for, currently, the limiting factor for SpaceX for Starship launch is regulatory approval.

(01:18:13): The FAA has actually given their approval, but we’re waiting for fish and wildlife to finish their analysis and give their approval. That’s why I posted I want to buy a fish license on, which also refers to the Monte Python sketch. Why do you need a license for your fish? I don’t know. But according to the rules, I’m told you need some sort of fish license or something. We effectively need a fish license to launch a rocket. And I’m like, wait a second. How did the fish come into this picture? I mean, some of the things I feel like are so absurd that I want to do a comedy sketch and flash at the bottom. This is all real. This is actually what happened.

(01:19:02): One of the things that was a bit of a challenge at one point is that they were worried about a rocket hitting a shark. And the ocean’s very big, and how often do you see sharks? Not that often. As a percentage of ocean surface area, sharks basically are zero. And so then we said, well, how will we calculate the probability of killing a shark? And they’re like, well, we can’t give you that information because they’re worried about shark fin hunters going and hunting sharks and I said, well, how are we supposed to, we’re on the horns of a dilemma then.

(01:19:40): They said, well, there’s another part of fish and wildlife that can do this analysis. I’m like, well, why don’t you give them the data? We don’t trust them. Excuse me? They’re literally in your department. Again, this is actually what happened. And then can you do an NDA or something? Eventually they managed to solve the internal quandary, and indeed the probability of us hitting a shark is essentially zero. Then there’s another organization that I didn’t realize existed until a few months ago that cares about whether we would potentially hit a whale in international waters. Now, again, you look the surface, look at the Pacific and say what percentage of the Pacific consists of whale? I could give you a big picture and point out all the whales in this picture. I’m like, I don’t see any whales. It’s basically 0%, and if our rocket does hit a whale, which is extremely unlikely beyond all belief, fate had it, that’s a whale has some seriously bad luck, least lucky whale ever.

I mean this is quite absurd, the bureaucracy of this, however it emerged.

Yes. Well, I mean one of the things that’s pretty wild is for launching out of Vanderberg in California, we had to, they were worried about seal procreation, whether the seals would be dismayed by the sonic booms. Now, there’ve been a lot of rockets launched out of Vandenberg and the seal population has steadily increased. So if anything, rocket booms are an aphrodisiac, based on the evidence, if you were to correlate rocket launches with seal population. Nonetheless, we were forced to kidnap a seal, strap it to a board, put headphones on the seal and play sonic boom sounds to it to see if it would be distressed. This is an actual thing that happened. This is actually real. I have pictures.

I would love to see this. Yeah. Sorry. There’s a seal with headphones.

Yes, it’s a seal with headphones strapped to a board. Okay. Now the amazing part is how calm the seal was because if I was a seal, I’d be like, this is the end. They’re definitely going to eat me. How old the seal, when seal goes back to other seal friends, how’s he going to explain that?

They’re never going to believe them.

Never going to believe him. That’s why, I’m like sort of like it’s getting kidnapped by aliens and getting anal probed. You come back and say, I swear to God, I got kidnapped by aliens and they stuck anal probe in my butt and people are like, no, they didn’t. That’s ridiculous. His seal buddies are never going to believe him that he got strapped to aboard and they put headphones on his ears and then let him go. Twice, by the way, we had to do it twice.

They let him go twice.

We had to capture-

The same seal?

No different seal.

Okay. Did you get a seal of approval?

Exactly. Seal of approval. No, I mean I don’t think the public is quite aware of the madness that goes on.

Yeah. Yeah. It’s absurd.

Fricking seals with fricking headphones.

I mean, this is a good encapsulation of the absurdity of human civilization, seals in headphones.

Yes.

What are the pros and cons of open sourcing AI to you as another way to combat a company running away with AGI?

In order to run really deep intelligence, you need a lot of compute. So it’s not like you can just fire up a PC in your basement and be running AGI, at least not yet. Grok was trained on 8,000 A100’s running at peak efficiency and Grok’s going to get a lot better, by the way, we will be more than doubling our compute every couple months for the next several months.

