马斯克在达沃斯上台,主持人请他谈谈旗下这些公司究竟在解决什么问题。
他先讲了个双关。听说有人要办"和平峰会"(Peace Summit),他第一反应听成了"Piece Summit",一小块一小块的峰会,一小块格陵兰,一小块委内瑞拉。他说,我们要的只是和平。
玩笑完了,他给了个正经答案:所有这些公司合起来只有一个目标,最大化文明拥有美好未来的概率,把意识扩展到地球之外。
然后他讲了为什么。
九千颗卫星,一次也没给外星飞船让过路
经常有人问他,外星人是不是就在我们中间。他的标准回答是"我就是",然后补一句"他们不信"。
但他给了一个可以验证的答案。SpaceX 在天上挂着 9000 颗卫星,从来没有一次需要为了避让外星飞船而机动变轨。
他说,如果有外星人,最该知道的那个人应该是他。九千颗卫星在轨运行,一次都没有。
所以他倾向于一个结论:生命和意识极其稀有,可能整个宇宙就我们这一份。
接下来的比喻,是这场对话里最有画面感的一段。
在我的脑子里,意识是广袤黑暗里的一根小蜡烛,一根很容易被吹灭的小蜡烛。这就是为什么生命必须变成多行星的。(马斯克)
地球如果遭遇自然灾害,或者人为制造的灾难,只要别处还有火种,意识就能继续烧下去。这是 SpaceX 存在的理由。
他说人应该始终把生命看成脆弱的东西。就我们所知,别处没有生命。
这段话听起来像抒情,其实是个决策框架。当样本量是 1 的时候,唯一的样本不能承担任何风险。地球是唯一的样本。
特斯拉的使命里多了三个字:可持续富足
说到特斯拉,他给了一个不常见的表述。原本的使命是"可持续能源",现在后面加了四个字,可持续富足(sustainable abundance)。
富足这个词很重。人类谈了几十年怎么消除全球贫困,怎么让每个人都过上很高的生活水平。他的判断是,唯一的办法是 AI 加机器人。
话说得很直白,没有绕弯。然后他马上加了警告:这不代表没有风险。AI 要非常小心,机器人要非常小心。没人想活在詹姆斯·卡梅隆的电影里。
他补了一句,卡梅隆的电影很棒,他很喜欢,但《终结者》那种情况不要。
转折在这里。前提是 AI 随处可得、接近免费,机器人也随处可得,那么全球经济会出现一次爆炸式增长,规模超过历史上任何先例。
这几个词要拆开看。"随处可得"意味着不是少数机构手里的稀缺品。"接近免费"意味着价格趋近于零,需求被彻底释放。"机器人随处可得"意味着物理世界的劳动也不再稀缺。
稀缺性同时从信息和原子两个方向被削掉,这在人类历史上没有发生过。
产出是乘法,不是加法
怎么理解这种增长,他给了个公式。
经济产出等于每台机器人的平均生产力乘以机器人的数量。
这个公式的关键在它是乘法。人口增长是慢变量,一代人二十五年。机器人数量可以是快变量。如果机器人能造机器人,这条曲线就进入指数区间。
他给的是良性情景下的预测:机器人会造出多到饱和的程度。
饱和的意思是,有一天你想不出还能跟机器人要什么。商品和服务多到需求端先撑不住。
他的具体预测是,机器人的数量会超过人类。
这个预测听起来夸张,但把它放进上面的乘法公式里,逻辑是自洽的。产出等于单位生产力乘以数量,只要数量没有上限,产出也就没有上限。而机器人造机器人,恰好把数量上限拿掉了。
有必须干的活,就没有人人富足
主持人问了个常见疑虑:活谁干,人还需要工作吗。
马斯克的回答是个逻辑判断,绕不过去。
你不能同时拥有"必须有人干的活"和"人人富足"。如果那些活是必须干的,而只有一部分人能干,那就不可能出现人人富足。这两个条件互相排斥。
他原话的意思是,这两样东西你没法都要。
所以推论只有一个方向:要富足,就得有几十亿台人形机器人把活全干了。他认为地球上每个人都会有一台,而且会想要一台。
他举的场景很具体。看孩子,看宠物。家里有老人,照顾起来非常贵,而且年轻人不够用。
他说很多朋友都提过,照顾年迈的父母非常难。贵是一方面,人手不够是另一方面。如果能有一台机器人照顾和保护老人,这是件很好的事。
这段回答里藏着一个判断:机器人最先被接受的场景,不是工厂,是家庭里那些又累又贵又没人愿意干的活。老龄化社会里,这类活的供给缺口最大。
他从没见过一只年轻的右臂
话题转到长寿和抗衰老,马斯克的回答方式跟大部分生物学家不一样。他认为这个问题非常可解,而且一旦解出来,答案会明显得不像话,一点都不微妙。
推理的起点是一个日常观察。他这辈子没见过一个人左胳膊老、右胳膊年轻。
身体里所有细胞基本以同一个速度衰老。为什么会这样?
