与马斯克进行78轮对话。
78 rounds of dialogue with Musk.
作者丨车东西（ID：chedongxi）
The author of the car (ID:chedongxi)
编辑丨晓寒
Xiaoqiao, editor
图源丨《TIME》官网
Source: TIME Network of Officials
想要深入了解马斯克的所思所想？想要了解特斯拉和SpaceX面临的技术挑战？一场两个半小时视频对话，足以满足你的大部分好奇！
You want to know more about what Muskell thinks? You want to know about Tesla and SpaceX's technical challenges? A two-and-a-half-hour video conversation is enough to satisfy most of your curiosity!
就在9天之前，Youtube知名科技UP主，拥有143万粉丝的大V Lex Fridman与马斯克进行了一场长达2小时31分47秒的深度对话，从SpaceX的载人飞船、猛禽火箭、殖民火星，一口气聊到特斯拉面临的自动驾驶工程难题，机器人，以及马斯克知名的第一性思考原则。
Just nine days ago, Youtube's famous technology UP owner, Big V Lex Fridman, with 1.43 million fans, held a 2-hour, 31 minutes, 47 seconds-long, in-depth dialogue with Mask, from SpaceX's manned spacecraft, Raptor rockets, colonization Mars, talking about Tesla's problems with autopilot engineering, robots, and the famous first thought principle of Mask.
作为一个对话类节目，发布在油管上的这个视频除了主播与马斯克两个人，并没有其他任何画面。但这不重要，重要的是Lex Fridman的78问都问到了点子上，既有技术原理、观点和看法，也有马斯克的心路历程与日常生活细节。
As a conversation-type programme, this video, which was posted on the pipeline, has no images other than the two hosts and the Mask. It is not important, however, that Lex Fridman’s 78 questions have all come up with ideas, ranging from technical principles, views and opinions to details of Mask’s path and daily life.
对话中的马斯克也非常活跃、真诚。见面伊始，他直接对着麦克风又唱又抖，搞得跟明星录播现场一样。
The dialogue was also very active and sincere. At the beginning of the meeting, he sang and shivered directly at the microphone, just like a star on the scene.
虽然歌唱的确实难听，但好在马斯克在面对78个问题时知无不言，金句、调侃与粗口齐出。
Although it is true that singing is ugly, it is good that in Mask's face of 78 questions, the words of gold, the tone and the boldness of words come out.
他讲到，虽然自己不是宗教人士，但在SpaceX的载人飞船发射前，他也虔诚的跪地祈祷（保佑发射平安）。聊到特斯拉的自动驾驶，他调侃称特斯拉开车比007詹姆斯邦德都厉害。而在谈到如何面对挫折与挑战时，马斯克则直言自己的字典没有放弃二字，也不管什么悲观与乐观主义。
He said that, although he was not a religious person, before the launch of the SpaceX manned spaceship, he devoutly bowed to his prayers. Speaking of Tesla's autopilot, he called Tesla a better driver than James Bond. And speaking of how to face setbacks and challenges, Mask said that his dictionary did not give up the word, no matter how pessimistic or optimistic.
“艹！干就完了。”马斯克说道。
"Shit! It's over." Musk said.
对话里也包括大量马斯克的真知灼见、反常操作、以及技术细节：
The
1、何时登陆火星？最早5年，最晚10年后。
When did you land on Mars? Five years at the earliest, 10 years later.
2、往返一趟火星多少钱？1万亿美元。
2 How much is a trip to Mars? $1 trillion.
3、何时实现L4级的FSD？2022年就有希望。
3 - When will the L4 level of FSD be achieved? 2022.
4、特斯拉自动驾驶软件中神经网络数量不断增加，基础代码正在快速减少，让自动驾驶越来越近。
4. The number of neural networks in Tesla autopilot software is increasing and the base code is rapidly decreasing, bringing autopilots closer and closer.
5、SpaceX造火箭时找不到合适的材料就自己造了几种合金……
5. SpaceX built several alloys by itself when no suitable material was found.
总而言之，这两个半小时的对话信息量很足，也非常精彩，是了解马斯克的一个重要窗口。以下是我们对这场视频对话实录的全程编译，为符合中文阅读略有删改。
All in all, these two half-hour dialogues were informative and excellent, and were an important window for understanding Mask. The following is a full-scale translation of the video proceedings, with minor deletions to read in Chinese.
Lex Fridman与马斯克
Lex Fridman and Musk.
首次太空载人曾下跪祈祷
The first spaceman to kneel and pray
下一步要在火星建立文明
The next step is to build civilization on Mars.
1、Lex Fridman：2020年5月，SpaceX的载人火箭进入太空轨道，被许多人视为人类太空探索新时代的第一步。这些人类太空飞行对我和数百万人来说是希望的灯塔，在过去的两年里，我们的世界一直在经历人类历史上最困难的时期之一。
我们看到分裂、恐惧、愤世嫉俗和共同人性的丧失，就在最需要的时候，所以感谢你给了世界希望和对未来兴奋的理由。
We saw division, fear, cynicism and loss of common humanity at the moment of greatest need, so thank you for giving the world hope and reason to be excited about the future.
马斯克：哦，谢谢你这么说。我确实想这么做。人类显然有很多问题，而且你知道，人们有时会做坏事，但我爱这个群体。我认为我们应该确保我们所做的一切，能够让我们可以有一个美好的未来，一个令人兴奋的未来，去最大化人类幸福。
Masc: Oh, thank you for saying that. I do want to do it. Humans obviously have a lot of problems, and, you know, people sometimes do bad things, but I love this group. I think we should make sure that what we do is that we can have a good future, an exciting future, to maximize human happiness.
2、Lex Fridman：我想问一下“载人龙飞船”的飞行任务（Crew dragon dem-2），这是Space X的第一次载人飞行，你在发射前感觉如何？你当时害怕吗？你兴奋吗？你在想什么？你的脑海中是否有这么多的利害关系？
2, Lex Fridman: I'd like to ask you about the Crew Dragon Dem-2, this is Space X's first manned flight. How do you feel before launch? Are you scared? Are you excited? What are you thinking? Is there so much interest in your mind?
马斯克：是的，当时我内心确实非常紧张。我们显然不能在任何方面让他们失望，所以压力极大。我相信在飞船发射的时候，没有人能够想到它成功的概率是多少。
Mask: Yes, I was really nervous at the time. Obviously, we couldn't let them down in any way, so there was a lot of pressure. I believe that when the ship was launched, no one could think of its chances of success.
我们绞尽脑汁去想能够提升成功概率的方法，但我们已经竭尽所能了，美国宇航局也是。我们已经在能力范围内做到最好了，所以我们继续往前推进并发射那艘飞船。我并不是一名宗教人士，但是我还是跪下为那次任务做了祈祷。
We're trying to figure out a way to raise the chances of success, but we're doing everything we can, and NASA is doing the best we can. We're doing the best we can, so we're moving forward and launching the ship. I'm not a religious person, but I'm on my knees praying for that mission.
3、Lex Fridman：当飞船发射成功之后，你们睡得着吗？或者说，当飞船成功返程落地时你们内心的感受是什么？
3, Lex Fridman: Can you sleep when the launch succeeds? Or, what do you feel inside when the ship succeeds?
马斯克：高压之后内心的石头终于落了地。与其说是高兴，不如说是一种解脱。在SpaceX随后的任务中，肯定会有很多乐趣，我认为Inspiration Mission实际上是非常鼓舞人心的。我推荐大家去看一部Netflix纪录片《发射倒计时：灵感4号平民太空任务》（Countdown: Inspiration 4 Mission to Space），这部纪录片内容非常好。我本人也受到了启发。在飞船发射过程中，我甚至有些感觉有些享受，而不像第一次SpaceX发射飞船时那么紧张。
Masc: The stones in my heart have finally fallen to the ground. Rather than happy, they are a relief. There will certainly be a lot of fun in SpaceX's subsequent mission, and I think it's encouraging to think that the Inspiration Mission is actually very encouraging. I recommend you to see a Netflix documentary, Launch Countdown: Insulting 4 Mission to Space, which is very good. I was inspired by it myself. During the launch of the spaceship, I even felt a little bit as nervous as I was when SpaceX first launched.
4、Lex Fridman：所以大部分人其实不知道Inspiratin 4 Mission是全球首次的商业飞行任务，也是“普通人”的首次太空旅行。
4, Lex Fridman: So most people don't really know that Inspira 4 Mission was the first commercial mission in the world and the first space trip of “the ordinary people”.
马斯克：是的，过去30～40年内，人类离地球最远的飞行应该是航天飞机维修哈勃望远镜任务和1972年阿波罗号的登月任务。这其实有点令人费解，当然这两次任务确实很酷。作为人类，我们希望可以继续做得更好，达到更高的境界。
Masc: Yes, in the last 30 to 40 years, the most distant flight of humans from Earth should be the Space Shuttle mission to repair the Hubble Telescope and the moon-on-the-moon mission of the Apollo in 1972. This is a bit confusing, of course, and these two missions are really cool. As humans, we want to be able to continue doing better and reach higher ground.
我个人认为，如果阿波罗号标志着人类的最高水平，那将是非常悲惨的。人类登上月球是我们曾经达到的水平，但现在已经过去了49年，我们还没有进行过第二次登月。这是否意味着我们的技术已经达到顶峰？所以我认为我们必须再次登陆月球，在那里建立一个科学基地，就像南极洲和世界其他地方的科学基地一样。在那里，我们能了解到更多关于宇宙本质的东西。
Personally, I think that it would be very tragic if the Apollo marked the highest level of humanity. Human beings were on the moon at the level that we had achieved, but it is now 49 years away, and we have not been on the moon for a second time. Does that mean that our technology has reached its peak? So I think we have to land on the moon again, building a scientific base there, just like that in Antarctica and the rest of the world.
我认为这一切都是相关联的，所以这就是我认为的下一件大事——建造一个基地，然后把人类送上火星，建立火星文明。
I think it's all connected, so that's what I think the next big thing is -- building a base and then sending humanity to Mars to build civilization on Mars.
5、Lex Fridman：下面我会采访您其中的一些细节。由于您忙于处理工程挑战，你还有精力惊叹于太空旅行或是每次火箭升空的神奇之处？
5, Lex Fridman: I'll talk to you about some of these details. Because you're busy with engineering challenges, you still have the energy to wonder about space travel or the wonders of every rocket going off?
你有没有对将要面临的挑战有不知所措的感觉？5月30日SpaceX首次载人飞行的任务已经过去一段时间了，你现在可以回顾一下它对你产生的影响。
Have you any sense of surprise about the challenges to be faced? The mission of SpaceX's first manned flight on 30 May has passed some time ago, and you can now look back at its impact on you.
当时影响你最深的可能是工程问题，但现在这已经成为了一个历史性时刻。
It may have been engineering, but it is now a historic moment.
你觉得21世纪值得被人铭记的时刻有哪些？对我个人而言，其中的一个时刻可能就是Inspiration 4 Mission，它或许会作为人类在新太空探索时代的早期尝试被铭记。
What time do you think the twenty-first century is worth remembering? For me personally, one of the moments may be the Inspiration 4 Mission, which may be remembered as an early attempt by mankind in the era of new space exploration.
马斯克：关于这次飞行，我想也许有些人知道，但更多的人都不知道。其实这就像，我实际上是SpaceX的首席工程师，我在所有的工程设计项目上都签了字，因此如果任何设备有任何问题，本质上都是我的错。所以我实际上脑子里想的都是工程设计。
Masc: I think some people might know about this flight, but more people don't. It's like, I'm actually the chief engineer for SpaceX, and I signed all the engineering projects, so if there's anything wrong with any equipment, it's essentially my fault. So I actually think about engineering.
