大会演讲·中文精读

我们为 FSD 模拟了整个世界:马斯克 2021 访谈

We Simulated the WORLD for Full Self Driving. ELON MUSK Interview 2021

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

一辆私家车,一周 168 小时里,平均只有 10 个小时在跑。剩下的 158 小时,它停在车位上,一边贬值,一边占地方。

马斯克在这段访谈里算的就是这笔账。他给出的判断是,一旦车能自己开,使用时长会从 10 小时涨到 50 到 60 小时,接近一辆出租车的强度。同一台车,同样的制造成本,效用直接翻五倍。

这个数字比任何关于自动驾驶的宏大叙事都更有说服力。买车的人付的是全价,用的是一小部分。自动驾驶要做的事,说到底就是把那 158 小时的闲置时间盘活。

围绕这个目标,马斯克在这次访谈里聊了很多东西:为什么特斯拉坚持不用激光雷达、为什么他在奥斯汀已经能放手让车开去机场、算力每六个月翻十倍意味着什么、芯片荒之后下一个卡脖子的是降压变压器、车载电压为什么要从 12 伏改成 48 伏、以及他被 AI 搞得有点抑郁之后是怎么想通的。

奥斯汀的那台车,已经能自己开到机场

访谈里有人问他,未来五到十年 AI 会带来什么。马斯克的回答从车上开始,没绕到大模型去。

AI 会以非常深刻的方式改变世界,其中最显著的一种,就是自动驾驶。我们在这上面投入了非常多。你在欧洲不太能感受到,因为我们的策略是先把美国跑通,再去增加世界其他地方的复杂度。(马斯克)

他给了一个具体的例子。他当时人在奥斯汀,如果要从住处开车去机场,车可以全程零干预把他送过去。

这台车用了什么,值得单独说。没有激光雷达,没有毫米波雷达,没有超声波,车上只有摄像头,加一套数字神经网络。

这套方案的逻辑非常朴素:人会开车,人靠两只眼睛和一个生物神经网络。那么机器开车,对应的就是摄像头加数字神经网络。生物的和数字的,结构上是同构的。

这个类比听着简单,背后是一整套技术路线的赌注。整个行业里坚持只走视觉方案的主流玩家,基本只剩特斯拉一家。其他人都认为激光雷达是必要的安全冗余,因为摄像头会受逆光、雨雾、夜间影响,测距也不如激光准。

马斯克的反驳一直没变过:人只靠眼睛就能安全驾驶,说明视觉信息本身是充分的。如果视觉不够,那问题出在智能不够,该补的是大脑,不是眼睛。加激光雷达相当于给一个不够聪明的人配一副拐杖,成本上去了,融合多传感器数据的麻烦也来了。

这个观点在过去几年一直被嘲笑,直到端到端神经网络上车之后,争论的声音才小下去。

为了开车,你得先造一个"婴儿版通用人工智能"

视觉方案为什么这么难?马斯克给了解释。

要让这套东西跑起来,车必须真的足够智能。它得学会读懂一切,还得判断司机和行人的意图。本质上,你在造一个婴儿版的通用人工智能,用它来解决开车这件事。(马斯克)

这句话值得拆开看。

开车看起来是个操作问题,方向盘、油门、刹车,三个动作而已。真正难的部分全在感知和预测上。前面那辆车刹车灯亮了,它是要减速还是要变道?路边那个人站在斑马线边上,身体朝着马路,他是在等车流过去,还是在犹豫要不要迈步?一个骑电动车的人回头看了一眼,接下来他要干什么?

这些问题没有规则可写。你没法在代码里写"如果行人身体朝向马路且距离小于 5 米则减速",因为现实里的组合是无穷的。唯一能处理这种无穷组合的东西,是一个从海量数据里自己总结规律的大规模神经网络。

所以做自动驾驶,绕不开做通用感知。你得让机器理解这个世界是怎么运转的,人是怎么想的。这个能力的名字叫常识,而常识恰恰是传统软件工程最难编码的东西。

这也是为什么马斯克说这件事"极其深刻"。一台能读懂行人意图的车,和一台只会识别红绿灯的车,中间隔着的不是一个功能,是一个智能层级。

副产品是,这套能力可以迁移。同一套视觉加神经网络的架构,从四个轮子上挪到两条腿上,就是人形机器人。特斯拉的 Optimus 项目走的就是这条路。马斯克在访谈里提到,他们做的机器人目标是"能干人类能干的任何事"。他也顺带区分了一下:工厂里的工业机器人是显式编程的,不会走也不会看,那是另一个物种。