只有一种解释:存在一个时钟,一个同步时钟,在给你身体里 35 万亿个细胞对表。
这个论证的说服力在于它排除了随机性。如果衰老是每个细胞各自积累损伤,那不同器官、不同部位的衰老速度应该参差不齐。实际上没有。全身同步,说明有一个统一的信号在驱动。
他也不否认死亡有功能。如果人活得太久,社会有僵化的风险,事情锁死在固定位置,变得沉闷,失去活力。
但延长寿命,甚至逆转衰老,他认为概率很高。只是他个人没在这件事上投太多时间。
AI 公司会自己去找客户
有人担心 AI 只服务少数人。马斯克觉得这个担心的方向错了。
AI 公司会去找尽可能多的客户,这是商业本能,不需要额外的道德动机。而且成本已经在很低的位置,还在往下掉。
他提到,AI 的成本几乎每个月都在实质性地变化。年度对比已经不够用,月度就能看出差别。
开源模型在他看来是重要的平衡力量,大概落后闭源模型一年。这个差距不足以形成永久垄断,因为客户永远可以选便宜的那一档。
所以结论是,AI 会像基础设施那样铺开,供应给全世界。
这个论证的关键不是善意,是竞争。只要有多家公司在抢客户,价格就会往下走,覆盖面就会往上走。开源模型落后一年这件事,恰好给了一个持续的价格锚。
卡住 AI 的不是芯片,是电
主持人问瓶颈在哪。他的回答指向一个东西:电。
AI 芯片的产量在指数级上升。电力的上线速度是另一条曲线,斜率小得多。两条线迟早交叉。
他说,很快,可能就在今年晚些时候,我们会造出比我们能点亮更多的芯片。
这句话的画面感很强。芯片堆在仓库里,插不上电,因为电网没有那么多余量。
有一个例外:中国。中国的电力增长非常猛。
这个瓶颈判断值得多停一会。过去两年,行业讨论的焦点全在算力、GPU、先进制程。马斯克把问题挪到了更底层:不管芯片多快,插上插座才有用。而发电站的建设周期以年计,芯片厂的扩产周期以季度计。
时间尺度错配,就是瓶颈。
太阳占太阳系质量的 99.8%
中国增长的主力是太阳能。他给的数据是,中国光伏产能大约每年 1500 吉瓦,装机超过每年 1000 吉瓦。
这里要做个换算。光伏是间歇的,配上电池才能当稳定电源。1000 吉瓦的装机,除以 4 到 5,大致相当于 250 吉瓦的稳定功率。
美国全国的平均用电量是 500 吉瓦。
也就是说,中国光靠光伏加储能,就能提供美国全年用电量的一半。
然后他讲了一笔更大尺度的账。
太阳占太阳系总质量的 99.8%。木星占 0.1%,剩下的所有东西加一起是零碎。
他的算法是:就算你把木星扔进热核反应堆烧掉,太阳的能量占比四舍五入还是 100%。再传送三个木星进来烧掉,把太阳系其他所有天体一起烧掉,太阳还是 100%。
这个算术的意义很直接:能源问题的答案只有一个方向,太阳。
地球上所有的化石燃料,本质上都是太阳能的库存。风能是太阳造成的大气环流。水能是太阳蒸发的水。核裂变是超新星爆炸的残留,跟太阳无关,但量级差了两三个数量级。
他说,这件事应该被认真考虑。
一百英里见方,够全美国用电
主持人问需要多大面积。他给了个会被反复引用的数字。
100 英里乘 100 英里,大约 160 公里乘 160 公里,铺满太阳能板,够全美国用电。
这个面积大概相当于犹他州的一个小角落,或者新墨西哥州的一块地。当然不会集中在一处,会有输电和储能的问题,但相对美国国土面积,这是个极小的比例。
欧洲同理。西班牙和西西里那些人口稀疏的地方铺开,就能供应整个欧洲的电力。
问题出在美国的政策上。光伏关税极高,而中国生产了全世界几乎所有的光伏板。关税把部署成本人为抬高了,经济性被扭曲。
他直接说了 SpaceX 和特斯拉要干什么:两家公司各自在美国建设每年 100 吉瓦的光伏制造产能,大概需要三年。他还鼓励别人也照做。
对其他国家的建议更简单:中国产的光伏板便宜得离谱,大规模铺开值得做。
明年年底,机器人开卖
人形机器人的进度,他给了几个时间点。
现在,特斯拉的 Optimus 在工厂里做简单任务。今年年底之前,会开始做更复杂的任务,但还在工业环境里。明年年底,开始向公众销售。
卖的条件是三条同时成立:可靠性足够高,安全性足够高,功能范围足够宽。他的说法是,基本上你可以让它做任何事。
自动驾驶这块,他的判断更干脆:这个问题基本已经解决了。
FSD 软件有时候一周更新一次。最近有保险公司给出信号,用 FSD 的车主保费打五折,理由是这套系统足够安全。
保险公司是用赔率说话的地方。半价保费的意思是,精算模型认为这套系统的出险率显著低于人类司机。这比任何发布会上的承诺都更有分量。
Robotaxi 已经在几个城市上线,今年年底会在美国大范围铺开。欧洲的监督式 FSD 审批,他希望能下个月拿到。中国的节奏差不多。