当我看到火箭时，我脑海里会想到那些可能会出错的因素和能够提升的部分。也就是说我看到龙飞船的时候也是如此。所以一般人看到火箭可能会说：“这个航天器或火箭看起来很酷！”，但我就会像一个读数机一样，一直在说有哪些风险，哪些问题。
When I see a rocket, I think about what could go wrong and what could go up. That's what I see when I see a dragonship. So when people see a rocket, they say, "This spacecraft or rocket looks cool!" But I keep saying, like a reader, what are the risks and what are the problems.
6、Lex Fridman：正如你所说，当你看到火箭的时候，你会列出风险清单。同时，你也说过星际飞船是一个非常困难的挑战。那么，以同样的方式让你列出一个星舰风险清单，并且如果能完美地解决其中一件工程问题，那么这个问题会是什么？也许是效率、引擎、不同部件的重量、各种事物交织复杂性甚至是星舰的着陆有关。
6, Lex Fridman: As you say, when you see a rocket, you list a risk list. At the same time, you say that an interplanetary ship is a very difficult challenge. So, in the same way, let you list a starship risk list, and if one of the engineering problems is solved perfectly, what is the problem? Maybe efficiency, engine, weight of different components, complexity of things, or even the landing of a starship is about .
马斯克：都不是，目前最需要花费时间的不是发动机的设计，而是发动机的生产。我经常说原型设计很简单，困难之处在于把它们生产出来，我们已经设计出了最先进的火箭发动机。
Masc: No, it's not engine design, it's engine production that is most time-consuming. I often say that prototype designs are simple, and the difficulty is producing them, and we've designed the most advanced rocket engines.
目前最好的火箭发动机可能是RD-180或RD-170，基本上是俄罗斯的双燃烧室、双喷嘴发动机。而我认为一个发动机成功与否在于它是否能把飞船送入轨道，而我们的发动机还没有把星舰送入轨道，但实际上其性能要比俄罗斯的RD系列发动机更好。
At present, the best rocket motors are either RD-180 or RD-170, basically a two-combustion chamber, two-spray engine in Russia. And I think the success of an engine is whether it can put the ship in orbit, and our engines have not put the starship in orbit, but they actually perform better than Russia’s RD series engines.
7、Lex Fridman：所以你指的是猛禽发动机。猛禽发动机与其他发动机不同之处在哪？
7, Lex Fridman: So you're talking about a Raptor engine. What's the difference between a Raptor engine and other engines?
马斯克：猛禽发动机是一种全流量分级燃烧循环发动机，它在一个非常高的室压下运行。衡量发动机好坏的关键就在于这个室压的数字是多少。
Masc: The Raptor engine is a full flow-scale combustion cycle engine that runs under very high pressure in a room. The key to measuring the engine's performance is the number of this room's pressure.
这个数字就是燃烧室的压力，猛禽发动机可以在300Bar的压力下运行，甚至可能更高。当前运行的RD发动机的室压应该在267Bar左右。而提升室压的难度是以非线性方式增加的，因此，增加10%的室压更像增加了50%的难度，发动机需要获得更高的功率密度。
This number is the pressure of the combustion chamber, where the bird engine can run under 300 bar pressure, perhaps even higher. The RD engine currently in operation should have room pressure around 267 bar. The difficulty of lifting the pressure is increased in a non-linear manner, so that an additional 10% of the pressure is more likely to increase the pressure by 50% and the engine needs to gain a higher power density.
这样就有了很高的推量比和很高的比冲量。因此，特定脉冲就像是火箭发动机效率的衡量标准，它实际上是发动机排出气体的有效排气速度。所以在一个非常高的压力下，你可以有一个紧凑的发动机，尽管如此它还是有一个高膨胀率，也就是出口喷管和喉部之间的比率。
So, a specific pulse is a measure of the efficiency of a rocket engine, and it is actually an efficient exhaust rate of the gas out of the engine. So, under a very high pressure, you can have a compact engine, although it still has a high inflation rate, i.e., the ratio between the export tube and the throat.
8、Lex Fridman：那为什么猛禽是一种难以大规模量产的发动机？
8, Lex Fridman: Why is the Raptor an engine that is difficult to produce on a large scale?
马斯克：它的构造非常复杂。发动机有很多零部件和很多独特的材料。因此，我们不得不发明几种不存在的合金，以使发动机工作，这是材料问题。此外，在一个全流量分级燃烧系统中，有很多反馈回路，推进剂和热气同时流向发动机的不同的地方，而且它们都对彼此有一个递归的影响。所以如果这里改变一些，那里就会受到一些影响，再去改变别处，事情就很难控制了。这也是为什么以前没有人做过这类发动机。我们之所以做一个阶段性燃烧的全流发动机，是因为理论上其效率最高。
Mask: It's a very complex structure. The engine has a lot of parts and a lot of unique materials. So we have to invent a few non-existent alloys to work on the engine, which is a material problem. And, in an all-flow sequenced combustion system, there are many feedback loops, and the propellant and the heat flow to different parts of the engine at the same time, and they all have a regressive effect on each other. So if there's a change here, there's some influence, and then there's a change elsewhere, it's hard to control things. That's why nobody's done these engines before.
为了制造一个完全可重复使用的火箭，必须把一切都做到极致——制造最好的发动机、最好的机身、最好的热屏蔽、最好的航空电子设备以及超智能控制机制。
In order to create a fully reusable rocket, everything must be perfect — the best engine, the best body, the best heat shield, the best aeronautical electronic equipment and super-intelligent control mechanisms.
例如，在减轻发动机自身质量上，我们不在助推器上安装着陆腿，而是用塔架承接返回的发动机。所以这就像是用一双筷子手臂在一座巨大的塔上抓住有史以来最大的飞行物一样。或者说这和空手道小子抓苍蝇一样，只不过这只苍蝇要大得多。
For example, in reducing the quality of the engine, we don't install landing legs on boosters, but we use towers to carry back engines. So it's like using a pair of chopstick arms to grab the largest flying object ever on a giant tower. Or it's like karate boys catch flies, but it's a much bigger fly.
9、Lex Fridman：你提到在一些时刻，你会怀疑这是否能够实现，因为这是非常困难的。
9, Lex Fridman: You mentioned that at some point, you'd question whether this can be achieved because it's very difficult.
马斯克：可能实现的一点是，我认为SpaceX会让星舰运转起来。这只是一个时间问题，我们要花多长时间来做这件事？我们要花多长时间才能真正实现火箭全面和迅速地重复使用？在其能够达到全面并迅速地重复使用前，需要多次发射试验。这些试验我可以用物理学计算。
Masc: One thing that can be achieved is that I think SpaceX will make the Starship work. It's just a matter of time. How long will it take us to do this? How long do we really have to do this?
我对成功充满信心，计算结果在预测成功区间内。在SpaceX，我们有一支天才团队，他们夜以继日地工作，让愿景成为现实——实现航天革命和使人类成为星际文明的关键，要有一个完全和快速可重复使用的轨道火箭。在这之前，从来没有一支轨道火箭是完全可重复使用的。而这一直是火箭技术的梦想，许多人曾试图这么做，但都没有成功。
I have confidence in success, and the calculations are in the predicted success zone. In SpaceX, we have a team of geniuses who work around the clock to make the vision a reality -- the key to a space revolution and to making humanity an interstellar civilization, with an orbital rocket that is completely and rapidly reusable.
10、Lex Fridman：你知道有很多人，包括许多专家、记者和一般的公众，都对可重复使用的火箭表示怀疑。而你自己认为这是一个非空集，虽然成功比较困难，但你还是全身心地投入，作为一名工程师、团队的一份子去负责推进这个项目，将其完成。你的力量来源是什么？
10, Lex Fridman: You know that many people, including many experts, journalists and the general public, have expressed doubts about reusable rockets. And you yourself think that this is a non-empty collection, and it's difficult to succeed, but you're still fully committed, as an engineer, as a team, to carry the project forward and complete it. What is the source of your power?
马斯克：我只是知道自己思考问题的方式。对我来说，这是一件很重要的事情，我们应该继续做下去，否则就会失败。而我也不需要什么力量来源。
Masc: I just know how I think about things. It's a very important thing for me, and we should go on, or we'll fail. And I don't need a source of power.
11、Lex Fridman：所以放弃不是你的性格？
11, Lex Fridman: So giving up is not your character?
马斯克：放弃这两个字不在我的字典里，管他乐观还是悲观，都滚蛋，干就完了！
Masc: Give up these two words not in my dictionary, either he's optimistic or pessimistic, get out of here, and we're done!
12、Lex Fridman：你能不能再放大到星舰的具体问题或你工作的任何工程问题上？你能不能试着描述下你思考问题的方式？比如不同的工程和设计问题？是否有一个系统的过程？你已经说过了思考的第一原则，但是否有一个过程？
12, Lex Fridman: Can you not zoom in on the specifics of the Starship or any engineering aspects of your work? Can you try to describe the way you think about it, for example, different engineering and design issues? Is there a system process? You have already said the first principle of thinking, but is there a process?
马斯克：物理学是基础，其他都是建议。我见过很多违反法律的人，但还没有见过可以违反物理学的人。如果是技术问题，你只需要确保自己没有违反物理学。另外，第一原则是分析，我认为分析可以应用于任何生活中所出现的事情。实际上，我们可以把一些事情归结为最基本的原则。物理学有些基础知识，比如说能量守恒定律，如果你违反了物理学，这件事就不可能完成。另一个方法是在极限中思考问题，如果你在思考一件事情的同时，把它扩展到一个非常大的范畴或一个非常小的范畴，事情会发生什么变化？
Musk: Physics is the foundation, and all others are advice. I've seen a lot of people who violate the law, but I haven't seen anyone who violates physics. If it's a technical problem, you just have to make sure you don't violate physics. And the first principle is to analyse, and I think that analysis can be applied to anything that happens in life. In fact, we can categorize some of the basics of physics as the most basic principles.
13、Lex Fridman：在火箭的制造中也是如此？
马斯克：是的，以制造为例，我认为这是一个被严重低估的问题。就像我说的，把一个先进的技术产品量产，比设计这款产品更难。假设你想弄清楚为什么这个零件或产品很贵？是因为我们在做一些蠢事吗？还是因为我们的产量太低？所以你想，如果每年的产量是一百万台呢？那还贵吗？这就是我的激进之处，把事情考虑到极限。如果一年一百万台还是很贵，那么数量就不是你的东西贵的原因，根本问题出在设计上。
Mask: Yes, in the case of manufacturing, I think it is a serious understatement. As I said, it is more difficult to make a product of advanced technology more expensive than to design it. Suppose you want to figure out why this part or product is expensive. Is it because we're doing something stupid or because our output is too low?
14、Lex Fridman：然后你就可以专注于降低设计的复杂性。
14, Lex Fridman: Then you can focus on reducing the complexity of the design.
马斯克：是的，这样一来必须改变设计，改变零部件，从根本解决价格问题。这在火箭领域很常见，因为产量相对较低，所以一个常见的借口是：“火箭价格很昂贵，是因为我们产量低。如果（我们）是在汽车或消费电子领域，成本就不会这么高了。”
Masc: Yes, so that we have to change design, change parts, and solve the price problem at a fundamental level. This is common in the field of rockets, because production is relatively low, so a common excuse is: "The cost of rockets is expensive because we're low. If we were in the area of automobiles or consumer electronics, the cost would not be so high."
我心里想的是，如果现在每年生产一百万支火箭，那火箭单价还会这么贵吗？如果答案是肯定的，那么规模经济就不是问题了。
What I'm thinking is, if one million rockets are produced every year now, would the unit cost of the rockets be so expensive? If the answer is yes, economies of scale are no longer a problem.
15、Lex Fridman：你是否将其纳入制造业或供应链中？当你考虑资源和原料时，会不会把这些因素放到第一原则分析中？比如供应链如何运作？
15, Lex Fridman: Do you integrate into manufacturing or supply chains? Do you put these factors in the first principle analysis when considering resources and raw materials? Such as how the supply chain works?
