Kyle Vogt: Cruise Automation #14

Transcript

00:00:00 The following is a conversation with Kyle Vogt.

00:00:02 He’s the president and the CTO of Cruise Automation,

00:00:05 leading an effort to solve one of the biggest robotics challenges of our time,

00:00:09 vehicle automation. He’s a cofounder of two successful companies, Twitch

00:00:13 and Cruise, that have each sold for a billion dollars.

00:00:16 And he’s a great example of the innovative spirit that flourishes

00:00:20 in Silicon Valley, and now is facing an interesting and exciting challenge of

00:00:26 matching that spirit with the mass production and the safety centric

00:00:31 culture of a major automaker like General Motors. This conversation is

00:00:35 part of the MIT Artificial General Intelligence series

00:00:38 and the Artificial Intelligence podcast. If you enjoy it,

00:00:42 please subscribe on YouTube, iTunes, or simply connect with me on Twitter

00:00:46 at Lex Friedman, spelled F R I D. And now here’s my conversation with Kyle Vogt.

00:00:54 You grew up in Kansas, right? Yeah, and I just saw that picture you had hidden

00:00:57 over there, so I’m a little bit a little bit worried about that now.

00:01:00 Yeah, so in high school in Kansas City, you joined Shawnee Mission

00:01:04 North high school robotics team. Yeah. Now that wasn’t your high school.

00:01:09 That’s right, that was that was the only high school in the area that had a

00:01:13 like a teacher who was willing to sponsor our first robotics team.

00:01:16 I was gonna troll you a little bit. Jog your memory a little bit.

00:01:19 Yeah, I was trying to look super cool and intense, because you know this

00:01:23 was BattleBots. This is serious business. So we’re standing there with a welded

00:01:28 steel frame and looking tough. So go back there. What is that drew you

00:01:32 to robotics? Well, I think I’ve been trying to figure

00:01:36 this out for a while, but I’ve always liked building things with Legos. And

00:01:38 when I was really, really young, I wanted the Legos that had motors and

00:01:41 other things. And then, you know, Lego Mindstorms came out, and for the

00:01:45 first time you could program Lego contraptions. And I think things

00:01:49 just sort of snowballed from that. But I remember

00:01:54 seeing, you know, the BattleBots TV show on Comedy Central and thinking that is

00:01:58 the coolest thing in the world. I want to be a part of that.

00:02:01 And not knowing a whole lot about how to build these

00:02:04 200 pound fighting robots. So I sort of obsessively poured over

00:02:09 the internet forums where all the creators for BattleBots would sort of

00:02:12 hang out and talk about, you know, document their build progress and

00:02:16 everything. And I think I read, I must have read like,

00:02:20 you know, tens of thousands of forum posts from basically

00:02:23 everything that was out there on what these people were doing. And eventually

00:02:26 like sort of triangulated how to put some of these things together.

00:02:30 And I ended up doing BattleBots, which was, you know, I was like 13 or 14, which

00:02:34 was pretty awesome. I’m not sure if the show is still

00:02:37 running, but so BattleBots is, there’s not an artificial

00:02:41 intelligence component. It’s remotely controlled. And it’s

00:02:44 almost like a mechanical engineering challenge of building things

00:02:47 that can be broken. They’re radio controlled. So,

00:02:50 and I think that they allowed some limited form of autonomy.

00:02:54 But, you know, in a two minute match, you’re, in the way these things ran,

00:02:58 you’re really doing yourself a disservice by trying to automate it

00:03:01 versus just, you know, do the practical thing, which is drive it yourself.

00:03:04 And there’s an entertainment aspect, just going on YouTube, there’s like an,

00:03:08 some of them wield an axe, some of them, I mean, there’s that fun.

00:03:12 So what drew you to that aspect? Was it the mechanical engineering?

00:03:15 Was it the dream to create like Frankenstein and

00:03:19 sentient being? Or was it just like the Lego,

00:03:22 you like tinkering with stuff? I mean, that was just building something.