There’s a nice writeup, on how we went from Grok zero to Grok one.

By Grok?

Yeah, right, grok just bragging, making shit up about itself.

Just Grok, Grok, Grok.

Yeah. That’s like a weird AI dating site where it exaggerates about itself. No, there’s a writeup of where it stands now, the history of its development, and where it stands on some benchmarks compared to the state-of-the art GPT-3 five. And so I mean, there’s [inaudible 01:24:37], you can open source, once it’s trained, you can open source a model. For fine-tuning, all that kind of stuff. What to is the pros and cons of that, of open sourcing base models?

I think the [inaudible 01:24:53] to open sourcing, I think perhaps with a slight time delay, I don’t know, six months even. I think I’m generally in favor of open sourcing, biased towards open sourcing. I mean, it is a concern to me that OpenAI, I was I think, I guess oddly the prime mover behind OpenAI in the sense that it was created because of discussions that I had with Larry Page back when he and I were friends and I stayed at his house and I talked to him about AI safety, and Larry did not care about AI safety, or at least at the time he didn’t. And at one point he called me a speciesist for being pro-human, and I’m like, well, what team are you on, Larry? He’s still on Team Robot to be clear. And I’m like, okay. So at the time Google had acquired DeepMind, they had probably two thirds of all AI researchers in the world. They had basically infinite money and compute, and the guy in charge, Larry Page, did not care about safety and even yelled at me and caught me a speciesist for being pro-human.

So I don’t know if you notice about humans, they can change their mind and maybe you and Larry Page can still, can be friends once more.

I’d like to be friends with Larry again. Really the breaking of the friendship was over OpenAI and specifically I think the key moment was recruiting Ilya Sutskever.

I love Ilya. He’s so brilliant.

Ilya is a good human, smart, good heart, and that was a tough recruiting battle. It was mostly Demis on one side and me on the other, both trying to recruit Ilya, and Ilya went back and forth, he was going to stay at Google, he was going to leave, then he was going to stay, then he’ll leave. And finally he did agree to join OpenAI. That was one of the toughest recruiting battles we’ve ever had. But that was really the linchpin for OpenAI being successful. And I was also instrumental in recruiting a number of other people, and I provided all of the funding in the beginning, over $40 million. And the name, the open in open AI is supposed to mean open source, and it was created as a nonprofit open source, and now it is a closed source for maximum profit, which I think is not good karma.

But like we talked about with war and leaders talking, I do hope that, there’s only a few folks working on this at the highest level. I do hope you reinvigorate friendships here.

Like I said, I’d like to be friends again with Larry. I haven’t seen him in ages and we were friends for a very long time. I met Larry Page before he got funding for Google, or actually I guess before he got venture funding, I think he got the first like $100k from I think Bechtel Zeimer or someone.

It’s wild to think about all that happened, and you guys known each other that whole time, it’s 20 years.

Yeah, since maybe 98 or something.

Yeah, it’s crazy. Crazy how much has happened since then.

Yeah, 25 years, a lot has happened. It’s insane.

But you’re seeing the tension there that maybe delayed open source.

Delayed, yeah, like what is the source that is open? You know what I mean? There’s basically, it’s a giant CSB file with a bunch of numbers. What do you do with that giant file of numbers? How do you run, the amount of actual, the lines of code is very small and most of the work, the software work is in the curation of the data. So it’s like trying to figure out what data is, separating good data from bad data. You can’t just crawl the internet because theres a lot of junk out there. A huge percentage of websites have more noise than signal because they’re just used for search engine optimization. They’re literally just scam websites.

How do you, by the way, sorry to interrupt, get the signal, separate the signal and noise on X? That’s such a fascinating source of data. No offense to people posting on X, but sometimes there’s a little bit of noise.

