本集简介
双语字幕
仅展示文本字幕，不包含中文音频；想边听边看，请使用 Bayt 播客 App。
以下是与乔治·豪茨的对话。
The following is a conversation with George Hautz.
他是Kama AI的创始人，这是一家基于机器学习的汽车自动化公司。
He's the founder of Kama AI, a machine learning based vehicle automation company.
他无疑是人工智能和整个科技领域直言不讳的代表人物。
He is most certainly an outspoken personality in the field of AI and technology in general.
他最初因成为首个携带并解锁iPhone的人而获得认可。
He first gained recognition for being the first person to carry or unlock an iPhone.
自那以后，他在硬件与软件结合领域做了不少有趣的事情。
And since then, he's done quite a few interesting things at the intersection of hardware and software.
这里是人工智能播客。
This is the artificial intelligence podcast.
如果你喜欢这个节目，请在YouTube订阅、在iTunes打五星好评、在Patreon支持我们，或者直接在Twitter上联系我@lexfreedman（拼写为f-r-I-d-m-a-n）。
If you enjoy it, subscribe on YouTube, give it five stars on iTunes, support it on Patreon, or simply connect with me on Twitter at lex freedman, spelled f r I d m a n.
我要特别感谢来自加拿大的Jennifer对播客在Patreon上的支持。
And I'd like to give a special thank you to Jennifer from Canada for her support of the podcast on Patreon.
詹妮弗。
Jennifer.
她是我研究生时期就结识多年的好友兼工程学同事。
She's been a friend and an engineering colleague for many years since I was in grad school.
你的支持意义重大，激励我持续制作这个系列节目。
Your support means a lot and inspires me to keep this series going.
现在请听我与乔治·霍茨的对话。
And now here's my conversation with George Hotts.
你认为我们生活在模拟世界中吗？
Do you think we're living in a simulation?
是的，但这可能是个无法证伪的命题。
Yes, but it may be unfalsifiable.
你说的'无法证伪'是什么意思？
What do you mean by unfalsifiable?
如果模拟系统的设计使得内外信息绝对隔离，且其硬件始终维持着模拟环境的运行规范，那么我们可能永远无法证实自己是否身处模拟之中。
So if the simulation is designed in such a way that they did like a formal proof to show that no information can get in and out, and if their hardware is designed to for the anything in the simulation to always keep the hardware in spec, it may be impossible to prove whether we're in a simulation or not.
所以他们将其设计成一个封闭系统，你无法突破这个系统。
So they've designed it such that it's a closed system, you can't get outside the system.
嗯，也许这是三种世界之一。
Well, maybe it's one of three worlds.
我们要么身处一个可被利用的模拟中，要么身处一个不仅无法被利用、而且虚拟机也遵循同样规则的模拟中。
We're either in a simulation which can be exploited, we're in a simulation which not only can't be exploited, but the same thing's true about VMs.
一个设计精良的虚拟机，你甚至无法检测自己是否在虚拟机中。
A really well designed VM, you can't even detect if you're in a VM or not.
这太妙了。
That's brilliant.
所以这个模拟是在虚拟机上运行的。
So the simulation's running on virtual machine.
是的，但在现实中，所有虚拟机都有检测方法。
Yeah, but now in reality, all VMs have ways to detect.
这正是关键所在。
That's the point.
我是说，你自己也做过不少黑客工作，应该知道任何复杂系统都会有进出的方法。
I mean, you've done quite a bit of hacking yourself, and so you should know that really any complicated system will have ways in and out.
所以这未必在未来也成立。
So this isn't necessarily true going forward.
我离开逗号公司期间，学习了COC。
I spent my time away from comma, I learned COC.
这是一种依赖类型语言，专门用于编写数学证明。
It's a dependently typed It's a language for writing math proofs in.
如果你用这类语言编写能通过编译的代码，那它在定义上就是正确的。
And if you write code that compiles in a language like that, it is correct by definition.
类型系统会验证其正确性。
The types check its correctness.
所以这个模拟系统可能是用类似语言编写的，如果是这样的话
So it's possible that the simulation is written in a language like this, in which case
是啊，但那种语言不可能具备足够的表现力。
Yeah, but that can't be sufficiently expressive a language like that.
哦，它可以的。
Oh, it can.
可以吗？
It can be?
哦，当然。
Oh, yeah.
好吧。
Okay.
嗯，所有
Well, all
没错，所以如果模拟有预定结束日期，它不必是图灵完备的。
right, so The simulation doesn't have to be Turing complete if it has a scheduled end date.
看起来实际上通过熵确实如此。
Looks like it does actually with Entropy.
我是说，我不认为一个能产生像宇宙这样复杂事物的模拟会有形式化的正确性证明。
I mean, I don't think that a simulation that results in something as complicated as the universe would have a formal proof of correctness.
对吧？
Right?
当然是有可能的。
It is possible, of course.
我们完全不了解他们的工具水平，也不清楚宇宙计算机到底有多复杂。
We have no idea how good their tooling is, we have no idea how complicated the universe computer really is.
它可能相当简单。
It may be quite simple.
只是规模非常庞大。
It's just very large.
对吧？
Right?
它确实非常非常庞大。
It's very it's definitely very large.
但基本规则可能超级简单。
But the fundamental rules might be super simple.
是的。
Yeah.
康威那种会撒谎的东西。
Conway's gonna lie kind of stuff.
对。
Right.
所以
So
如果你能入侵一个可被破解的模拟系统，你会如何改变它？你会采取什么方式来入侵这个模拟？
if you could hack so imagine simulation that is hackable, if you could hack it, what would you change about the like, how would you approach hacking a simulation?
我做那个演讲的原因
The reason I gave that talk
顺便说，我不太熟悉你做的那个演讲。
By the way, I'm not familiar with the talk you gave.
我只是看到你讲过关于逃离模拟之类的内容。
I just read that you talked about escaping the simulation or something like that.
也许你可以跟我讲讲那个主题和其中的信息。
So maybe you can tell me a little bit about the theme and the message there too.
那并不是一个关于如何实际逃离模拟世界的实用性演讲。
It wasn't a very practical talk about how to actually escape a simulation.
更多是关于重构'我们对抗他们'这种叙事方式。
It was more about a way of restructuring an us versus them narrative.
如果我们继续沿着当前的技术路线走下去。
If we continue on the path we're going with technology.
我认为我们会陷入大麻烦，不仅是作为物种，甚至作为这个物种中的个体成员。
I think we're in big trouble, like, as a species and not just as a species, but even as me as an individual member of the species.
所以如果我们能改变修辞方式，更多地向上思考，比如思考我们身处模拟世界以及如何逃脱。
So if we could change rhetoric to be more like to think upwards, like to think about that we're in a simulation and how we could get out.
这样我们就已经走在正确的道路上了。
Already we'd be on the right path.
当你真正做到之后要做什么，嗯，我想在这个过程中我会获得更多的智慧。
What you actually do once you do that, well, I assume I would have acquired way more intelligence in the process of doing that.
那我就直接问了。
So I'll just ask that.
那么这种向上思考的方式，你认为能激发什么样的突破性想法？
So the thinking upwards, what kind of ideas what kind of breakthrough ideas do you think thinking in that way could inspire?
你为什么说‘向上’？
And why did you say upwards?
向上。
Upwards.
指向太空吗？
Into space?
你是指各种形式的探索吗？
Are you thinking sort of exploration in all forms?
对现代主义一代人有效的太空叙事，在后现代一代人身上就不那么适用了。
Space narrative that held for the modernist generation doesn't hold as well for the postmodern generation.
什么是太空叙事？
What's the space narrative?
我们说的是同一个空间吗？
Are we talking about the same space?
三维空间吗？
The three-dimensional space?
就是，向上的那个空间。
Like, going up space.
就像，埃隆·马斯克那样搞建设。
Like, building like, Elon Musk.
我们要造火箭。
We're gonna build rockets.
我们要去火星。
We're gonna go to Mars.
我们要殖民宇宙。
We're gonna colonize the universe.
你提到的叙事——我出生在苏联。
And the narrative you're referring I was born in The Soviet Union.
你指的是太空竞赛。
You're referring to the race to space.
太空竞赛。
The race to space.
是的。
Yes.
探险者。
Explorer.
好的。
Okay.
那是个伟大的现代主义叙事。
That was a great modernist narrative.
是啊。
Yeah.
在当今文化中，它似乎不再具有同样的分量了。
It doesn't seem to hold the same weight in today's culture.
我希望能有好的后现代叙事来取代它。
I'm hoping for good postmodern narratives that replace it.
让我们想想，你经常和人工智能打交道。
So let's think, so you work a lot with AI.
所以人工智能是那种叙事的一种表现形式。
So AI is one formulation of that narrative.
可能还有...我不太清楚你在虚拟现实和增强现实方面做多少。
There could be also, I don't know how much you do in VR and AR.
是的。
Yeah.
那是另一个领域，我了解不多，但每次在研究中接触时，那个虚拟世界都令人着迷。
That's another I know less about it, but every time I play with it in our research, it's fascinating, that virtual world.
你对虚拟世界感兴趣吗？
Are you interested in the virtual world?
我想搬到虚拟现实里去生活。
I would like to move to virtual reality.
是指你的工作方面吗？
In terms of your work?
不是。
No.
我是想实际搬到那里去住。
I would like to physically move there.
我在云端能租到的公寓比现实世界里能租的好太多了。
The apartment I can rent in the cloud is way better than the apartment I can rent in the real world.
嗯，这都是相对的，不是吗？
Well, it's all relative, isn't it?
因为别人也会拥有很棒的公寓，所以你在虚拟世界里还是会处于劣势
Because others will have very nice apartments too, so you'll be inferior in the virtual world
但这不是我看待世界的方式。
as But that's not how I view the world.
对吧？
Right?