00:03:25 I think the idea of, you know, this radio controlled machine that

00:03:29 can do various things, if it has like a weapon or something was pretty

00:03:32 interesting. I agree it doesn’t have the same

00:03:34 appeal as, you know, autonomous robots, which I,

00:03:37 which I, you know, sort of gravitated towards later on. But it was definitely

00:03:40 an engineering challenge because everything you did in that

00:03:44 competition was pushing components to their limits. So we would

00:03:49 buy like these $40 DC motors that came out of a

00:03:54 winch, like on the front of a pickup truck or something, and we’d

00:03:57 power the car with those and we’d run them at like double or triple their

00:04:01 rated voltage. So they immediately start overheating,

00:04:04 but for that two minute match you can get,

00:04:06 you know, a significant increase in the power output of those motors

00:04:09 before they burn out. And so you’re doing the same thing for your battery packs,

00:04:12 all the materials in the system. And I think there’s something, something

00:04:16 intrinsically interesting about just seeing like where things break.

00:04:20 And did you offline see where they break? Did you

00:04:23 take it to the testing point? Like how did you know two minutes? Or was there a

00:04:26 reckless let’s just go with it and see? We weren’t very good at

00:04:30 BattleBots. We lost all of our matches the first round.

00:04:34 The one I built first, both of them were these wedge shaped robots because

00:04:38 wedge, even though it’s sort of boring to look

00:04:40 at, is extremely effective. You drive towards another robot and

00:04:43 the front edge of it gets under them and then they sort of flip over,

00:04:46 kind of like a door stopper. And the first one had a

00:04:49 pneumatic polished stainless steel spike on the front that would shoot out about

00:04:53 eight inches. The purpose of which is what? Pretty,

00:04:56 pretty ineffective actually, but it looks cool.

00:04:58 And was it to help with the lift? No, it was, it was just to try to poke holes

00:05:02 in the other robot. And then the second time I did it, which is the following,

00:05:07 I think maybe 18 months later, we had a, well a titanium axe with a, with a

00:05:12 hardened steel tip on it that was powered by a hydraulic

00:05:16 cylinder which we were activating with liquid CO2, which was,

00:05:21 had its own set of problems. So great, so that’s kind of on the hardware side.

00:05:26 I mean at a certain point there must have been born a fascination

00:05:31 on the software side. So what was the first piece of code you’ve written?

00:05:35 Go back there, see what language was it?

00:05:38 What, what was that? Was it Emacs? Vim? Was it a more

00:05:41 respectable modern IDE? Do you, do you remember any of this?

00:05:45 Yeah, well I remember, I think maybe when I was in

00:05:49 third or fourth grade, the school I was at, elementary school, had a bunch of

00:05:52 Apple II computers and we’d play games on those. And I

00:05:56 remember every once in a while something would,

00:05:58 would, would crash or wouldn’t start up correctly and it would dump you out to

00:06:02 what I later learned was like sort of a command prompt.

00:06:05 And my teacher would come over and type, I actually remember this to this day for

00:06:08 some reason, like PR number six or PR pound six, which is

00:06:12 peripheral six, which is the disk drive, which would fire up the disk and load the

00:06:15 program. And I just remember thinking wow, she’s

00:06:17 like a hacker, like teach me these, these codes, these error codes, that is what

00:06:21 I called them at the time. But she had no interest in that, so it

00:06:23 wasn’t until I think about fifth grade that I had a school where you could

00:06:28 actually go on these Apple IIs and learn to program. And so it was all in basic,

00:06:31 you know, where every line, you know, the line numbers are all number, that every

00:06:34 line is numbered and you have to like leave enough space

00:06:37 between the numbers so that if you want to tweak your code you go back and

00:06:41 the first line was 10 and the second line is 20. Now you have to go back and

00:06:44 insert 15 and if you need to add code in front of

00:06:47 that, you know, 11 or 12 and you hope you don’t run out of line numbers and have

00:06:50 to redo the whole thing. And there’s go to statements? Yeah, go to

00:06:54 and it’s very basic, maybe hence the name, but a lot of fun.

00:06:58 And that was like, that was, you know, that’s when, you know,

00:07:01 when you first program you see the magic of it.

