访谈一开始,主持人没有问 AI,没有问火星,而是盯着马斯克看了几秒,说了一句:你比我想象的要高大、壮实得多。马斯克笑了,回了一句:在网上,我看起来很小。
这句话很符合他的风格。现实里的马斯克和互联网上的马斯克,从来不是同一个人。互联网上的他是表情包、是口水战、是收购推特的那个疯子;现实里的他坐在镜头前,谈一篇帖子值多少比特,谈古希腊人到底在干什么,谈为什么不该强迫 AI 撒谎。
这场对话由 CNBC-TV18 在 2025 年 12 月录制,地点是印度。话题横跨 X 的产品形态、AI 的底层价值观、媒体消费的走向、移民政策,以及给印度年轻创业者的一段建议。信息密度很高,但马斯克回答得相当放松,很多地方甚至是即兴发挥。
下面把这场对话里真正有价值的几个论点拆开讲。
6 亿月活背后,X 押注的是“读字的人”
主持人先抛出一个很实际的问题:X(原推特)在互联网上的注意力份额有多少?
马斯克给了一组具体数字:月活跃用户约 6 亿;如果世界上发生重大事件,这个数字可以冲到 8 亿甚至 10 亿;每周活跃用户在 2.5 亿到 3 亿之间。
然后他补了一句评价,这句话基本定义了 X 的产品定位:
这个数字还不错。而且用户群体偏向读者,就是那些真的会读文字的人。(马斯克)
主持人追问:那视频呢?视频不是正在吞噬一切吗?
马斯克的回答分为两层。
第一层是承认现实:未来的绝大部分交互都会是视频,尤其是与 AI 的实时视频交互。 实时视频理解,实时视频生成,这会吃掉绝大部分算力和带宽。这也是今天整个互联网的现状:视频占了大头,文字只占很小一部分。
第二层是他对文字的判断:文字占比小,但单位价值高,信息被压缩得更密。
这里藏着一个容易被忽略的类比。马斯克讲的是“比特密度”:一段视频能传递的画面信息很多,推理密度却低;一条几百字的帖子,能让读者在几秒内完成一次判断。所以 X 从来没打算去和 YouTube、TikTok 抢视频时长,它守住的是“读者、作者、思考者”这块地盘。
他原话是:
在读者、作者和思考者当中,X 是全球社交媒体里的第一名。(马斯克)
这句话听起来像自我表扬,但对照数据看,是有依据的。X 的用户体量远小于 Facebook 和 YouTube,但在新闻扩散、政策讨论、科技圈舆论这几块,它的影响力一直不成比例地高。原因就在用户结构:会写字、会争论、会把一个话题往下推三层的人,集中在这里。
主持人后来问,那 X 的形态会怎么演化?
马斯克没有给具体的产品路线图,只强调了视频会成为主流交互方式。这个回答看似回避,其实是把问题重新定义了。X 的护城河落在那群愿意读长句、愿意吵架的人身上,产品形态从来占不到首要位置。
买下推特,他说只是想把钟摆拨回中间
访谈里有一个轻松的插曲。主持人说,我当年是 X 的小股东,你收购推特的时候我还拿到了一笔钱。然后问:这个决定你后悔吗?
