访谈·中文精读

马斯克对话萨姆·奥特曼(2016.9.15)

Elon Musk Interview with Sam Altman (2016.9.15)

中文精读由 AI 基于原始视频与文稿改写,非官方翻译,可能存在偏差或错误;表述以原视频为准。说明

2016年9月,马斯克在台上说了一个数字:5厘米每秒。

这是特斯拉工厂流水线的平均速度,S型和X型混在一条线上跑。一辆车从这头走到那头,比蜗牛散步快不了多少。马斯克说,他有信心把这条线提速20倍,做到1米每秒。

1米每秒是什么概念?人慢悠悠走路的速度。快走大概1.5米,跑起来能到10米以上。

就算他兑现了这个20倍的提升,汽车生产线还是走得比人慢。

他把这件事称为自己2016年最大的顿悟:真正重要的是那个制造机器的机器,是工厂本身。造工厂比造车难至少两个数量级。

这个细节值得先摆出来,因为它几乎是马斯克思维方式的一个缩影。他不盯最后那个光鲜的产品,他盯的是产品背后那条最慢、最笨、最难改的链条,然后问:这条链能不能快20倍。

这场2016年的对话里,他谈得更多的是别的东西:一个22岁的年轻人今天该去解决什么问题,技术为什么会退步,火星还差几年,AI 到底危险在哪里。

如果今天22岁,他会挑三个问题

访谈一开场,奥特曼就把这个问题抛了出来:你年轻时说过有五个问题最重要,如果今天你22岁,会选哪几个?

马斯克先给了一个缓冲。他说,一个人做的事情对社会有用,这就够了,不一定非要改变世界。哪怕只是一个小游戏,或者照片分享功能的一点改进,只要对大量的人有一点点好处,那也挺好。

东西不需要非得改变世界才算好。

这个前提说完,他才开始列清单。真正可能影响人类未来的东西,排在最前面的是AI。他用了一个很重的词:近期内影响人类的最大单一事项。他强调的重点落在“以什么方式发生”上:AI 的到来必须是好的那种,如果他能看一眼水晶球,他会希望看到那样的结局。这件事有走错的可能,所以必须确保它走对。这是他当下认为最紧迫的工作。

第二项是基因学。如果能真正解决遗传病,如果能用基因重编程预防痴呆、阿尔茨海默症这类疾病,那会是非常好的事。

第三项是人脑的高带宽接口。他在这里讲了一个当时听起来很超前、现在回头看相当准确的判断:人类其实已经有了一个数字化的“第三层自我”,你的邮箱、电脑、手机、应用,都是它的延伸。某种意义上人类已经是超人了,卡住的地方在接口。大脑皮层和这个数字自我之间的那条通道,带宽太低。把这个带宽瓶颈解决掉,对未来的重要性不亚于前两项。

他还补了一句容易被忽略的话:他当年在大学里想的那五件事,是生命多行星化、加速向可持续能源转型、互联网、基因学和AI。他从来没指望自己五件全干。那时候他以为自己会从电动汽车的电气化入手,实习做的就是先进超级电容,想看它能不能相对电池形成突破。去斯坦福读研,方向也是电动车用的先进储能技术。

然后他1995年把这个计划搁下了,去开互联网公司。原因是他看到了一个时间窗口:某些技术正处在那条曲线最陡的一段,过了就过了。他不想在斯坦福读完博士,然后眼看着这一切在别处发生。而且他不确定自己研究的技术最终能不能成功。靠很多最终对现实世界毫无影响的东西,也能拿到博士学位。他当时的优化目标很朴素:我能做什么,才是真的有用。

“有用”可以被算出来:效用差乘以受益人数

奥特曼追问了一个更实操的问题:一个人怎么判断自己能不能成事、能有多大影响?马斯克的回答里藏着一个公式。

他说,不管你想创造的是什么,先算两件事:它相对现有技术水平带来的效用增量是多少,再乘以它能影响多少人。

一个带来巨大改变、但只影响少量到中等数量人的东西,很好。一个只带来很小改变、但影响海量人的东西,同样好。这两者的曲线下面积其实差不多。(马斯克)