他还开了个玩笑,说希望机器人对我们好一点。这话听着轻松,但它是这场对话里唯一一次流露出不安的地方。

算力每六个月翻十倍,英伟达市值超过加拿大

聊完车,话题转到 AI 本身。马斯克给了一个他在产业里观察到的数字。

就 AI 算力而言,我从没见过任何技术发展得这么快。上线运行的 AI 计算能力,看起来每六个月就涨十倍。显然这个速度不可能永远持续,否则会超过宇宙的质量。但我确实没见过这样的东西。(马斯克)

六个月十倍,意味着两年是十六倍,三年是六十四倍。这个斜率放在任何工业史上都找不到对照物。

他顺手解释了英伟达的市值为什么这么夸张:因为目前最好的神经网络芯片在它手上。他说英伟达的市值已经超过了加拿大的 GDP,还补了一句,芯片热潮比历史上任何一次淘金热都大。

卖铲子的人比挖金子的人赚得多,这个规律在一百多年前的加州成立,在今天的人工智能行业里再次成立。区别在于,这次铲子的技术门槛高得多,能造的人少得多。

指数曲线不可能无限延伸,这一点马斯克很清楚。他关心的问题是:它会在哪里撞墙。

芯片荒之后是变压器荒,变压器荒之后是电荒

马斯克的预测链条很有意思,因为每一环都是被上一环挤出来的。

一年多以前我预测过这件事。当时短缺的是芯片,神经网络芯片。然后很容易推断出,下一个短缺的会是降压变压器,因为你得把电送到这些机器里。从电网出来的电压是 100 到 300 千伏,芯片要的是 0.6 伏,中间要降很多级。再下一个短缺就是电。我认为明年你就会看到,找不到足够的电来跑所有的芯片。(马斯克)

这段话里藏着一个容易被忽略的工程常识。

一座大型数据中心从接到电网那一刻起,电要走完一整条降压路径:高压输电网进来,经过变电站降到几十千伏,再经过配电变压器降到几百伏,进机房之后经过 UPS 和不间断电源,到机架上是 48 伏或者 12 伏,最后到主板上,经过多相供电模块,降到 1 伏以下喂给 GPU。

每降一级,就要一批设备。芯片产量可以靠扩产线在一年内拉起来,变压器不行。大型电力变压器的交付周期在过去两年从几个月拉长到三四年,因为它需要硅钢片、铜绕组、绝缘油和大量手工装配,扩产慢得像造船。

于是出现了一个很荒诞的局面:全世界都在抢 GPU,结果卡住进度的是一台几十吨重的铁疙瘩。

马斯克在这里讲了个谐音梗,他自己说这个笑话不太好笑。

他们需要变压器来跑 transformer。AI 里有个东西叫 transformer,是一种神经网络结构,而现在他们缺变压器来驱动这些 transformer。(马斯克)

英文里 transformer 既是电力变压器,也是 2017 年那篇论文提出的注意力网络架构。今天所有主流大模型都建立在这个架构上。一个词,两个意思,正好撞在了一起。

从 6 伏到 48 伏,一根铜线的七十年

访谈里有人问到特斯拉 Semi 卡车用的 48 伏架构,问这会不会再一次改变汽车行业。

马斯克的回答从历史讲起。汽车的低压系统最早是 6 伏,后来启动电机需要更大功率,就翻倍到 12 伏。他说其实真实值更接近 13.7 伏,12 伏只是个粗略叫法,而且还不是恒定的,会随电池状态浮动,台下坐着的电气工程师都懂。

这个 12 伏一用就是几十年。为什么现在要改?

算一下就明白。功率等于电压乘以电流。同样的功率,电压提高四倍,电流就降到四分之一。导线里流过的电流小了,需要的铜线截面积就可以小很多。马斯克说,48 伏架构下用的铜大概只有 12 伏架构的四分之一。

车上有几公里的线束,减掉四分之三的铜,省下来的是重量、成本和装配工时。对一台要跑量的车来说,这是实打实的钱。

那为什么不干脆用更高的电压?马斯克说,再往上就会碰到监管问题。业界选 48 伏,是因为它刚好卡在安全低压的门槛之下,类似以太网供电(PoE)用的那个电压区间。高了就要满足更严格的绝缘和防护要求,成本和复杂度又会反弹回来。