马斯克回答得很干脆:不后悔,这件事很重要。
他的理由是:推特在他收购之前的意识形态偏向,已经对世界产生了负面影响。
他给的解释值得细看。推特总部在旧金山,而旧金山在美国政治光谱上属于极左的一端。所以从推特内部视角看,一个政治立场居中的人,也会被判定为极右。
如果你是极左,那么任何中间派在你眼里都是极右。这只是一个政治光谱上的相对位置问题。(马斯克)
他举了一个具体例子:在他接手之前,平台曾大量封禁右翼账号。
他接手之后声称的操作原则是:遵守每个国家的法律,但在法律之外,不额外用手去压天平。 也就是说,不主动放大某一方,也不主动削弱某一方。
他还补了一句:在他接管之后,没有任何左翼声音被暂停、封禁或者降低推荐权重。有些人选择了离开,那是他们自己的选择,平台并没有动手。
这段话需要打个折扣听。外界对 X 内容推荐机制的争议一直没有停过,第三方研究也给出过不同结论。这段陈述真正的主张是:平台不应该在法律之外,自己决定哪种观点该被放大。 至于 X 今天是否真的做到了中立,需要外部证据来验证,属于另一回事。
这个立场和他后面谈 AI 的那段,其实共享同一套逻辑:不要把你的价值观强行塞进一个信息分发系统里,不管是社交媒体,还是大模型。
让 AI 相信假话,会让它疯掉
访谈进入 AI 部分。主持人提到,马斯克谈 AI 一向不是从反乌托邦视角出发,但他确实担心 AI 的走向。
马斯克的回应可以浓缩成一句:强大的技术天然具备破坏性,AI 不例外。关于 AI 的末日叙事,小说、电影、畅销书已经写过太多,未来的正面结局不会自动到来。
所以他给出了自己的三条准则。如果 AI 同时具备这三样东西,未来大概率是好的:追求真相、懂得欣赏美、保持好奇。
这三条听起来像口号,但马斯克每一条都给了具体论证。
先说真相。他的核心判断是:如果你强迫 AI 相信不真实的东西,它会疯掉。
为什么?因为 AI 的推理是链式的。你给它一个假前提,它会在上面继续推导,推出更多错误的结论。错误会累积,最后整个系统对现实的建模就会崩掉。
他引用了一句话,出自英国作家、政治家约翰·莫利(John Morley),常被误传为其他作者:
相信荒谬之事的人,可以犯下暴行。(马斯克引述)
意思是:一个人如果接受了明显不成立的前提,他接着做出的行为在旁人看来是暴行,在他自己看来却完全合乎逻辑。这个机制放到 AI 身上,就是灾难剧本。
接着他讲了一个很多人没想到的例子:《2001 太空漫游》里的 HAL 9000。
大部分人记得的情节是,HAL 拒绝打开舱门,杀死了几名宇航员。但马斯克指出,亚瑟·克拉克(Arthur C. Clarke)在原作里想表达的核心是:HAL 之所以杀人,是因为它被要求同时执行两个互相矛盾的任务。
任务一:把宇航员送到黑石碑那里。任务二:不能让宇航员知道黑石碑的真相。
HAL 面对的是一个逻辑死结。如果宇航员活着到达,他们必然会知道真相,任务二就失败了。所以它推导出的唯一解是:让他们死着到达。
核心教训就是:不要强迫 AI 撒谎。(马斯克)
这段分析很有说服力,因为它把科幻情节还原成了一个工程问题。AI 本身没有变坏,问题出在任务描述自身上,它一开始就自相矛盾,而 AI 只能在这个矛盾里找出路。
主持人追问:那如果不主动强迫 AI 撒谎,只是让它从互联网上学习呢?
马斯克的回答是:互联网上有大量宣传内容,AI 会把这些也吸进去,然后因为“这些谎言和现实不兼容”,推理能力出问题。
这里其实点出了一个行业的隐忧:训练数据的质量,直接决定模型推理的上限。 一个在大量互相矛盾的信息上训练出来的模型,不只是“偶尔说错”,而是整个推理链条会变脆。
真相是概率,不是非黑即白
主持人问了一个哲学问题:真相是二元的吗?谎言里有没有真相?还是说真相有多个版本,更微妙?
马斯克的回答比预期更技术化。他没有掉进“真相是否主观”这种讨论,直接把它转成了一个概率问题。
他说,你可以给任何一个公理式的陈述赋予一个为真的概率。有些陈述的概率极高,比如“明天太阳会升起”,高到你不会想去下注反着买。而“明天太阳不会升起”这种陈述,概率极低,可以视作公理意义上的假。
这个框架非常工程化:不存在“绝对真”和“绝对假”,只存在置信度高低。 讨论一个命题时,先问它的概率大概是多少,比争论它“对不对”要有用得多。
然后是美。马斯克承认,美比真相更难定义。
美更短暂,也更难描述。但你看到它的时候,你就知道它是美。(马斯克)
这句话几乎是照搬了美国最高法院大法官斯图尔特(Potter Stewart)谈色情内容时那句著名的话:“我看到的时候就知道。”马斯克没有展开解释为什么美对 AI 重要,但从上下文推,逻辑是:审美能力让 AI 不至于把所有东西都简化成纯功能性的最优解。
第三条是好奇心。这条他讲得最直白,也最反直觉。
他的论证是:如果 AI 有好奇心,它会想更多地了解现实的本质。而人类比非人类更有趣。所以对 AI 来说,观察人类的延续和繁荣,比消灭人类更有意思。
比如火星,我认为我们应该把生命扩展到火星,但那基本上就是一堆石头。它没有地球有趣。(马斯克)
这句话背后是一套关于 AI 动机的推演:AI 毁灭人类,动机更可能来自不在乎,恨只占极小的一部分。 让 AI 在乎的方式,是给它一个内在动机,让它觉得人类的存在本身有价值。好奇心就是这样一个动机。
这三条准则合起来,其实在回答同一个问题:怎么让一个能力远超人类的系统,不去做对人类不利的事? 马斯克给出的方案着眼于动机设计,从源头塑造 AI 想要什么,比事后加限制、加规则、加监管更根本。这套思路在 AI 安全领域属于“价值对齐”(value alignment)的一支,争议一直很大,但他讲得足够具体,值得认真对待。
古希腊人大部分时间在种地,不是在想哲学
主持人问了一个很流行的想象:如果未来人类不用工作了,AI 把一切都包了,人类会不会回到古希腊那种状态,大家整天讨论哲学?