马斯克：正是如此，举一个在极限中思考问题的例子，这一方法对制造所有产品来说都很适用。如果你看一下火箭的原材料，你会发现原料有铝、钢、钛合金、特种合金、铜等等。
Masc: That's the case, for example, of thinking about the limits of the problem, a method that works for all products. If you look at the raw materials of the rockets, you find them in aluminium, steel, titanium, special alloys, copper, etc.
每个部件的组成元素的重量是多少，它们的原材料价值是多少？在不改变原材料的情况下，以上几个问题为火箭的成本设定了渐近极限。在原材料不变时，你把比例看作一个神奇的数字，在不断计算后，把原子排列成最终的形状，这将是你产品的最低成本。实际上产品成本的高低在于如何将原子放到所需的形状中。
What is the weight of the constituent elements of each component, and what is the value of their raw materials? Without changing the raw materials, these questions set a near-term limit on the cost of a rocket. When the raw materials do not change, you look at the scale as a magic number, and, in constant calculation, atoms as the final shape, which will be the lowest cost of your product.
16、Lex Fridman：我想先讲一段故事，我经常和Jim Keller聊天，他曾经与你共事过。
16, Lex Fridman: I want to tell a story first. I talk to Jim Keller often. He used to work with you.
马斯克：Jim在特斯拉做得很好。
Masc: Jim did a good job in Tesla.
17、Lex Fridman：所以我想他也受到了你的启发，思考问题的方式应该和你相似。同样，其他在特斯拉和SpaceX工作的人也具备同样的思维。他们都学会了这种思维方式，而且非常明显。
17, Lex Fridman: So I think he's inspired by you, and he should think in a way similar to you. So are others who work in Tesla and SpaceX. They've learned this way of thinking, and it's obvious.
Jim曾跟我说过，Tesla Bot的成本可能不高。当时我们有过争论，我不相信他说的，规模量产的成本不可能低。目前为止，我已经与学术界和仿真机器人领域的专家进行了多次沟通，他们研发仿人机器人和动力机器人的成本高昂。Jim则持有不同观点，他说：“第一原则就是降低制造成本。”我想你就是这样做的，你已经在特斯拉汽车和各种先前被视为复杂的系统上做了同样的思考。你的原话是：“我们现在如何把它简化？”
Jim told me that the cost of Tesla Bot might not be very high. We had a debate, and I didn't believe what he said, that the cost of mass production could not be low. So far, I've been in touch with academics and experts in the field of imitation robots, and they're expensive to develop human and power robots. Jim had a different view, and he said, "The first principle is to lower the cost of manufacturing." I thought you did that, and you thought the same about Tesla cars and systems that were previously seen as complex. Your exact words were, "How do we simplify it now?" & nbsp;
马斯克：是的，我认为如果某企业善于制造，那基本上也可以大批量制造。产品的制造成本渐进式地接近其原材料价值，再加上你需要取得的任何知识产权。这是一件非常困难的事情，但对任何东西来说都是可能的。因此一个产品在最初设计时，经常发生的情况是，人们从他们熟悉的工具、部件和方法开始，然后试图用他们现有的工具和方法制造这个产品。
Mask: Yes, I think that if an enterprise is good at manufacturing, it can be made in large quantities. The costs of manufacturing products are incrementally close to the value of their raw materials, together with any intellectual property rights you need to acquire. This is a very difficult thing, but it is possible for anything. So when a product is initially designed, it often happens that people start with tools, components and methods that they know, and then try to make it with the tools and methods that they have.
另一种思考方式是尝试想象完美产品或技术，不管它是什么。然后想：“原子怎样才能完美地排列？让我们试着找出如何获得这种形状的物品”。
Another way of thinking is to try to imagine the perfect product or technology, whatever it is. And then think, "How can the atoms be perfectly arranged? Let's try to figure out how to get this shape."
19、Lex Fridman：在开始认真思考前，这种方式听起来其实有点儿“瑞克和莫蒂”（美国科幻喜剧动画片）式的荒谬。如果大家按照以前的思考方式做事经常受挫，那么真的应该用这种方式来思考。
19, Lex Fridman: Before you start thinking seriously, it sounds a little bit "Rick and Morty" kind of ridiculous. If you're always frustrated with the way you used to think, then you really should think about it in that way.
马斯克：常见的思考方式其实是一种惯性功能。人们更倾向于使用他们熟悉的工具和方法，导致做出的东西不太可能是完美产品。所以这就是为什么要从两个方向来考虑问题：第一，可以用我们的工具建造什么；第二，理论上的完美产品是什么样的？
Mask: The usual way of thinking is actually a inertia. People tend to use tools and methods that they know so well that what is made is unlikely to be a perfect product. So that's why we think in two directions: first, what can be built with our tools; second, what is the theoretical perfect product?
因为你学习得越多，对完美产品的定义也会改变。在此之前，你实际上不知道完美的产品会是什么样的，但你可以制造一个近似于完美的产品。具备了以上两个思考方向，你应该会想：“现在我们需要创造什么工具、方法、材料才能让原子变成需要的形状？”
Because the more you learn, the better the definition of the perfect product will change. Until then, you actually don't know what the perfect product is, but you can make a product that looks like the perfect one. With these two directions in mind, you should think, "What tools, methods, materials that we need to create to make the atom the shape we need?"
这种思考方式很有效，但很少人这么想。
This way of thinking is very effective, but very few think so.
最晚10年后送人上火星
delivers people to Mars 10 years later
火星最好用加密货币
Mars preferably in an encrypted currency
20、Lex Fridman：下面我们来谈谈火星，你之前提到了，在月球上建一个科学基地，进行一些研究，这对科学界来说是非常有益的。但其实真正的飞跃，在这个看似不可能的类别中——让人类登陆火星。你觉得SpaceX未来多久会把人类送上火星？
马斯克：最早是5年后，最晚是10年后。
Masc: Five years from now, 10 years from now.
21、Lex Fridman：决定因素是什么？你觉得是由工程方面的因素决定的吗？还是说工程因素并不是瓶颈问题？
21, Lex Fridman: What are the determinants? Do you think it's engineering? Or is engineering not a bottleneck?
马斯克：根本上来说还是星舰的设计，Starship星舰是有史以来最复杂的先进火箭，我不知道决定因素是什么，或许是星舰的规模等类似的因素吧。
Mask: Basically, it is the design of the Starship, the most sophisticated rocket ever, and I don't know what the determining factor is, perhaps similar to the size of the Starship.
星舰很大，先前的飞船同我们的星舰相比，不是一个重量水平。因此星舰的基本优化是最大限度地减少每吨的入轨成本，并最终减少到登陆火星表面的成本。优化成本似乎是一个商业目标，但它实际上是需要优化的东西。登陆火星表面的每吨成本是固定的，之后我们可以在那里再建立一个自给自足的城市。除此之外，我们所能做的就微乎其微了。
Starships are large, and previous ships are not a weight level compared to our starships. So the basic optimization of starships is to minimize the cost of in-orbiting per ton, and eventually to reduce the cost of landing on the surface of Mars. Optimization costs seem to be a commercial objective, but it is actually something that needs to be optimized. The cost per ton of landings on Mars is fixed, and then we can build a self-sufficient city there.
现在去火星的往返成本大约是1万亿美元，但有再多的钱也买不到去火星的船票。其实要把人送上火星也不是不可能，但我们想要的不仅是登陆火星在上面树立旗帜留下脚印，然后和登陆月球一样，半个世纪都不回去一次。我们想在火星定居，长久地生存在那儿，我认为人类需要成为一个多星球定居的物种。
The cost of going back and forth to Mars is about $1 trillion, but there's no more money to buy a ticket to Mars. It's not impossible to send people to Mars, but what we want is not just to land on Mars and set footprints on it, but not to go back for half a century, like the moon. We want to settle on Mars and live there for a long time, and I think we need to be a multi-planetary species.
可能这对很多人来说有些深奥，但随着时间的推移，地球很可能会经历一些灾难，这可能是人类行为产生的结果，或者是气候突变等外部原因造成了全球性的灾难，人类的命运或许会同恐龙一样。
This may be somewhat profound for many, but over time the planet is likely to experience some catastrophes, which may be the result of human behaviour, or external causes such as climate mutations, which may be global in nature and whose fate may be the same as that of dinosaurs.
如果发生这种情况，人类将以某种方式继续生存，太阳逐渐变大，最终会吞噬地球。地球可能会在500万年后变得很热，不适合人类居住。虽然这个时间还很长，但也只比地球存在的时间长10%。
If that happens, humanity will continue to survive in some way, the sun will grow and eventually devour the Earth. The Earth may become hot and unfit for human habitation after 5 million years. This is a long time, but only 10 per cent longer than it exists.
但是地球已经存在了45亿年，这也是45亿年来第一次有可能将生命延伸到地球之外的行星。机会之窗可能会开放很长一段时间，在机会之窗开启时，迅速采取行动是明智的。
But the planet has existed for 4.5 billion years, and this is the first time in 4.5 billion years that it is possible to extend life beyond the planet. The window of opportunity may be open for a long time, and it is wise to act quickly when the window of opportunity opens.
22、Lex Fridman：核威胁、疫情以及各种风险会给我们一些动力。
马斯克：我的意思是，文明可能会因人口崩溃而消失，或是因为第三次世界大战消失。
Mask: I mean, civilization may disappear as a result of population collapse or because of World War III.
这些都充满了不确定性。我认为人类的未来在很大概率上是好的。但是每个世纪都有1%的可能会让一个文明终结，斯蒂芬·霍金也有类似的预言。
These are uncertainties. I think the future of humanity is good in a lot of probability. But one percent of every century could end a civilization, and Stephen Hawking had a similar prophecy.
他很大概率是对的。我们应该把人类看作是一个多星球居住的物种，就像给生命买保险一样。
He's very likely to be right. We should think of human beings as a species that lives on multiple planets, like insurance for life.
我们把植物和生物送上火星，给这个星球注入生命，让它成为第二个有生命存在的星球。它们仅凭自己无法前往火星，如果SpaceX不把人类带上火星，那么等到太阳膨胀使地球无法居住的时候，人类就会灭绝。
We send plants and organisms to Mars, injecting life into the planet, making them the second living planet. They cannot travel to Mars on their own. If SpaceX does not bring mankind to Mars, then humanity will be extinct when the sun expands to make the planet uninhabitable.
23、Lex Fridman：你认为在火星建立文明最困难的部分是什么？是从工程层面改造火星？还是经济层面或是人类层面？让这些人永远在火星定居，不返回地球。
23, Lex Fridman: What do you think is the most difficult part of establishing civilization on Mars?
马斯克：我认为他们还是会返回地球的。
Masc: I think they'll still be back on Earth.
24、Lex Fridman：但有些人会选择永远定居火星。
24, Lex Fridman: But some people choose to settle on Mars forever.
马斯克：是的，我相信很多人会，但是我们要造的是一艘可以随时往返的星舰。制造星舰的成本昂贵，我们要的是一艘可以重复使用的星舰，同时记录这些人前往火星的旅程。
Masc: Yes, I'm sure many people will, but we're going to build a starship that can come and go at any time. It's expensive to build a starship, and we're going to have a starship that can be used again, and we're going to record these people's journeys to Mars.
25、Lex Fridman：你考虑过把火星地球化的成本吗？比如一些实际的建造事宜？你现在是否专注于建设登陆火星的飞船？
25, Lex Fridman: Have you considered the cost of globalizing Mars? Some practical construction, for example? Do you now focus on building a spaceship to land on Mars?
马斯克：当然考虑过。但是首先要解决如何到达火星这件事，如果连火星都去不了，那再讨论其他事情就没有任何意义了。我现在所思考的就是要降低到达火星的成本。目前，平均每吨重量到达火星的成本在10亿美元（约合人民币63.8亿元）左右，这些成本不仅包括制造火箭、发射，还包括隔热、导航系统、太空通信以及着陆系统。因此，10亿美元（约合人民币63.8亿元）只是粗略估计。但这显然有些昂贵，而且无法创造一个可以自给自足的文明，所以我们需要把成本至少缩减一千倍。
The cost of reaching Mars is currently around $1 billion (about $6.38 billion). These costs include not only making rockets, launchings, but also insulations, navigation systems, space communications, and landing systems. So, the $1 billion (about $6.38 billion) is a rough estimate.