他的判断是,长期看所有车的低压架构都会走到 48 伏。这是低压电气架构的下一步,没什么可犹豫的。

电动车和 AI 抢的是同一批变压器

这场对话里有一个很好的追问:很多人喜欢电动车,但被充电基础设施劝退了。

马斯克的回应是,特斯拉建全球超充网络(Supercharger)的初衷就是这个。用户需要确信自己可以跑长途,可以在任何地方充上电。他还提了一句,电力其实无处不在,除非你跑到很偏远的乡下去。

然后他把两条线并到了一起。

电动车和 AI 同时在增长,两者都需要电,两者其实都需要电压变压器。我认为这正在给电力设备和发电能力制造巨大的需求。(马斯克)

这个观察比它听起来更重要。

过去十年,电力需求在发达国家基本是平的,甚至略有下降,因为制造业外迁、家电效率提升。发电厂和电网的投资长期偏低,整个行业习惯了需求不增长的世界,做规划的人手边只有很小的增量假设。

现在同时来了两个吃电大户。一座大型 AI 数据中心的功耗已经可以顶上一座中小城市。而全面电动化意味着交通领域的能源消耗要从加油站搬到电网上。两件事在同一时间窗口发生,电网和设备制造商完全没有准备。

所以那些关于"AI 会不会取代工作"的讨论,短期内可能被一个更朴素的问题盖过去:电从哪来。

"愿你生活在有趣的时代"

访谈中间有一段马斯克少见的坦白。

有句话据说是中国诅咒:愿你生活在有趣的时代。我们正好活在最有趣的时代。有一阵子我确实有点沮丧,我会想,它们会不会接管?我们会不会变得没用?后来我想通的方式是问自己:我宁愿活着看到 AI 末日,还是不愿意?我想我还是想看到。(马斯克)

这段话的信息量在于它承认了恐惧。一个正在造这个东西的人,也曾经在半夜想过自己会不会亲手把人类送上失业的末路。

他用来自我开解的方式其实挺人类:好奇心压过了恐惧。哪怕结局不好,也想亲眼看看会怎么收场。

关于生产力,他的判断是正面的。AI 会大幅提升生产率,从体力劳动到供应链物流,每个领域都会被推着往前走。他提到客服场景已经跑在前面,复杂的聊天机器人可以回答相当绕的问题,而且已经在替代一部分人工。

他还说了一句,我们确实站在有史以来最大的一次技术革命的边上。这句话在访谈里没被展开,但它和前面那条算力曲线是同一件事的两种说法。

下一代车:更便宜,而且默认自己会开

有人问特斯拉接下来的产品规划,特别是欧洲市场缺 2 万到 2.5 万欧元的入门电动车这件事。

马斯克的回答很直接:下一代车型是低成本车型,而且高度围绕自动驾驶来设计。

后半句比前半句重要。一辆为自动驾驶设计的车,内饰布局、座位朝向、储物逻辑、甚至车身比例都会变。当方向盘不再是必需品,车就不再只是驾驶舱,它会变成一个移动的房间。造一台便宜的车不难,难的是在便宜的同时把感知硬件、算力和冗余做够,因为自动驾驶的成本是硬性的。

他对这套系统的描述也挺有意思。

AI 越好,它开起来就越像人。它会做一些你不会期待计算机会做的事。它不像机器人在跳舞,它开得平顺、直觉。(马斯克)

"机器人在跳舞"这个比喻很准。早期的辅助驾驶就是这样,动作是对的,但节奏是僵的,像照着节拍器做动作。人开车不这样,人会预判、会让、会在模糊地带做微调。这种平滑感,恰恰是端到端神经网络最擅长带来的东西,因为它学的就是人怎么开。

他还顺带提了一台不讲道理的车:新一代 Roadster,零到百公里加速可以做到一秒以内。

量产车百公里加速进一秒,靠的不只是电机功率,还有抓地力、空气动力学和下压力。这个数字在工程上几乎是贴着物理极限在做。他把它称作蛋糕上的樱桃,跟前面那些关于电网和变压器的严肃话题放在一起,显得格外轻松。

把这几件事串起来看

这场对话表面上很散,从自动驾驶聊到人形机器人,从算力聊到变压器,从铜线粗细聊到中国谚语。但底下有一条线。

马斯克在做的事情,本质上是把现实世界数字化。车用摄像头把道路变成神经网络能吃的张量,机器人的视觉系统做的是同一件事,AI 数据中心的算力是支撑这一切的底座,而底座最终要落回物理世界:铜、硅、铁、电。