00:07:03 It’s like, it just, just like this world opens up with, you know, endless

00:07:07 possibilities for the things you could build or

00:07:09 or accomplish with that computer. So you got the bug then, so

00:07:12 even starting with basic and then what C++ throughout,

00:07:15 what did you, was there computer programming, computer science classes in

00:07:19 high school? Not, not where I went, so it was self

00:07:22 taught, but I did a lot of programming. The thing that, you know, sort of

00:07:27 pushed me in the path of eventually working on self driving cars is actually

00:07:31 one of these really long trips driving from my

00:07:34 house in Kansas to, to I think Las Vegas where we did the BattleBots competition

00:07:39 and I had just gotten my, I think my learner’s permit or

00:07:43 early driver’s permit and so I was driving this,

00:07:47 you know, 10 hour stretch across western Kansas where it’s just,

00:07:50 you’re going straight on a highway and it is mind numbingly boring. And I

00:07:53 remember thinking even then with my sort of mediocre programming

00:07:57 background that this is something that a computer can do, right? Let’s take a

00:08:00 picture of the road, let’s find the yellow lane markers and,

00:08:03 you know, steer the wheel. And, you know, later I’d come to realize this had been

00:08:06 done, you know, since, since the 80s or the 70s or even

00:08:10 earlier, but I still wanted to do it and sort of

00:08:13 immediately after that trip switched from sort of BattleBots, which is more

00:08:16 radio controlled machines, to thinking about building,

00:08:21 you know, autonomous vehicles of some scale. Start off with really small

00:08:24 electric ones and then, you know, progress to what we’re

00:08:27 doing now. So what was your view of artificial intelligence at that point?

00:08:30 What did you think? So this is before, there’s been waves in artificial

00:08:35 intelligence, right? The current wave with deep learning

00:08:39 makes people believe that you can solve in a really rich deep way the computer

00:08:44 vision perception problem, but like in

00:08:48 before the deep learning craze, you know, how do you think about,

00:08:52 how would you even go about building a thing that perceives itself in the

00:08:56 world, localizes itself in the world, moves around the world?

00:08:59 Like when you were younger, I mean, what was your thinking about it? Well,

00:09:02 prior to deep neural networks or convolutional neural

00:09:05 analysis, these modern techniques we have, or at least

00:09:06 ones that are in use today, it was all a heuristic space and so like old school

00:09:11 image processing and I think extracting, you know, yellow lane markers out of an

00:09:16 image of a road is one of the problems that lends itself

00:09:21 reasonably well to those heuristic based methods, you know, like

00:09:24 just do a threshold on the color yellow and then try to

00:09:27 fit some lines to that using a Huff transform or something and then

00:09:31 go from there. Traffic light detection and stop sign detection, red, yellow, green.

00:09:35 And I think you could, I mean, if you wanted to do a full,

00:09:39 I was just trying to make something that would stay in between the lanes on a

00:09:43 highway, but if you wanted to do the full,

00:09:46 the full, you know, set of capabilities needed for a driverless car,

00:09:50 I think you could, and we’d done this at cruise, you know, in the very first days,

00:09:54 you can start off with a really simple, you know, human written heuristic just to

00:09:58 get the scaffolding in place for your system. Traffic light detection,

00:10:01 probably a really simple, you know, color thresholding on day one just to

00:10:04 get the system up and running before you migrate to, you know, a deep

00:10:08 learning based technique or something else.

00:10:10 And, you know, back in when I was doing this, my first one, it was on a Pentium

00:10:14 203, 233 megahertz computer in it and I think I wrote the first version in

00:10:19 basic, which is like an interpreted language. It’s

00:10:21 extremely slow because that’s the thing I knew at the time.

00:10:24 And so there was no, no chance at all of using,

00:10:27 there was no, no computational power to do any sort of reasonable

00:10:32 deep nets like you have today. So I don’t know what kids these days are doing. Are

00:10:35 kids these days, you know, at age 13 using neural networks in

00:10:38 their garage? I mean, that would be awesome.

00:10:40 I get emails all the time from, you know, like 11, 12 year olds

00:10:44 saying I’m having, you know, I’m trying to follow this TensorFlow tutorial

00:10:48 and I’m having this problem. And the general approach

00:10:53 in the deep learning community is of extreme optimism of, as opposed to,

00:11:00 you mentioned like heuristics, you can, you can, you can

00:11:03 separate the autonomous driving problem into modules and try to solve it sort of

00:11:06 rigorously, or you can just do it end to end.