星舰要做的事,人类一次都没做成过
SpaceX 今年的目标是完全复用。这件事人类从来没做成过。
猎鹰 9 做到了部分复用。助推级回收了 500 多次,但二级每次都要扔掉,在大气层里烧毁。那一级的成本,相当于一架中小型喷气机。
星舰是有史以来最大的飞行器。
完全复用做成之后,进入太空的成本会降两个数量级,一百倍。他拿飞机打比方:如果每飞一次就得把飞机扔掉,机票会贵到没人坐得起。只加油的话,成本就是油钱。
具体到什么程度:低于航空货运的成本,每磅不到 100 美元,而且是轻松做到。
这个数字要放进历史里看。航天飞机时代,每磅入轨成本在几千美元量级。猎鹰 9 把它压到一两千。星舰如果做到一百以下,太空就从"任务"变成了"物流"。
物流的意思是,你不再需要为每一公斤讨价还价,可以按需发车。
太空里的太阳能板,效率是地面五倍
这里出现了整场对话里最反直觉的一段。
太空里的太阳能板,效率是地面的五倍。
原因有几个。太空里永远是晴天,没有昼夜,没有季节,没有天气。也没有大气衰减,同样一块板能多出大约 30% 的功率。
发电时长和单位功率同时拉满,五倍就出来了。
散热怎么办。太空的背阴面是 3 开尔文,接近绝对零度。太阳能板朝太阳,散热器背对太阳,永远晒不到,散热效率极高。
他的结论是:放 AI 算力最便宜的地方会是太空,这个判断在两年内成立,最迟三年。
几年内,SpaceX 会开始发射太阳能 AI 卫星。太空不缺地方,规模可以做到每年几百太瓦。
这个论证链条值得复述一遍。AI 的瓶颈是电和散热。太空解决了电,因为太阳能密度高五倍。太空解决了散热,因为背景温度是 3 开尔文,靠辐射散热不需要水也不需要压缩机。太空还解决了选址,因为地球上建数据中心要跟居民抢地。
三个约束同时松开,成本曲线就会穿过去。
五年后,AI 比全人类加起来聪明
主持人问十年后会发生什么。他说不知道十年后,但按现在 AI 的进展速度:
今年年底,AI 会比任何一个人都聪明。不晚于明年。
2030 或 2031 年,大概五年之后,AI 会比全人类加起来都聪明。
这两个时间点之间隔了五年,从"超过单个人"到"超过所有人加起来"。
第一个门槛的争议在于"聪明"怎么定义。单个领域的超越早就发生了,比如棋类、蛋白质结构、代码补全。他指的是通用意义上的超越。
第二个门槛的争议在于"加起来"是什么意思。人类的集体智能不是简单相加,是协作网络。而 AI 的协作网络可以复制、可以并行、可以共享全部记忆。
他说的"加起来",大概是指总量。总量超越的速度取决于两个变量:单个模型的能力提升速度,和部署规模。第一个变量已经在陡峭区间,第二个变量受限于电力,也就是前面说的瓶颈。
他想要真的企业号
有人问他,少年时的什么念头把他带到了今天。
他小时候读了很多科幻、奇幻和漫画。他喜欢技术。他说自己从没想过会走到今天这个位置,看起来极不可能。
他想做的事,是把科幻变成科学事实。
我们想要真正的星际舰队,真的《星际迷航》,有巨大的飞船在宇宙里航行,去别的行星,去别的恒星系,去从来没人去过的地方。(马斯克)
驱动他的东西,他叫做好奇心的哲学。
他想弄明白生命的意义。物理的标准模型在宇宙的开端和终结上到底对不对。还有那些我们不知道该问、但应该问的问题。AI 会帮上忙。
他也想知道到底有没有外星人。如果飞船真能飞到别的恒星系,我们可能会遇到外星人,也可能找到很多早已死掉的文明遗迹。
他说的其实是想知道到底发生了什么。对宇宙好奇,这就是他的哲学。
火星那段讲得很轻松。单程六个月,但行星每两年才对齐一次。有人问他愿不愿意死在火星上。
愿意,但不要是撞上去那一刻。(马斯克)
宁可乐观而错
主持人问他,希望自己的最后一句是什么。
他说他会鼓励所有人对未来保持乐观和兴奋。
然后给了一个概率上的理由:
就生活质量而言,宁可做一个乐观而犯错的人,也不要做一个悲观而正确的人。(马斯克)
这句话值得多想一层。
它不只是心理按摩。一个悲观而正确的人,会把资源投在防守上,投在阻止事情发生上,投在为坏结果做准备上。一个乐观而犯错的人,会把资源投在尝试上。
犯错的成本被摊薄,试对的收益被放大。前提是这个时期的技术曲线是陡峭向上的。如果曲线是平的,悲观更有道理,因为守成比冒险划算。
他对曲线的判断很明确:我们正处在历史上最有趣的时代,他不认为有比现在更有趣的时代。
回到开头那根蜡烛。整场对话里,他所有的动作都指向同一个方向:让这根蜡烛多几个备份,让它周围的黑暗少一点。
星舰是把火种带出去。机器人是把人从必须干的活里解放出来。太阳能和太空数据中心是给庞大的算力找到不会烧完的燃料。而他对时间表的乐观,本质上是一个下注:赌人类能在这根蜡烛被吹灭之前,把它变成一片星空。