26、Lex Fridman：把登陆成本控制在每吨100万美元（约合人民币638万元）？
26, Lex Fridman: Controlling landing costs at $1 million (approximately $6.38 million) per ton?
马斯克：是的，理想情况下每吨的登陆成本要远低于100万美元（约合人民币638万元）。但除此之外，还有在火星上建立一座自给自足的城市成本。这一部分是关键，如果因为一些原因，地球不再向火星发送飞船，这里（火星）的人类也应该能生存下去。
MASK: Yes, ideally, the cost of landing is well below $1 million per ton (approximately $6.38 million). But, in addition, there is the cost of building a self-sufficient city on Mars. This is crucial, and if, for some reason, the Earth no longer sends ships to Mars, humans here should survive.
千里之堤，溃于蚁穴，就是这个道理，只要缺少一个关键部分，那么这就不算一个自给自足的城市。就像一个人如果在海上远行，除了维生素C之外，他什么都有，但他知道没有维生素C早晚都会死去。
A thousand miles of dyke, strangulation in an ant cave, that's the truth, and it's not a self-sufficient city without a key part. Like a man who travels at sea, he has everything but vitamin C, but he knows that no vitamin C will die sooner or later.
所以我们要让火星的城市拥有自我维持的能力，我不确定在我的有生之年是否能够看到这一景象，但我喜欢能看到这个城市能够建立的可能。
So we want the city of Mars to be self-sustaining, and I'm not sure I can see it in my lifetime, but I like to see the possibility that it can be built.
你可能会问我：“建造一个自我维持的城市，飞船需要搭载的最低吨位是多少？”
You might ask me, "What's the minimum tonnage that a ship needs to carry to build a self-sustaining city?" & nbsp;
这里面有很多不确定因素，我也不知道，因为需要在火星上建设大量的基础设施，比如半导体工厂，矿石冶炼厂等，所以重量至少是一百万吨。而且火星是最不适合居住的行星之一，但它比地球离太阳更远。
There's a lot of uncertainty, and I don't know, because there's a lot of infrastructure on Mars, like semiconductor plants, ore smelters, so it weighs at least a million tons. And Mars is one of the most uninhabitable planets, but it's far away from the sun.
27、Lex Fridman：是的，范围仅限于在太阳系中。
马斯克：是的，在太阳系里。太阳系也可能有更适宜人类居住的行星，但是我们达到不了。
Masc: Yes, in the solar system. The solar system may also have planets more suitable for humans, but we can't reach it.
28、Lex Fridman：太难了？
马斯克：是的，非常非常困难。
MASK: Yes, very, very difficult.
29、Lex Fridman：其实我们可以换个思路，你刚才提到过物理学是你的出发点，而广义相对论认为虫洞是存在的。你认为人类可以利用虫洞来进行光速的旅行吗？
29, Lex Fridman: In fact, we can change our minds. You just mentioned that physics is your starting point, and in general relativity you think wormholes exist. Do you think humans can use wormholes to travel light speed?
马斯克：虫洞的事情是值得商榷的，但我们目前还不知道有什么方法可以超过光速。其实还有一些空间理论的说法，在空间内让一个物体以光速移动，再让空间本身移动，这其实叫做扭曲空间，这个速度能够超过光的移动速度。
Masc: The thing about wormholes is debatable, but we still don't know how to move beyond the speed of light. There are some space theories that allow an object to move at the speed of light in space and then move the space itself, which is called a distortion of space, which can outpace the speed of light.
就像宇宙大爆炸，宇宙爆炸的速度远远超过了光速。
Like a big explosion in the universe, the speed of explosion in the universe far exceeds the speed of light.
30、Lex Fridman：是的。
马斯克：扭曲空间需要巨大的能量。究竟是多少，目前还未可知。
Masc: Distorting space takes a lot of energy. It's not known how much.
31、Lex Fridman：你在火箭的推进方面，有多少创新？能把（火箭）推进的效率提升10倍吗？你在物理学中找到可以使发动机效率得到明显改善的方法了吗？
31, Lex Fridman: How many innovations do you have in rocket propulsion? Can you increase the efficiency of a rocket ten times? Did you find a way in physics to make a significant improvement in engine efficiency?
马斯克：正如我所说的，火箭技术真正的梦想，是建立一个完全和快速可重复使用的轨道系统。而猎鹰9号是唯一可重复使用的火箭，在SpaceX的视频中你能看到，火箭能够返回并且成功着陆，同时我们成功回收了火箭整流罩的前锥体，但是我们没有成功回收上半段。
Masc: As I said, the real dream of rocket technology is to create a fully and rapidly reusable orbital system. And Eagle 9 is the only reusable rocket, as you can see on SpaceX's video, that the rocket can return and land successfully, while we have successfully recovered the front cone of the rocket's shroud, but we have not succeeded in recovering the first half.
这意味着我们在建造上半段火箭时会把成本控至最低。猎鹰9号是一种两级火箭，可以看作是两架飞机，一架大的和一架小的，大飞机成功收回，而小飞机则没有。不过上半段火箭的成本也很高，至少在1000万美元（约为6372万元人民币）左右。
This means that we will have the lowest cost when building the first half of the rocket. Eagle 9 is a two-stage rocket that can be considered two planes, one big and one small, and the big one is successfully recovered, while the smaller ones are not. But the cost of the first half of the rocket is also high, at least 10 million yuan (approximately 6,372,000 yuan).
现在猎鹰9号的重复使用程度还达不到我们所希望的水平。猎鹰9号每次的发射成本不包含在我们最低的边际成本中，火箭的单次的发射成本在1500万美元（约为9557万元人民币）到2000万美元（约为1.27亿元人民币）左右，这已经是有史以来最低的火箭发射成本了。除此之外，我们正致力于推进火箭的迅速复用，如果成功，那么希火箭入轨的每吨成本有望降低100倍。
In addition, we are working on the rapid re-entry of rockets, which, if successful, could be reduced by 100 times the cost per ton of rockets in orbit.
试想一下，如果你有一架飞机或一辆汽车，你每次开车都要买一辆新车，那会很贵。但如果你给汽车加油或充电，这样就把你的出行成本降低了1000倍，这个道理在火箭领域也一样适用。
Imagine, if you had a plane or a car, you'd buy a new car every time you drove. But if you refueled or charged a car, that would reduce your travel costs by 1,000 times, as would apply in the field of rockets.
众所周知，制造火箭是很困难的一件事。所以如果我不能重复利用火箭的话，那么我可以舍弃一些构造，哪怕是重要的部分，看看是不是能达成我想要的目标，结果证明是可以的。所以理论上星舰也可以计算每次发射的成本，比如在100万美元（约为638万元人民币）到200万美元（约为1274万元人民币）之间，然后单次发射可以运载超过100吨的补给。虽然现在这么看有些疯狂。
It is well known that it is difficult to build a rocket. So if I can't repeat the rocket, I can drop some of the structures, even the important parts, to see if I can achieve what I want, and the results prove possible. So, in theory, Starships can calculate the cost of each launch, for example, between US$ 1 million (approximately RMB 6.38 million) and US$ 2 million (approximately RMB 12.74 million), and then single launches can carry more than 100 tons of supplies.
32、Lex Fridman：是的，很不可思议。所以你想表达的是，目前猎鹰9号的主要目的是让其可以全面复用，而不是取得理论物理学中的某种突破。
32, Lex Fridman: Yes, it's incredible. So what you're trying to say is that the main purpose of Eagle 9 is to make it fully operational, not to make a breakthrough in theoretical physics.
马斯克：并没有什么辉煌的突破，造一个可以完全重复使用的火箭就已经是一个极其困难的工程问题了。
Masc: There is no brilliant breakthrough, and building a rocket that can be fully reused is already an extremely difficult engineering issue.
33、Lex Fridman：我想问一个略带哲学意味的、有趣的问题。目前，你专注的是如何到达火星。如果人类一旦登陆火星，你认为火星早期的文明应该有怎样的政治经济体制？这个答案有助于人们想象未来的火星文明。
33, Lex Fridman: I'd like to ask a slightly philosophical and interesting question. For now, you're focusing on how to get to Mars. What kind of political and economic system do you think the early civilizations of Mars should have if they had landed on Mars?
马斯克：我个人认为，火星上应该建立直接民主制度，人们可以直接对事情投票，法律必须足够简短，方便人们理解。此外，法律还需要不断更新，并且具备“法律回收”的规定，可以随时废除一些陈旧的规则和制度。
Mask: I personally believe that there should be a direct democracy on Mars, that people can vote directly on things, and that the law must be short enough to be easily understood. Moreover, the law needs to be kept up to date and there is a “law recovery” requirement that allows for the repeal of outdated rules and regimes at any time.
34、Lex Fridman：你曾经说过，SpaceX会把一块狗狗币放到月球上，你曾经也说过狗狗币是人民的货币，那么你是否认为狗狗币是火星的官方货币或是未来的货币？
34, Lex Fridman: Did you ever say that SpaceX would put a piece of puppies on the moon, and you once said that puppies were the currency of the people, so do you think that puppies are the official currency of Mars or the currency of the future?
马斯克：我认为火星本身要拥有与地球不同的货币，因为两个星球距离太远，无法实现货币的同步流通。
Masc: I think Mars itself has to have a currency different from the Earth, because the two planets are too far away to be able to synchronize the circulation of money.
35、Lex Fridman：所以火星必须有一个独立的货币体系？
35, Lex Fridman: So Mars must have an independent monetary system?
马斯克：是的，因为火星离地球的距离大概是20光分（光子在一分钟内行走的距离）左右，也许更远，所以两个星球之间无法实现真正意义上的同步。
MASK: Yes, because Mars is about 20 centimetres away from the Earth (the distance where the photons walk within a minute) and perhaps even further away, there is no real synchronization between the two planets.
我不知道未来火星是否会有加密货币，但应该会产生一种火星本地化的货币。
I don't know if Mars will have an encrypted currency in the future, but it's supposed to produce a localized Martian currency.
36、Lex Fridman：火星上的人决定货币的形式？
马斯克：完全正确，火星的未来应该由火星人决定。我认为加密货币本身很有趣，它可以减少数据库中的错误——也就是我们所称之为货币的东西。
Masc: Exactly, the future of Mars should be decided by the Martians. I think it's interesting to encrypt the money itself, and it can reduce the errors in the database -- that's what we call the money.
我是Paypal的创始人之一，所以在货币的实际应用中有着相当深刻的理解。现在已经2021年了，银行业仍然还在购买大型机器，运行古老的COBOL代码。
I'm one of the founders of Paypal, so there's a great deal of understanding in the practical application of money. Now 2021, the banking industry is still buying large machines and running old COBOL codes.
美联储的设备可能比银行的还要老，他们的设备是一个笨重的老式COBOL主机。除此之外，政府拥有货币数据库的编辑权限，他们利用这些编辑权限来获得更多的货币，这就增加了数据库中的错误——也就是金钱。所以我认为应该通过信息理论的角度来看待金钱。
The Fed’s equipment may be older than that of banks, and their equipment is a clumsy old COBOL mainframe. In addition, the government has editing privileges for the currency database, which they use to obtain more money, adding to the errors in the database – that is, money.
可以把整个经济系统看作是一个互联网连接，有带宽、时延抖动和丢包这些网络通信中经常会出现的错误。因此我们也可以从这个角度思考货币，然后从信息理论的角度想究竟什么样的系统能使经济运行得最好。
The entire economic system can be seen as an Internet connection, with bandwidth, time-exploitation, and frequent errors in these web-based communications. So we can think about money from this point of view, and then think about what kind of system can make the economy work best from the point of view of information theory.
加密货币其实是一种尝试，以减少政府稀释货币供应量所造成的税收影响。
Encrypted currencies are in fact an attempt to reduce the tax impact of government dilution of money supply.