这个说法看着朴素,但它把“我该干什么”这个模糊的问题,拆成了两个可以估的量。很多人卡在第一步,总想找一个惊天动地的点子;也有人卡在第二步,做出了很好的东西却只有几十个用户。两头都不占,面积就小。

马斯克自己的选择,恰好是在做面积最大的那一类。电动车影响的是全球每年几千万辆的汽车市场,可持续能源影响的是整个能源结构,火星殖民影响的是物种存续,AI 影响的是所有人。

他自己也承认,这套算法里少算了“成功概率”这一项,而这一项会极大地改变计算结果。他后来补了一句:想做博士生的人,大多不该去读。这个判断要放到概率的语境里才成立。

SpaceX 的成功概率,他自己估的是不到一成

奥特曼提到,当年马斯克决定做 SpaceX,在外人看来是彻底的疯癫。马斯克直接承认:是的,非常疯。

他补了一个很关键的区分。如果目标是把风险调整后的回报做到最优,那开火箭公司是彻头彻尾的愚蠢。他的目标另有其事。

推理链条是这样的:如果没有人去做点什么改进火箭技术,人类就会一直被困在地球上。而当时那些大型航空航天公司,对激进创新毫无兴趣,只想让手里的老技术每年变好一点点,有时候甚至会更糟。

他举的例子很有说服力。1969年,人类用土星五号登上了月球。后来航天飞机只能把人送到近地轨道。再后来航天飞机退役了。这条趋势线往下走,终点是零。

很多人默认技术会自动一年比一年好。它不会。技术只有在聪明人拼命推动的时候才会变好。没人管它,它就退步。

他讲了两个历史案例。古埃及人建起了那些不可思议的金字塔,然后他们忘了怎么建金字塔,后来连象形文字都不会读了。罗马人修出了那些了不起的道路、引水渠和室内管道,然后他们也忘了怎么干这些事。

“熵不站在你这边。”这是他的原话。

年轻人该不该读博:大多不该

访谈里有一段问答短得像刀。

奥特曼问:你觉得今天想让自己有用的人,应该去读博士吗?

马斯克:大多不该。

他给的理由要回到1995年那个决定。当时他在斯坦福读研,方向是电动车储能技术,他要把这个计划搁下来去创业。理由是时间的成本:一段技术最陡的爬坡就那么几年,人在学校里待着,会错过整段。

更麻烦的是产出本身。一个人可以在一条没有现实杠杆的路径上,把最宝贵的十年投进去,出来发现世界已经走过去了。他那时的算法就一条:哪些事能真的落地,我就去做哪个。

这段话不是说学术训练没价值,他讲的是另一种风险:方向选错,努力就白费。反过来看,他自己后来做的事,每一件都踩在拐点上。2002年做 SpaceX,2004年投特斯拉,2015年底参与创办 OpenAI。时间点都不算早,但都在那条曲线开始变陡的位置。

恐惧是真的,宿命论能救你

奥特曼说马斯克“异常无畏”,能在所有人都说这是疯主意的时候往前走。他问这种力量从哪来。

马斯克的回答很直接,他并不无畏。

我其实觉得我恐惧感挺强的。我属于会害怕的那类人,不是没有恐惧的那种,我感受得很强烈。只是有些事情足够重要,你足够相信它,你就会带着恐惧去做。(马斯克)

他加了一句:人不该因为感到恐惧就认定自己不该做这件事。感到恐惧是正常的,不感到恐惧才说明精神出了问题。

然后他给了一个听起来有点奇怪、但很实用的方法:宿命论。

如果你干脆接受那些概率,恐惧就会减弱。他做 SpaceX 的时候,自己估的成功率不到10%。他接受了这个判断,大概率会输光一切。但他想,也许能往前推一点。哪怕自己死了,也许别的公司能接过接力棒继续往前跑。那也算有点用。

特斯拉也一样。他当时认为一家汽车公司成功的概率极低。

“接受概率”这一步,把恐惧从一个拦路的东西,变成了一个已经付过账的成本。

火星:从“也许根本不可能”到“确定有一条路”

奥特曼问了一个当时几乎算玩笑的问题:火星殖民地现在成功的概率有多大?