我并不是那样看待世界的，那几乎是一种零和博弈的世界观。
I don't view the world I mean, it's a very like almost zero sum ish way to view the world.
比如说，我的豪华公寓之所以豪华，并不是因为邻居也有一套。
Say like, my great apartment isn't great because my neighbor has one too.
不是的。
No.
我的公寓之所以豪华，是因为——你看这个洗碗机，老兄。
My great apartment is great because like, look at this dishwasher, man.
你只要碰一下盘子，它就自动洗干净了。
You just touch the dish and it's washed.
对吧？
Right?
这种美好是独立存在的，无论我是唯一拥有这种公寓的人，还是大家都拥有。
And that is great in and of itself if I have the only apartment or if everybody had the apartment.
我根本不在乎。
I don't care.
所以你拥有根本性的感恩之心。
So you have fundamental gratitude.
世界首次认识乔·哈特（乔治·哈茨）是在2007年8月，或许更早，但肯定是在2007年8月，当时你成为首个破解并携带解锁iPhone的人。
The world first learned of Joe Hart, George Hartz in August 2007, maybe before then, but certainly in August 2007 when you were the first person to unlock, carry unlock an iPhone.
你是怎么进入黑客领域的？
How did you get into hacking?
你发现漏洞并入侵的第一个系统是什么？
What was the first system you discovered vulnerabilities for and broke into?
那确实算是第一件事了。
So that was really kind of the first thing.
我在2006年买了本叫《灰帽黑客》的书。
I book had in 2006 called Greyhat Hacking.
我想我当时意识到，如果获得这类能力，就能控制世界。
And I guess I realized that if you acquired these sort of powers, could control the world.
但那时候我对计算机其实了解不多。
But I didn't really know that much about computers back then.
我是从电子设备入门的。
I started with electronics.
第一次破解iPhone是物理层面的。
The first iPhone hack was physical.
卡带破解。
Cardware.
你得拆开它，把地址线拉高。
You had to open it up and pull an address line high.
这是因为我当时并不懂软件漏洞利用。
And it was because I didn't really know about software exploitation.
我在随后几年学会了这些，并且变得非常擅长。
I learned that all in the next few years, and I got very good at it.
但那时候，我只知道内存芯片怎么连接处理器这些东西。
But back then, I knew about like how memory chips are connected to processors and stuff.
但你懂软件和编程啊。
Well, you knew about software and programming.
你只是不知道而已。
You just didn't know.
哦，真的吗？
Oh, really?
所以你对世界和计算机的认知是物理层面的，是硬件方面的？
So your view of the world and computers was physical, was hardware?
实际上，如果你看我2007年8月发布的那段代码，简直惨不忍睹。
Actually, if you read the code that I released with that in August 2007, it's atrocious.
是用什么语言写的？
What language was it?
C语言。
C.
C？
C?
不错。
Nice.
而且是用一种支离破碎的状态机风格的C语言写的，我当时根本不会编程。
And in a broken sort of state machine esque C, I didn't know how to program.
是啊。
Yeah.
那你是怎么学会编程的？
So how did you learn to program?
你的学习历程是怎样的？
What was your journey?
因为我是说，我们可以聊聊这个。
Because mean, we'll talk about it.
你还直播过一些编程过程。
You've livestreamed some of your programming.
没错。
Yeah.
这种混乱而美妙的局面，你是怎么达到的？
This chaotic, beautiful mess, how did you arrive at that?
多年不断的练习。
Years and years of practice.
我在iPhone解锁后的那个夏天去了谷歌实习。
I interned at Google after the summer after the iPhone unlock.
我为他们完成了一个合同项目，为街景搭建硬件设备，并编写了与之交互的软件库。
And I did a contract for them where I built a hardware for Street View, and I wrote a software library to interact with it.
那些代码简直糟糕透顶。
And it was terrible code.
生平第一次，我收到了来自我敬重之人的反馈：'别再这样写代码了'。
And for the first time I got feedback from people who I respected saying, Don't write code like this.
当然，仅仅得到反馈是不够的。
Now, of course, just getting that feedback is not enough.
真正让我进步的是，我想编写一个能模拟并可视化ARM二进制文件的工具。
The way that I really got good was I wanted to write this thing that could emulate and then visualize ARM binaries.
因为我想更好地破解iPhone。
Because I wanted to hack the iPhone better.
我不满意的是无法看到处理器内部状态，也无法单步执行，因为上面没有调试器，特别是对于引导RAM和引导程序这样的底层操作。
And I didn't like that I couldn't see what the I couldn't single step through the processor because I had no debugger on there, especially for the low level things like the boot RAM and the bootloader.
所以我尝试开发这个工具来实现它。
So I tried to build this tool to do it.
我第一次构建这个工具时，效果很糟糕。
And I built the tool once and it was terrible.
我第二次构建这个工具，依然很糟糕。
I built the tool second times, it was terrible.
我第三次构建了这个工具。
I built the tool a third time.
这次是在Facebook工作期间，工具勉强能用。
This by the time I was at Facebook, it was kind of okay.
然后2014年我再次以谷歌实习生身份第四次构建这个工具时，终于觉得它可用了。
And then I built the tool a fourth time when I was a Google intern again in 2014, and that was the first time I was like, this is finally usable.
这个怎么发音，Kira？
How do you pronounce this, Kira?
Kira，是的。
Kira, yeah.
所以这本质上是以最有效的方式可视化程序运行时计算机状态的变化，这就是你所说的调试器。
So it's essentially the most efficient way to visualize the change of state of the computer as the program is running, that's what you mean by debugger.
是的，它是一个无时间限制的调试器，你可以像前进一样轻松地回退。
Yeah, it's a timeless debugger, so you can rewind just as easily as going forward.
想想看，如果你使用GDB，必须对变量设置监视点才能观察变量变化。
Think about it, if you're using GDB, have to put a watch on a variable, if you want to see if that variable changes.
在Cura中，只需点击那个变量，它就会显示该变量每次被修改或访问的所有时间点。
In Cura, can just click on that variable and then it shows every single time when that variable was changed or accessed.
可以把它想象成计算机运行日志的Git版本控制。
Think about it like Git for your computer's run log.
所以就像程序运行时计算机状态的深度日志记录，而且你可以回退查看。
So there's like a deep log of the state of the computer as the program runs and you can rewind.
为什么这种调试方式不常见呢？也许确实存在，能给我解释一下原因吗？
Why isn't that, maybe it is, maybe you can educate me, why isn't that kind of debugging used more often?
因为工具太差劲了。
Because the tooling's bad.
嗯，有两个原因。
Well, two things.
第一，如果你想调试Chrome浏览器，Chrome是一个200MB大小的二进制文件，在桌面设备上运行都很慢。
One, if you're trying to debug Chrome, Chrome is a 200 megabyte binary that runs slowly on desktops.
所以用它来调试会非常困难。
So that's going to be really hard to use for that.
但它非常适合用于CTF比赛和引导ROM以及小型代码片段。
But it's really good to use for like CTFs and for boot ROMs and for small parts of code.
因此调试大型系统会很困难。
So it's hard if you're trying to debug massive systems.
什么是CTF？什么是引导ROM？
What's a CTF and what's a boot ROM?
引导ROM就是iPhone通电后立即执行的第一段代码。
A boot ROM is the first code that executes the minute you give power to your iPhone.
好的。
Okay.
CTF是我参加过的一些竞赛，全称是夺旗赛。
And CTF were these competitions that I played, Capture the Flag.
夺旗赛啊，我正想问你关于这个呢。
Capture the Flag, I was going to ask you about that.
那是什么比赛？
What are those?
我在YouTube上看过几个相关视频，看起来非常有意思。
I watched a couple of videos on YouTube, those look fascinating.
从这些比赛中，你对系统漏洞的高层认知有哪些收获？
What have you learned about maybe at the high level of vulnerability of systems from these competitions?
我觉得在CTF的黄金时期，汇集了全世界最顶尖的安全专家互相挑战，他们不断设计出新的可攻击玩具模型，然后大家就开始比拼谁能攻破它。
I feel like in the heyday of CTFs, you had all of the best security people in the world challenging each other, and coming up with new toy exploitable things over here, and then everybody, okay, who can break it?
成功攻破后，服务器里会有一个名为flag的文件作为凭证。
And when break it, you get like there's like a file in the server called flag.
然后有一个程序在运行，监听一个存在漏洞的套接字。
And then there's a program running, listening on a socket that's vulnerable.
所以你编写一个漏洞利用程序，获得一个shell，然后执行cat flag命令，再把flag输入到基于网页的记分板上，就能得分。
So you write an exploit, you get a shell, and then you cat flag, and then you type the flag into, like, a web based scoreboard, and you get points.
所以目标本质上是在系统中找到一个漏洞，让你能运行shell来执行任意代码。
So the goal is essentially to find an exploit in a system that allows you to run shell to run arbitrary code on that system.
这是其中一类挑战。
That's one of the categories.
这就像是可攻破类别的挑战。
That's like the ponable category.
可攻破？
Ponable?
对，可攻破。
Yeah, ponable.
就像你知道的，你攻破了那个程序。
It's like, you know, you pwn the program.
这是一个程序。
It's a program.
是啊。
Yeah.
首先我要道歉，我可能会说这是因为我是俄罗斯人，但也许你能帮我科普一下。
You know, first of all, I apologize, I'm going to say it's because I'm Russian, but maybe you can help educate me.
就像某些电子游戏早年把'own'拼错那样。
Some video game like misspelled own way back in the day.
对。
Yeah.
我在想是否有官方定义，可能得去'Urban Dictionary'查查。
And there's just, I wonder if there's a definition, I'll have to go to Urban Dictionary for it.
没错，应该
Yeah, it'd be
看看它会怎么解释会很有趣。
interesting to see what it says.
好的，顺便问一下，CTF的鼎盛时期是什么时候？
Okay, so what was the heyday of CTF, by the way?
我们说的是哪个十年？
What decade are we talking about?