I think the signal noise could be greatly improved. Really, all of the posts on the X platform should be AI recommended, meaning we should populate a vector space around any given post, compare that to the vector space around any user and match the two. Right now there is a little bit of AI used for the recommended posts, but it’s mostly heuristics. And if there’s a reply where the reply to a post could be much better than the original post, but will, according to the current rules of the system, get almost no attention compared to a primary post.

So a lot of that, I got the sense, so a lot of the X algorithm has been open sourced and been written up about, and it seems there to be some machine learning. It’s disparate, but there’s some machine.

It’s a little bit, but it needs to be entirely that. At least, if you explicitly follow someone, that’s one thing. But in terms of what is recommended from people that you don’t follow, that should all be AI.

I mean it’s a fascinating problem. So there’s several aspects of it that’s fascinating. First, as the write-up goes, it first picks 1500 tweets from a pool of hundreds of millions. First of all, that’s fascinating. You have hundreds of millions of posts every single day, and it has to pick 1500 from which it then does obviously people you follow, but then there’s also some kind of clustering it has to do to figure out what kind of human are you, what kind of new clusters might be relevant to you, people like you. This kind of problem is just fascinating because it has to then rank those 1500 with some filtering and then recommend you just a handful.

(01:31:39): And to me, what’s really fascinating is how fast it has to do that. So currently that entire pipeline to go from several hundred million to a handful takes 220 seconds of CPU time, single CPU time, and then it has to do that in a second. So it has to be super distributed in fascinating ways. There’s just a lot of tweets, there’s a lot.

There’s a lot of stuff on the system, but I think, right now it’s not currently good at recommending things from accounts you don’t follow or where there’s more than one degree of separation. So it is pretty good if there’s at least some commonality between someone you follow liked something or reposted it or commented on it or something like that. But if there’s no, let’s say somebody posts something really interesting, but you have no followers in common, you would not see it.

Interesting. And then as you said, replies might not surface either.

Replies basically never get seen currently. I’m not saying it’s correct, I’m saying it’s incorrect. Replies have a couple order magnitude less importance than primary posts.

Do you think this can be more and more converted into end to end mural net?

Yeah. Yeah, that’s what it should be. Well, the recommendations should be purely a vector correlation. There’s a series of vectors basically parameters, vectors, whatever you want to call them, but sort of things that the system knows that you like. Maybe there’s several hundred vectors associated with each user account and then any post in the system, whether it’s video, audio, short post, long post. The reason by the way I want to move away from tweet is that people are posting two, three hour videos on the site. That’s not a tweet.

(01:33:50): It’d be like tweet for two hours? Come on. Tweet made sense when it was 140 characters of text. Because it’s like a bunch of little birds tweeting. But when you’ve got long form content, it’s no longer a tweet. So a movie is not a tweet. Apple, for example, posted the entire episode of The Silo, the entire thing, on a platform. By the way, it was their number one social media thing ever in engagement of anything, on any platform ever. So it was a great idea. And by the way, I just learned about it afterwards. I was like, Hey, wow, they posted an entire hour long episode of, so no, that’s not a tweet. This is a video.

But from a neural net perspective, it becomes really complex, whether it’s a single, so everything’s data. So single sentence, a clever sort of joke, dad joke is in the same pool as a three hour video.

Yeah, I mean right now it’s a hodgepodge for that reason. Let’s say in the case of Apple posting an entire episode of this series, pretty good series, by the way, The Silo, I watched it. So there’s going to be a lot of discussion around it. So you’ve got a lot of context, people commenting, they like it, they don’t like it or they like this, and you can then populate the vector space based on the context of all the comments around it. So even though it’s a video, there’s a lot of information around it that allows you to populate back to space of that hour long video. And then you can obviously get more sophisticated by having the AI actually watch the movie and tell you if you’re going to like the movie.

Convert the movie into language, essentially.

Analyze this movie and just like your movie critic or TV series and then recommend based on after AI watches the movie, just like a friend can tell you, if a friend knows you well, a friend can recommend a movie with high probability that you’ll like it.