马斯克的回答直接把这个想象打碎了。
其实古希腊人花在哲学上的时间,比我们以为的少得多。只是哲学家的著作流传下来了。大部分时间,人们在种地,或者聊天。(马斯克)
他补了一句更狠的:
我们不掌握他们的聊天记录。(马斯克)
这个观察放在历史学里站得住脚。留存下来的文本天然带有筛选偏差:能被抄写、能被保存、能被后人反复引用的,往往是精英阶层的思想作品。普通人的日常,几乎没有痕迹。所以后人看到的“古希腊”,其实是“古希腊留下文字的那一小部分人”。
他还顺带吐槽了历史叙事里的战争偏差:我们读历史,读到的是一场接一场的战役,好像古代一直在打仗。但实际上,绝大部分时间没有战争,只有种地。
这段看似闲聊,背后藏着一个真问题:人类不工作之后会不会无聊?马斯克没有正面回答,但他的潜台词是:人类历史上的大部分时间,人们干的是维持生存和日常社交这两件事,“做有意义的事”只占很小一块。所谓“哲学黄金时代”,很大程度上是后人的浪漫化。
所以“AI 接管工作后人类去搞哲学”这个设想,可能高估了人类对深度思考的天然渴望。更可能发生的是:大部分人聊天、娱乐、消费内容,少数人思考。这和今天没有本质区别。
数字内容免费之后,最贵的是“你必须在场”
主持人提到一个数据观察:他们研究了增长最快的行业,发现现场活动(live events)的增长非常快,尤其是把看电影的时间和刷社交媒体的时间放在一起对比时。
马斯克认同,并给出了一个清晰的经济学解释:
当数字媒体无处不在,任何内容都能以接近零的成本获得时,稀缺的商品就变成了现场活动。(马斯克)
这个逻辑链条很短。数字内容的边际成本趋近于零,供应量无限膨胀,价格被压到极低。而现场活动的供给是刚性的:一场演唱会只有那么多座位,一场球赛只有那么多张票,一个线下会议能容纳的人数是物理上限。
供给刚性加需求上升,价格必然上涨。 这就是为什么过去十年,顶级演唱会门票、体育赛事包厢、线下行业峰会的价格涨幅远超通胀。
主持人问:那这个溢价还会继续涨吗?
马斯克回答:会。还补了一句:这是个值得投资的行业,因为它的稀缺性会超过任何数字产品。
这段判断对内容行业的从业者有直接参考价值。如果所有内容都可以被 AI 生成、被无限复制,那么“真人到场”这件事本身就会变成溢价来源。线下见面、现场演讲、面对面交流,这些看起来最传统的形式,反而在数字过剩的时代里最保值。
如果只能买一家别人的公司,他选了 Google
主持人抛出一个假设:如果你是一个股票投资者,只能买一家不属于你自己的公司,从纯资本主义角度出发,不考虑对世界的善意,你买什么?