我觉得，可能因为我玩的那个时代让我有偏见。
I think like, I mean, maybe I'm biased because it's the era I played.
大概是2011到2015年，因为现代CTF比赛和现代编程竞赛很像。
But like 2011 to 2015, because the modern CTF scene is similar to the modern competitive programming scene.
有些人喜欢做训练。
You have people who like do drills.
有些人会进行练习。
You have people who practice.
当你完成这些后，它就从通用的计算机技能游戏，变成了针对这五个类别的专项训练。
And then once you've done that, you've turned it less into a game of generic computer skill and more into a game of, okay, you drill on these five categories.
在那之前，它并没有像后来那样受到那么多关注。
And then before that, it didn't have like as much attention as it had.
我不知道。
I don't know.
他们就像，我曾经在韩国的一次比赛中赢得了3万美元。
They were like, I won $30,000 once in Korea for one of these competitions.
天啊。
Holy crap.
是啊。
Yeah.
他们...他们...我是说，钱就是钱，但这说明那里可能有很多高手。
They were they were that means I mean, money is money, but that means there was probably good people there.
没错。
Exactly.
是的。
Yeah.
这些挑战是人设计的，还是基于真实系统中的某些漏洞？
Are the challenges human constructed, or are they grounded in some real flaws in real systems?
通常都是人为设计的，但通常会受到真实漏洞的启发。
Usually, they're human constructed, but they're usually inspired by real flaws.
什么样的系统，我猜主要是针对移动端的？
What kind of systems, I imagine, is really focused on mobile?
比如，现在存在漏洞的？
Like, has vulnerabilities these days?
主要是像安卓这样的移动系统吗？
Is it primarily mobile systems like Android?
哦，所有系统都还存在漏洞。
Oh, everything does still.
是的，当然。
Yeah, of course.
价格某种程度上涨了，因为能发现漏洞的人越来越少，而且安全领域现在的状况是，如果你想越狱一部iPhone，不再需要一个漏洞利用，而是需要九个。
The price has kind of gone up because less and less people can find them, and what's happened in security is now if you want to like jailbreak an iPhone, you don't need one Xplo anymore, you need nine.
九个串联起来？
Nine chained together?
是啊，怎么了？
What do Yeah.
你还好吧？
You Okay.
那么从更高层次的哲学角度来说，黑客行为究竟有什么好处？
So it's really What's benefit, speaking higher level philosophically about hacking?
我的意思是，从我对你的了解来看，你只是热爱挑战，并不想真正做什么，不想把那些漏洞公之于众任其肆虐。
I mean, sounds from everything I've seen about you, you just love the challenge, and you don't want to do anything, you don't want to bring that exploit out into the world and do any actual, let it run wild.
你只想解决问题，然后继续下一件事。
You just wanna solve it, and then you go on to the next thing.
哦，没错。
Oh, yeah.
我是说，干违法的事真的不值得。
I mean, doing criminal stuff's not really worth it.
实际上，我不做犯罪活动的理由和我不做防御工作的理由是一样的。
And I'll actually use the same argument for why I don't do defense for why I don't do crime.
如果你想防御一个系统，假设系统有10个漏洞。
If you want to defend a system, say the system has 10 holes.
对吧？
Right?
作为防御者，即使你发现了其中9个漏洞，只要攻击者从最后一个漏洞攻入，你仍然会输。
If you find nine of those holes as a defender, you still lose because the attacker gets in through the last one.
但作为攻击者，你只需要找到10个漏洞中的1个。
If you're an attacker, you only have to find one out of the 10.
但如果你是罪犯，即使你10次中有9次使用VPN登录，只要有一次忘记，你就完了。
But if you're a criminal, if you log on with a VPN nine out of the 10 times, but one time you forget, you're done.
因为你被抓了，好吧。
Because you're caught Okay.
因为作为罪犯，你只要失手一次就会被抓住？
Because you only have to mess up once to be caught as a criminal?
是的。
Yeah.
这就是为什么我不是罪犯。
That's why I'm not a criminal.
不过好吧，让我说说，因为我刚和某人进行了一场高层次讨论，实际上是关于核武器的，为什么我们还没把自己炸飞。
But okay, let me, because I was having a discussion with somebody just at a high level about nuclear weapons actually, why we haven't blown ourselves up yet.
我的感觉是，世界上所有聪明人，如果你看看聪明人的分布，聪明人通常都是好人。
And my feeling is all the smart people in the world, if you look at the distribution of smart people, smart people are generally good.
然后另一个人在和物理学家肖恩·卡罗尔交谈时，他说，不，好人和坏人是均匀分布在所有人中的。
And then this other person, was talking to Sean Carroll, the physicist, and he was saying, no, good and bad people are evenly distributed amongst everybody.
我的直觉是，好的黑客大体上都是好人，他们不想搞乱世界。
My sense was good hackers are in general good people, and they don't want to mess with the world.
你怎么看？
What's your sense?
我甚至都不确定这一点。
I'm not even sure about that.
比如，我过着不错的生活。
Like, I have a nice life.
犯罪不会给我带来任何好处。
Crime wouldn't get me anything.
但如果你品行端正又拥有这些技能，你可能也会过得不错。
But if you're good and you have these skills, you probably have a nice life too.
对吧？
Right?
对。
Right.
可以用于其他方面。
Can use it for other things.
是否存在一种道德观念，你脑海里是否有个微弱的声音在说：'是啊，如果你能黑进某个系统去伤害别人，虽然可能不是身体上的伤害，但能以某种方式扰乱他们的生活，同时还能赚大钱'
Is there an ethical, is there a little voice in your head that says, well, yeah, if you could hack something to where you could hurt people, and you could earn a lot of money doing it though, not hurt physically perhaps, but disrupt their life in some kind of way.
难道没有这样一个微弱的声音在说
Isn't there a little voice that says
嗯，有两方面。
Well, two things.
第一，我真的不在乎钱。
One, I don't really care about money.
所以金钱不会成为我的动力。
So like the money wouldn't be an incentive.
刺激感或许会是个诱因。
The thrill might be an incentive.
但在我19岁时，我读了《罪与罚》。
But when I was 19, I read Crime and Punishment.
没错。
Right.
很好。
Good.
这本书是另一个让我彻底打消犯罪念头的重要因素。
That was another great one that talked me out of ever really doing crime.
因为感觉就像——那将会是我的下场。
Because it's like, that's gonna be me.
我可能会逍遥法外，但这件事会一直萦绕在我心头。
I'd get away with it, but it would just run through my head.
即使我能逃脱惩罚，你懂吗？
Even if I got away with it, you know?
犯罪时间够长的话，你终将难逃法网。
And then you do crime for long enough, you'll never get away with it.
说得对。
That's right.
归根结底，这是选择向善的好理由。
In the end, that's a good reason to be good.
我不会说自己是个好人。
I wouldn't say I'm good.
我只敢说自己不算坏人。
I would just say I'm not bad.
你是个天才程序员，也是这个词积极意义上的黑客。
You're a talented programmer and a hacker in a good positive sense of the word word.
你曾经尝试过，发现了各种系统中的漏洞。
You've played around, found vulnerabilities in various systems.
从整个过程中，你对系统设计等方面有了哪些广泛的认识？
What have you learned broadly about the design of systems and so on from that from that whole process?
你
You
学会不轻信人们表面上的说法，而是看清事物的本质。
learn to not take things for what people say they are, but you look at things for what they actually are.
是啊。
Yeah.
我明白你告诉我的内容，但它实际
I understand that's what you tell me it is, but what does
上能做什么？
it do?
没错。
Right.
而且你们有很棒的可视化工具，能真正了解它在做什么。
And you have nice visualization tools to really know what it's really doing.
哦，希望我现在比2014年时更擅长编程了。
Oh, wish I'm a better programmer now than I was in 2014.
我说过，Kira是我编写的第一个真正可用的工具。
I said, Kira, that was the first tool that I wrote that was usable.
我不会说那段代码写得有多好。
I wouldn't say the code was great.
我现在依然不认为自己的代码有多出色。
I still wouldn't say my code is great.
那么除了练习之外，你的编程能力是如何提升的？
So how was your evolution as a programmer except practice?
你从C语言开始，是什么时候开始学习Python的？
You started with C, at which point did you pick up Python?
因为你现在Python用得很溜了。
Because you're pretty big in Python now.
对，是在大学的时候。
Now, yeah, in college.
我22岁时去了卡内基梅隆大学。
I went to Carnegie Mellon when I was 22.
我又回去读书了。
I went back.
我当时想，我要选你们最难的计算机课程，看看我能学得怎么样。
I'm like, I'm gonna take all your hardest CS courses, and we'll see how I do.
对吧？
Right?
我因为没接受正规本科教育而错过了什么吗？
Did I miss anything by not having a real undergraduate education?
我选了操作系统、编译器、人工智能，还有他们给新生设置的数学淘汰课程。
Took operating systems, compilers, AI, and their freshman Weeder math course.
你提到的有些课程其实相当难。
Some of those classes you mentioned are pretty tough, actually.
这些课程很棒。
They're great.
至少2012年左右的2012操作系统和编译器课程
At least 2012 circa 2012 operating systems and compilers
是
were
其中两门它们是我一生中上过的最好的课。
two of the they were the best classes I've ever taken in my life.
因为你既要写操作系统又要写编译器。
Because you write an operating system and you write a compiler.
我用C语言写了操作系统，用Haskell写了编译器，但
I wrote my operating system in C, and I wrote my compiler in Haskell, but
不知怎么的
somehow
我那个学期还学会了Python。
I picked up Python that semester as well.
实际上我开始用Python是为了参加CTF比赛。
I started using it for the CTFs actually.
那才是我真正开始接触CTF的时候。
That's when I really started to get into CTFs.
CTF就是与时间赛跑，所以我不能用C语言来写代码。
CTFs, you're all it's a race against the clock, so I can't write things in C.