马斯克的回答出人意料。他说,说来奇怪,他觉得概率相当不错。

然后他给了一个很重要的区分,这个区分比那个概率数字更有价值。

到这一步,我确定存在一条路。我确定成功是可能的结果之一,一个能自我维持、而且能持续增长的火星殖民地。几年前我还不确定成功是不是哪怕可能的结果之一。(马斯克)

这里的变化跟数值无关。它跟成功率从5%变成50%没有关系,变化发生在更底层:从“不确定这件事在物理和工程上到底可不可行”,变成“可行性已经确认,剩下的是执行和存活问题”。对一个工程师来说,这是两种完全不同的处境。

时间表上,他说让相当数量的人去火星,大概10年能做成,也许更快,9年。前提是两个:SpaceX 在那之前不能死掉,他自己也不能死,如果他死了,得有人接班继续干。

奥特曼建议他别坐第一次发射。他说第一次发射本来就是无人的。他想去,唯一的顾虑是网络延迟。

他顺便算了一笔账。火星距太阳大约12光分,地球距太阳8光分。最近的时候火星离地球4光分,最远的时候20光分。而且信号还不能直接穿过太阳。

4到20分钟的延迟意味着什么?意味着在火星上刷网页、打视频电话、开远程会议,全部体验极差。这也解释了他后来去做卫星互联网,以及为什么他反复强调人机接口的带宽问题。带宽这件事,他从大脑想到了行星之间。

AI 真正的危险,来自身后拿着它的那个人

说到 AI,马斯克先做了一个免责声明。他强调自己讲的属于预测,代表他能想到的所有选项里最好的那一个。别人可能想出更好的方案。

这个最好的方案叫:AI 技术的民主化。

意思是,没有任何一家公司、任何一小群人,能控制先进的 AI 技术。他给出的理由跟涨价、垄断这类常规担忧没关系。他担心的是安全。一旦有某个极其强大的 AI 存在,你根本不知道最终谁会控制它。它可能被某个坏家伙偷走,某个邪恶的独裁者或者国家派情报机构把它弄到手。局面会变得极不稳定。

他在这里做了一个很关键的澄清,这个澄清在2016年说得很早,后来被大量误读:

我不认为风险是 AI 一上来就发展出自己的意志。我担心的更多是,有人可能用它来做坏事。哪怕那个人本来不打算做坏事,别人也可能从他手里把它夺走,然后拿去做坏事。(马斯克)

这个判断解释了 OpenAI 为什么以那种方式成立。奥特曼和团队创办 OpenAI 的目的,就是把 AI 技术扩散出去,别让它集中在少数人手里。

但他紧接着说,民主化必须配上另一件事:解决大脑皮层的高带宽接口。

人类太慢了

访谈里有一段对话特别能说明问题。

马斯克说,我们需要解决皮层的高带宽接口。停顿了一下,他自己接了一句:人类太慢了。

奥特曼跟着重复:人类太慢了。

马斯克:对,就是这样。

他接着解释。你今天已经在用一个数字化的自我了。你的邮件、你的电脑、你的手机、你的应用,都是你的延伸。用他的说法,人类实际上已经是超人的水平,因为这些工具极大扩展了能力。卡住的地方在接口:从大脑皮层到这个数字自我之间的通道,带宽极低。

想想人是怎么输出的。打字,一分钟几十个词。说话,一分钟一百多个词。而人脑内部的运算量,远远超过这个输出速率。所有的想法、判断、直觉,最后被压成一条窄窄的语音或文字流,传给机器。这是整套系统里最窄的那一段。

皮层和边缘系统已经给过答案

为了说明人机融合为什么可行,马斯克用了一个现成的例子:人脑自己。

人脑里有皮层,负责思考,是后来进化出来的上层结构。也有边缘系统,负责本能和情绪,是更原始的那部分。这两个系统一起工作,配合得相当好。偶尔会打架,但总体运转良好。

他说,他没见过任何一个人希望自己去掉皮层,或者去掉边缘系统。

这个类比指向的结论是:如果能把皮层和人的数字延伸之间的连接带宽提上去,人就会变成一个 AI 与人类的共生体。如果这种能力普及开来,任何想要的人都能拥有,那么控制问题也就顺带解决了。

他的原话大意是:我们不用担心某个邪恶的独裁者 AI,因为那时候我们集体就是那个 AI。

这个想法的争议很大,但逻辑是自洽的。如果超级智能只有一份,谁拿到手谁说了算,那是不稳定的。如果超级智能有几十亿份,每个人类都带着一份,权力结构就回到了人类自己手里。

OpenAI 六个月,四十个人

奥特曼有点不好意思地问:作为一家成立六个月的公司,你觉得 OpenAI 现在怎么样?