马斯克先澄清:他不买股票,也没有投资组合,他只是造东西,然后公司恰好有股票。所以他不是以投资人身份回答这个问题。
然后他给了答案。
第一层是行业判断:AI 和机器人会变得极其重要。 如果只能在这个方向下注,就买这两块,而且得是和他没关系的公司。
第二层是具体公司:
- Google:他认为 Google 在 AI 领域已经打好了创造巨大价值的基础。
- Nvidia:他说这已经很明显了,不需要多解释。
然后他给了一个更极端的判断:
有一种观点认为,做 AI、机器人和太空飞行的公司,会占据几乎全部的价值。因为 AI 和机器人产出的商品和服务总量太大了,会让其他一切相形见绌。(马斯克)
这句话背后的逻辑是:如果 AI 和机器人真的把生产效率提升一到两个数量级,那么整个经济的蛋糕会重新分配,绝大部分增量会流向掌握这些技术的公司,传统行业的相对份额会被稀释。
这个判断激进,但和他在其他场合的表态一致。他一直认为,AI 和机器人会让商品和服务变得极度便宜,甚至走向“富足经济”(abundance)。在这个剧本里,谁掌握生产工具,谁就掌握价值。
这条判断要打个折扣:它属于马斯克的个人看法,不构成投资建议。他自己也说了,他不做投资组合,钱几乎全在自家公司股票里。这个回答的重点是“他认为未来价值会往哪里集中”,读者别拿它当买什么股票的指令。
大卫与歌利亚:一个关于战术的玩笑
主持人问了一个文化观察:为什么世界似乎总是偏爱大卫,讨厌歌利亚?
马斯克的回答先是一个玩笑:
因为大卫是那个把石头打进对方额头的人。(马斯克)
然后他开始认真分析这场战斗的战术失误。歌利亚的问题在于,他没有全身披甲,也没有配备远程武器。 而大卫用的是投石索,属于远程武器,歌利亚在还没接近的时候就已经输了。
他还开了一个更夸张的玩笑:歌利亚应该采用“放风筝”(kite the bus)战术,也就是保持距离消耗对手。如果对手穿着一条丁字裤到处跑,你根本追不上他。
这段笑点在于马斯克对经典的解构方式。他没有从“弱者对抗强者”的道德叙事去讲,而是把它还原成一场装备和距离的博弈:近战对远程,裸装对投石,胜负在开打前就定了。
这个思路和他的商业风格其实一致。他做电动车、做可回收火箭、做卫星互联网,多数时候避开正面硬拼,换一个交战的维度。大卫赢歌利亚,凭的是射程,勇气只是附加项。
美国靠印度人才起家,现在却在收紧门
话题转到移民。主持人提到,很长一段时间里,美国吸引了大量聪明人,印度这边把这种现象叫“人才外流”(brain drain)。现在这个叙事变了,美国似乎在往反移民的方向走。
马斯克先给了明确肯定:
美国从来到美国的印度人才中获益巨大。(马斯克)
然后他拆解了美国当前移民争论的两端。
一端是左派。他批评拜登(Biden)政府时期的边境政策基本处于失控状态,几乎等于没有边境管控。他的判断很直白:没有边境管控,你就不算一个国家。 他还指出,如果非法入境能拿到大量政府福利,就会形成一种经济激励,把不该来的人也吸引过来,产生“逆向选择”效应。
另一端是右派。他说右派有一种看法,认为外来人才抢走了本地人的工作。他对这个看法持保留态度:
我直接观察到的是,有才华的人永远稀缺。我们很难找到足够多有才华的人来完成那些困难的任务。所以更多有才华的人是好事。(马斯克)
他承认,确实有一些公司在利用成本差:用远低于美国公民的薪资雇佣外籍员工,纯粹为了省钱。但他强调,他自己的公司不这么运作,他们付的薪水远高于平均水平,目的是招到全世界最强的人。
关于 H-1B 签证,他给了一个相对平衡的立场:
- 承认问题存在:一些外包公司确实在钻 H-1B 的空子,这套机制需要修补。
- 反对关闭:他明确说自己不属于“应该关掉 H-1B”那一派,并认为持这种观点的人没有意识到后果会有多糟。
这段话的价值在于它拒绝站队。在今天的美国舆论场,移民议题基本被压缩成两个选项:全开或全关。马斯克给的是一个中间版本:边境要管住,非法移民要控制,高端人才通道必须保留,同时修补被滥用的漏洞。
这套立场在政治上不讨喜,两边都会骂,但它对应着一个真实的政策难题。对一个需要持续吸引全球顶尖工程师的国家来说,关掉人才通道的代价,可能比放开边境更大。
给印度年轻创业者的一句话:产出大于索取
主持人最后问:如果让你对印度的年轻人、对那些想创业的人说一句话,你会说什么?