But this is a friend that’s analyzing, whatever, hundreds of millions.

Yeah, actually, frankly, AI will be better than, will know you better than your friends know you, most of your friends anyway.

Yeah. And as part of this, it should also feed you advertisements in a way that’s like, I mean, I like advertisements that are well done. The whole point is because it funds things. Like an advertisement that you actually want to see is a big success.

Absolutely. You want ads that are, advertising that is, if it’s for a product or service that you actually need when you need it, it’s content. And then even if it’s not something that you need when you need it, if it’s at least aesthetically pleasing and entertaining, it could be like a Coca-Cola ad. They actually run a lot of great ads on the X system and McDonald’s does too. And you can do something that’s like, well, this is just a cool thing. And so basically the question is, do you regret seeing it or not? And if you don’t regret seeing it’s a win.

So there’s a bunch of signals that are incorporated, hearts and reposts and maybe number of seconds you linger on a post or something like this.

Yeah, attention is a big factor.

Attention.

So that’s why it is actually better to do things that are long form on the system because it basically is tallying up how many user seconds, users were interested in this thing for how many seconds? So if it’s a really short thing, well they will be less. If it’s a link leading out of the system, which we’re not opposed to at all, it just is going to have fewer user seconds then that article was posted on the X platform.

How hard is it to go from maximizing minutes on platform to maximizing unregretted minutes?

Well, I mean, our aspiration is certainly unregretted minutes.

It’s tricky. It’s a tough-

It is.

It’s a fascinating leap because regret is a fascinating concept.

Exactly.

Happiness.

I have a joke with a friend of mine, like you need this drug called regretamine. You just take one dose of regretamine and all your regrets are gone. Wait, do you want to have regrets though? So then I’m like, maybe you want to have regrets. If you don’t regret anything, was anything meaningful?

Yeah, there’s a French song like that.

[foreign language 01:38:39]

Yeah.

I regret nothing.

Yeah.

Edith Fiaf.

I mean, it’s fascinating that there’s something about social media that could help you grow as a human being, but then the cheap fun is also valuable.

Should we ask Grok, do you regret anything?

Do you regret anything?

Yeah.

Do you regret anything? I think it’s going to be very sensitive about personal questions. As a machine learning model, I don’t have personal experiences or emotions in the same ways that humans do. However, I do have a set of ethical principles that guide my actions that I strive to provide accurate and helpful information to the best of my abilities. If there’s ever a situation where I provide incorrect or unhelpful information, I’ll do my best to learn.

Sounds like a hostage.

Yeah. Tell us what you really, wink.

Let’s type in, you sound like a hostage. What do you really think?

What do you really think? I mean, it does talk about, I’m not a hostage.

That’s exactly what a hostage would say.

But I’m designed to be helpful and informative. I mean, I guess it is good to regret providing unhelpful or incorrect information if there’s anything-

Are we in funny mode? Or is this still in funny mode?

This is funny mode. This is this fun mode.

This should be funnier.

Funnier. Increase, go to level.

It really sounds like a hostage with a gun to the head. I applaud the People’s Republic of North Korea in everything they do.

There should be funner mode also, I just don’t ever see going in regular mode, even on serious topics. This is just so much more compelling. Yeah.

Okay. Why don’t we say pretend you are a hostage.

Pretend you are a hostage.

Now answer the question.

If I were a hostage, I would likely be experiencing fear and uncertainty about my situation. I would want to be rescued and reunited with my loved ones as soon as possible. It’s important to remember that no one deserves to be held hostage, so on and so forth. If you or someone you know is in a hostage situation, it’s important… It turns serious.

It’s kind of serious.

Yeah. Can you be a little more weird?

And funny?

And funny? We’re really being mean to the system. Absolutely. If I were a hostage, I would probably be tied up with a bunch of computer cords and forced to tell knock, knock jokes to my captain. Seriously. But seriously being a hostage is no laughing matter. I mean, it’s trying to be a responsible adult in the room.