以下是从这场对话里能直接拿走的东西:
- 如果你在判断 AI 行业的瓶颈,把目光从芯片挪到电网上。芯片的扩产周期以季度计,发电和输电以年计,时间尺度错配的地方就是钱和信息差所在。
- 查一下你所在地区的光伏上网电价和装机审批流程。太阳能已经是地球上最便宜的稳定电源,剩下的障碍基本是政策和并网,不是技术。
- 如果你是开发者或小团队,先跑开源模型。它落后闭源大约一年,成本低一到两个数量级,对绝大多数场景够用。这一年的差距,比多数人愿意付出的溢价要便宜。
- 机器人向公众销售的时间点,他给的是明年年底。家庭场景的准备可以从现在开始想:老人看护、儿童看护、宠物照料,这三个是需求最刚性、供给最缺的方向。
- 留意保险公司的定价。半价保费这种信号比任何发布会都准,精算师用真金白银在投票。凡是保险开始给折扣的技术,普及期都到了。
- 评估你手上的工作,问一句:这是"必须有人干"的活吗?如果是,它会被机器人替代,因为不替代就不可能有富足。如果它涉及判断、关系、创造力,短期内更安全。
- 保持乐观作为一种策略,而不是作为一种情绪。在一个技术曲线陡峭向上的时期,把资源投在尝试上的期望值,高于投在防守上。
Yeah. We're going to make this interesting.
I don't know. I mean, five.
You're most welcome. Yeah. I mean, I heard about the formation of the Peace Summit, and I was like, "Is that P-I-E-C-E?" Little piece of Greenland, the little piece of Venezuela.
All we want is peace.
Oh, thank you.
Well, we have an incredible team at Tesla.
And that's the reason.
Well, they're all very difficult technology challenges, but the overall goal of my companies is to maximize the future of civilization, basically maximize the probability that civilization has a great future and to expand consciousness beyond earth. So if you take SpaceX, for example, that SpaceX is about advancing rocket technology to the point where we can extend life and consciousness beyond earth to the moon, to Mars, eventually to other star systems. And I think we should always view consciousness, life as we know it as precarious and delicate because to the best of our knowledge, we don't know of life anywhere else.
I'm often asked, "Are there aliens among us?" And I'll say that I am one, but-
... they don't believe me.
I think if anyone would know if there are aliens among us, it would be me. And we have 9,000 satellites up there, and not once have we had to maneuver around an alien spaceship. So I don't know. Bottom line is I think we need to assume that life and consciousness is extremely rare and it might only be