00:11:08 And most people just kind of love the idea that, you know, us humans do it end

00:11:12 to end. We just perceive and act. We should be able to use that, do the

00:11:17 same kind of thing when you’re on nets. And that,

00:11:19 that kind of thinking, you don’t want to criticize that kind of thinking because

00:11:23 eventually they will be right. Yeah. And so it’s exciting and especially

00:11:26 when they’re younger to explore that as a really exciting approach. But yeah,

00:11:30 it’s, it’s changed the, the language,

00:11:35 the kind of stuff you’re tinkering with. It’s kind of exciting to see when these

00:11:38 teenagers grow up. Yeah. I can only imagine if you,

00:11:42 if your starting point is, you know, Python and TensorFlow at age 13

00:11:46 where you end up, you know, after 10 or 15 years of that,

00:11:49 that’s, that’s pretty cool. Because of GitHub, because the state tools for

00:11:53 solving most of the major problems in artificial intelligence

00:11:56 are within a few lines of code for most kids.

00:12:00 And that’s incredible to think about also on the entrepreneurial side.

00:12:04 And, and on that point, was there any thought about entrepreneurship before

00:12:10 you came to college? Is sort of doing, you’re building this

00:12:14 into a thing that impacts the world on a large scale? Yeah. I’ve always

00:12:18 wanted to start a company. I think that’s, you know, just a cool concept of

00:12:22 creating something and exchanging it for value or creating value, I guess.

00:12:28 So in high school, I was, I was trying to build like, you know, servo motor

00:12:32 drivers, little circuit boards and sell them online

00:12:35 or other, other things like that. And certainly knew at some point I wanted to

00:12:39 do a startup, but it wasn’t really, I’d say until college, until I felt

00:12:42 like I had the,

00:12:45 I guess the right combination of the environment, the smart people around you

00:12:48 and some free time and a lot of free time at MIT.

00:12:52 So you came to MIT as an undergrad 2004. That’s right. And that’s when the first

00:12:58 DARPA Grand Challenge was happening. Yeah. The, the timing of that is

00:13:02 beautifully poetic. So how did you get yourself involved in that one?

00:13:05 Originally there wasn’t a official entry. Yeah, faculty sponsored thing. And so a

00:13:09 bunch of undergrads, myself included, started meeting and got together and

00:13:14 tried to haggle together some sponsorships. We got a vehicle donated,

00:13:18 a bunch of sensors and tried to put something together. And so we had,

00:13:22 our team was probably mostly freshmen and sophomores, you know, which, which was

00:13:26 not really a fair, fair fight against maybe the,

00:13:30 you know, postdoc and faculty led teams from other schools. But

00:13:33 we, we got something up and running. We had our vehicle drive by wire and

00:13:36 you know, very, very basic control and things. But

00:13:41 on the day of the qualifying, sort of pre qualifying round, the one and

00:13:47 only steering motor that we had purchased,

00:13:50 the thing that we had retrofitted to turn the steering wheel on the truck

00:13:55 died. And so our vehicle was just dead in the water, couldn’t steer.

00:13:58 So we didn’t make it very far. On the hardware side. So was there a software

00:14:02 component? Was there, like, how did your view of autonomous

00:14:05 vehicles in terms of artificial intelligence

00:14:09 evolve in this moment? I mean, you know, like you said from the 80s has been

00:14:13 autonomous vehicles, but really that was the birth of the modern wave.

00:14:16 The, the thing that captivated everyone’s imagination that we can actually do this.

00:14:21 So what, how were you captivated in that way?

00:14:25 So how did your view of autonomous vehicles change at that point?

00:14:28 I’d say at that point in time it was, it was a

00:14:32 curiosity as in, like, is this really possible?

00:14:35 And I think that was generally the spirit and

00:14:39 the purpose of that original DARPA Grand Challenge, which was to just get

00:14:44 a whole bunch of really brilliant people exploring the space and pushing the

00:14:49 limits. And I think, like, to this day that

00:14:52 DARPA Challenge with its, you know, million dollar prize pool

00:14:56 was probably one of the most effective, you know, uses of taxpayer

00:15:00 money dollar for dollar that I’ve seen, you know, because that,

00:15:04 that small sort of initiative that DARPA put,

00:15:07 put out sort of, in my view, was the catalyst or the tipping point

00:15:12 for this, this whole next wave of autonomous vehicle development. So that

00:15:16 was pretty cool. So let me jump around a little bit on that point.