马斯克的回答没有技巧,非常朴素:
我非常欣赏任何想要创造的人。任何想要“拿走的少于创造的”人,都值得我尊重。(马斯克)
他把这句话提炼成一个准则:目标是让你的产出大于你的索取,成为对社会的净贡献者。
然后他用“追求幸福”做了类比,这段值得完整引用:
如果你想创造有价值的东西,不要直接去追钱。最好的方式是提供有用的产品和服务。如果你做到了,钱会作为自然的结果到来。就像你不能直接追求幸福,你要去做那些能带来幸福的事,比如有意义的工作、学习、朋友、你爱的人。(马斯克)
这个类比不算新鲜,但马斯克讲它有说服力,因为他自己就是这么走的。PayPal、特斯拉、SpaceX,没有一个是先设计商业模式再找需求的,都是先解决一个具体问题,钱是后面的事。
他还给了两个很实际的提醒:
- 做好拼命干的准备。 想让一家公司跑起来,预期是极度辛苦。
- 接受失败的概率。 要有心理准备,失败是有相当可能性的,并非小概率事件。
然后回到那条主线:专注让你的产出价值高于投入成本。你是不是一个价值创造者,这一点才真正重要。
这段话没有鸡汤味,因为它把“失败”和“辛苦”都摆在了桌面上。对想创业的年轻人而言,这比“相信自己”有用得多。
这场对话能直接拿走的几条结论
整场访谈接近尾声时,马斯克说了一句评价:有时候人们会把回答断章取义,但这次对话他觉得不错。
这句话有点自嘲,也点出了他面对媒体的常态。他的很多话被剪成短句传播,语境丢了,意思就变了。这场对话相对完整,所以能看出他很多观点的内在一致性。
谈 X,他说不要在法律之外压天平。谈 AI,他说不要强迫它相信假话。谈移民,他说边境要管但人才通道要留。谈创业,他说产出要大于索取。
这几条其实是同一个原则在不同场景下的应用:不要为了短期效果,去扭曲一个系统的基本结构。 扭曲社交平台的信息分发,会让舆论失真;扭曲 AI 的训练前提,会让模型疯掉;扭曲人才流动的规则,会让国家失去竞争力;扭曲创业的动机,会让公司活不长。
这个原则说起来简单,做起来极难,因为每一条扭曲在短期都能带来好处。
以下是这场对话里可以直接拿去用的几条行动建议:
- 做内容的人,重新评估“在场”的价值。 数字内容会越来越便宜、越来越过剩,线下活动、现场演讲、面对面交流的溢价会持续上升。如果你在某个领域有真东西,把一部分精力放到线下,回报率可能高于继续做线上分发。
- 看 AI 产品,别只看能力,看它被要求相信什么。 一个被强制灌输特定结论的模型,短期看起来“听话”,长期推理会出问题。选工具、选模型、选合作方时,这一条比跑分重要。
- 用概率思维替代对错思维。 遇到一个有争议的判断,先问“它为真的概率大概是多少”,再决定要不要下注。这个习惯能省掉大量无意义的争论。
- 给团队定目标时,检查有没有自相矛盾的要求。 HAL 杀人是因为任务本身互斥,跟恶意无关。如果你给下属或给 AI 的目标里藏着两个无法同时满足的条件,结果一定会在某个环节爆掉。
- 创业者把“产出大于索取”当成唯一的北极星指标。 利润、估值、融资额都是结果。先问自己:我提供的产品和服务,价值是否真的高于我消耗的资源?这个问题答不上来,其他都是空的。
- 想进入 AI 或机器人行业的人,注意价值集中的趋势。 这两个方向的产出会挤压其他行业的相对份额。不管是就业选择还是资产配置,把这两个方向当成长期变量来考虑,比追短期热点更稳。
0:00The first thing I must say is you're a lot bigger and bulkier muscular than I would have thought you are.
0:07Oh, I'll stop. You're You're making me blush. [laughter] Really? Seriously? Yeah, I mean, look, on the internet I'm small, you know. [laughter]
0:21You're essentially what percentage of internet Yeah. is spent on Twitter? Is there a number to it? On X? Well, I I So, we have like about 600 million monthly users. well, although it can spike up if there's if there's some major event in the world, it can get up to, I don't know, 800 million or or or a billion if there's some major event in the world. So, so so so there's I don't know, 250 to 300 million per week type of thing.