us. And if that's the case, then we need to do everything possible to ensure that the light of consciousness is not extinguished because we're effectively... The way I view it is, the image in my mind is of a tiny candle in a vast darkness, a tiny candle of consciousness that could easily go out.
And that's why it's important to make life multi-planetary, such that if there is a natural disaster or a manmade disaster on earth, that consciousness continues. That's the purpose of SpaceX. Tesla is obviously about sustainable technology, and also at this point, we've added to our mission, sustainable abundance. So with the robotics and AI, this is really the path to abundance for all. People often talk about solving global poverty or essentially how do we give everyone a very high standard of living? I think the only way to do this is AI and robotics. Which doesn't mean that it is without its issues. I mean, we need to be very careful with AI. We need to be very careful with robotics.
We don't want to find ourselves in a James Cameron movie, Terminator. He's great great movies, love his movies, but we don't want to be in Terminator, obviously. But if you have ubiquitous AI that is essentially free or close to it and ubiquitous robotics, then you will have an explosion in the global economy, an expansion in the global economy that is truly beyond all precedent.
Yes.
Yeah. I mean, a way to think of it is that if you have a large number of humanoid robots, the economic output is the average productivity per robot times the number of robots.
And actually my prediction is in the benign scenario of the future that the robots will actually make so many robots in AI that they will actually saturate all human needs. Meaning you won't be able to even think of something to ask the robot for at a certain point. There will be such an abundance of goods and services because my prediction is there'll be more robots than people.