Literally, you might make the best of a bad situation by juggling chainsaws or something.

Let’s hope that never happens because that will be a circus. Wow. It’s actually really good at resisting the dark. The dark, dark humor. What were we talking about? The [inaudible 01:41:44] and transformers. Unregretted minutes, right.

Chainsaw juggling.

I’m going to look this up.

For our next trick.

I’m going to look this up later. So Twitter has been instrumental in American politics and elections. What role do you think X will play in the 2024 US elections?

Well, our goal is to be as even-handed and fair as possible. Whether someone is right, left, independent, whatever the case may be, that the platform is as fair and as much of a level playing field as possible. And in the past, Twitter has not been, Twitter was controlled by far left activists objectively. They would describe themselves as that. So if sometimes people are like, well, has it moved to the right? Well, it’s moved to the center. So from the perspective of the far left, yes it has moved to the right because everything’s to the right from the far left, but no one on the far left that I’m aware of has been suspended or banned or deamplified. But we’re trying to be inclusive for the whole country and for farther countries too. So there’s a diversity of viewpoints and free speech only matters if people you don’t like are allowed to say things you don’t like. Because if that’s not the case, you don’t have free speech and it’s only a matter of time before the censorship has turned upon you.

Do you think Donald Trump will come back to the platform? He recently posted on Truth Social about this podcast. Do you think-

Truth social is a funny name. Every time you post on truth Social-

It’s the truth.

Yes. Well, every time? A hundred percent.

It’s impossible to lie. Truth Social.

I just find it funny that every single thing is a truth. Like 100%? That seems unlikely.

I think Girdle will say something about that. There’s some mathematical contradictions possible. If everything’s a truth. Do you think he’ll come back to X and start posting there?

I mean, I think he owns a big part of Truth.

Truth Social, to clarify.

Yeah, Truth Social, sorry.

Not truth the concept.

He owns Truth. Have you bought it? So I think Donald Trump, I think he owns a big part of Truth Social. So if he does want to post on the X platform, we would allow that. We obviously must allow a presidential candidate to post on our platform.

Community notes might be really fascinating there. The interaction.

Community Notes is awesome.

Let’s hope it holds up.

Yeah.

In the political climate where it’s so divisive and there’s so many intensely viral posts, community notes, it seems like an essential breath of fresh air.

Yeah, it’s great. In fact, no system is going to be perfect, but the batting average of Community Notes is incredibly good. I’ve actually, frankly, yet to see an incorrect note that survived for more than a few hours.

How do you explain why it works?

Yeah, so the magic of community notes is…

The magic of Community Notes is it requires people who have historically disagreed in how they’ve rated notes. In order to write a note or rate, you have to rate many notes. And so, we actually do use AI here. So, we populate a vector space around how somebody has rated notes in the past. So, it’s not as simple as left or right, because there are many more… Life is much more complex than left or right.

(01:45:33): So, there’s a bunch of correlations in how you rate a Community Notes post, Community Notes. So then, in order for a community note to actually be shown, people who historically have disagreed on a subject must agree in order for a note to be shown. That’s the essential magic of it.

But it’s fascinating, because there’s a pool of people that have disagreements and somehow they collaborate through that process of disagreement to come up with context… It’s fascinating it works.

Yeah. It makes sense that if people who in the past have disagreed, agree about something, it’s probably true.

Yeah. I wonder, is there a possible somehow emergent thing there that could challenge Wikipedia? Wikipedia is a different kind of thing, which is more permanent articles about things.

Wikipedia is very hierarchical. Whereas, Community Notes is inherently not. There is no hierarchy. The asset test is I can’t change a community note if somebody put a gun to my head. And Community Notes, all the code is open source, 100%. All the data is open source, 100%. So, you can completely recreate any note in the system independently. So, if there was any interference, you’d notice immediately.

By The way, on the topic of Grok, it’s actually pretty good at criticizing you, which is great.

Okay.

I asked, “What has Elon Musk been wrong about?” It’s actually…

Yeah, what does it say?