哦，有时间限制的，所以你们得选用能最快完成任务的编程语言。
Oh, there's a clock component, so you really want to use the programming languages so you can be fastest in them.
四十八小时内，尽可能攻破更多挑战。
Forty eight hours, pwn as many of these challenges as you can.
攻破。
Pwn.
对。
Yeah.
每个挑战大概100分，哪个队伍得分最高就赢了。
You got like a 100 points a challenge, whatever team gets the most.
你们俩都曾在Facebook和Google短暂工作过。
You were both at Facebook and Google for a brief stint.
是的。
Yeah.
实际上在Google的Project Zero待了五个月，期间你们开发了Kira。
With Project Zero actually at Google for five months where you developed Kira.
Project Zero主要是做什么的？
What was Project Zero about in general?
只是对这些公司的安全工作感到好奇。
Just curious about the security efforts in these companies.
嗯，Project Zero和我去那里的时间差不多。
Well, Project Zero started the same time I I went there.
你在那里的具体年份是？
What what what years are you there?
2015年。
2015.
展开剩余字幕（还有 480 条）
2015年。
2015.
那正是Project Zero刚开始的时候。
So that was right at the beginning of of of Project Zero.
规模很小。
It's small.
这是谷歌的进攻性安全团队。
It's Google's offensive security team.
我会尽量给出我能提供的最好的公开解释。
I'll try to give the best public facing explanation that I can.
基本理念是这些漏洞确实存在于世界上。
So, the idea is basically these vulnerabilities exist in the world.
国家机构掌握着它们。
Nation states have them.
一些高能力的恶意行为者也掌握着它们。
Some high powered bad actors have them.
有时人们会发现这些漏洞，并通过漏洞赏金计划提交给公司。
Sometimes people will find these vulnerabilities and submit them in bug bounties to the companies.
但很多公司并不真正在意。
But a lot of the companies don't really care.
他们甚至不修复这些漏洞。
They don't even fix the bug.
存在漏洞对他们来说并无损害。
There's no It doesn't hurt for there to be a vulnerability.
所以零日计划的做法与众不同。
So Project Zero is like, we're gonna do it different.
我们会公布漏洞，并给予90天的修复期限。
We're going to announce a vulnerability and we're gonna give them ninety days to fix it.
无论他们是否修复，我们都会公开这个零日漏洞。
And then whether they fix it or not, we're gonna drop the drop the zero day.
哦，哇。
Oh, wow.
我们要公开这个漏洞了
We're gonna drop the weapon on
这技术太酷了
That's the tech so cool.
简直太酷了
That is so cool.
我就喜欢这种最后期限
I love that deadlines.
哇，这真的太酷了
Oh, that's so cool.
给他们真正的截止期限
Give them real deadlines.
没错
Yeah.
然后
And
我认为这对推动行业发展起到了很大作用。
I think it's done a lot for moving the industry forward.
我看了你在网上直播的编程过程。
I watched your coding sessions that you streamed online.
你从零开始编写代码，完成基础项目。
You code things up, basic projects, usually from scratch.
作为一个程序员，我得说看着你编程时，你打字速度非常快，而且你的思维既出色又有些天马行空。
I would say, sort of as a programmer myself just watching you, that you type really fast, and your brain works in both brilliant and chaotic ways.
我不确定这是否总是成立，但对于直播来说确实如此。
I don't know if that's always true, but certainly for the livestreams.
这让我觉得很有趣，因为我更倾向于缓慢、系统和谨慎，而你行动的速度可能比我快一个数量级。
So it's interesting to me because I'm more, I'm much slower and systematic and careful, and you just move, I mean, probably in order of magnitude faster.
所以我很好奇，你的这种疯狂是否有方法可循？
So I'm curious, is there a method to your madness?
这是后天习得的，还是你天生如此？
Is is this or is this just who you are?
有利有弊。
There's pros and cons.
我的编程风格有利有弊，我自己也清楚。
There's pros and cons to my programming style, and I'm aware of them.
比如，如果你让我快速搭建一个使用未充分文档化API的功能，我能极速完成，因为我会不断尝试直到成功。
Like, if you ask me to get something up and working quickly with an API that's kind of undocumented, I will do this super fast because I will throw things at it until it works.
但如果你让我对一个向量进行90度旋转，再在xy平面上翻转，我可能会折腾两小时都搞不定。
If you ask me to take a vector and rotate it 90 degrees, and then flip it over the x y plane, I'll spam program for two hours and won't get it.
哦，因为这种问题本可以用纸笔设计解决，而你习惯直接尝试——而且由于经验丰富通常都能成功。
Oh, because it's something you could do with a sheet of paper, think through design, and then just you really just throw stuff at the wall, and you get so good at it that it usually works.
我也应该提升另一方面的能力。
I should become better at the other kind as well.
有时我也会有条不紊地处理问题。
Sometimes I'll do things methodically.
只是这种风格在Twitch直播上就没什么看点了。
It's nowhere near as entertaining on the Twitch streams.
我在Twitch直播时也确实会夸张一点。
I do exaggerate it a bit on the Twitch streams as well.
我是说Twitch直播，你想看什么游戏？
The Twitch streams I mean, what do you wanna see a game?
你是想看每分钟操作数吗？
Are you wanna see action per minute?
对吧？
Right?
我也会向你展示编程的APM（每分钟操作数）。
I'll show you APM for programming too.
是啊。
Yeah.
推荐大家去看看。
Recommend people go to it.
我想我大概看了你好几个小时，实际上我编程时把你放在后台播放，因为你让我感觉就像在看一个顶级游戏玩家，那种高速操作让人充满能量，太棒了，非常鼓舞人心，而且让我对自己的编程速度感到自愧不如。
I think I watched watched probably several hours of you, like I've actually left you programming in the background while I was programming because you made me, it was like watching a really good gamer, it's like energizes you because you're like moving so fast, it's so it's awesome, it's inspiring, and it made me jealous that, like, because my own programming is inadequate in terms of speed.
哦，哇。
Oh, wow.
因为我在直播时比平时要慌乱两倍，是的。
Because I was like So I'm twice as frantic on the live streams Yeah.
比起没有直播时写代码的状态。
As I am when I code without them.
哦，这超级有趣。
Oh, it's super entertaining.
所以我甚至没注意你在写什么代码，这很棒。
So I I wasn't even paying attention to what you were coding, which is great.
就是看着你在窗口和Vim之间切换，我想这是最...是的。
It's it's just watching you switch Windows and Vim, I guess, is the most Yeah.
用的是Vim和screen。
There's Vim and screen.
我在Facebook养成了一套工作流程并坚持了下来。
I've developed a workflow at Facebook and stuck with it.
你最近是怎么学习新的编程工具、理念和技术的？
How do you learn new programming tools, ideas, techniques these days?
你学习新东西的方法论是什么？
What's your, like, methodology for learning new things?
我为Comma写过文章。
So I wrote for Comma.
目前世界上分布式文件系统极其复杂。
The distributed file systems out in the world are extremely complex.
如果你想安装类似Ceph这样的系统。
If you want to install something like Ceph.
Ceph我认为是开源基础设施的分布式文件系统。
Ceph is I think the open infrastructure distributed file system.
或者有更新的如seaweedfs，但这些项目都超过1万行代码。
Or there's newer ones like seaweedfs, but these are all 10,000 plus line projects.
我觉得有些甚至达到10万行代码。
I think some of them are even 100,000 line.
光是配置它们就是场噩梦。
And just configuring them was a nightmare.
所以我自己写了一个。
So, I wrote one.
只有200行代码，用NGINX连接存储服务器，还有个我用Go写的小型主控服务器。
It's 200 lines and it uses like NGINX to the volume servers and has this little master server that I wrote in Go.
而我用Go的方式...
And the way I In Go.
哇。
Wow.
如果说我对哪行代码感到自豪的话，可能有些精妙的漏洞利用我觉得很美，但最自豪的还是这个。
If I would say that I'm proud per line of any code I wrote, maybe there's some exploits that I think are beautiful and then this.
这个只有200行。
This is 200 lines.
而且我对它的思考方式，我觉得非常棒。
And just the way that I thought about it, I think was very good.
它之所以很好，是因为这是我写的第四个版本，而我放弃了之前的三个版本。
And the reason it's very good is because that was the fourth version of it that I wrote, and I had three versions that I threw away.
你刚才提到是用Go语言写的吗？
You mentioned did you say Go?
是用Go语言写的，没错。
I wrote in Go, yeah.
而且
And
Go。
Go.
所以那是
So that's Is
那是一种函数式语言吗？
that a functional language?
我忘了Go是什么语言了。
I forget what Go is.
Go是谷歌开发的语言。
Go is Google's language.
没错。
Right.
它不是函数式语言。
It's not functional.
某种程度上说，它是更简单的C++。
In a way it's C plus plus but easier.
它是强类型的。
It's strongly typed.
它有一个良好的生态系统。
It has a nice ecosystem around it.
当我第一次接触它时，我觉得它有点像Python，但做任何事情都要花两倍的时间。
When I first looked at it, I was like, this is like Python, but it takes twice as long to do anything.
是的。
Yeah.
我发现OpenPilot正在迁移到C语言，但它仍然包含大量Python组件，现在我明白了为什么Python不适合大型代码库，以及为什么需要像Go这样的语言。
That I've OpenPilot is migrating to C, it still has large Python components, I now understand why Python doesn't work for large code bases and why you want something like Go.
有意思。
Interesting.
那么为什么Python不适用呢？至少就我个人而言，我做了很多自动驾驶相关的演示级工作，大部分都是用Python开发的。
So why why doesn't Python work for so even most speaking for myself at least, like, do a lot of stuff, basically demo level work with autonomous vehicles, and most of the work is Python.
是啊。
Yeah.
为什么Python不适合大型代码
Why doesn't Python work for large code
库？
bases?