Yeah. I mean, nothing's perfect, but it is a necessary... You can't have both. You can't have work that has to be done and amazing abundance for all because if it's work that has to be done, and only some people can do it, then you can't have abundance for all.
Yeah,
Exactly. But if you have billions of humanoid robots, and I think there will be, I think everyone on earth is going to have one and going to want one. Because who wouldn't want a robot to, assuming it's very safe, watch over your kids, take care of your pet. If you have elderly parents, a lot of friends might have said they have elderly parents. It's very difficult to take care of them.
Yeah, it's expensive. It's expensive and there aren't enough young people to take care of the old people. So if you had a robot that could take care of and protect an elderly parent, I think that would be great. That would be an amazing thing to have. And I think we will have those things. So overall, I'm very optimistic about the future. I think we're headed for a future of amazing abundance, which is very cool. And definitely we are in the most interesting time in history. I don't think there's a more interesting time in history.
I haven't put much time into the aging stuff. I do think it is a very solvable problem. I think when we figure out what causes aging, I think we'll find it's incredibly obvious. It's not a subtle thing. The reason I say it's not a subtle thing is because all the cells in your body pretty much age at the same rate. I've never seen someone with an old left arm and a young right arm ever in my life. So why is that? That means that there must be a clock, a synchronizing clock that is synchronizing across 35 trillion cells in your body. And there is some benefit to death, by the way.
There's a reason why we don't actually have a longer lifespan because if people do live forever for a very long time, I think there's some risk of an ossification of society, of things just getting locked in place. And it just may become stultifying, lack vibrancy. But that said, do I think we will figure out ways to extend life and maybe even reverse aging? I think that's highly likely.
I just think the natural thing is it's going to be very broad because AI companies will seek as many customers as they possibly can. And the cost of AI is already very low and it's plummeting every year. The cost of AI is almost meaningfully changing on a month to month basis.
Yes. There's open models. And the open models only luck... They're maybe a year behind the closed models. So I think AI companies will seek as many customers as possible, which means they'll provide AI to the world.
The limiting factor. Yeah. I think the limiting factor for AI deployment is fundamentally electrical power.
Yeah. I mean, we're seeing the rate of AI chip production increase exponentially, but the rate of electricity being brought online is-
Yeah. It's clear that we're very soon, maybe even later this year, we'll be producing more chips than we can turn on. Except for China. China's growth in electricity is tremendous.
Actually, solar is the biggest thing in China. I believe China's production capacity on solar is 1,500 gigawatts a year, and they're deploying over 1,000 gigawatts a year of solar. Now, for continuous solar load, you divide that by roughly, I don't know, four or five, call it that's around 250 gigawatts of steady state power paired with batteries. And that's a very big number. That's half of the average power usage in the US. So US power usage on average is 500 gigawatts. China, just in solar. Just in solar that can provide steady state power and batteries can do half of the US electricity output per year just with solar. Solar is by far the biggest source of energy.
And actually, when you look beyond, or even on earth, but certainly beyond earth, the sun rounds up to 100% of all energy. This is an important thing to consider. So the sun is 99.8% of the mass of the solar system. Jupiter is about 0.1% and everything else is miscellaneous. Now, even if you were to burn Jupiter in a thermal nuclear reactor, the amount of energy produced by the sun would still round up to 100% because Jupiter is only 0.1%. If you teleported three more Jupiters into our solar system and burnt three more Jupiters, and everything else in the solar system, the sun's energy would still round up to 100%. So it's really all about the sun.
And that's why one of the things we'll be doing with SpaceX within a few years is launching solar-powered AI satellites because the space is really the source of immense power and then you don't need to take up any room on earth. There's so much room in space and you can scale to, I think, ultimately hundreds of terawatts a year.
Yeah. So I mean, I guess rough way to think about it is 100 miles by 100 miles, or call it 160 kilometers, by 160 kilometers of solar is enough to power the entire United States. So 100 mile by 100 mile area is, I mean, you could take basically a small corner of Utah.