00:15:20 They also did the Urban Challenge where it was in the city, but it was very

00:15:23 artificial and there’s no pedestrians and there’s very little human

00:15:27 involvement except a few professional drivers. Yeah.

00:15:31 Do you think there’s room, and then there was the Robotics Challenge with

00:15:34 humanoid robots. Right. So in your now role is looking at this,

00:15:38 you’re trying to solve one of the, you know, autonomous driving, one of the

00:15:42 harder, more difficult places in San Francisco.

00:15:45 Is there a role for DARPA to step in to also kind of help out,

00:15:49 like, challenge with new ideas, specifically pedestrians and so on, all

00:15:54 these kinds of interesting things? Well, I haven’t, I haven’t thought about it

00:15:56 from that perspective. Is there anything DARPA could do today to further

00:15:59 accelerate things? And I would say, my instinct is that that’s maybe not the

00:16:04 highest and best use of their resources and time,

00:16:07 because, like, kick starting and spinning up the flywheel is, I think, what

00:16:11 what they did in this case for very, very little money. But today this has become,

00:16:16 this has become, like, commercially interesting to very large companies and

00:16:19 the amount of money going into it and the amount of

00:16:22 people, like, going through your class and learning about these things and

00:16:25 developing these skills is just, you know, orders of magnitude

00:16:29 more than it was back then. And so there’s enough momentum and inertia

00:16:33 and energy and investment dollars into this space right now that

00:16:37 I don’t, I don’t, I think they’re, I think they’re, they can just say mission

00:16:41 accomplished and move on to the next area of technology that needs help.

00:16:46 So then stepping back to MIT, you left MIT during your junior year.

00:16:50 What was that decision like? As I said, I always wanted to do

00:16:54 a company in, or start a company, and this opportunity landed in my lap, which

00:16:59 was a couple guys from Yale were starting a

00:17:02 new company, and I googled them and found that they had

00:17:05 started a company previously and sold it actually on eBay for

00:17:09 about a quarter million bucks, which was a pretty interesting story, but

00:17:13 so I thought to myself, these guys are, you know, rock star entrepreneurs, they’ve

00:17:17 done this before, they must be driving around in Ferraris

00:17:20 because they sold their company, and, you know, I thought I could learn

00:17:25 a lot from them, so I teamed up with those guys and,

00:17:27 you know, went out during, went out to California during IAP, which is MIT’s

00:17:33 month off, on a one way ticket and basically never went back.

00:17:37 We were having so much fun, we felt like we were building something and creating

00:17:40 something, and it was going to be interesting

00:17:42 that, you know, I was just all in and got completely hooked, and that

00:17:47 that business was Justin TV, which is originally a reality show about a guy

00:17:51 named Justin, which morphed into a live video

00:17:56 streaming platform, which then morphed into what is Twitch

00:17:59 today, so that was, that was quite an unexpected journey.

00:18:04 So no regrets? No. Looking back, it was just an obvious, I mean,

00:18:09 one way ticket. I mean, if we just pause on that for a second,

00:18:12 there was no, how did you know these are the right guys, this is the

00:18:17 right decision, you didn’t think it was just follow the

00:18:21 heart kind of thing? Well, I didn’t know, but, you know, just trying something for a

00:18:25 month during IAP seems pretty low risk, right?

00:18:28 And then, you know, well, maybe I’ll take a semester off, MIT’s pretty flexible

00:18:31 about that, you can always go back, right? And then after two or three cycles of

00:18:35 that, I eventually threw in the towel, but, you know, I think it’s,

00:18:40 I guess in that case I felt like I could always hit the undo button if I had to.

00:18:44 Right. But nevertheless, from when you look

00:18:48 in retrospect, I mean, it seems like a brave decision,

00:18:51 you know, it would be difficult for a lot of people to make. It wasn’t as

00:18:54 popular, I’d say that the general, you know, flux of people

00:18:59 out of MIT at the time was mostly into, you know, finance or consulting jobs in

00:19:04 Boston or New York, and very few people were going to

00:19:07 California to start companies, but today I’d say that’s,

00:19:10 it’s probably inverted, which is just a sign of,

00:19:14 a sign of the times, I guess. Yeah. So there’s a story about

00:19:18 midnight of March 18, 2007, where TechCrunch, I guess, announced

00:19:24 Justin.TV earlier than it was supposed to, a few hours.