马斯克说,进展不错,团队非常有才华,工作很拼。

他提了一个结构上的细节。OpenAI 当时是一家501(c)(3)的非营利机构。他说,很多非营利组织没有紧迫感,这没什么问题,它们本来也不需要。但 OpenAI 有紧迫感,因为团队真的相信这个使命。这个使命是降低未来出现生存性伤害的风险。

当时公司四十个人。他说自己相当佩服这些人做的事和他们的水平,而且公司一直在找相信这个使命的人加入。

这段对话放在2016年看很平常。放在今天回头看,它是整个访谈里信息量最大的一段。九个月之后,2017年,关于 OpenAI 方向和治理的分歧开始公开化。马斯克2018年离开了董事会。再往后就是众所周知的那条路线:2019年成立带利润上限的 OpenAI LP,微软进来,技术不再完全开放。

马斯克在访谈里说的那套“民主化”,和 OpenAI 后来的实际路径几乎朝着两个方向走。他当时把判断标准说得很清楚:不能让任何一家公司或一小群人控制先进的 AI,那非常危险。

他80%的时间在做工程和设计

访谈快到尾声,奥特曼问了一个很基本的问题:你现在每天的时间怎么分配?

马斯克的回答,把很多人对他的想象推翻了。

很多人以为我肯定花很多时间跟媒体打交道,或者处理商业上的事。实际上几乎所有时间,大概80%,都花在工程和设计上,做下一代产品。(马斯克)

具体分工是这样的。

SpaceX 这边,格温·肖特维尔(Gwynne Shotwell)是首席运营官,管法务、财务、销售和一般商业活动。马斯克自己的时间几乎全在工程团队,改进猎鹰9号和龙飞船,开发火星殖民的架构。

特斯拉这边,他在设计工作室大概每周半天,处理审美、外观和手感这些事。剩下大半个星期,全在过整车的工程,以及工厂的工程。

奥特曼提起一段往事。很多年前马斯克带他参观 SpaceX,最让他震撼的,是马斯克知道那枚火箭的每一个细节,每一处工程的来龙去脉。奥特曼说,很多人对马斯克的印象跟这个差得很远。

马斯克回了一句很淡的话:很多人觉得我是个生意人,这也没关系,做生意没什么不好,但真的不是那样。

这段对话值得停一下。外界给马斯克贴的标签,从“营销天才”到“资本玩家”到“话题制造者”,全都在商业层面。而他自己描述的日常,是80%的时间在跟工程师抠细节。这个比例和他做的那些“疯决定”是配套的。只有真的钻到工程细节里,人才能判断一件事在物理上可不可行,而不是听行业里的人说“这做不成”。

真正难的是制造机器的机器

回到开头那个5厘米每秒。

奥特曼看着车间里的机器人干活,说这些车就这么“变”出来了,挺神奇的。马斯克马上泼了盆冷水:这里的自动化水平,跟未来的超级工厂(Gigafactory)和 Model 3 的生产线比,还很低。

他讲了自己这一年的顿悟。真正重要的是那个制造机器的机器,是工厂。造工厂比造车难至少两个数量级。

速度对比很残酷。当时那条混线,S 和 X 一起跑,平均速度大概5厘米每秒。目标是多少?至少1米每秒,20倍。

他把这个速度放到了人的尺度里做参照。1米每秒是慢走,快走1.5米每秒,人全力跑能超过10米每秒。现在的产线是0.05米每秒,慢得离谱。就算做到1米每秒,你走路还是比产线快。

这个对比里藏着一个很少被讲透的道理。人们看制造业,看的是产品:这辆车设计得好不好看,加速几秒,续航多少。马斯克看的是产能爬坡的速度。一家车企能不能活,最后卡在产线上:那条线能不能稳定地、以足够快的节拍把车吐出来。这件事的难度,比产品本身高两个数量级。