他反复强调的那些短缺,芯片、变压器、电力,全都不是软件能解决的问题。它们提醒我们,这一轮技术浪潮的尽头是物理约束。写代码的人很多,能造出大型电力变压器的人很少。

自动驾驶本身,他的态度是罕见的笃定。这次访谈里他没给时间表,也没用"明年就能实现"这类说法,只说"我们相当接近了"。对一家在这个问题上放了好几次鸽子的公司来说,这种措辞本身就是一种收敛。

剩下的事情,就是让物理世界跟上。

  • 如果你做出行、物流或车队生意,现在就该按"车每周跑 50 小时"重算模型:资产周转率、停车需求、保险定价、维修周期全都会跟着变。用 10 小时的旧假设做的财务模型,两年内会失效。
  • 如果你在做 AI 基础设施,把变压器和配电当成一级风险来管理:GPU 的采购周期已经在缩短,电力设备的交付周期还在拉长。提前锁变压器、锁并网容量、甚至锁发电资源,比多买几千张卡更能决定项目能不能按期上线。
  • 如果你是硬件工程师,认真评估 48 伏架构的机会:线束、连接器、保险丝、DC-DC 转换器、功率器件,整条供应链都会被重做一遍。这个迁移一旦开始就不会回头,早进场的人在成本曲线上占位置。
  • 如果你只是普通车主,别急着为"完全自动驾驶"付溢价:功能要靠软件迭代兑现,硬件只是门票。真正该看的是这套系统在你常走的路上表现如何,发布会上的演示视频说明不了什么。
  • 如果你在关注 AI 投资,把视线从芯片往上游挪一格:算力的瓶颈正在从硅转移到铜和铁。电力设备、输配电、发电,这些看起来很传统的行业会在这轮周期里被重新定价。
  • 如果你对 AI 的恐惧是真的,试试马斯克那一招:把问题换成"我愿不愿意活着看到结局"。答案通常是愿意。恐惧不会因为回避而消失,好奇心能让你在恐惧里保持清醒,继续做事。
语音识别自动转录(Whisper),含全部发言人,可能有识别错误。已按语句合并成段落,并去掉 um、uh 等口头填充词。

0:00Welcome, big hander, Elon Musk. Thank you. Action. Oh, how are you? Good. Fantastic. We have found it. We have two days of AI and Gen AI and what-so-ver, but you look real. My head is very big on the screen. I had a reality too, please. We have had so much talk about AI in these two days, so we need your opinion.

0:26What do you think about the next five to ten years on AI? Please give us a view artificial insemination But try try audience and I say Yes artificial intelligence is It's certainly a profoundly changed world. One most significant ways is self-driving So we've been putting a lot of effort into self-driving Technology or cars you don't see it quite as much in Europe because we first have to make it work in the US before we

0:57we increased the complexity of trying to make it elsewhere in the world. But I think we're quite close to having the car be fully autonomous. So for example, like right now, I'm in Austin, and if I went to drive, say to the airport, the car could take me to the airport with no interventions.

1:13Using and all it's using are digital neural nets, in other words, artificial intelligence and cameras. So there's no LIDAR radar, nothing. And in fact, if you think about how human-stripe cars, humans are biological neural nets and we use eyes. So it's the eyes and biological neural nets.

1:30The analog is digital. The analog for digital is cameras and digital neural nets. It doesn't work remarkably well. But it has been quite difficult to do this because, in order to use that, in order for this to work, the car has to really be quite fully intelligent. As a subset, for example, it has to learn how to read everything and how to assess intention among drivers and pedestrians.

1:54So you're creating a baby artificial general intelligence to solve this. But this will obviously be extremely profound. And obviously I've talked about what about if you're all good into the broader question in a minute. But the average use of a passenger vehicle is only about 10 hours per week out of 100 hours.

2:14The point at which it is autonomous, I think you would see probably a third of the hours being used because it would be like a taxi. Maybe my guess is like 50 to 60 hours, which means that the utility of a passenger vehicle would increase by a factor of five. And this is gigantic.

2:31But we'll see. So that's the word motive. Then there's what people think I've certainly experienced online where you can ask it a question and ask for an essay or a picture or to understand a picture. And that is progressing rapidly. What I'm seeing in terms of AI compute is I've never seen any technology advanced faster than this.

2:49The artificial intelligence computer coming online appears to be increasing by a factor of 10 every six months. Like obviously that can't continue at such a high rate for forever or to exceed the best of the yearboats. But I've never seen anything like it. And this is why you see NVIDIA's market cap being so gigantic because they currently have the best neural net chips.