因为缺乏类型检查是个大问题。
Because, well, lack of type checking is a big one.
所以错误会悄悄潜入。
So errors creep in.
是的，你根本不知道，编译器什么也告诉你不了。
Yeah, and you don't know, The compiler can tell you nothing.
一切要么是语法错误，要么就没问题。
Everything is either syntax error is fine.
如果你在Python中拼错变量名，编译器不会捕捉到这一点。
If you misspell a variable in Python, the compiler won't catch that.
有些时候可以用linter工具来捕获这类错误。
There's linters that can catch it some of the time.
没有类型系统，这确实是最大的缺点。
There's no types, this is really the biggest downside.
另外，Python运行速度慢，不过这与此无关。
Then, well, Python's slow, but that's not related to it.
好吧，可能有点关系，主要是缺乏...
Well, maybe it's kind of related to it, so it's lack of
那你最近在用哪些工具呢？
So what's in your toolbox these days?
是Python吗？
Is it Python?
还有什么？
What else?
Go语言？
Go?
得换点别的了。
Need to move to something else.
我正在探索依赖类型语言，我太爱这类语言了。
My adventure into dependently typed languages, I love these languages.
就是它们的语法还停留在80年代风格。
They just have like syntax from the eighties.
你觉得
What do you think about
JavaScript怎么样？
JavaScript?
ES6，现代版的TypeScript？
ES6, the modern TypeScript?
JavaScript整个生态系统简直令人困惑到难以置信。
JavaScript is the whole ecosystem is unbelievably confusing.
没错。
Right.
NPM把包从0.22更新到0.25，结果破坏了你的Babel linter——那个把ES5转译成ES6的工具，然后代码就跑不起来了...我为什么又要编译JavaScript？
NPM updates a package from 0.22 to 0.25, and that breaks your Babbel linter, which translates your ES5 into ES6, which doesn't run on why do I have to compile my JavaScript again?
不过它可能是未来的趋势。
It may be the future though.
如果你仔细想想，我最近开始接受JavaScript，就像我一直接受PHP一样，似乎这些最糟糕的语言反而活得最久。
If you think about, I mean, I've embraced JavaScript recently just because just like I've continually embraced PHP, it seems that these worst possible languages live on for the longest.
就像蟑螂永远不死一样。
It's like cockroaches never die.
是啊。
Yeah.
嗯，它在浏览器里运行而且速度很快。
Well, it's in the browser and it's fast.
确实很快。
It's fast.
是啊。
Yeah.
它在浏览器里运行，计算可能会持续发展，你知道的，未来浏览器在分布式计算中的角色还不明确。
It's in the browser and compute might stay become, you know, the browser it's unclear what the role of the browser is in terms of distributed computation in the future.
所以JavaScript肯定会长久存在。
So JavaScript is definitely here to stay.
是啊。
Yeah.
有趣的是，说不定哪天自动驾驶汽车也会用JavaScript运行。
It's interesting if autonomous vehicles will run on JavaScript one day.
我是说，你得考虑这些可能性。
I mean, you have to consider these possibilities.
我们所有的调试工具都是基于JavaScript的。
Well, all our debug tools are JavaScript.
实际上我们刚刚将它们开源了。
We we actually just open source them.
我们有一个工具浏览器，可以用来标注你的分离操作，还有一个工具Cabana，可以用来分析车辆的总线数据。
We have a tool explorer, which you can annotate your disengagements, And we have tool Cabana, which lets you analyze the can traffic from the car.
所以基本上，当你查看日志的可视化内容时，都是在用JavaScript。
So basically, anytime you're visualizing something about the log, you're using JavaScript.
是的。
Yeah.
目前来看，网页是最优秀的UI工具包。
Well, web is the best UI toolkit by far.
没错。
Yep.
所以你看，你现在就是在用JavaScript编程。
So, and then, you know what, you're coding in JavaScript.
我们有个React开发员，他很棒。
We have a React guy, he's good.
React，不错。
React, nice.
我们开始吧。
Let's get into it.
那我们来聊聊自动驾驶汽车。
So let's talk autonomous vehicles.
你创立了Kama AI。
You founded Kama AI.
从高层次来说，你是怎么进入车辆自动化这个领域的？
At a high level, how did you get into the world of vehicle automation?
你能否也为不了解的人讲讲Calm AI的故事？
Can you also just, for people who don't know, tell the story of Calm AI?
当然。
Sure.
当时我在一家AI初创公司工作，有个朋友找到我说，兄弟，虽然不知道未来会怎样，但现在最酷的AI应用就是自动驾驶汽车。
So I was working at this AI startup and a friend approached me and he's like, dude, I don't know where this is going, but the coolest applied AI problem today is self driving cars.
我当时就说，那当然。
I'm like, well, absolutely.
他说，你想见见埃隆·马斯克吗？
Do you want to meet with Elon Musk?
他正在找人开发自动驾驶的视觉系统。
And he's looking for somebody to build a vision system for autopilot.
那时特斯拉还在用AP1系统，仍在使用Mobileye的方案。
This is when they were still on AP1, they were still using Mobileye.
埃隆当时正在寻找替代方案。
Elon back then was looking for a replacement.
他邀请我过去，我们谈了个合同：如果我开发的系统能达到Mobileye的性能水平就能成交。
And he brought me in and we talked about a contract where I would deliver something that meets Mobileye level performance.
合同约定：如果明天就能交付，我能拿到1200万美元；每延迟一个月交付，就要倒扣100万美元。
I would get paid $12,000,000 if I could deliver it tomorrow, and I would lose $1,000,000 for every month I didn't deliver.
是啊。
Yeah.
我当时就想，好吧，这交易不错。
So, I was like, okay, this is a great deal.
这是个超级刺激的挑战。
This is a super exciting challenge.
你知道吗，就算花十个月赚200万美元也不错，说不定五个月就能搞定，也可能完全失败一分钱都拿不到，白干十二个月。
You know what, even if it takes me ten months, I get $2,000,000 it's good, maybe I can finish up in five, maybe I don't finish it at all and I get paid nothing and I'll work for twelve months for free.
要不先暂停一下这个话题。
So maybe just take a pause on that.
我其实很好奇，因为我在机器人领域工作很久了，看到像你这样带着点天真但才华横溢的人直接入局很有意思。
I'm also curious about this because I've been working in robotics for a long time, I'm curious to see a person like you just step in and sort of somewhat naive, but brilliant, right?
这种状态其实最好，因为你能全力以赴解决问题。
So that's the best place to be because you basically full steam take on a problem.
你当时——现在懂得多了——但那时候你觉得解决自动驾驶的难度有多大？
How confident, how, from that time, because you know a lot more now, at that time, how hard do you think it is to solve all of autonomous driving?
我记得在会议上向埃隆建议过，在每个摄像头后面装一个GPU，让计算本地化处理。
I remember I suggested to Elon in the meeting putting a GPU behind each camera to keep the compute local.
这真是个极其愚蠢的主意。
This is an incredibly stupid idea.
会议结束十分钟后我就离开了，当时我就想，我本可以多花点时间思考这个问题
I leave the meeting ten minutes later, and I'm like, I could have spent a little bit of time thinking about this problem
的想法。
for a idea.
哦，直接把所有摄像头信号都传到一个大GPU上处理就行了。
Oh, just send all your cameras to one big GPU.
这样处理效果会好得多。
You're much better off doing that.
哦，抱歉。
Oh, sorry.
你说的是每个摄像头后面都装
You said behind every camera of every
配备GPU。
have GPU.
一个小型GPU。
A small GPU.
我当时想，哦，我可以把通信的前几层放在那里。
Was like, oh, I'll put the first few layers of my comms there.
我为什么要那么说？
Like, why'd I say that?
那是有可能的。
That's possible.
虽然可能，但这是个糟糕的主意。
It's possible, but it's a bad idea.
这并不明显是个坏主意。
It's not obviously a bad idea.
明显很糟糕。
Pretty obviously bad.
但不管这主意到底好不好，我离开与埃隆的那场会议时，一直在自责。
But whether it's actually a bad idea or not, I I left that meeting with Elon, like, beating myself up.
我就想，我为什么要说这么蠢的话？
I'm like, why did I say something stupid?
是啊。
Yeah.
你至少没有全面考虑过每个方面。
You haven't like, you haven't at least, like, thought through every aspect fully.
他反应也非常敏锐。
He's very sharp too.
生活中我通常说蠢话都能蒙混过关，但这次他立刻指出了我的问题。
Like, usually in life, I get away with saying stupid things, and then kind of course Oh, right right away, he called me out about it.
生活中我通常说蠢话都能蒙混过关。
And like usually in life, get away with saying stupid things.
很多时候人们甚至根本不会注意到。
And then like people will, you know, a lot of times people don't even notice.
我会试着纠正并引导对话回到正轨。
And I'll like correct it and bring the conversation back.
但在埃隆面前，根本行不通。
But with Elon, it was like, nope.
好吧，那根本不是合同告吹的原因。
Like, okay, well, that's not at all why the contract fell through.
第二次见他时我准备充分多了。
I was much more prepared the second time I met him.
是啊，但总体而言你觉得难度有多大？
Yeah, but in general, how hard did you think it is?
十二个月的时间线确实很紧张。
Like twelve months is a tough timeline.
噢，我当时就想直接克隆Mobileye IQ3的方案。
Oh, I just thought I'd clone Mobileye IQ three.
我可没想过要解决L5级自动驾驶这种难题。
I didn't think I'd solve level five self driving or anything.
所以当时的目标是
So the goal there was to
实现车道保持，良好的车道保持功能。
do lane keeping, good lane keeping.
我朋友给我看了Mobileye的输出结果，基本上就是两条车道和前方车辆的位置。
Saw My friend showed me the outputs from a Mobileye, and the outputs from a Mobileye was just basically two lanes and a position of a lead car.