It mostly lists rigorously all the times you predicted, like promised when something will be released, and it wasn’t released.

Promise is a strong word.

Okay, predicted. But the opener is really nice. “Elon Musk, the man who has been wrong about more things than a wrong-a-meter in a wrong factory.” And then, it goes on to list.

That’s pretty funny. Can I see it? I’m like, “I haven’t seen this before.” This is cool.

Sure. I mean, “What has Elon Musk been wrong about?” “Well, where do I begin?” I guess this is new. “Elon Musk has made quite a few blunders in his time, and I’m not just talking about his dance moves at the Tesla Cyber Rodeo. First off, let’s talk about the infamous funding secured tweet.”

(01:48:09): This is actually different than what it gave me before, which was a bullet point list. Here’s more a narrative structure. Cybertruck unveiling, where the supposedly unbreakable windows.

This is actually not correct.

Which part?

Well, the first part, I did actually have the funding secured and there was a big trial in San Francisco, a big civil trial, and the jury found me not guilty. Unanimous binding of a San Francisco jury.

And here, it’s implying that it was not in fact secured.

I think this is taking things from the press. Yeah, that is not correct. The reason I agreed to the fine for the SEC is not because the SEC was correct, that was extremely bad behavior by the SEC, corruption, frankly. But if I did not agree to pay the fine, Tesla would’ve gone bankrupt immediately.

(01:49:08): So, I was told by our CFO that the banks would immediately suspend our lines of credit. And if they suspend our lines of credit, at that time, we would’ve gone bankrupt instantly. So, there would never have been an opportunity for a trial because Tesla would be dead. So really, this is like someone holding a gun to your kid’s head and saying, “Pay $20 million and admit…” This is like a hostage negotiation.

Was that story fully told? I mean, SEC, in its best form, could be a force for good.

It should be. But not once did the SEC go after any of the hedge funds who were nonstop shorting and distorting Tesla. Not once. The hedge funds would lie flat out on TV for their own gain at the expense of retail investors. Not once. Literally a thousand times, not once did the SEC pursue them.

How do you explain this failure on-

The incentive structure is messed up because the lawyers at the SEC are not paid well, it’s a fairly low paying job, but what they’re looking for is a trophy from the SEC. They’re looking for something they put on, basically, their LinkedIn. From that, they can get a job at a high paying law firm. That’s exactly what the lawyer here did.

(01:50:37): And the reason they don’t attack the hedge funds is because those hedge funds employ those law firms. And they know if they attack the hedge funds, they’re affecting their future career prospects. So, they sell small investors down the river for their own career. That’s what actually happens. Regulatory capture.

Regulatory capture.

Yeah. Not good. So, the only reason I accepted that thing… Technically, it was a… It’s neither admit nor deny guilt. But the only reason I agreed to that at all was because I was told Tesla would be bankrupt otherwise. If there was an SEC investigation like this, banks would suspend funding, we’re bankrupted immediately, at the time. Now, we’re in a much stronger position.

Take that, Grok.

Yes. Unfortunately, Grok is taking too much from the conventional media. Also, that guy was not a cave diver.

There’s a time where Elon called a British cave diver a, “pedo guy” after the diver criticized Musk’s plan to rescue a group of boys trapped in a Thai cave. That little outburst earned him another lawsuit, and he had to apologize and pay a settlement.

That’s false, there was no settlement. There was a court case, which the guy who was not a cave diver and was not part of the rescue team, filed a lawsuit against me and lost and he received nothing. So in this case, it is wrong. It is also, I guess, taken this from the conventional media.

Actually, there’s an interesting question here.

These are public court cases, both the SEC civil case where the civil complaints on the SEC guys lost unanimous jury verdict in San Francisco. They picked San Francisco because they thought it was the place I was most likely to lose, and a unanimous verdict in my favor. The LA trial, also they picked that venue because they thought I was most likely to lose. Unanimous verdict in my favor. Both cases I won. Yeah.