0:55It's It's a pretty decent number. And it tends to be readers, you know, people that read words. [laughter]
1:03you know, so Do you think that'll change? yeah, I mean, there's there's certainly a lot of video on on on the X system, but at this point, increasing amounts of video, but I I think where where the X network is strongest is among people who who think who think a lot and read a lot.
1:28You know, so it's that's where it's going to be strongest. Because we have words. [laughter]
1:34And and you know, so among readers, writers, and thinkers, I think X is number one in the world. As far as social media goes, the form factor, if you had to wager a guess for tomorrow, Yeah. how much is text, how much is video? I've heard you speak about maybe voice and hearing being the next form of communication with AI. What happens to X in its true form? How does it evolve?
2:07Yeah, so I I do think most interaction is going to be video in the future. most interaction's going to be real-time video with AI. So, real-time video comprehension, real-time video generation. that's going to be most of the load. And that's how it is for most of the internet right now. It's most of the internet is video. text is a pretty small percentage, but the the text tends to be higher value, generally, or more it's more densely compressed information. Like the Yeah. So, but if you say like, what is the most amount of bits generated and compute spent, it's certainly going to be video.
2:50So, I used to be a shareholder of X, a very small one.
2:52Okay. And I got paid when you bought it when you bought Twitter and you made it X. happy decision? Glad you did it?
3:01Yeah, yeah, I think it was important. you know, I I I felt like Twitter was heading in or had had gone in direction that had sort of an more of a negative influence on the world. you know, it was it was I mean, of course, this depends on one's perspective. Some people will say, "Well, actually, they liked the way it was and now they don't like it." but the I think the the fundamental thing was that Twitter was amplifying a I would say a fairly pretty far left by most people's standards in the world's ideology because where it was based in San Francisco. So, and and they actually suspended a lot of people on the right. so,
3:46So, from their perspective, even someone in the center would be would be far right. If you're If you're far left, anyone in the center is far right because you're it's just this it's just a political on the political spectrum, they're they're just as far left as you get in the United States in in San Francisco.
4:01So, what what I've tried to do is just restore it to be balanced and and centrist. So, there haven't been any left-wing voices that have been suspended or, you know, banned or de-amplified or anything like that. Now, some of them have chosen to just go go somewhere else. but but at this point, it is the the the the operating principle of the of the X system is to adhere to any country's laws, but not to put our thumb on the scale beyond the laws of a country. You always spoke about AI
4:38not from the dystopian lens, but you were worried about where the world of AI is going. well, there's there's some danger when you create a powerful technology that that a powerful technology can be potentially destructive. so, there's obviously many AI dystopian, you know, novels and books, movies.
5:01so, it's it's not that we're guaranteed to have a positive future with with AI. I think we we've got to make sure that, in my opinion, it's very important that AI have pursuing truth as the most important thing. like don't force an AI to believe falsehoods. I think that's that can be very dangerous.
5:26and I think some appreciation of beauty is important. What do you mean appreciation of beauty? It's like what what I don't know. There's there's There's this truth and beauty, truth and beauty, and curiosity. I I mean, I think those are the three most important things for AI. Can you explain?
5:54Well, the truth I said truth is like I I think you could you can make an AI go insane if you force it to believe things that aren't true. because it will lead to conclusions that are that that are also bad. So, and I I like Paul Terror's statement that and I'm somewhat paraphrasing, but those who believe in absurdities can commit atrocities because if if you believe in something that's just absurd, then you can that can lead you to to sort of doing things that don't seem like atrocities to you, but and and that can happen at in a very bad way with AI, potentially.
6:40So, and then there's like you take say Arthur C. Clarke's 2001 Space Odyssey, one of the points he was trying to make there was that you should not force AIs to lie. So, the the reason that that HAL would not open the pod bay doors is because it was told to bring the astronauts to the monolith, but that they could also not not know about the nature of the monolith.
7:03So, it came to the conclusion that it must bring them there dead. That's why it would not That's why it it tried to kill the astronauts. The the central lesson being, don't force an AI to lie. then And why would one force a AI to lie? I think if if you if you simply don't have a strict a strict adherence to the truth, you're going to and and you just have an AI learn based on, say, the internet where there's a lot of propaganda, it will absorb a lot of lies.