3:11I mean, I think the NVIDIA's market cap exceeds the GDP of Canada or something recently, and it was quite high. So, yeah, you know, it may go high, you know. The chip rush is bigger than any gold rush that has ever existed. So, and then there will also be robots, like humanoid robots, not just sort of industrial robots, we are of course familiar with you, use industrial robots, we use industrial robots every one does, but they are programmed explicitly and they don't walk around.

3:39But Tesla, for example, the optimist program is making a humanoid robot that is capable of doing, when we say anything, a human can do. I mean, I hope, I just hope the robots are nice to us, you know? We hope so too, you know, that the robots will be nice to us. And maybe, let's say, in this regard, because you are working on what can we

4:03expect of, let's say, support which is going to come, maybe also a little bit with regard to production, because in the Bausch connected world we were also discussing application of AI, gen AI into production in order to improve productivity, maybe a couple of thoughts on this one too.

4:26Well, the plus side of AI is that I think productivity will increase dramatically. So across every field, whether it be manual labor to supply chain logistics, there's already a lot of movement to use chatbots, like sophisticated chatbots for customer service, for example, a way they can answer, quite complex questions already.

4:46But I mean, I think we really are on the edge of one of the biggest technology revolution that has ever existed. You know, there's supposedly a sort of a Chinese curse, may you live in interesting times. Well, we live in the most interesting of times, the most interesting. And for a while, I was a little bit, I was making a bit depressed, frankly, because I was like, well, will they take over?

5:07Will we be useless? But then the way that I reconciled myself to this question was would I rather be alive to see AI apocalypse or not? I guess I'd like to see this. That's a good thought. That's a good thought because you mentioned it every six months, that's an exponential curve.

5:27Right? Exponential curves are beautiful. There's kind of a question, where does it ever end? Right? Do you see any end yet or will it go on for three, four, five years like this? Because then we talk hundred phones, right? A thousand phones of time. Yeah, that's why I said obviously.

5:41Yeah, I mean, you can't grow at that rate without quickly exceeding the mass of the universe. So clearly, it's going to hit some constraints. The constraints that AI compute, I'm very predictable. I've actually predicted this over a year ago. So over a year ago, the shortage was chips, neural net chips.

5:58Then I was very easy to predict that the next shortage will be voltage step-down transformers. So you've got to feed the power to these things. So if you've got 100 to 300 kilobolts coming out of your utility, and it's got to step all the way down to 0.6 volts. That's a lot of stepping down.

6:15So now we're going to step down. And this is my not that funny joke, which is that they need transformers to run transformers. Because the AI is like, there's this thing called a transformer in AI. It's a neural, I don't know, it's a combination of sort of neural nets. And they're running out of transformers to run transformers.

6:31Then the next shortage will be electricity. So I think next year you'll see the electricity that just can't find enough electricity to run all the chips. I mean, during the chip crisis, we certainly all did have many contacts. Fortunately, we were able to master it together. Of course, we all would like to avoid a second chip crisis.

6:50But with regard to your comment on power demand, we all know today, vehicles have three to four kilowatt power demand for the peripherals. But this is continuously going up. And Tesla has recently launched the Simon truck, which I believe is mostly coming with 48 volt. Do you see this as a let's say, further disruption of automotive industry, another technology change within automotive.

7:19Yeah, I think, you know, the automotive industry has been living with 12 volts for a very long time, essentially roughly. I think it started off as six volts, and then they needed a more powerful starter voter, and then they doubled it to 12 volts. Although it's actually more like 13.7 volts.

7:36You're right. Absolutely. Yes. And the bus voltage actually varies a lot for those since the price of electrical engineers in the audience. Like 12 volts is a very rough approximation. But it's really kind of an arbitrary low voltage. And if you look at say power ethernet, it's running at 50 volts or roughly 48 volts.

7:56And we said it at 48 volts because anything really above that, you start to run into regulatory concerns. And power ethernet is like I said, roughly at that level. But the advantage of obviously quarter-blaying the voltage is that you could substantially reduce how much copper or conductor is used in the car.

8:12So you can use very small copper wires for roughly a quarter, not exactly a quarter, but roughly a quarter as much copper is needed for 48 bolts as 412. So I think it's a logical move. That's really the next step for the low voltage architecture for the car. I think ultimately long-term wall costs will be 48.