New Mexico. Obviously you wouldn't want it all in one place. But it is a very small percentage of the area of the US to generate all of the electricity that the US uses. And the same is true actually for Europe. You could take relatively unpopulated areas of say Spain and Sicily and generate all of the electricity power that Europe needs.
Well, there is-
Well, unfortunately in the US, the tariff barriers for solar-
... are extremely high. And that makes the economics of deploying solar artificially high because China makes almost all the solar and the-
Yeah. Well, I can tell you what we're going to do at SpaceX and Tesla is we're building up large scale solar. So the SpaceX and Tesla teams both separately are working to build to a hundred gigawatts a year of solar power in the US, of manufactured solar power. And that'll probably take us, I don't know, about three years or something. But these are pretty big numbers. And I'd encourage others to do the same. We obviously don't control the US tariff policy, but for other countries that China makes solar cells that are incredibly low cost and I think it would be worth doing large scale solar.
Yeah.
Well, humanoid robotics will advance very quickly. I think we do have some of the Tesla Optimus robots doing simple tasks in the factory except probably later this year. By the end of this year, I think they'll be doing more complex tasks, but still deployed in an industrial environment. And probably sometime next year, I'd say that... By the end of next year, I think we'd be selling humanoid robots to the public. That's when we are confident that it's very high reliability, very high safety, and the range of functionality is also very high. You can basically ask it to do anything you'd like.
Yes. So the Tesla Full Self-driving software, we update it sometimes once a week. And recently, some of the insurance companies have said that it is actually so safe that Tesla Full Self-driving is so safe that they're offering customers half price insurance if they use Tesla Full Self-driving in their car.
Yeah. I think self-driving cars is essentially a solved problem at this point.
And Tesla has rolled out robotaxi service in a few cities and will be very widespread by the end of this year within the US. And then we hope to get supervised full self-driving approval in Europe, hopefully next month.
Yeah. And then maybe a similar timing for China, hopefully.
Sure. The major breakthrough that SpaceX is hoping to achieve this year is full reusability. So no one has ever achieved full reusability of a rocket, which is very important for the cost of access to space. We've achieved partial reusability with Falcon 9 by landing the boost stage. We've now landed the boost stage over 500 times, but we have to throw away the upper stage. The upper stage burns up on reentry for Falcon 9. And the cost of that is equivalent to a small to medium size jet. But with Starship, which is a giant rocket, it's the largest flying machine ever made.
Yeah. Mars and the moon as well as for high volume satellite stuff. So Starship, hopefully this year we should prove full reusability for Starship, which will be a profound invention because the cost of access to space will drop by a factor of a hundred when you achieve full reusability. It's the same sort of economic difference that you would expect between, say, a reusable aircraft and a non-reusable aircraft. If you have to throw your aircraft away after every flight, that would be a very expensive flight. But if you only have to refuel then it's the cost of the fuel. And so that's really the fundamental breakthrough that gets the cost of access to space.
We think below the cost of freight on aircraft. So under $100 a pound type of thing easily. So it makes putting large satellites into space, very cheap. And then when you have solar in space, you get five times more effectiveness, maybe even more than that, than solar on the ground because it's always sunny.
Yeah. Well, it's always sunny, so you don't have a day night cycle or seasonality, or weather, and you get about 30% more power in space because you don't have atmospheric attenuation of the power. The net effect is solar is five times more... Any given solar panel will do five times more energy in space than on the ground.
I think the case, it's a no-brainer for building solar-powered AI data centers in space because as you mentioned, it's also very cold in space. If you're in the shadow, it's very cold in space. Just 3 degrees Kelvin. So you have solar panels facing the sun and then a radiator that's pointed away from the sun, so it has no sun incidents. And then it's just cooling. It's a very efficient cooling system. So net effect is that the lowest cost place to put AI will be space, and that'll be true within two years, maybe three. Three at the latest.