我当时就想，我几周内就能收集数据集并训练这个网络。
I'm like, I can gather a dataset and train this net in weeks.
然后我确实做到了。
And I did.
说实话，第一次体验特斯拉搭载的Mobileye系统时，它的优秀程度让我很惊讶。
Well, first time I tried the implementation of Mobileye in a Tesla, I was really surprised how good it is.
它好得不可思议，虽然我做过很多计算机视觉项目，但没想到构建一个稳定的系统会这么难。
It's quite incredibly good, because I thought it's, just because I've done a lot of computer vision, I thought it'd be a lot harder to create a system that that's stable.
我个人确实很意外，必须承认之前我持怀疑态度，以为系统会频繁退出或断开，但实际上它相当稳定可靠。
So I was personally surprised, just, you know, have to admit it because I was kind of skeptical before trying it, because I thought it would go in and out a lot more, it would get disengaged a lot more, and it's pretty robust.
那么当你着手解决这个问题时，它有多难呢？
So how hard is the problem when you tackled it?
我认为AP1非常出色。
So I think AP1 was great.
埃隆提到在洛杉矶405号公路上车道标记有些模糊时会出现系统脱手的情况，Mobileye系统会失效。
Elon talked about disengagements on the 405 down in LA where the lane marks were kind of faded, and the mobile eye system would drop out.
我在三个月内就做出了一个我认为质量相当的系统。
Like, I had something up and working that I would say was like the same quality in three months.
质量相当，但你怎么知道？
Same quality, but how do you know?
你总是这样说话，是啊。
You you say stuff like that Yeah.
很自信，但你无法确定，我喜欢这点，但问题在于你其实更多是靠感觉来判断，因为你...
Confidently, but you can't and I love it, but Well The question is you can't you're you're kinda going by feel because you You're
靠感觉。
going feel.
完全正确。
Absolutely.
是的。
Yeah.
绝对如此。
Absolutely.
就像我借用了朋友的特斯拉。
Like like I would take I had I borrowed my friend's Tesla.
对。
Yeah.
我会开着AP1出去兜风，然后也会试驾我的系统。
I would take AP one out for a drive, and then I would take my system out for a drive.
看起来两者似乎差不多。
And it seems reasonably like the same.
那么405高速路段，要开发出一个能真正作为产品部署的系统有多难？
So the four zero five, how hard is it to create something that could actually be a product that's deployed?
我是说，我读过一篇文章，埃隆回应说，你提到构建自动驾驶系统比乔治·霍德级别的工作要复杂得多。
I mean, I've read an article where Elon responded, said something about you saying that to build autopilot is more complicated than a single George Hod's level job.
要开发一个能在全球范围内运行的系统，这项工作有多难？
How hard is that job to create something that would work across globally?
我不认为全球化是主要挑战，但埃隆随后补充说，在一个10人的公司里需要两年时间，而我现在已经花了四年时间，公司有12个人，我觉得还需要再花两年。
I don't think globally is the challenge, but Elon followed that up by saying it's going to take two years in a company of 10 people, and here I am four years later with a company of 12 people, and I think we still have another two to go.
两年？
Two years?
那么，你对特斯拉Autopilot V2、V3的进展怎么看？
So yeah, so what you think about how Tesla is progressing with Autopilot V2, V3?
我认为我们跟他们的进度保持得相当好。
I think we've kept pace with them pretty well.
我觉得Navigator和Autopilot表现很糟糕。
I think Navigator and Autopilot is terrible.
我们内部测试过一些相同功能的演示版本。
We had some demo features internally of the same stuff, and we would test it.
我当时就想，我甚至不会把这个作为开源软件发布给用户。
And I'm like, I'm not shipping this even as, like, open source software to people.
你为什么觉得它很糟糕？
Why do you think it's terrible?
《消费者报告》对此有很好的描述。
Consumer Reports does a great job of describing it.
比如在变道时，它的表现比人类司机还差。
Like, when it makes a lane change, it does it worse than a human.
嗯。
Mhmm.
像Autopilot、Open Pilot这种系统，它们的车道保持能力应该比人类更强才值得发布。
You shouldn't ship things like Autopilot, Open Pilot, they lane keep better than a human.
如果你在高速公路上开启它一小时，它绝不会压到车道线。
If you turn it on for a stretch of highway, like an hour long, it's never going to touch a lane line.
人类司机可能会压线两次。
Human will touch probably a lane line twice.
你刚刚启发了我。
You just inspired me.
我不确定你这个说法是否有数据支撑。
I don't know if you're grounded in data on that.
我读过你的好吧。
I read your Okay.
但那不。
But that no.
不过这挺有意思的。
But that's interesting.
我其实好奇我们平时到底会多频繁地轻微触碰到车道线，因为
I wonder actually how often we touch lane lines in general, like a little bit, because
这个问题我可以用通用数据集轻松解答。
it is I could answer that question pretty easily with the common dataset.
是啊，我很好奇。
Yeah, I'm curious.
我从未回答过这个问题，我不知道。
I've never answered it, I don't know.
我只是，二就像我个人
I just, two is like my personal
感觉是对的。
It feels right.
这很有趣，因为每次你碰到车道线时，都会带来一点压力，而车道保持功能正是要消除这种压力。
That's interesting, because every time you touch a lane, that's a source of a little bit of stress, and kind of lane keeping is removing that stress.
老实说，最终最大的价值就是消除了必须保持在车道内的压力。
That's ultimately the biggest value add, honestly, is just removing the stress of having to stay in lane.
而且我认为，说实话，我不认为人们完全意识到，首先
And I think, honestly, I don't think people fully realize, first of all,
这是一个巨大的增值，但也仅此而已。
that that's a big value add, but also that that's all it is.
不仅如此，我发现这是一个巨大的增值。
And that not only I find it a huge value add.
我们搬到圣地亚哥时，我开着租来的Enterprise车南下，结果我竟然怀念起了这个系统。
I drove down when we moved to San Diego, I drove down in enterprise rent a car, and I missed it.
我太想念拥有这个系统的感觉了。
I missed having the system so much.
没有它驾驶要疲劳得多。
It's so much more tiring to drive without it.
车道居中功能才是关键所在。
It is that lane centering that's the key feature.
从某种程度上说，这是目前自动驾驶汽车中唯一真正为人们生活带来价值的功功能。
And in a way, it's the only feature that actually adds value to people's lives in autonomous vehicles today.
Waymo并没有为人们的生活创造价值。
Waymo does not add value to people's lives.
它只是一个更贵、更慢的优步。
It's a more expensive, slower slower Uber.
也许有一天会突然出现价值跃升，但我通常不相信这种说法。
Maybe someday, it'll be this big cliff where it adds value, but I don't usually believe that.
知道吗，这很有趣，我还没和Is聊过这个。这很好，因为我虽然直觉上这么觉得，但现在我们把它明确表达出来了。
Know, that's fascinating, I haven't talked to Is that This is good, because I haven't, have intuitively, but I think we're making it explicit now.
我确实认为优秀的车道保持功能是购车的理由，将会成为购车的理由，而且能带来巨大的附加价值。
I actually believe that really good lane keeping is a reason to buy a car, will be a reason to buy a car, and is a huge value add.
直到我们刚才开始讨论这个话题，我才真正意识到我一直觉得马斯克追求四级自动驾驶是错误的追求方向。
I've never, until we just started talking about it, I haven't really quite realized that I've felt with Elon's chase of level four is not the correct chase.
因为你应该直接说，从特斯拉的角度来看，特斯拉拥有最好的车道保持功能。
Because you should just say, Tesla has the best, as if from a Tesla perspective, say, Tesla has the best lane keeping.
Kama AI也应该说，Kama AI拥有最好的车道保持功能，仅此而已。
Kama AI should say, Kama AI has the best lane keeping, and that is it.
是的。
Yeah.
没错。
Yeah.
所以你认为
So do you think
你还得处理纵向控制。
You have to do the longitudinal as well.
不能只做车道保持。
You can't just lane keep.
必须实现自适应巡航（ACC），但ACC比车道保持宽容得多，尤其在高速上。
You have to do ACC, but ACC is much more forgiving than lane keep, especially on the highway.
顺便问下，CAM AI是纯视觉方案吗？
By the way, are you CAM AI is camera only.
对吗？
Correct?
不是。
No.
我们用了雷达。
We use the radar.
从车上你们能获取到...好的。
We From the car, you're able to get the okay.
我们现在可以只用摄像头了。
We can do a camera only now.
虽然技术上已经可以做到，但我们还是保留了雷达，现在是融合系统。
It's gotten to the point, but we leave the radar there as like a it's it's it's fusion now.
好的。
Okay.
那我们来谈谈硬件方面的系统规格吧。
So let's maybe talk through some of the system specs on the hardware.
你们提供的硬件配置是什么？
What are what's what's the hardware side of what you're providing?
软件方面OpenPilot有哪些功能？
What's the capabilities on the software side with So OpenPilot and so
OpenPilot运行的设备就是我们销售的盒子，其实就是装在塑料壳里的手机。
OpenPilot, the box that we sell that it runs on, it's a phone in a plastic case.
嗯。
Mhmm.
这没什么特别的。
It's nothing special.
我们出售的是不含软件的版本。
We sell it without the software.
就像你买手机一样，非常简单。
So you're like, you know, you buy the phone, it is just easy.
设置会很方便，但出售时不包含软件。
It'll be easy setup, but it's sold with no software.
OpenPilot即将发布0.6版本。
OpenPilot right now is about to be 0.6.
等达到1.0版本时，我认为就能面向消费者市场了。
When it gets to one point o, I think we'll be ready for consumer product.
我们不会添加任何新功能。
We're not gonna add any new features.
我们只专注于把车道保持功能做到极致。
We're just gonna make the lane keeping really, really good.
好的。
Okay.
明白了。
I got it.
那我们现在有什么配置？
So what do we have right now?