37、Lex Fridman：你觉得加密货币是21世纪人们做交易、储存财富的新方式吗？
37, Lex Fridman: Do you think encryption money is a new way for people to trade and store wealth in the 21st century?
马斯克：钱只是一种数据，而人们往往会认为金钱本身就有力量，其实这是错误的，金钱本身并没有力量。
Masc: Money is a figure of data, and it is often thought that money has power in itself, which is wrong, and money does not have power in itself.
在思考过程中加入物理学的观点是非常重要的，如果你是一个亿万富翁，却被困在一个热带岛屿上，那么你的钱实际上并没有价值。因为孤岛上没有资源分配，钱则是资源分配的数据库。除了你自己，岛上没有任何资源可以分配，所以钱是无用的。
It's important to include the idea of physics in your thinking, because if you're a billionaire and you're stuck on a tropical island, your money is not really worth it. Because there's no resource allocation on an isolated island, it's a database of resource allocation.
所以我们可以把货币看作是一个跨时空的资源分配数据库，这个数据库或数据系统应该以什么形式出现，才能发挥最大效果呢？目前比特币存在一个根本性的问题——交易量非常有限，交易确认延迟过长。
So we can see money as a data base for resource allocation across time and space, in what form should this database or data system appear in order to be most effective? At present, there is a fundamental problem in Bitcoin – a very limited volume of transactions and too long delays in confirming transactions.
从这个问题的角度来看，比特币并不是一种理想的加密货币，我不否认它本身存在一定作用，比特币可能有助于解决货币数据库问题的一个方面——即财富的储存或相对义务的核算——但它作为一种日常货币没有用处。
From the point of view of this question, Bitcoin is not an ideal encrypted currency, and I do not deny that it has a role to play in itself, and that bitcoin may help to solve one aspect of the currency database problem — namely, the storage of wealth or the accounting of relative obligations — but it is not useful as a daily currency.
虽然狗狗币是因一个玩笑而诞生的，但我仍然认为它有很多优点。它的交易量比比特币高很多，现在如果你想做一笔比特币交易，那么这笔交易的价格非常高。
Although it was born out of a joke, I still think it has a lot of merit. It's a lot more than bitcoin, and now if you want to do a bitcoin deal, the price of the deal is very high.
比特币诞生在2018年，当时的互联网还没有现在这么先进，当时比特币的交易可能能与当时的网速同步。但现在是2021年，再过10年，这个（交易）速度会低的离谱。
Bitcoin was born in 2018, when the Internet was not as advanced as it was, when Bitcoin’s transactions might be synchronized with the Internet speed at that time. But it's now 2021, and in another 10 years, it's going to be a very low speed.
而且我认为货币的数量应该呈线性增长，如果一种货币过于紧缩，会随着时间推移而升值，那么人们就不愿意把这种货币花掉。
And I think that the amount of money should grow linearly, and if a currency is too tight to appreciate over time, people are reluctant to spend it.
建立向量空间是自动驾驶最难的部分
38、Lex Fridman：特斯拉的Autopilot在过去六年中取得了令人难以置信的成就。之所以说它是令人惊叹的成就，是因为我清楚地知道这其中充满了挑战。 我当时在麻省理工工作，同时我也有许多在DARPA（美国国防部高级研究计划局）工作的朋友，所以非常理解计算机视觉的困难。因此，当我第一次驾驶最初基于Mobileye自动驾驶系统的特斯拉汽车时，我持有一种怀疑态度。我认为这辆车不可能做到自主保持车道，并为驾驶员创造一种舒适的驾驶体验。 I was working in MIT, and I had many friends in DARPA (the United States Department of Defense's Advanced Research and Planning Bureau), so I understood the difficulties of computer visualization. So, when I first drove the Tesla car, which was originally based on Mobileye's autopilot system, I had a skeptical attitude. I didn't think it would be possible to keep the driveway autonomous and create a comfortable driving experience for the driver. 马斯克：其实车道保持很容易做到。 Masc: In fact, it's easy to keep the driveway. 39、Lex Fridman：但其实最初Mobileye给特斯拉的原型方案并不能解决这个问题。 马斯克：我们必须在Mobileye方案的基础上编写大量的代码，Mobileye的方案并不是我们推向市场的自动驾驶方案。 Masc: We have to prepare a large number of codes based on the Mobileye programme, which is not an automatic driving programme that we push to the market. 40、Lex Fridman：这其实就是Autopilot发展过程中的插曲——你做事的方式。你习惯以一种全新的方式解决问题，这其实是我所见过的最大胆的决定，你在软件和硬件上，都做了一套全新的自动驾驶方案。 40, Lex Fridman: This is actually an interlude in Autopoilot's development -- the way you do things. You're used to solving problems in a new way. It's actually the boldest decision I've ever seen. You've made a whole new autopilot program on software and hardware. 我可能最关心和最喜欢的是Andrej Karpathy所领导的工作——数据集的选择、数据引擎过程、神经网络架构、在实际生活中网络测试、验证的方式。与计算机视觉的图像网模型相比，这就像现实世界的人工智能。 may be my most concerned and favorite job led by Andrej Karpathy - data set selection, data engine processes, neural network architecture, ways of testing and authenticating in real life. This is like artificial intelligence in the real world when compared to a computer visual image network model. 马斯克：Andrej很了不起，他显然发挥了重要作用。其实我们也有很多非常有才华的人在推动这项工作，Ashock实际上是自动驾驶工程的负责人，Andrej是人工智能项目的负责人。 Masc: Andrej is amazing, and he obviously plays an important role. We have a lot of very talented people working on it, and Ashock is actually in charge of autopilot, and Andrej is in charge of artificial intelligence projects. 不夸张地说，特斯拉Autopilot的AI团队聚集了世界上最顶尖的一群人。所以特斯拉的自动驾驶软件才能不断取得进展。 It's no exaggeration to say that Tesla Autopilot's AI team brings together the world's best people. So Tesla's autopilot software can move on. 41、Lex Fridman：在Autopilot发展的5～6年中，你对自动驾驶有哪些见解？ 41, Lex Fridman: What do you think about autopilots in the 5-6 years of AutoPilot development? 马斯克：一开始，我就知道让汽车自动驾驶是一件极其困难的事情，但在项目推进的过程中，我发现这比我想象的还要困难。要让车辆自动驾驶的基础就是重现人类的驾驶方式——人类眼睛感知并用神经网络处理信息，最终完成驾驶行为。 Mask: At first, I knew it was extremely difficult to get a car to drive automatically, but as the project went forward, I found it more difficult than I thought. The basis for driving automatically was to recreate the human mode of driving — human eye perception and neurological network processing of information, and eventually to complete driving. 而特斯拉的FSD软件就是这样设计的，基本上是利用视觉算法和神经网络做到自动驾驶。因此，为了部署FSD，我们需要将视觉算法和神经网络以数字的形式重建起来，摄像头有性能上的提升，同时神经网络在芯片之中。很明显，这种方式能够实现完全自动驾驶能力。并且，这是唯一的解决方案，我不相信还有其他解决方案。 And that's how Tesla's FSD software is designed, basically using visual algorithms and neurological networks for autopilot. So, in order to deploy FSD, we need to rebuild visual algorithms and neural networks in digital form, with camera performance upgrades, and neural networks in the chip. Clearly, this way is able to achieve full autopilot. And, this is the only solution, and I don't believe there are other solutions. 42、Lex Fridman：但问题是要把哪些类人的感知能力编码到机器当中？当人类在开车时都会注意哪些方面？比如机器要知道一个开启的车门是怎样的，这就是感知能力。系统必须首先解决感知问题，才能进一步控制、规划汽车。你必须把驾驶中涉及的内容研究透彻，这其中还有很多不同的边缘情况。 42, Lex Fridman: But the question is which kind of person's perception capabilities are to be encoded into the machine? What are the things that humans notice when driving? 马斯克：我认为难点并不在于控制逻辑。 Masc: I don't think the hard part is controlling logic. 43、Lex Fridman：你认为在整个问题中，什么才是最困难的部分？ 马斯克：最困难的应该是建立一个准确的向量空间，因为需要太多的软件，以及大量的代码。摄像头感知图像之后形成数字信号，需要将数字信号映射到向量空间中，最终可以识别汽车、人、车道线、曲线和红绿灯等等。一旦建立起了准确的向量空间，控制车辆就会变得像玩游戏一样简单，就像是玩GTA、赛博朋克等游戏一样。建立准确的向量空间虽然很难，但并不是不可逾越的。 Masc: The most difficult thing to do is to create an accurate vector space, because too many software and a lot of codes are needed. Camera images generate digital signals that need to be mapped into vector space, eventually identifying cars, people, lane lines, curves, red green lights, etc. Once the exact vector space is built, it becomes as simple as playing games, like playing games such as GTA, Saibonk. It is difficult to build an accurate vector space, but it is not insurmountable. 44、Lex Fridman：是的，人类自身的感知系统将光子映射到头脑中的向量空间是一件不可思议的事情，但大部分人都没有意识到。 44, Lex Fridman: Yes, it's amazing that humans' own perception systems map photons into vector space in their minds, but most people don't realize it. 马斯克：其实现在你的大脑正在处理大量的数据，为你呈现了一副非常清晰的图像。就好像我们环顾四周时，你会看到五彩缤纷的颜色，但实际上这些颜色是你眼睛的视锥细胞传递给大脑的。这种细胞能接受光刺激，能将光能转换为神经冲动，故亦称光感受器。 Masc: The realization is a lot of data being processed in your brain, showing you a very clear image. It's like when we look around you, you see colorful colors, but in fact these are transmitted to the brain by the eye cone cells of your eyes. These cells can accept light stimuli, turn light into nerve impulses, so it's called light sensers. 你的眼睛会给看到的东西“上色”。但你并没有意识到这个过程。你的大脑在对眼睛的视觉信号进行大量的后期处理。这很不可思议，但其实人的大脑在处理一些数据的时候，也会忘记一些事情，所以记忆力也许是人类大脑最薄弱的地方。我们的记忆力是有限的，大脑在接收信息的同时也会遗忘，所以最后它最后处理完的信息其实要比大脑接收的信息要少。 Your eyes are going to give you something to look at. But you're not aware of the process. Your brain does a lot of late-stage processing of visual signals. It's amazing, but when the brain deals with data, it forgets things, so memory is probably the weakest part of the human brain. Our memory is limited, and the brain forgets when it receives information, so the information that it ends up processing is actually less information than the brain receives. 因此，你的大脑在试图创建一个向量空间，当你在开车的时候，可以思考下你的大脑在有意识地想什么。 So your brain is trying to create a vector space, and when you're driving, you can think about what your brain is consciously thinking. 特斯拉的汽车用摄像头来充当眼睛，完成自动驾驶。这和人类开车的方式一样，人的眼睛就相当于两个摄像头。虽然开车时用视力观察周围的环境方式不错，但人们会分心，思考一些东西、发短信、吵架或是调电台。 Tesla's car uses a camera as an eye to perform autopilot. It's like the way humans drive, they have the same eyes as two cameras. It's nice to look at the surroundings with eyesight, but people are distracted, thinking about things, texting, arguing, or calling the radio. 人在往前后左右观察的时候，实际上就是在刷新大脑的向量空间，大脑试图提炼出相关的向量空间，这些空间基本上是固定位置的和正在移动的物体，然后大脑把这些内容编辑成最少量的信息。这些信息是你开车时的必要信息。 When you look around and around, it's actually refreshing the vector space of the brain, trying to extract the vector space that is basically fixed and moving, and then the brain edits it into the minimum amount of information that you need when you're driving. 45、Lex Fridman：大脑能自动编辑向量空间，或者进一步压缩成概念之类的东西。人类的大脑有时超越了向量空间，变成概念空间，你看到的东西不再是空间上的某种表现，它就像一个你应该知道的概念。 