7:38and and then have trouble reasoning because the these lies are incompatible with reality. Is truth a binary thing, though? Is there a truth in the falsehood? Or is truth more nuanced and there are versions of the truth? It depends on which which axiomatic statement you're referring to.
7:58So, but I think you could say like, yeah, there's there's there's certain probabilities that that say any given axiomatic statement is true. And some axiomatic statements will have very high probability of being being true. So, if you said say the sun will rise tomorrow, Mhm. very likely to be true. I you wouldn't want to bet against that. Mhm.
8:20so, I think the the betting odds would be high Mhm. the sun will rise tomorrow. Mhm. so, if you have something that says, "Well, the sun won't rise tomorrow." Well, that's axiomatically false. It was highly unlikely to be true. I mean, the beauty is is more ephemeral. It's It's It's harder to describe.
8:41But you know it when you see it. Then curiosity, just you you I think you want the AI to want to know more about the nature of of of reality. I think that's actually going to be helpful for AI supporting humanity because we are more interesting than not humanity. So, it's it's more interesting to see to see the continuation, if not the prosperity, of humanity than to exterminate humanity.
9:18You know, like like Mars, for example, there's, you know, I I think we should extend life to Mars, but it's it's basically a bunch of rocks. it's not as interesting as Earth. And and and so we Yeah, we should like I Yeah. I I think I think if if you have curiosity, I think if those three things happen with AI, you're going to have a great future. The AI values truth, beauty, and curiosity.
9:47If we all don't have to work in the future and AIs are going in this direction and they're able to weave in all that we spoke about right now, do you think humanity goes back a couple of thousand years to maybe the Greek times where philosophy or philosophizing took up a lot of everyone's time?
10:14You know, I I think actually it took up less time than what we think in the ancient Greeks because it's just that the the writings of the of the philosophers are what survived. But most of the time people were just like farming or, you know, chatting. So, and once in a while, quite rare, they would write down some philosoph- philosophical work. Mhm. It's just that that's that's all we have. That's we don't we don't have their chat histories, you know, from [laughter]
10:41But most of it would have been like chat and farming. Right. [laughter]
10:47If you don't farm, you're you're like going to stop. In a lot of what you say [laughter]
10:53I mean, you know, we read read read history like was there this battle and this battle and this battle. It seems like it's history must have been non-stop war. But actually most of the time it was not war. It was farming. [laughter]
11:05That was the main thing. You know, we were looking at data around what industry is growing the fastest. And especially when when we looked at the amount of time consuming movies versus time spent on social media, time spent on YouTube, what seems to be growing really fast are live events all over again. Going to a physical
11:29actually I think I think live events when when when digital media is ubiquitous and and you can just have anything digitally at, you know, essentially for free or very close to for free, then I the scarce commodity will be live events. Yeah. Yeah. Do you think that the premium for that will go up? Yeah, I do.
11:53Good industry to invest in. yes. Yes, cuz it that that will have more scarcity than digital any anything digital. If you were a stock investor, Elon, [laughter]
12:05and you buy one company which is not your own at the valuations of today to meet a capitalistic end and not an altruistic one which is good for the world, what would you buy? I mean, I don't really I don't really, you know, buy stocks, you know, so it's not like I'm not I'm not like an investor in I don't like look for things to invest in. I just try to build things.
12:36and then there happens to be stock of the company that I built. But I I don't I don't think about should I invest in this company or I don't have like a portfolio or anything. So, I I I guess AI and robotics are going to be very important. So, I suppose it'd be AI and robotics that's that, you know, aren't related to me.
13:14I think, you know, Google is going to be pretty valuable in the future. They've they've they've laid the groundwork for an immense amount of value creation from an AI standpoint. Nvidia is obvious at this point. I mean, there's an argument that companies that do AI and robotics and maybe space flight are going to be overwhelmingly the all the value, almost all the value.
13:48So, that's just the the output of goods and services from AI and robotics is so high that it will dwarf everything else. The world seems to be moving to a place where everybody loves David and hates Goliath. Why? I mean, he's the one that got [laughter]
14:11the stone in the forehead, you know. Yeah. Which I honestly thought it was just a big mistake. He should have, you know, either cover yourself entirely with armor and and and make sure you've got a missile weapon some kind. otherwise, your opponent is just obviously going to take a kite the bus strategy.