这也解释了他后来为什么在“产能地狱”里睡工厂地板。他认定那一环是整个系统里最难、最该被盯住的地方,所以人得在场。

他给你的一套判断标准

把这场访谈从头到尾串一遍,马斯克给的东西可以当一套判断框架来用,跟鸡汤没关系。下面几条是可以直接上手的。

  • 用“效用差乘以受益人数”给想法排序。拿张纸,左边写你的方案相对现状能带来多大增量,右边写它能触达多少人。两个数乘一下。改变巨大但只服务少数人,和改变微小但服务所有人,面积可能一样大。真正该避开的是两头都小的方案。
  • 先确认可行性,再谈概率。马斯克谈火星时的用语值得学。他之前的状态是“不确定成功是不是可能的结果之一”,现在是“确定存在一条路”。在你投入之前,先分清自己卡在哪一层:物理上根本做不到,还是做得到但成功率低。这两者的应对方式完全不同。
  • 用接受概率来消解恐惧。把最坏结果写下来,承认它可能发生,然后问自己:即使这样,我能不能至少把球往前推一点。马斯克做 SpaceX 时估的成功率不到10%,他接受了输光一切的可能,然后就动手了。
  • 别默认技术会自动变好。金字塔、罗马水道、土星五号,都是退步的实例。任何你正在依赖的技术能力,如果没有人持续推动,它会衰减。反过来,这也意味着很多领域存在被低估的机会:去找那些“明明可以更好、但一直没人动”的地方。
  • 别把读博当默认路径。先问自己:我研究的东西,最终对现实世界有杠杆吗?如果答案模糊,去一线做,比在象牙塔里做五年更可能让你有用。
  • 盯住那个“制造机器的机器”。不管你在做软件还是硬件,最难的往往在稳定产出产品的那套系统上。找到你所在领域里那条“产线”,它的改进空间通常比产品本身大一个数量级。
  • 检查你的带宽。你脑子里想的,和你实际输出给世界的,中间有一条很窄的通道。打字、说话、开会,都是低带宽接口。想清楚哪一段最堵,然后优先拓宽它。

最后补一句。这场对话发生在2016年9月,那时候特斯拉还在产能地狱的门口,火星计划还只是一张架构图,OpenAI 只是个四十人的非营利组织。九年过去,回头看他对 AI 风险的判断、对工厂难度的判断、对带宽瓶颈的判断,几乎每一条都兑现了。

他当年那句“我确定存在一条路”,现在听,分量不太一样了。

YouTube 平台字幕(自动或人工),可能有识别错误,人名与数字请以视频为准。已按语句合并成段落,并去掉 um、uh 等口头填充词。 文稿来源:YouTube(自动/人工字幕)。

0:00Today we have Elon Musk. Elon, thank you for joining us. Yeah, thanks for having me. So, we want to spend the time today talking about your view of the future and what people should work on. So, to start off, could you tell us you famously said when you were younger there were five problems that you thought were most important for you to work on. if you were 22 today, what would the five problems that you would think about working on be?

0:22well, first of all, I think I think if somebody is doing something that is useful to the rest of society, I think that's a good thing. Like, it doesn't have to change the world. Like, you know, if you're doing something that has high value to to people, and and frankly, even if it's something if it's like just a little game or you know, the some improvement in photo sharing or something. If it If it has a If it has a small amount of of good for a large number of people, that's I mean, I think that's that's fine. Like, stuff doesn't need to be change the world just to be good.

1:00but it you know, in terms of things that I think are most likely to affect the the future of humanity, I think AI is probably the single biggest item in the near term that's likely to affect humanity. So, it's very important that we have the advent of AI in a good way that that's is something that if you were if you could look at the crystal ball and and see the future, you would like you would like that outcome.

1:28because it is something that could go could go wrong as we've talked about many times. and so, we really need to make sure it goes right. That's that's I think AI work working on AI and making sure it's great future. That's that's the most important thing I think right now. the most pressing item. Sec- then I would say anything in to do with with genetics. if you can actually solve

1:58genetic diseases, if you can prevent dementia, or Alzheimer's or something like that, that with genetic reprogramming, that would be wonderful. So, I think this genetics it might be the sort of second most important item. I I think having a high bandwidth interface to the brain, like we're currently bandwidth limited. We we we have a digital tertiary self in the form of our email capabilities, our computers, phones, applications.