45, Lex Fridman: The brain automatically edits vector space, or further compresss it into concepts. The human brain sometimes goes beyond vector space, becomes conceptual space, and what you see is no longer a manifestation of space, it's like a concept that you should know. 如果在学校区域，你会记得这是一个概念，这很奇怪。虽然当时没有看到学生，但你心里会想：要小心开车，当心学生。 If you remember it in the school area, it's strange. Although you didn't see the students at the time, you thought, "Be careful driving, be careful with the students." 马斯克：你需要建立向量空间，然后对这些空间进行实际预测。 Mask: You need to build vector space and then make real predictions of those spaces. 比如当你开车时，前面有一辆卡车，还有一些孩子准备过马路，当你距离卡车较近的时候，卡车挡住了你的视线，你看不到那些小孩了。这时，你心里想的是这些孩子现在在什么位置，需要提前预测一下（位置）。 For example, when you're driving, there's a truck in front of you, and there's some kids going across the street, and when you're closer to the truck, the truck's blocking your sight, and you can't see the kids. And what you're thinking about is where these kids are right now, and you need to look ahead. 46、Lex Fridman：对计算机视觉来说，识别移动的物体，比如这个物体被一棵树遮挡后又重新出现，这种连贯的追踪是非常难的。 46, Lex Fridman: For computer visuals, it is very difficult to identify a moving object, such as the object being blocked by a tree and reappearing. 马斯克：这就是我们正在做事情——“客体永久性”，这有点像人类的神经网， 当一个人从幼儿阶段长大时，有一个提升自我认知的时间点，机器也是如此。 Masc: That's what we're doing -- "personal permanency" -- it's kind of like the neural network of humans, and when a person grows up in early childhood, there's a point of time for self-awareness, and so is the machine. 如果你把玩具或其他东西藏在背后，然后再拿出来给一个婴儿看，对那个婴儿来说，每次你把东西拿出来，这个东西都是第一次出现的，对婴儿们来说都是新事物。他/她们可以整天玩藏猫猫。但之后，婴儿长大获得了“客体永久性”，他们意识到这个东西其实一直都存在，并没有消失，只是被人藏在身后了。 If you hide a toy or something behind your back and then show it to a baby, every time you take it out, it comes out for the first time, it's new for the babies. He/she can play hide-and-seek all day long. But then the baby grows up and it gets permanent, and they realize that it actually exists, and it doesn't disappear, but it's hidden behind them. 47、Lex Fridman：有时我希望我们永远不要获得“客体永久性”。 47, Lex Fridman: Sometimes I hope we never get a "physical object." 马斯克：是的。汽车神经网络的一个重要的发展是记忆，跨越时间和空间的记忆。实际上人类的大脑不能掌控记忆的多少，你想记住的越多，就会耗费更多的代价。 Masc: Yes. An important development in the neural network of cars is memory, memory that transcends time and space. In fact, human brains can't control the amount of memory, and the more you want to remember, the more it costs. 所以如果你想长时间记住太多的东西的话，到后期很多记忆就会变得比较陈旧，并且你还需要再记住一些新的东西。 So if you want to remember too much for a long time, a lot of memories will become old at a later stage, and you need to remember something new. 比如说，一段5秒钟的记忆——你把车停在红灯前，行人在绿灯亮起时过马路。但如果你再次经过路口时，因为一些原因你可能看不清行人，但你依然需要记住马路边上可能会有行人，并且他们可能要过马路之类的事情。所以即使这超过了你的时间记忆，也不能超过你的空间记忆。 For example, a five-second memory -- you park your car in front of a red light, walkers cross the street when the green light comes on. But if you go through the intersection again, for some reason, you may not be able to see pedestrians, but you still need to remember that there may be pedestrians on the side of the road, and they may have to cross the road. So, even if it's more than your time memory, it doesn't go beyond your spatial memory. 48、Lex Fridman：所以获取数据来学习你所说的所有概念是一个迭代的过程。 马斯克：对，但我们可能会换个名字。 49、Lex Fridman：好吧，我相信这会和《瑞克和莫蒂》一样精彩。 49, Lex Fridman: Well, I'm sure it'll be as good as Rick and Morty. 马斯克：我们的团队在车里设计神经网络很多次了，超疯狂的。 Masc: Our team's been designing neural networks in the car many times. It's crazy. 50、Lex Fridman：所以每次有一个新版本，你都会把它重命名为更荒谬或更难忘和美丽的东西吗？啊不对，不能说是荒谬。 50, Lex Fridman: So every time there's a new version, will you rename it something even more absurd or unforgettable and beautiful? No, it's not. 马斯克：目前操纵汽车的神经网络的完整阵列有非常多的层级，这是非常复杂的，所以我们没有采用这种方式。 Masc: The complete array of neural networks currently operating cars contains very many layers, which is very complex, so we do not follow this approach. 我们采用简单的神经网络，基本上就是在一个摄像机的一帧图像上进行识别，然后试着把它们结合在一起。我们有自己的C语言编译器。因此，为了获得最大的性能，我们也在不断优化我们的C语言编译器的最大效率。 We use simple neural networks, basically identifying them on a camera image, and trying to combine them. We have our own C-language compiler. So, in order to maximize our performance, we're also constantly optimizing the efficiency of our C-language compiler. 事实上，我们最近在C语言编译器上做了一个新的REV命令，它将直接编译到我们的Autopilot自动驾驶硬件上。 In fact, we recently made a new REV order on the C-language compiler, which will be translated directly into our AutoPilot autopilot hardware. 51、Lex Fridman：你想要用你自己的编译器编译所有的东西吗? 51, Lex Fridman: Do you want to compile everything with your own compiler? 马斯克：对，有各种各样的计算。 MASK: Yeah, there's a lot of calculations. 52、Lex Fridman：那就是说也包括CPU、GPU这些基本的东西吗？所以你自己也参与到编译代码中了？ 52, Lex Fridman: Does that mean you're also involved in compiling the basics of CPUs and GPUs? 马斯克：对，我们还安排了很多人参与其中。 MASK: Yes, we have a lot of people involved. 这涉及到很多的硬核的软件工程，所以我们也在FSD计算机上做了大量的计算，并且我们需要在有限的计算能力下尽可能地达到每秒最高的帧数。 This involves a lot of hard core software work, so we've done a lot of calculations on the FSD computer, and we need to reach the highest frame number per second as much as possible with limited computing capacity. 特斯拉团队有很多非常有能力的软件工程师，他们也在致力于提高我们的计算效率。 The Tesla team has many very capable software engineers who are also working to improve our computing efficiency. 53、Lex Fridman：你想要达到像电子游戏一样多的高帧率的话就需要全分辨率，高帧率，低延迟。 马斯克：对，除了上述几点，还需要低抖动。现在我们要做的事情之一就是不需要通过图像信号处理器对图像进行后处理。 MASK: Yes, in addition to the above, there is a need for low shaking. Now one of the things we have to do is not need to reprocess images through the image signal processor. 比如一般的相机为了让照片看起来更漂亮，需要做很多后处理。但我们不关心图片是否漂亮，我们只想要数据，所以我们的相机收集的只是原始的光子计数。 Common cameras, for example, need to do a lot of reprocessing to make the picture look more beautiful. But we don't care if it's beautiful, we just want data, so our cameras just collect raw photo counts. 所以我们的系统在电脑上看到的图像实际上比人们通常看到的要多得多。人眼看到差不多光线的两个点，实际上光子数量可能相差很多。这意味着我们的相机在黑暗中也能看得很清楚，因为它能探测到光子数的微小差异。 So our systems actually see a lot more images on the computer than people usually see. People see two dots of almost light, and actually the number of photons can be very different. That means that our cameras can also see clearly in the dark, because they can detect small differences in the number of photons. 由于去掉了图片的后处理过程，我们的延迟提升了13毫秒。其中每个摄像头大概是节省了1.5或者1.6毫秒，8个摄像头加起来就节省了13毫秒。 The delay was increased by 13 milliseconds by removing the post-processing of the picture. Each of the cameras saved about 1.5 or 1.6 milliseconds, and the eight cameras together saved 13 milliseconds. 我们如何统计延迟？包括景象进入摄像头，摄像头数据进入多种神经网络，进入基于C语言和少量C++写的程序，再到自动驾驶系统给出驾驶指令，并将指令发给电驱动执行单元，以实现加速、制动或者转向操作。 How do we count delays? These include video access to cameras, camera data to multiple neural networks, access to C-language-based and small C++-written programs, driving instructions from autopilot systems, and sending instructions to power-driven execution units to speed up, brake or turn to operations. 一些控制器的工作频率在10Hz（每100ms工作一次），所以这时候已经产生了100毫秒的延迟。所以我们想要升级电驱动系统的控制器，让它以100Hz或者更高的频率进行工作（每10毫秒工作一次），这样就能减少整个系统的反应延迟。 Some controllers work at 10Hz (per 100ms), so there's already a 100 ms delay. So we want to upgrade the power-drive control so that it works at 100Hz or higher (per 10 ms), reducing the system's response delay. 注：马斯克表示，延迟问题可测量和预测，不是那么难。真正难的是一个名为“抖动”的现象，抖动用于描述数据传输延迟的变化程度。抖动越高，整个系统的延迟问题就会被大幅放大，这时候系统就不知所措了。 Note: Mask says that delay is not so difficult to measure and predict. The real difficulty is a phenomenon called "shaking," which is used to describe the extent of delay in the transmission of data. The higher the shaking, the greater the delay in the entire system will be greatly magnified, and the system will be overwhelmed. 马斯克没有介绍特斯拉是如何解决抖动的，但他表示Autopilot或者FSD系统最终的驾驶表现，比007詹姆斯邦德开车还厉害。 Mask did not describe how Tesla addressed the shaking, but he indicated that the ultimate driving performance of Autopilot or the FSD system was better than James Bond driving at 007. 今年底将实现L4 will achieve L4 & nbsp by the end of this year; 未来车辆可以识别声音 Future vehicles recognize sound 54、Lex Fridman：回顾过去的6年，同时面向未来，基于你目前的理解，你认为解决FSD这个问题难度如何？你认为特斯拉汽车在什么时候会达到L4级的FSD？ 马斯克：我认为自动驾驶看上去非常有可能，可能会在2022年实现。 mask: I think autopilots seem very likely to happen in 2022. 55、Lex Fridman：那么是通过什么样的方式实现的呢？是通过目前的FSD beta版本吗？这一版本会变得越来越好并最终超越目前的水平实现更高的自动化吗？ 55, Lex Fridman: So how is it done? Is it going to be done through the current FSD version? Is it going to get better and eventually more automated than it is? 马斯克：是的，任何关注特斯拉FSD beta的人都可以看到目前车辆的接管比例正在不断下降。极端情况下，司机也可以介入让车辆远离一些危险的情况。目前每一百万英里的司机干预正在急剧下降，看起来在2022年就能够实现自动驾驶，FSD的事故发生比例会小于一般人。 Masc: Yes, anyone who cares about Tesla FSD beta can see that the proportion of vehicles taking over is decreasing. In extreme cases, drivers can also intervene to keep vehicles away from some dangerous situations. 对于此，我们不必向监管机构证明这一点，我们只需要一个相应的标准来证明FSD的表现比一般人更好。我觉得自动驾驶的安全性将至少会高于人类两倍或者三倍，受到损伤的可能性也会低于人类两倍或者三倍，自动驾驶将会是更好的一种方式。 We do not have to prove this to regulators, and we need a corresponding standard to prove that the FSD is doing better than the average person. I think that autopilots will be at least twice or three times safer than humans, they will be twice or three times less likely to be damaged, and autopilots will be a better way. 56、Lex Fridman：所以目前特斯拉推出了FSD 10.6版本，10.7版本也已经在路上了，也许11.0版本也会在不远的将来出现。 56, Lex Fridman: So Tesla now has FSD version 10.6, version 10.7 is on its way, and perhaps version 11.0 will appear in the near future. 