8:32It's interesting that many people love electric vehicles and want to have them, but some are harmed by a missing infrastructure. That would be the same, actually, as the iPhone. We will have lots of chips ready, but we will have the power to operate them. There's a race between infrastructure and technology right now.

8:49And you're also certain technology may win, which is a problem because, actually, infrastructure is a win. We should have all the infrastructure we need. Is that correct? Yeah, are you talking about charging infrastructure because and that kind of thing? Well, the reason Tesla developed the sort of global supercharger network was

9:06specifically to address that point in the point you're making, which is people need to be comfortable that they can travel long distances and charge their car wherever they go. Unfortunately, electricity is almost everywhere. And so it's not all not all that rare unless you're really far on the countryside or something.

9:22So yeah, you need the charging infrastructure and the electricity to charge electric cars. In fact, the simultaneous growth of electric cars and AI, both of which need electricity, both of which need voltage transformers, actually, I think, is creating a tremendous demand for electrical equipment and for electrical power generation.

9:42Maybe one more question, Elon, with regard to new vehicle designs coming to the market. Of course, you cannot share with us confidential secrets. But anyways, what are the things you believe, let's say, we will see over the next years in the market and maybe just a small comment.

10:01In Europe we see a shortage of affordable battery electric vehicles in the offerings because many people in Europe are only able to spend like 20,000, 25,000 euros for their vehicles. So what are the new vehicle types, vehicle designs you are thinking about at Tesla? Well, our next generation vehicle will be a low cost vehicle.

10:27Also, that is very focused on autonomy. Like I said, we really are getting to the point where the car can drive itself well. When you're in the US, I would recommend giving it, trying out our car and see how it works. It's remarkable, you know. As the AI gets better and better, it actually feels quite human-like in the way that it drives.

10:47It does things that you wouldn't expect a computer to do. It's not like a robot dancing, you know. It's like it feels smooth and intuitive. So our next generation vehicle, which we're quite far along on, is very much anticipating autonomy. And then of course, we've got some fun things as well like on the new Tesla roads, which we'll be able to do zero to 100 kilometers in under one second.

11:11Great, thank you very much. That's what we need. That's good news. That's a good space of fun stuff too, you know. That's the cherry on the cake. But it's fun. You're going to have fun stuff, too. Yeah. Right. Well, that's very good news, because in the end, there's a lot of people here in the room that's still on their list to buy an electric vehicle, and you just announced something which they probably want to buy.

11:32So thank you very much. Thank you very much. And thank you extremely for 15 years of partnership and work together. We're inspired always by your speed, attitude, and graphite and things come out. Thank you very much, Jina Must. All right. And thank you for your support. We're very much appreciated.

从现场发言里挑出的 9 句原话,附中文译文(AI 翻译,以英文原话为准)。点时间戳可跳到视频对应位置。

但我认为,我们已经非常接近让汽车实现完全自动驾驶了。

But I think we're quite close to having the car be fully autonomous.

▶ 0:39

我猜大概是50到60小时,这意味着乘用车的效用会提高五倍。

Maybe my guess is like 50 to 60 hours, which means that the utility of a passenger vehicle would increase by a factor of five.

▶ 2:00

芯片热潮比史上任何一场淘金热都更大。

The chip rush is bigger than any gold rush that has ever existed.

▶ 2:56

我的意思是,我希望,我只是希望机器人能善待我们,你懂的。

I mean, I hope, I just hope the robots are nice to us, you know?

▶ 3:24

但我认为,我们确实正处于有史以来最大的技术革命之一的前沿。

But I mean, I think we really are on the edge of one of the biggest technology revolution that has ever existed.

▶ 4:26

好吧,我们生活在最有趣的时代,最有趣的时代。

Well, we live in the most interesting of times, the most interesting.

▶ 4:52

下一个短缺将是电力。

Then the next shortage will be electricity.

▶ 6:13

事实上,电动汽车和AI同时增长,二者都需要电力,也都需要变压器,我认为这正在对电气设备和发电产生巨大的需求。

In fact, the simultaneous growth of electric cars and AI, both of which need electricity, both of which need voltage transformers, actually, I think, is creating a tremendous demand for electrical equipment and for electrical power generation.

▶ 9:17

嗯,我们的下一代汽车将是一款低成本汽车。它也非常专注于自动驾驶。

Well, our next generation vehicle will be a low cost vehicle. Also, that is very focused on autonomy.

▶ 10:14

资源与出处

相关视频

同频道、同主题或相邻年份