2:30we're effectively superhuman, but we're extremely bandwidth constrained in that interface between the cortex and your sort of that that tertiary digital form of yourself, and having solved that bandwidth constraint would would be, I think, very important for the future as well. So, one of the I think most common questions I I hear young people at ambitious young people ask is, "I want to be the next Elon Musk. How do I do that?" obviously, the next Elon Musk will work on very different things than than you did, but what have you done or what did you do when you were younger that you think sort of set you up to have a big impact?

3:08Well, I think first of all, I should say that I do not expect to be involved in all these things. So, the the the the the the five things that I thought about the time in in college, quite a long time ago, 25 years ago, you know, being you know, making life multi-planetary, accelerating the accelerating the transition to sustainable energy, the the internet, broadly speaking, and and then genetics and AI, I think I didn't expect to be involved in in in all of those things. I actually, at the time in college, I I sort of thought helping with electrification of of of cars was how I would start out, and that's that's actually what I worked on as an intern was

3:53advanced ultra capacitors with to see if they they would be a breakthrough relative to batteries for energy storage in in cars. And then when I came out to go to Stanford, that's what I was going to be doing my grad studies on is is was working on advanced energy storage technologies for electric cars. And then I put that on hold to start an internet company in in '95 because the the the there does seem to be like a time for particular technologies when they're at a steep point in the inflection curve. And and I didn't want to you know, do a PhD at Stanford and then and watch it all happen and then a- a- and I wasn't entirely certain that the technology I'd be working on would

4:37actually succeed. I can get you can get a you know, doctorate on many things that ultimately are not do not have a practical bearing on the world. and I wanted to you know, just I I really was just trying to be useful. That's the optimization. It's like what what are the what can I do that would actually be useful?

4:57Do you think people that want to be useful today should get PhDs? mostly not. So what What is the best way to be useful? but mostly not. How should someone figure out how they can be most useful? whatever this thing is that you're trying to create, what would what would be the utility delta compared to the current state of the art times how many people would affect.

5:21So that's why I think having something that has a that's that has a makes makes a big difference, but affects a sort of small to moderate number of people is great. A- as is something that makes e- even a small difference, but it but affects a vast number of people. Like the area Yeah.

5:36Un- you know, under the curve. Yeah, exactly. The under the area under the curve is would actually be roughly similar for those two things. So it's actually really about yeah, I trying to be useful and And then when you're trying to estimate probability of success, so you say this thing will be really useful, good area under the curve. I guess to use the example of SpaceX. Mhm.

5:57When you made the go decision that you were actually going to do that, this was kind of a very crazy thing at the time. Very crazy, for sure. Yeah. They were not sure about saying that. but I kind of agree I agreed with them that it was quite crazy. Crazy if if if the objective was to achieve the best risk-adjusted return, starting a rocket company is insane.

6:19but that was not that was not my objective. I I I suddenly come to the conclusion that if something didn't happen to improve rocket technology, we'd be stuck on Earth forever. and and the big aerospace companies had just had no interest in radical innovation. all they wanted to do was try to make their old technology slightly better every year. And in fact, sometimes it would actually get worse.

6:46and particularly in rockets, it's pretty bad. Like the in in '69 we were able to go to the moon with a Saturn V, and then the space shuttle could only take people to low Earth orbit, and then the space shuttle retired. I mean, that that trend is basically trends to zero. It if you will sometimes think technology just automatically gets better every year, but it actually doesn't. It only gets better if smart people work work like crazy to make it better. That's how any technology actually gets better.

7:12And by itself, technology if if people don't work on it, actually will decline. I mean, you could look at look at the history of civilizations, many civilizations, and and look at say, ancient Egypt where they were able to build these incredible pyramids, and then they they basically forgot how to build pyramids.

7:30and and then even hieroglyphics, they forgot how to read hieroglyphics. Or you look at Rome and how they were able to to build these incredible roadways and aqueducts and indoor plumbing, and they forgot how to do all of those things. And there are many such examples in in history.