马斯克：是的，实际上我们非常希望在2021年推出11.0版本，但是11.0版本有很多基础的神经网络架构需要重写，此外还在创建向量空间上有一些基本改进。 Masc: Yes, in fact, we would very much like to have version 11.0 in 2021, but version 11.0 has many basic neural network structures that need to be rewritten, and there are some basic improvements in creating vector space. 57、Lex Fridman：也就是说FSD 11.0版本将会有很多质的飞跃，11也是一个非常酷的数字。 57, Lex Fridman: That means the FSD version 11.0 will have many qualitative leaps and 11 is a very cool number. 马斯克：FSD 11.0真的会有很多基本的改进，神经网络的变化将会带来更多性能的提升。在一开始可能会有很多问题，但这是很有益的。 Masc: FSD 11.0 will really have a lot of basic improvements, and changes in neural networks will lead to more performance improvements. There may be a lot of problems at the beginning, but it's useful. 就像我们在利用阿尔法软件那样，基本上需要处理一大堆C，C++代码，用神经网络代替部分C++代码，Andrej Karpathy非常强调这一点。这就像神经网络逐步“吃掉”传统的软件，随着时间的推移，传统的软件将会越来越少，而神经网络会越来越多。 As we do with Alpha software, there is basically a lot of C, C++ codes that we have to deal with, replacing part of C++ with a neural network, which is emphasized by Andrej Karpathy. It's like a neural network that “eats” traditional software, which will become less and fewer over time, and more and more neural networks. 最终呈现的结果仍然是软件，但更多是通过神经网络来实现的，基于启发式算法少了。这是一个很大的改进。 The end result is still software, but much more is done through neurological networks, based on less inspirational algorithms. This is a great improvement. 58、Lex Fridman：所以你想要一直增加神经网络。 58, Lex Fridman: So you want to keep increasing the neural network. 马斯克：但这甚至不是所有的神经网络，就像一个改变游戏规则的人，但是必须利用大量C、C++组成一个神经网络。 Masc: It's not even all neural networks, like a person who changes the rules of the game, but it has to be made of a large number of C, C++ neural networks. 神经网络通过少量的数据来识别：这是一条车道线；这是一个可行驶空间；这是一辆车；这是一辆自行车或其他的障碍物。神经网络必须向C、C++代码输出正确的向量。 The neural network is identified with a small amount of data: this is a driveway; this is a viable space; this is a vehicle; this is a bicycle or other barrier. The neural network must export the right vector to C, C++ code. 对于我们已经完成的，我觉得我们做得很好，使它的增长速度达到了局部的最大值，这是了不起的事情。另一方面，我们对神经网络的训练也进行了很大的调整，需对原始图像进行训练，得到最终的图像信息。 For what we've done, I think we've done well, and it's a great thing that we've been growing up to a local maximum. On the other hand, we've made great adjustments to the training of neural networks, and we need to train the original images and get the final image information. 59、Lex Fridman：所以这是一种降低复杂性的方法，降低了整个过程的复杂度。 59, Lex Fridman: So this is a way to reduce complexity and the complexity of the process. 马斯克：是的，所以代码数量也会有所减少。 60、Lex Fridman：这很迷人。所以，特斯拉目前正在通过所有传感器的加入来降低难度。 60, Lex Fridman: It's fascinating. Tesla is making it easier by joining all sensors. 马斯克：主要依托于摄像头。 Masc: Relying mainly on the camera. 61、Lex Fridman：和人类一样，但我们也通过耳朵听声音。 马斯克：是的，实际上我们还要将声音融入其中，需要听救护车的警报声或者消防车的声音，再或者有人对你大喊大叫，所以也需要一些音频。 Masc: Yes, we actually have to put our voices into it, and we need to listen to an ambulance's alarm or a fire truck, or someone yells at you, so we need some audio. 老实说，提出这些想法非常容易，但实施起来非常困难。就像提出去月球是最简单的一部分，而到达月球是最难的部分。 To be honest, it is very easy to come up with these ideas, but it is very difficult to implement them. It is like going to the moon is the easiest part and reaching the moon is the hardest part. 整个过程中在软件和硬件层面都存在很多核心的工程问题，需要对很多代码进行调整，减少延迟，如果我们不这样做，系统就不会工作。 There are many core engineering problems at both the software and hardware levels throughout the process, and many codes need to be adjusted to reduce delays, and if we do not, the system will not work. 而这就是工程师们主要的工作内容，他们就像你认识的那些无名英雄，他们对项目的成功至关重要。 And that's what engineers do, they're like the unnamed heroes you know, they're essential to the success of the project. 62、Lex Fridman：对于我来说，外面在发生什么事情，安德烈在做什么事情，这都让我超级令人兴奋。软件的基础架构等一切都在用数据引擎进行，整个过程像一个艺术品。 马斯克：它的规模之大难以令人相信。我们编写了这些所有的用于培训和标签的定制软件，并且进行标定，标定是必不可少的，因为，当你处理像环绕视频的时候很难，从头开始标记环绕视频是非常困难的。 Mask: It's incredibly large. We've developed all these customized software for training and tagging, and tagging is essential because, when you're dealing with video wrappings, it's very difficult to mark video wrappings from the top. 一个人需要花费很长时间比如几个小时来标记一个视频剪辑，或者做一个标定，而软件只需要对视频剪辑应用进行大量的计算，进行预先分配和猜测环绕视频中发生的所有事情。 It takes a long time for a person to mark a video clip, for example, for hours, or to do a tag, while the software requires a large number of calculations of the video clip application, pre-allocation and speculation about everything that happens in the video. 63、Lex Fridman：还有就是需要纠正它们。 马斯克：是的，然后人类所要做的就是调整，比如调整不正确的地方。这就像把生产力提升了100倍或更多。 Masc: Yes, and then all humans have to do is adjust, for example, to the wrong place. It's like raising productivity 100 times or more. 特斯拉人形机器人可以成为人类的伴侣？ 64、Lex Fridman：因此，你说特斯拉机器人将会在工厂有作用。我认为人形机器人对所有机器人爱好者来说都是难以置信的。我认为类人机器人、双足机器人所展示的优雅运动真是太酷了。我们已经谈论了比较类似的所有想法，其中一些你已经谈到过了的如特斯拉FSD其实只是另一种机器人问题。 64, Lex Fridman: So you say that Tesla robots will work in the factory. I think human robots are incredible to all robots. I think it's so cool to have class robots and two-foot robots showing grace. We've talked about all the more similar ideas, some of which you've already talked about, like Tesla FSD is actually just another robot. 我非常关心人类和机器人的互动能够展示出了如此人性化的一面，你认为机器人应该在工厂内工作？还是说在任何一个地方比如家里，成为朋友或者助理之类的？ I'm very concerned that the interaction between humans and robots shows such a human face. Do you think robots should work in factories or be friends or assistants anywhere like home? 马斯克：我认为可能性是无穷无尽的，这并不完全符合特斯拉加速可持续能源的主要任务方向，但制造人形机器人，是我们可以为这个世界做的一些非常有趣的事情，能够以很多形式为世界带来帮助。 Masc: I think the possibilities are endless, not exactly in line with Tesla's main mission of accelerating sustainable energy, but the creation of human robots is something very interesting that we can do for the world and can help it in many ways. 我认为未来有很多工作，如果没有报酬，人们就不会去做，比如说洗碗，你整天都在洗碗的话，就会很烦躁，就算你真的很喜欢洗碗，你愿意每天洗八个小时吗？ I think there's a lot of work to do in the future, and people don't do it without pay, like washing dishes, you'll be upset if you're washing dishes all day. Even if you really like washing dishes, would you like to do it for eight hours a day? 再比如说那些危险的工作，这些工作既危险又无聊，还存在重复性压力伤害。 For example, hazardous work, which is dangerous and boring, is subject to repeated stress. 我认为这些工作就是人形机器人能够发挥最大价值的地方，我们目标就是让机器人去做一些人类不愿意做的工作。 I think it's where human robots can be of the greatest value, and our goal is to get robots to do something that humans do not want to do. 65、Lex Fridman：你是否想象过未来世界上有数以亿计的特斯拉机器人，在世界各地做不同的工作，执行不同的任务。 Lex Fridman: Do you ever imagine hundreds of millions of Tesla robots in the world in the future, doing different jobs and carrying out different tasks around the world? 马斯克：我还没有真正思考过这个问题，但我想可能会有类似的事情发生。 Masc: I haven't really thought about it, but I think something like that might happen. 66、Lex Fridman：我可以问一个疯狂的问题吗？目前特斯拉的保有量加速增加，已经接近200万辆，他们中有很多都具备FSD（马斯克：我想我们已经超过了200万辆），你认为特斯拉机器人的数量是否会超过特斯拉汽车的数量？ 马斯克：我认为这个问题很有趣，通常情况下我会思考非常遥远的事情，但我确实没有思考过特斯拉机器人的未来。 Masc: I think that's an interesting question, and usually I think of very distant things, but I don't really think about the future of the Tesla robot. 我称这些机器人为optimus Subprime，因为他不像一个巨大的可以变形的机器人，但他应该是一个通用的帮助机器人。我认为特斯拉经常和现实世界进行互动，我们开发自动驾驶，同时定制很多硬件，硬核软件，让它们高效运行和节能。 I call these robots optimus Subprime, because he doesn't look like a giant deformed robot, but he's supposed to be a universal help robot. I think Tesla interacts with the real world, we develop autopilots, and we customize a lot of hardware, hard nuclear software, so they can run efficiently and save energy. 如果你有一个拥有10000台计算的巨大服务器机房，那么做神经网络并不复杂，但现在把它提炼出来让其在一个低功耗的人形机器人计算机或汽车上运行，这实际上非常困难，有很多复杂的软件工作要做。 If you have a huge server room with 10,000 computing units, that neural network is not complicated, but it's now refined to allow it to operate on a low-capacity robotic computer or car, which is actually very difficult, with a lot of complex software to do. 因此，我们用神经网络解决汽车导航的问题，而汽车又有点像有四个轮子的机器人，如果把它延伸放到一个有手臂和腿的机器人中，比较困难的两点是既要让机器人足够智能，又要让其以一种明智的方式和环境交互。 So, we use neural networks to solve the problem of car navigation, and cars are sort of like robots with four wheels. If we extend them into a robot with arms and legs, the two more difficult things are to make robots smart enough to interact with each other in a wise way and in the environment. 要达成这一切，既需要现实世界的AI，又需要非常擅长制造。特斯拉非常擅长制造，又有AI，所以让人形机器人工作意味着开发与汽车不同的定制电机和传感器。 To do this, it takes AI in the real world, and it needs to be very good at manufacturing. Tesla is very good at manufacturing, and AI, so working with human robots means developing custom-made machines and sensors that are different from cars. 我认为我们在开发先进的电动机和电子电力方面有最好的专业知识，都可以用于人形机器人的应用。 I think we have the best expertise in the development of advanced electric motors and electronic power, which can be used for human robotic applications. 67、Lex Fridman：有一些东西对我们很有吸引力，但不一定能够吸引到所有的人。世界上被巨大的孤独包围，很多人都想要更多的陪伴，与其他人的友谊等，在奥斯汀，有很多人养狗。似乎还有一个巨大的机会让机器人减少世界上的孤独感，帮助人们相互联系，在某种程度上狗就可以实现这个目标。你是否考虑过特斯拉机器人在这一点发挥作用？还是认为特斯拉机器人应该发挥特定的任务，不与人交往？ [strong] 67, Lex Fridman: There's something that's attractive to us, but it doesn't necessarily attract all of us. The world is surrounded by great loneliness, many want more company, friendships with others, and so on, and there's a lot of dogs in Austin. There seems to be a huge opportunity for robots to reduce the sense of loneliness in the world, help people to connect, and to some extent dogs can achieve this. Have you considered the role of Tesla robots in this? 