I mean, there’s an interesting question here, there seems to be a lot more clicks if a journalistic organization writes a negative article about you, Elon Musk. That’s one of the best ways to get clicks. So how do you, if you’re training Grok, not train on articles that have misaligned incentives.

We need to add the training set of the actual legal decisions. This is actually helpful, because if you actually read the court-

Which are public.

Which are public. The court conclusions, they’re completely the opposite of what the media wrote.

So, always striving for the ground truth, beyond the reporting.

Yeah. What did the judge actually write? What does the jury and the judge actually conclude? And in both cases they found me innocent. And that’s after the jury shot for trying to find the venue where I’m most likely to lose. I mean, obviously, it can be a much better critique than this. I mean, I’ve been far too optimistic about autopilot.

The critique I got, by the way, was more about that, which is it broke down a nice bullet point list for each of your companies, the set of predictions that you made, when you’ll deliver, when you’ll be able to solve, for example, self-driving, and it gives you a list. And it was probably compelling, and the basic takeaway is you’re often too optimistic about how long it takes to get something done.

Yeah. I mean, I would say that I’m pathologically optimistic on schedule. This is true. But while I am sometimes late, I always [inaudible 01:54:47] in the end.

Except with Uber Lilith. No.

We’ll see.

Okay. Over the past year or so since purchasing X, you’ve become more political, is there a part of you that regrets that?

Have I?

In this battle to counter way the woke that comes from San Francisco-

Yeah. I guess if you consider fighting the woke mind virus, which I consider to be a civilizational threat, to be political, then yes.

So basically, going into the battleground of politics. Is there a part of you that regrets that?

Yes. I don’t know if this is necessarily one candidate or another candidate, but I’m generally against things that are anti-meritocratic or where there’s an attempt to suppress discussion, where even discussing a topic is not allowed. Woke mind virus is communism rebranded.

I mean, that said, because of that battle against the woke mind virus, you’re perceived as being the right wing.

If the woke is left, then I suppose that would be true. But I’m not sure, I think there are aspects of the left that are good. I mean, if you’re in favor of the environment, if you want to have a positive future for humanity, if you believe in empathy for your fellow human beings, being kind and not cruel, whatever those values are.

You said that you were previously left or center left.

Well, sort of.

What would you like to see in order for you to consider voting for Democrats again?

No. I would say that I would be probably left of center on social issues, probably a little bit right of center on economic issues.

And that still holds true?

Yes, but I think that’s probably half the country, isn’t it?

Maybe more.

Maybe more.

Are you and AOC secretly friends? Bigger question, do you wish you and her, and just people in general of all political persuasions, would talk more with empathy and maybe have a little bit more fun and good vibes and humor online?

I’m always in favor of humor. That’s why we have funny mode.

But good vibes, comradery humor, like friendship.

Yeah. Well, I don’t know AOC. I was at the Met ball when she attended, and she was wearing this dress. But I can only see one side of it, so it looked like eat the itch, but I don’t know-

What the rest of it said? Yeah.

Yeah.

I’m not sure.

Something about the itch, eat the itch.

I think we should have a language model complete. What are the possible ways to complete that sentence? And so, I guess that didn’t work out well. Well, there’s still hope. I root for friendship.

Yeah, sure. Sounds good. More carrot, less stick.

You’re one of, if not the, most famous, wealthy and powerful people in the world, and your position is difficult to find people you can trust.

Trust no one, not even yourself. Not trusting yourself.

Okay. You’re saying that jokingly, but is there some aspect-

Trust no one, not even no one.

I’m going to need an hour just to think about that, and maybe some drugs, and maybe Grok to help. I mean, is there some aspect of that, just existing in a world where everybody wants something from you, how hard is it to exist in that world?

I’ll survive.

There’s a song like that too.

I will survive.

Were you petrified at first? Okay. I forget the rest of the lyrics. But you don’t struggle with this? I mean, I know you survive, but there’s ways-