7:49So I I think you should always bear in mind that you know entropy is not on your side. Yeah. One thing I I really like about you is you're unusually fearless and willing to go in the face of other people telling you something that's crazy. And I know a lot of pretty crazy people you still stand out. where does that come from or how do you think about making a decision when everyone tells you this is a crazy idea? Or where do you get the internal strength to do that?

8:17Well, first of all I'd say I I actually think I I think I fear fear quite strongly. so it's not as though I just have the absence of fear. I have I feel it quite strongly. but there are just times when something is important enough you believe in it enough that you you do it in spite of fear.

8:37So speaking of important things. Like people shouldn't think I I I I people shouldn't think well, I feel fear about this and therefore I shouldn't do it. it's normal to be to feel fear. Like you'd have to There'd have to be something mentally wrong if you didn't feel fear. So you just feel it and let the importance of it drive you to do it anyway.

8:57Yeah. You know, I actually something that can be helpful is fatalism to some degree. if you just if you just accept the probabilities then that diminishes fear. so when I started SpaceX I thought the odds of success were less than 10%. and I just accepted that actually probably I would just lose lose everything.

9:22But that maybe we would make some progress. If we could just move the ball forward, even if we died, maybe some other company could pick up the baton and move and keep moving it forward. so that would still do some good. yeah, same with Tesla. I thought you know, the odds of a car company succeeding were extremely low. What do you think the odds of the Mars colony are at this point today?

9:45Well, oddly enough, I actually think they're pretty good. So, like when can I go? Okay. at this point, I am certain there is a way. I'm certain that success is one of the possible outcomes for establishing a self-sustaining Mars colony, in fact, a growing Mars colony. I'm certain that that is possible.

10:07whereas until maybe a few years ago, I was not sure that success was even one of the possible outcomes. So, a meaningful number of people going to Mars, I I think this is potentially something that can be accomplished in about 10 years. maybe sooner. maybe 9 years. I need to make sure that SpaceX doesn't die between now and then, and that I don't die, or if I do die, that someone takes over who will continue that.

10:35You shouldn't go on the first launch. Yeah, exactly. well, the first the first launch will be robotic anyway, so I want to go except for the internet latency. Yeah, the internet latency is going to be pretty significant. Mars is roughly 12 light minutes from the sun, and Earth is eight light minutes, so at closest approach, Mars is four light minutes away. At furthest approach, it's 20.

10:57A little more cuz you have to you can't sort of talk directly through the sun. Speaking of really important problems, AI. So, you have been outspoken about AI. could you talk about what you think the positive future for AI looks like and how we get there? Okay, I I mean, I do want to emphasize that this is not really

11:20something that I advocate or or this is not prescriptive. This is simply pre- hopefully predictive. cuz people will sometimes say, like, well, like like this is something that I want to occur instead of So, this is something I think that probably is the best of the available alternatives.

11:42the best of the available alternatives that I can come up with and maybe someone else can come up with a better approach or better outcome is that we achieve democratization of AI technology, meaning that no one company or small set of individuals has control over advanced AI technology. Like that that's very dangerous.

12:07It could also get stolen by somebody bad, you know, like some evil dictator or country could send their intelligence agency to go steal it and gain control. It just becomes a very unstable situation, I think if you've got any any incredibly powerful AI. you just don't know who's who's going to control that. So, it's not as I think that the risk is that the AI would develop a will of its own right off the bat. I think it's more It's the concern is that some someone may use it in a way that is bad.

12:39and even if they weren't going to use it in a way that's bad, that somebody could take it from them and use it in a way that's bad. That that I think is quite a big danger. So, I think we must have democratization of AI technology and make it widely available. and that's you know, the reason that obviously Yumin the rest of the team you know, created OpenAI was to help with the democrat- help help spread out AI technology so it doesn't get concentrated in the hands of a few.

13:10and and but then that of course that needs to be combined with solving the high-bandwidth interface to the cortex. Humans are so slow. Humans are so slow. Yes, exactly. But you know, we we already have a situation in our brain where we've got the cortex and limbic system. And the limbic system is is kind of the I mean, that's that's the primitive brain. It's kind of like the your your instincts and whatnot.