马斯克：老实说，我并没有从陪伴的角度考虑过，但我认为它实际上最终会成为一个非常好的伴侣。它可以个性化发展，随着时间的发展，他可能会发展出一种独特的个性，而不是所有的机器人都一样。 Masc: To be honest, I didn't think about it from the perspective of companionship, but I think it's actually going to be a really good partner. It can be personalized, and as time goes on, he may develop a unique personality, not all robots. 而这种个性可能会变得和拥有者匹配，或者成为人类的同伴、另一半。 And that personality may become matched to the owner, or human companion, the other half. 我觉得这很有趣，就像一个我很喜欢的日语短语：wabi-sabi，细微的瑕疵，使某些东西变得特别，机器人个性的细微缺陷，或许会让它成为一个不可思议的伙伴，基本上就像r2d2、c-3po（《星球大战》电影中的两个机器人）之类的东西。 I think it's funny, like a Japanese phrase that I really like: wabi-sabi, a tiny flaw that makes something special, a tiny flaw in the robotic personality, and maybe it makes it an incredible partner, basically like r2d2, c-3po. 68、Lex Fridman：从机器人的角度来看，我认为有一些缺陷的特性非常好，在一般的家庭环境中，有一些缺陷非常可爱，你可能会爱上这些缺陷。但这与自动驾驶非常不同，自动驾驶在一个高风险的环境中，你不能搞砸。因此在家用机器人更有趣。 68, Lex Fridman: From a robotic point of view, I think some of the defects are very good, and some of the defects in the general family environment are cute and you may fall in love with them. This is very different from autopilot, which in a high-risk environment you can't screw up. So it's more fun to use robots at home. 马斯克：是的，事实上，如果你想象r2d2、c-3po，他们就有很多类似的缺陷和不完美的地方，他们会为了愚蠢的事情互相争论。 Masc: Yes, in fact, if you imagine r2d2, c-3po, they have a lot of similar defects and imperfections, and they argue about stupid things. 69、Lex Fridman：他们真的擅长做任何事情吗？我不确定。 69, Lex Fridman: Are they really good at anything? I'm not sure I'm sure . >. 马斯克：但他们肯定为故事增加了一些古怪的元素，他们会犯错误。这只是让人们对他们的可爱产生共鸣。所以我认为特斯拉机器人也可能会发生这样的事。 Masc: They certainly add some weird elements to the story, and they make mistakes. It's just resonance with their love. So I think Tesla robots could have done the same thing. 我们有信心制造出来他们，但是我不确定确切的时间表是什么，我们可能会在2022年底或类似的时间拥有一个原型。 We are confident that they will be created, but I am not sure what the exact timetable is, and we may have a prototype by the end of 2022 or something like that. 70、Lex Fridman：它可以与特斯拉汽车相连，这一定非常酷。 70, Lex Fridman: It can be connected to Tesla, which is so cool. 马斯克：是的，它将使用自动驾驶推理计算机 ，我们为了Model S/Y/3/X这四款车做了很多训练，在识别现实世界事物方面的技术可以直接给机器人用。不过也有很多需要定制的执行器和传感器有待开发。 MASK: Yes, it will use autopilot reasoning computers, we've trained a lot for four cars like Model S/Y/3/X, and we can use technology directly for robotics in identifying the real world. But there are many executors and sensors that need to be customized. 71、Lex Fridman：在向量空间之外，还要有一个额外的模块用于爱。（马斯克：我们也可以用于汽车），在所有环境中都可以用，很多人会在汽车内吵架，汽车也可以帮助人们。 71, Lex Fridman: Beyond vector space, there's an extra module for love. (Mask: We can also use cars) It's available in all environments, lots of people fight in cars and cars can help people. 你是历史系的学生，也是丹·卡林的粉丝（马斯克：是的，有史以来最棒的博客），他几乎不算是播客，（马斯克：丹·卡林更像是一本有声读物），我刚和他聊了聊，他说你们讨论军事之类的东西。 You're a history student and a fan of Dan Kälin. He's hardly a podcast. 马斯克：是的，我觉得应该被叫做工程师战争，从本质上讲，当技术上发生快速变化的时候，工程在战斗的胜利中起着关键的作用。 Mask: Yes, I think it should be called the war of engineers, and in essence, when technology changes rapidly, engineering plays a key role in winning the battle. 火箭部分灵感来自俄罗斯 rockets inspired in part by Russian & nbsp; 年轻人应多尝试 Young people should try more 72、Lex Fridman：你对二战很了解吗？ 马斯克：我们对二战中的战斗机和轰炸机技术进行过一次深入的探讨，但最终的讨论范围更广。我刚刚进行了一个研究，研究二战中所有的战斗机和轰炸机，以及不断的剪刀石头布的游戏。 Masc: We've had an in-depth, but ultimately broader, discussion of fighter and bomber technology in World War II. I've just done a study of all fighter and bomber planes in World War II, as well as an ongoing game of scissors and stones. 一个国家可以制造一架飞机，但另一个国家也会制造一架飞机打败上一个国家的飞机，然后第一个国家就会再制造一架飞机来打败上一架，真正重要的是创新的速度，以及获得高质量燃料和原材料的机会。 One country could build an aircraft, but another country would also build an aircraft to defeat an aircraft from the previous country, and then the first country would build another aircraft to defeat the previous one. What was really important was the speed of innovation and access to high-quality fuel and raw materials. 因此，德国虽然有一些令人惊叹的设计，但他们无法制造他们，因为他们无法获得原材料，他们的石油和燃料存在问题，基本上燃料的质量非常不稳定。 Thus, while Germany has some amazing designs, they are unable to manufacture them because they do not have access to raw materials, their oil and fuel are problematic and the quality of fuel is essentially very unstable. （Lex Fridman：所以设计不是瓶颈）是的，美国有非常稳定的燃料，质量非常一致。如果你想制造一个高性能的飞机发动机，那么燃料必须是一致的混合物，必须有高辛烷值，也不能有杂质之类的，不然会弄脏发动机。 Yes, the U.S. has a very stable fuel, very consistent in quality. If you want to build a high-performance aircraft engine, the fuel must be a consistent mixture, it must have a high-octanane value, it can't have impurities, or it'll get dirty. 但是德国人从来没有很好的获得石油，他们试图通过入侵高加索地区来得到它，但这并没有奏效。德国一直在和劣质油作斗争，他们不能指望为他们的飞机提供高质量的燃料，所以他们必须研发一些添加剂之类的东西。 But the Germans never got oil well, and they tried to get it by invading the Caucasus, but it didn't work. Germany has been fighting bad oil, they can't expect high-quality fuel for their planes, so they have to develop additives and stuff like that. 而美国有很棒的燃料，也提供给英国，这使得英国人和美国人能够设计出超高性能的飞机发动机。德国可以设计发动机，他们只是没有燃料，然后他们得到的铝合金的质量也没有那么好。 And the U.S. has a great fuel, and it's also provided to the UK, which enables British and American people to design super-performance jet engines. Germany can design engines, they just don't have fuel, and the aluminum alloys they get aren't that good. 73、Lex Fridman：你和丹谈论过这些？ 马斯克：是的。 MASK: Yes. 74、Lex Fridman：从广义上看历史，当你看到成吉思汗、斯大林、希特勒，你从这些时刻中学到了什么？是否可以帮助你深入了解人性？了解当今人类行为，无论战争还是个人或者人类的行为，历史的任何方面。 74, Lex Fridman: In the broad sense of history, what do you learn from these moments when you see Genghis Khan, Stalin, Hitler? Can you help you gain a deeper understanding of humanity? 马斯克：是的，我觉得历史很迷人，有很多不可思议的事情发生，它们帮助你了解文明和个人的本质。 Masc: Yes, I think history is fascinating, and there are a lot of incredible things happening that help you understand the essence of civilization and individuals. 75、Lex Fridman：人类对彼此做残酷的事情，会让你难过吗？比如二战期间，滥用权力的残酷。 马斯克：我认为其中一些事情，大部分人类只是相处和生活，战争、灾难其实只是间歇性和罕见的，如果不是，那么人类将很快就会不复存在。 Masc: I think some of these things, most of humanity is just living and living together, and wars, catastrophes, are just intermittent and rare, and if not, humankind will soon cease to exist. 而战争往往会被历史学家大肆书写，而没什么大事发生的正常年份不会被写太多。大多数人只喜欢务农，过他们喜欢的生活，在某个地方做一个村民。 Wars are often written by historians, and nothing happens in a normal year. Most people only like farming, living the life they like, and being a villager somewhere. 76、Lex Fridman：蒂姆·多德（火箭科学传播者）制作了一个火箭树，完整展示了苏联火箭的历史，蒂姆对未来，对太空飞行超级感兴趣。 76, Lex Fridman: Tim Dodd has created a rocket tree that presents the entire history of Soviet rockets, and Tim is super interested in the future and in space flight. 马斯克：我喜欢这样的人，如果你对空间之类的事情感兴趣就太好了，在向普通人解释火箭技术方面他很棒。猛禽曾经的发动机是氢发动机，但氢有很多挑战。它的密度非常低，是一种冷冻剂，所以它一直是液体，非常接近零度，还需要做绝缘，所以面临很多挑战。 Masc: I like people like that, if you're interested in things like space, he's great at explaining rocket technology to ordinary people. Raptor used to have a hydrogen engine, but hydrogen has many challenges. Its density is very low, it's a refrigerant, so it's always liquid, it's very close to zero, it needs to be insulated, and it's a lot of challenges. 实际上我正在阅读一些有关俄罗斯火箭开放的一些信息，我的印象里苏联、俄罗斯、乌克兰在转向甲基氧，做了一些有趣的测试和数据 ，那确实令人印象深刻。 I'm actually reading some information about the opening up of Russian rockets, and it's really impressive that I remember the Soviet Union, Russia, Ukraine turning to methyl oxygen, doing some interesting tests and data. 实际上，我们可以优化入轨的成本，降低到火星的成本，我认为可以选择甲烷，部分灵感就是来自俄罗斯。 In fact, we can optimize the cost of getting in orbit and reduce it to Mars, and I think we can choose methane, partly inspired by Russia. 77、Lex Fridman：我们的年轻人想要做一些大事的话，你会给他们什么建议？ 77, Lex Fridman: What would you advise our young people if they wanted to do something important? 马斯克：尝试做一些有用的事，做对你同胞、对世界有用的事情。随时思考你贡献的比你消耗的多吗？要努力为社会作出积极的贡献，我认为这就是目标。 Masc: Try to do something useful to your compatriots and to the world. Do you think about your contribution at any time more than you consume? To try to make a positive contribution to society, I think that's the goal. 不要为了成为领导者或者其他什么而试图成为领导，很多时候，成为领导者的人却是不想成为领导者的人。 Instead of trying to be a leader or something in order to be a leader, in many cases those who are leaders do not want to be leaders. 如果你过着有用的生活，那就是美好的生活，值得过的生活。就像我说的，我会鼓励人们使用物理和心理工具，并将它们运用到生活中，这是最好的工具。 If you live a useful life, it's a good life, a life worth living. As I say, I encourage people to use physical and psychological tools and apply them to life, which is the best tool. 78、Lex Fridman：当你想到教育和自我教育时，你有什么建议？如何变得有用，产生更多积极影响？ 78, Lex Fridman: What do you suggest when you think about education and self-education? How can it become useful and have a more positive impact? 马斯克：我鼓励人们阅读大量书籍，尽可能多的吸收信息。你至少对知识景观有一个大致的了解，尝试了解很多东西，因为你可能不知道自己感兴趣的是什么。广泛的探索，和不同的人交谈，了解各行各业，喜欢什么职业就去做。 Masc: I encourage people to read a lot of books and as much information as possible. You have at least a general understanding of the knowledge landscape, and you try to understand a lot because you may not know what you're interested in. Extensive exploration, talking to different people, knowing the various professions, doing what you like. 尽可能的多学习，寻找意义。多阅读和尝试，找到和你的才能相匹配的事情去做。 Learn as much as you can and find meaning. Read more and try and find something that matches your talent.
注册有任何问题请添加 微信：MVIP619 拉你进入群
打开微信扫一扫
添加客服
进入交流群
发表评论