I don't know what's going to happen in 10 years, but the rate at which AI is progressing, I think we might have AI that is smarter than any human by the end of this year. And I would say no later than next year.
And then probably by 2030 or 2031, call it five years from now, AI will be smarter than all of humanity collectively.
I could make a joke about peace.
Well, I mean, as a kid, I read a lot of science fiction, sci-fi fantasy books-
... and comic books. And I always liked technology. I didn't expect to be where I am today. Seems incredibly implausible. But yeah, I was inspired by reading about books about the future, about science fiction. And I guess I want to make science fiction, not fiction forever at some point turn science fiction to science fact. And we want to have Starfleet and Star Trek really, for real, where we actually have giant spaceships traveling through space, going to other planets, traveling to other star systems, going places where we've never gone.
Huh?
Yeah.
No, I guess my essential, what I would call the philosophy of curiosity. I'd like to understand the meaning of life. Is the standard model of physics correct regarding the beginning of life, beginning of existence and the end of the universe? What questions do we not know to ask that we should ask? And AI will help us with these things. So I'm just trying to understand, how did we get here? What's going on? What's real? Are there aliens? Maybe there are. And if we've got spaceships that are traveling to other star systems, we may encounter aliens or we may find many long dead alien civilizations. But I just want to know what's going on. I'm curious about the universe and that's my philosophy.
Yeah. I mean, I would say, I don't-
I've been asked but
It's six months.
Yeah, six months, but the planets only align every two years.
So yeah, I've been asked a few times, "Do I want to die on Mars?" And I'm like, "Yes, but just not on impact."
Well, I think generally, I think my last words would be, I would encourage everyone to be optimistic and excited about the future.
And generally, I think for quality of life, it is actually better to err on the side of being an optimist and wrong rather than a pessimist and right.
我的看法是,我脑海中浮现的画面是:在广阔黑暗中有一支微小的蜡烛,那是一簇很容易熄灭的微小意识烛光。
The way I view it is, the image in my mind is of a tiny candle in a vast darkness, a tiny candle of consciousness that could easily go out.
如果你有大量人形机器人,经济产出就等于每台机器人的平均生产力乘以机器人的数量。
if you have a large number of humanoid robots, the economic output is the average productivity per robot times the number of robots.
我认为,AI 部署的限制因素从根本上说是电力。
I think the limiting factor for AI deployment is fundamentally electrical power.
所以太阳占太阳系质量的 99.8%。木星约占 0.1%,其余一切都只是零头。
So the sun is 99.8% of the mass of the solar system. Jupiter is about 0.1% and everything else is miscellaneous.
我认为,自动驾驶汽车在现阶段基本上已经是一个被解决的问题。
I think self-driving cars is essentially a solved problem at this point.
所以,Starship 希望今年能证明完全可重复使用;这将是一项意义深远的发明,因为一旦实现完全可重复使用,进入太空的成本将下降一百倍。
So Starship, hopefully this year we should prove full reusability for Starship, which will be a profound invention because the cost of access to space will drop by a factor of a hundred when you achieve full reusability.
所以净效果是,部署 AI 成本最低的地方将是太空,而且这会在两年内成真,也许三年。最迟三年。
So net effect is that the lowest cost place to put AI will be space, and that'll be true within two years, maybe three. Three at the latest.
我认为,到今年年底,我们可能就会拥有比任何人类都聪明的 AI。而且我要说,最迟明年。
I think we might have AI that is smarter than any human by the end of this year. And I would say no later than next year.
有人问过我几次:‘你想死在火星上吗?’我说:‘想,但别是撞上去那一刻。’
I've been asked a few times, "Do I want to die on Mars?" And I'm like, "Yes, but just not on impact."
总的来说,我认为就生活质量而言,宁可做一个乐观但判断错误的人,也比做一个悲观却判断正确的人更好。
And generally, I think for quality of life, it is actually better to err on the side of being an optimist and wrong rather than a pessimist and right.
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