搭载的是高通骁龙820。
It's a Snapdragon eight twenty.
配备索尼IMX298前置摄像头作为驾驶员监测摄像头（其实就是手机的前置摄像头），还有一个CAN总线收发器。
It's a Sony IMX two ninety eight forward facing camera, driver monitoring camera, which is a selfie cam on the phone, and a CAN transceiver.
这个小东西叫PANDAS。
This little thing called PANDAS.
嗯。
Mhmm.
它们通过USB与手机通信，然后通过三个CAN总线与车辆进行数据交换。
And they talk over USB to the phone, and then they have three CAN buses that they talk to the car.
其中一条CAN总线是雷达CAN总线。
One of those CAN buses is the radar CAN bus.
一条是主车CAN总线，另一条是代理摄像头CAN总线。
One of them is the main car CAN bus, and the other one is the proxy camera CAN bus.
嗯。
Mhmm.
我们保留原有摄像头不动，这样不会关闭自动紧急制动功能。
We leave the existing camera in place so we don't turn AEB off.
目前如果使用我们的纵向控制，还是会关闭自动紧急制动，但我们会在1.0版本前解决这个问题。
Right now, we still turn AEB off if you're using our longitudinal, but we're gonna fix that before one point o.
明白了。
Got it.
哇。
Wow.
真酷。
That's cool.
所以说CAN总线是双向的。
So and it's CAN both ways.
那你们是如何控制车辆的呢？
So how are you able to control vehicles?
我们通过代理方式实现，与我们合作的车辆本身已配备车道保持辅助系统。
So we proxy, the vehicles that we work with already have a lane keeping assist system.
车道保持辅助系统的功能范围其实非常广泛。
So lane keeping assist can mean a huge variety of things.
有的系统只会在你偏离车道一英尺后对方向盘施加轻微扭矩（比如老款丰田），而特斯拉仍称之为车道保持辅助的系统则能在高速公路上让你始终保持在车道正中央。
It can mean it will apply a small torque to the wheel after you've already crossed a lane line by a foot, which is the system in the older Toyotas, versus like, I think Tesla still calls it lane keeping assist, where it'll keep you perfectly in the center of the lane on the highway.
就像操纵游戏杆控制车辆一样，这些车本身已具备线控驾驶能力。
You can control, like you would, the joystick, car, so these cars already have the capability of drive by wire.
那么对于一辆能与OpenPilot兼容的车辆来说，实现转向控制是否很容易？
So is it trivial to convert a car that it operates with, OpenPilot is able to control the steering?
噢，新车型或... 我们现在已支持45个不同品牌的车辆。
Oh, a new car or a car that we so we have support now for 45 different makes of cars.
具体是哪些车型
What are what are the cars
一般来说？
in general?
主要是本田和丰田。
Mostly Hondas and Toyotas.
我们支持今年生产的几乎所有本田和丰田车型。
We support almost every Honda and Toyota made this year.
还有大量通用汽车、斯巴鲁以及
And then bunch of GMs, bunch of Subarus, bunch of It
不一定要像普锐斯那样的车型。
doesn't have to be like a Prius.
也可以是卡罗拉。
It could be Corolla as well.
哦，那个
Oh, the
2020款卡罗拉是搭载Open Pilot的最佳车型。
2020 Corolla is the best car with open pilot.
它刚刚上市。
It just came out.
这款的执行器延迟比老款卡罗拉更小。
There the actuator has less lag than the older Corolla.
我觉得...我开始看你的视频了，你制作视频的方式太棒了。
I think I I started watching a video with your mean, the way you make videos is awesome.
你就直接在经销商那里直播。
You're just literally at the dealership streaming.
我当时
I had
把手机给了我朋友。
to my friend the phone.
我说：你可能想直播一小时吧？
I'm like, probably you wanna stream for an hour?
是啊，基本上如果出了点小问题，你就顺其自然。
Yeah, and basically, like if stuff goes a little wrong, you just like, you just go with it.
对，我超爱这种风格。
Yeah, I love it.
嗯，这是真实的。
Well, it's real.
没错，这是真实的。
Yeah, it's real.
这太美了，而且和其他公司制作那种视频的方式形成鲜明对比。
That's so beautiful and it's so in contrast to the way other companies would put together a video like that.
这就是为什么我喜欢这样做。
Kind of why I like to do it like that.
很好。
Good.
我是说，如果有一天你变得超级富有和成功，我希望你能保持这种方式，因为我觉得人们真正喜爱的就是这种真实感
I mean, if you become super rich one day and successful, I hope you keep it that way because I think that's actually what people love, that kind of genuine
哦，那才是我珍视的一切。
Oh, it's all that has value to me.
是啊。
Yeah.
金钱毫无意义，如果我为了赚钱而出卖自己，那就是背叛。
Money has no if I sell out to, like, make money, I sold out.
这无关紧要。
It doesn't matter.
我能得到什么？
What do I get?
游艇？
Yacht?
我不想要游艇。
I don't want a yacht.
我认为特斯拉在自动驾驶日上也隐约体现了这种精神。
And I think Tesla's actually has a small inkling of that as well with Autonomy Day.
他们确实透露了比我预期更多的信息，当然其中也有营销宣传的成分，但比其他公司愿意公开的内容要多，这正是我希望他们能继续深入的方向。
They did reveal more than I mean, of course, there's marketing communications, you could tell, but it's more than most companies would reveal, which is I hope they go towards that direction more.
其他公司，比如通用、福特
Other companies, GM, Ford
哦，特斯拉肯定会赢得五级自动驾驶的竞争。
Oh, Tesla's gonna win level five.
他们确实有这个实力。
They really are.
那我们就来聊聊这个话题吧。
So let's just talk about it.
所以你认为你们目前主要专注于二级自动驾驶？
You think so you're focused on level two currently currently.
我们在实现五级自动驾驶方面会比特斯拉晚一到两年。
We're gonna be one to two years behind Tesla getting to level five.
好的。
Okay.
嘿。
Hey.
我们是安卓系统。
We're Android.
对吧？
Right?
我们是安卓系统。
We're Android.
你们是安卓系统。
You're Android.
我只是说一旦特斯拉实现了，我们会落后一两年。
I'm just saying once Tesla gets it, we're one to two years behind.
我并不是在预测特斯拉何时能实现。
I'm I'm not making any timeline on when Tesla's gonna get it.
没错。
That's right.
你确实做到了。
You did.
这太棒了。
That's brilliant.
抱歉，各位投资者。
I'm sorry, investors.
如果你们认为年底前能拥有自动驾驶出租车车队，我敢打赌这不可能实现。
If you think you're gonna have an autonomous robotaxi fleet by the end of the year I'll bet against that.
那么你怎么看这个问题，大多数L4级公司基本上还是在用常规的安全驾驶员进行全自动驾驶测试，而特斯拉则试图从车道保持直接跨越到全自动驾驶。
So what do you think about this, the most level four companies are kind of just doing their usual safety driver, doing full autonomy kind of testing, and then Tesla does basically trying to go from lane keeping to full autonomy.
你如何看待这种策略？
What do you think about that approach?
它成功的可能性有多大？
How successful would it be?
这是个好得多的策略，因为特斯拉收集的数据规模是其他公司无法比拟的。
It's a ton better approach, because Tesla is gathering data on a scale that none of them are.
他们让真实用户坐在驾驶座上。
They're putting real users behind the the wheel of the cars.
我认为这是唯一可行的策略。
It's I think the only strategy that works.
渐进式的。
The incremental.
特斯拉的方法有几个组成部分，不仅仅是渐进主义。
Well, so there's a few components to Tesla approach that's more than just the incrementalist.
你提到的一个是软件部分，即空中软件更新。
What you spoke with is the one is the software, so over the air software updates.
必要性。
Necessity.
没错。
Right.
我是说，Waymo和Cruise也有这些功能，那些不算。
I mean, Waymo had Cruise have those too, those aren't.
这些区别于
Those differentiate from the
对
Right.
没有车道保持辅助系统的车辆——除了特斯拉，没有配备车道保持系统的车辆能做到这一点。
No lane keeping assist systems have No cars with lane keeping system have that except Tesla.
是的
Yeah.
另一个关键因素是数据，即查询数据的能力。
And the other one is the data, the other direction, which is the ability to query the data.
我认为他们实际收集的数据量没有人们想象的那么多，但他们具备随时开启和关闭数据收集的能力。
I don't think they're actually collecting as much data as people think, but the ability to turn on collection and turn it off.
所以我同时涉足机器人学领域和心理学人因工程领域。
So I'm both in the robotics world and the psychology human factors world.
许多人认为由于人因问题，二级自动驾驶是有缺陷的。
Many people believe that level two autonomy is problematic because of the human factor.
就像任务自动化程度越高，人的警觉性就越会下降，开始犯困、变得懈怠，玩手机的时间也越来越多等等。
Like the more the task is automated, the more there's a vigilance decrement, you start to fall asleep, you start to become complacent, start texting more and so on.
你担心这个问题吗？
Do you worry about that?
因为如果我们讨论的是从车道保持到完全自动驾驶的过渡，如果你80%的时间都不监督机器，你难道不担心这意味着什么吗？首先，或者说安全性方面——
Because if we're talking about transition from lane keeping to full autonomy, if you're spending 80% of the time not supervising the machine, do you worry about what that means One, or the safety of the
我们不认为OpenPilot达到1.0版本标准，除非我们能实现100%的驾驶员监控。
we don't consider OpenPilot to be one point o until we have 100% driver monitoring.
目前你确实可以绕过我们的驾驶员监控系统。
You can cheat right now our driver monitoring system.
有几种方法可以作弊，这些方法都很明显。
There's a few ways to cheat it, they're pretty obvious.
我们正在努力改进这个系统。
We're working on making that better.
在我们推出能自动驾驶汽车的消费级产品前，我必须确保开发出无法破解的驾驶员监控系统。
Before we ship a consumer product that can drive cars, I wanna make sure that I have driver monitoring that you can't cheat.