13:42And then the cortex is the thinking upper part of the brain. Those two seem to work together quite well. occasionally your cortex and limbic system may disagree, but they It generally works pretty well. Generally works pretty well. And it's like rare to find someone who I I've not found someone who wishes to either get rid of their cortex or get rid of their limbic system.

14:02Very true. Yeah, it's that's unusual. So, so I think if if we can effectively merge with AI, like improving that the the the neural link being between your cortex and the the the your digital extension of yourself, which already, like I said, already exists, just has a bandwidth issue.

14:26and then then effectively you become an an AI human symbiote. and and if that then is widespread with anyone who wants it can have it, then we solve the control problem as well. we don't have to worry about some sort of evil dictator AI, because kind of we are the AI collectively.

14:49That seems like the best outcome I can think of. So, you've seen other companies in the early days that start small and get really successful. hope I'm not over asking this on camera, but how do you think Open AI is going as a 6-month-old company? I think it's going pretty well. I think we've got a really talented group at Open AI, and likely. Yeah, really really talented team, and they're working hard.

15:11Open AI is structured as say a a 501c3 nonprofit. but, you know, many nonprofits do not have a sense of urgency. It's fine. they don't have to have a sense of urgency. but Open AI does. cuz I think people really believe in the mission. I think it's important and it's it's about minimizing the risk of existential harm in the future.

15:40And so I I I think it's going well. I'm pretty impressed with what people are doing and the talent level. And obviously we're always looking for great people to join who believe in the mission. 40 people now. It's Yeah. It's going well. all right, just a few more questions before we we wrap up. How do you spend your days now?

15:59Like what do what do you allocate most of your time to? My time is mostly split well, split between SpaceX and and and Tesla and of course I I try to spend it's a part of every week at Open AI. so I spend most I spend basically half a day at Open AI most weeks. and then and then I have some Open AI stuff that happens during the week.

16:25But other than that, it's really SpaceX and Tesla. or Tesla? Like what does your time look like there? yeah, so that's a good question. I think a lot of people think I I must spend a lot of time with media or or on businessy things, but actually almost almost all my time, like 80% of it is spent on engineering and design.

16:44Engineering and design, so it's developing next generation product. that's 80% of it. You probably don't remember this. A very long time ago, many many years, you took me on a tour of SpaceX. And the most impressive thing was that you knew every detail of the rocket and every piece of engineering that went into it.

17:03and I don't think many people get that about you. Yeah, I think a lot of people think I'm kind of a business person or something. Which is fine, like business is fine, but like I but really it's you know, it's like at SpaceX, Gwynne Shotwell is chief operating officer. She kind of manages legal, finance, sales,

17:25and kind of general business activity. And then my time is almost entirely with the engineering team working on improving a the Falcon 9 and the Dragon spacecraft and developing the Mars colonial architecture. and then at Tesla, it's working on the Model 3 and you know, some in the design studio, typically half a day a week.

17:50dealing with this is aesthetics and and look and feel things. And and then most of the rest of the week is just going through engineering of of of the car itself as well as engineering of the the factory. cuz the the biggest epiphany I've had this this year is that what really matters is the is the machine that builds the machine, the factory. and this is that is at least two orders of magnitude harder than the vehicle itself. It's amazing to watch the robots go here and these cars just happen.

18:23Yeah. No, this is actually is has a relatively low level of automation compared to what the Gigafactory will have and what Model 3 will have. What's the speed on the line of these cars? Actually, average speed of the line is incredibly slow. It's probably about it including both X and S.

18:45it it's maybe five you know, 5 cm per second. And what can you get This is very slow. Or what would you like to get to? I'm confident we can get to to at least 1 m per second, so 20-fold increase. That'll be very fast. Yeah. At least. I mean, I think quite a 1 m per second, just put that in perspective, is is a slow walk or or like a medium-speed walk. A fast walk could be 1.5 m per second.

19:15And And then the the fastest humans can run over 10 m per second. So, if we're only doing 0.05 m per second, that's very slow current current speed. And and at 1 m per second, you can still walk faster than the production line.

资源与出处

相关视频

同频道、同主题或相邻年份