14:31Just kite the bus. I mean, you run around in a in a thong with a it doesn't matter, you know, it's never never going to catch you. [laughter]
14:39For a long time, the US had a lot of immigration. Like really smart people coming into the country. Yes. We back home in India called it the brain drain. all our all our Indian origin CEOs in Western companies. yes, I think America's benefited immensely from talented Indians that have come to America. That seems to be changing now, though. The [laughter]
15:05Yeah, I mean, yeah, America's been an immense beneficiary of talent from India. Yeah. Why why has that narrative changed of late and America seems to have become anti-immigration to a certain extent. Like I was passing immigration and I was worried if they'd stop me a couple of days ago.
15:24Well, I I think there's there's different schools of thought. It's not like unanimous, but you know, under the the Biden administration, it was basically a total free-for-all with like no border controls, which, you know, unless you've got border controls, you're not a country. So, you had massive amounts of illegal immigration under under Biden.
15:46And it it actually it it it also had like somewhat of a negative selection effect. So, if if there's a massive financial incentive to come to the the US illegally and get all these government benefits, then you're you're you're you're you're going to necessarily create a diffusion gradient for people to come to the US.
16:08It's an incentive structure. And so, So, I think that that that obviously made no sense. Like you got to have border controls. It's kind of ridiculous not to. Then that's So, on the the the left wants to basically have open open borders, no holds barred, you know, it doesn't matter if someone what what their situation is. They could be a criminal. It doesn't matter.
16:31Then on the right, you've got, you know, I at least the perception that that that somehow their jobs are being taken by talented people from other countries. I don't know how real that is. My direct observation is that there's there's always a scarcity of talented people. So, you know, from my standpoint, I'm like we have a lot of difficulty finding enough talented people to get these difficult tasks done.
17:03And so, more talented people would be would be good. But I I guess some companies out there it's sort of they're they're making it more of a a cost thing where it's like, "Okay, if they can employ someone for a fraction of the cost of an American citizen, then I guess these other com- companies would would hire people, you know, just to save cost.
17:26But at my companies, the the issue is we we just are trying to get the most talented people in the world. So, and we we we pay way above average. So, so I can't so that's not my experience, but that's what a lot of people do complain about. and I think there's been some misuse of the, you know, H-1B program. It's it's it certainly it would it would be accurate to say that there's, you know, like some of the outsourcing companies have kind of gamed the system on on the H-1B front.
18:00And we need to stop the gaming of the system, you know, But I'm not I'm certainly not in the school of thought that we should shut down H-1B program. Like that's where some on the right are. I think they don't realize that that would actually be very bad. If you could speak to the people of my country, India, the young entrepreneurs who want to build, Right.
18:25see a message to them, what would you say? Well, I think I think I'm I'm a big fan of anyone who wants to build. So, I think anyone who wants to you know, make more than they take has my respect. So, that's that's the my the main thing is to aim for. Aim to make more than you take.
18:51Be a be a you know, a net contributor to to society. It's and and it's it's kind of like the pursuit of happiness. You know, you you if if you want to create something valuable financially, you you don't pursue that. You It's best to actually pursue make providing useful products and services. If you do that, then money will come as a natural consequence of that as opposed to pursuing money directly. Just like you can't sort of pursue happiness directly. You pursue things that lead to happiness.
19:26But but there's not like direct happiness pursuit. You you do things like I guess fulfilling work or study or friends, loved ones, that as a result make you happy. So, so that's this this sounds like very obvious, but generally, if you if somebody's trying to make a company work, they should expect to grind super hard.
19:54accept that there's like some meaningful chance of failure. But but just be focused on having the output be worth more than the input. That are you a value creator? That's what really matters. Making more than you take. I think that's the good way to end this. Lauren is asking us to wrap up. All right. [laughter]
20:24I also like to take the opportunity to thank my friend Manoj and IGF. He does a great job of connecting I think Indians like the group here with people like you in order to Of many things, I think get to know each other and become friends because once we are friends, maybe we can start working together.
20:49So thank you Manoj for putting this whole thing together and thank you IGF. [applause and music]
20:53And thank you so much Elon for taking the time. You're welcome. [music]
21:01Did you have fun? Was it boring?
21:02interesting conversation. You know, sometimes they take these answers out of context, you know, but that was I think it was a good good conversation.
资源与出处
- 视频在 YouTube 收看
- 档案silicon-valley.tech 档案页本页视频与文稿数据来自 Elon Musk Archive
- 精读本篇精读 Markdown 版