一个成功的驾驶员监控系统应该是什么样的？
What's like a successful driver monitoring system look like?
它是否仅仅关乎于让你一直盯着路面？
It's keep is is it all about just keeping your eyes on the road?
嗯，有几个方面。
Well, a few things.
这是我们最初在驾驶员监控方面采用的方法。
So that's what we went with at first for driver monitoring.
我正在检测你的头部实际朝向的位置。
I'm checking I'm actually looking at where your head is looking.
摄像头的分辨率没那么高。
The camera's not that high resolution.
眼睛的识别有点困难。
Eyes are a little bit hard to get.
嗯，头部有这么大。
Well, head is this big.
我是说，那是
I mean, that's
头部监测效果不错。
Head is good.
实际上从心理学角度来说，持续存在的监控提醒会不断警示你必须保持专注，但我们想做得更好。
And actually a lot of it, just psychology wise to have that monitor constantly there, it reminds you that you have to be paying attention, but we want to go further.
我们刚聘请了一位全职人员专门负责驾驶员监控系统。
We just hired someone full time to come on to do the driver monitoring.
我想要检测画面中的手机，还要确保你没有睡着。
I want to detect phone in frame I want to make sure you're not sleeping.
摄像头能拍到身体的多少部分？
How much does the camera see of the body?
这个摄像头拍得不够多。
This one, not enough.
不够多。
Not enough.
下一个版本会覆盖全部范围。
The next one, everything.
嗯，这很有趣，鱼眼镜头，因为我们只是做数据收集而非实时监控，但鱼眼镜头能很好地捕捉人体。
Well, it's interesting, fisheye, because we're doing just data collection, not real time, but Fisheye is a beautiful being able to capture the body.
而智能手机确实是最大的问题。
And the smartphone is really like the biggest problem.
我可以给你看看我们新系统的其中一张图片。
I'll I'll show you I can show you one of the pictures from from our our new system.
太棒了。
Awesome.
所以，你基本上是在说驾驶员监控将成为解决方案。
So, you're basically saying the driver monitoring will be the answer to that.
我认为你在论文中提出的另一点也很有道理。
I think the other point that you raised in your paper is good as well.
你不能要求人类在不给予他们控制权的情况下监督机器。他们必须能随时接管控制。
You're not asking a human to supervise a machine without giving them the They can take over at any time.
我们的安全模型允许你随时接管。
Our safety model, you can take over.
我们会在油门或刹车时解除控制。
We disengage on both the gas or the brake.
我们不会在转向时解除控制，我认为没必要这样做。
We don't disengage on steering, I don't feel you have to.
但我们会在油门或刹车时解除控制。
But we disengage on gas or brake.
因此你可以非常轻松地接管，也很容易重新介入。
So it's very easy for you to take over, and it's very easy for you to re engage.
这种切换应该极其便捷。
That switching should be super cheap.
即使是自动驾驶功能，这些车辆也需要双击确认。
The cars that require even autopilot requires a double press.
这几乎显而易见。
That's almost see.
我不喜欢那样。
I don't like that.
然后就是取消操作。
And then then the cancel.
要在自动驾驶模式下取消，你要么按取消键——但没人知道那是哪个键，所以他们就踩刹车。
To cancel in autopilot, you either have to press cancel, which no one knows what that is, so they press the brake.
但很多时候你其实并不想踩刹车。
But a lot of times you don't actually wanna press the brake.
你想踩油门。
You wanna press gas.
所以你应该通过踩油门或晃动方向盘来取消，但这样也不好。
So you should cancel on gas or wiggle the steering wheel, which is bad as well.
哇。
Wow.
这太棒了。
That's brilliant.
我还没听人这么清楚地阐述过这个观点。
I haven't heard anyone articulate that point.
哦，我这是
Oh, I it's
我 这是 思考过的。
I This is think about.
因为我认为特斯拉实际上比其他大多数汽车制造商做得更好，让这个过程更顺畅，但你刚才描述说明还有改进空间。
Because I think actually Tesla has done a better job than most automakers at making that frictionless, but you just described that it could be even better.
一旦启动，我非常喜欢超级巡航的体验。
I love Super Cruise as an experience once it's engaged.
不知道你用过没有，但要让它尝试启动。
I don't know if you've used it, but getting the thing to try to engage.
是的。
Yeah.
我经常使用超级巡航系统。
I've driven Super Cruise a lot.
那么你对Super Cruise系统总体有什么看法？
So what's your thoughts on the Super Cruise system in general?
当你退出Super Cruise时它会降级为ACC，所以我的车还在加速，感觉很奇怪。
You disengage Super Cruise and it falls back to ACC, so my car is still accelerating, it feels weird.
除此之外，当你在高速公路上真正启用Super Cruise时，它的表现非常出色。
Otherwise, when you actually have Super Cruise engaged on the highway, it is phenomenal.
我们买了那辆凯迪拉克，刚卖掉它，但我们买它就是为了体验这个功能。
We bought that Cadillac, we just sold it, but we bought it just to experience this.
我想让办公室里的每个人都明白：这就是我们要努力打造的标准。
I wanted everyone in the office to be like, This is what we're striving to build.
通用汽车在驾驶员监测方面是开创者。
GM pioneering with the driver monitoring.
你喜欢他们的驾驶员监测系统吗？
You like their driver monitoring system?
它还有些小毛病。
It has some bugs.
如果阳光从这里照射过来，系统就会对你视而不见。
If there's a sun shining back here, it'll be blind to you.
但总体来说，大部分情况下还是不错的。
But overall, mostly, yeah.
你懂这么多真是太酷了。
That's so cool that you know all this stuff.
我很少能和懂这些的人交流，因为这车实在太稀有了，真遗憾。
I don't often talk to people that because it's such a rare car, unfortunately,
他们买这辆车就是为了这个功能。
They bought one explicitly for this.
我们折旧损失了大约2.5万美元，但我觉得很值得。
We lost like 25 k in the deprecation, but I feel it was worth it.
让我非常惊喜的是，通用这套系统如此创新，却几乎没怎么宣传，也没多少人讨论。
I was very pleasantly surprised that GM system was so innovative and really wasn't advertised much, wasn't talked about much.
是啊。
Yeah.
我当时很担心这个系统会消亡，会彻底消失。
And I was nervous that it would die, that they would disappear.
嗯，他们把系统装错了车型。
Well, they put it on the wrong car.
他们应该把它装在Bolt上上，而不是装在那些没人买的凯迪拉克怪车上。
They should have put it on the Bolt and not some weird Cadillac that nobody bought.
我认为他们至少会把系统推广到整个车队。
I think that's gonna be into they're saying at least it's gonna be into their entire fleet.
既然我们谈到驾驶员监控，你对马斯克声称不需要驾驶员监控的说法怎么看？
So what do you think about as long as we're on the driver monitoring, what do you think about Elon Musk's claim that driver monitoring is not needed?
通常我很欣赏他的观点，但这个说法太愚蠢了。
Normally I love his claims, that one is stupid.
这个说法愚蠢至极，而且他年底前不可能实现L5自动驾驶车队。
That one is stupid and he's not going to have his level five fleet by the end of the year.
希望他能承认错误：好吧我错了，我会加装驾驶员监控。
Hopefully he's like, okay, I was wrong, I'm going to add driver monitoring.
因为当这些系统发展到每千英里才出错一次时，绝对需要驾驶员监控。
Because when these systems get to the point that they're only messing up once every thousand miles, you absolutely need driver monitoring.
那么让我来扮演一下，虽然我同意你的观点，但让我唱个反调。
So let me play, because I agree with you, but let me play devil's advocate.
当然。
Sure.
一种可能性是，在没有驾驶员监控的情况下，人们能够自我监督、自我调节。
One possibility is that without driver monitoring, people are able to monitor, self regulate, monitor themselves.
你
You
所以你的观点很明显——你知道所有在特斯拉车里睡觉的人吧？
so your ideas obviously You've know, all the people sleeping in Teslas?
是的，不过我对所有在特斯拉车里睡觉的人持保留态度，因为我已经不再关注这类事情了，我想看到真实数据，现在太多美化报道了，感觉不够科学。
Yeah, well, I'm a little skeptical of all the people sleeping in Teslas because I've stopped paying attention to that kind of stuff, because I want to see real data, there's too much glorified, it doesn't feel scientific to me.
所以我想知道，到底有多少人真的在特斯拉车里睡觉，而不是假睡。
So I want to know, you know, how many people are really sleeping in Teslas versus sleeping.
我开车过来时，在一辆没有自动驾驶功能的车上，困得差点睡着。
I was driving here, deprived in a car with no automation, I was falling asleep.
我同意这有点炒作成分。
I agree that it's hype y.
就像，你知道吗？
It's just like, you know what?
如果埃隆安装了驾驶员监控系统...我上次使用自动驾驶是在三月份租了辆Model 3试驾。
If Elon put driver monitoring, my last autopilot experience was I rented a Model three in March and drove it around.
方向盘检测功能很烦人。
The wheel thing is annoying.
方向盘检测烦人的原因是...我们也有类似功能，但不会因为方向盘操作就退出自动驾驶。
And the reason the wheel thing is annoying, we use the wheel thing as well, but we don't disengage on wheel.
特斯拉要求你施加刚好能触发扭矩传感器的力度握住方向盘，既不能太轻导致系统警告，又不能太重，这设计很矛盾。
For Tesla, you have to touch the wheel just enough to trigger the torque sensor to tell it that you're there, but not enough as to disengage it, which don't use it for two things.
根本不需要设置方向盘操作就退出自动驾驶，完全没必要。
Don't disengage on wheel, you don't have to.
关于 Bayt 播客
Bayt 提供中文+原文双语音频和字幕，帮助你打破语言障碍，轻松听懂全球优质播客。