00:19:28 The site didn’t work. I don’t know if any of this is true, you can tell me.

00:19:32 And you and one of the folks at Justin.TV,

00:19:36 Emmett Shearer, coded through the night. Can you take me through that experience?

00:19:41 So let me, let me say a few nice things that,

00:19:45 the article I read quoted Justin Kahn said that you were known for bureau

00:19:49 coding through problems and being a creative, quote, creative

00:19:52 genius. So on that night,

00:19:56 what, what was going through your head, or maybe I’d put another way,

00:20:00 how do you solve these problems? What’s your approach to solving these kinds of

00:20:04 problems where the line between success and failure seems to be pretty

00:20:08 thin? That’s a good question. Well, first of all, that’s, that’s a

00:20:12 nice of Justin to say that. I think, you know, I would have been

00:20:15 maybe 21 years old then and not very experienced at programming,

00:20:18 but as with, with everything in a startup, you’re sort of racing against

00:20:23 the clock. And so our plan was the second we had

00:20:28 this live streaming camera backpack up and running, where Justin could wear it

00:20:33 and no matter where he went in a city, it

00:20:35 would be streaming live video. And this is

00:20:36 even before the iPhones. This is like hard to do back then.

00:20:40 We would launch. And so we thought we were there and the backpack was working

00:20:45 and then we sent out all the emails to launch the,

00:20:47 launch the company and do the press thing. And then, you know,

00:20:51 we weren’t quite actually there. And then

00:20:54 we thought, oh, well, you know, they’re not going to announce it until

00:20:58 maybe 10 a.m. the next morning. And it’s, I don’t know, it’s 5 p.m. now. So

00:21:02 how many hours do we have left? What is that? Like, you know, 17 hours to go.

00:21:06 And, and that was, that was going to be fine.

00:21:10 Was the problem obvious? Did you understand

00:21:12 what could possibly, like, how complicated was the system at that point?

00:21:16 It was, it was pretty messy. So to get a live video feed that looked decent

00:21:22 working from anywhere in San Francisco, I put together this system where we had

00:21:27 like three or four cell phone data modems and

00:21:29 they were, like, we take the video stream and,

00:21:32 you know, sort of spray it across these three or four modems and then try to

00:21:36 catch all the packets on the other side, you know, with unreliable cell phone

00:21:39 networks. It’s pretty low level networking.

00:21:41 Yeah, and putting these, like, you know, sort of

00:21:44 protocols on top of all that to, to reassemble and reorder the packets and

00:21:47 have time buffers and error correction and all that kind of stuff.

00:21:50 And the night before it was just staticky. Every once in a while the image

00:21:55 would, would go to staticky and there would be this horrible,

00:21:58 like, screeching audio noise because the audio was also corrupted.

00:22:01 And this would happen, like, every five to ten minutes or so and it was

00:22:05 a really, you know, off putting to the viewers.

00:22:08 How do you tackle that problem? What was the, uh, you’re just freaking out behind a

00:22:12 computer. There’s, are there other, other folks working

00:22:16 on this problem? Like, were you behind a whiteboard? Were you doing, uh,

00:22:19 Yeah, it was a little, it was a little, yeah, it’s a little lonely because there’s four of us

00:22:23 working on the company and only two people really wrote code.

00:22:26 And Emmett wrote the website and the chat system and I wrote the

00:22:30 software for this video streaming device and video server.

00:22:34 And so, you know, it’s my sole responsibility to figure that out.

00:22:37 And I think, I think it’s those, you know, setting,

00:22:40 setting deadlines, trying to move quickly and everything where you’re in that

00:22:42 moment of intense pressure that sometimes people do their

00:22:45 best and most interesting work. And so even though that was a terrible moment,

00:22:48 I look back on it fondly because that’s like, you know, that’s one of those

00:22:51 character defining moments, I think. So in 2013, October, you founded

00:22:58 Cruise Automation. Yeah. So progressing forward, another

00:23:02 exceptionally successful company was acquired by GM in 16

00:23:06 for $1 billion. But in October 2013, what was on your mind?

00:23:14 What was the plan? How does one seriously start to tackle

00:23:19 one of the hardest robotics, most important impact for robotics

00:23:23 problems of our age? After going through Twitch, Twitch was,

00:23:27 was, and is today, pretty successful. But the, the work was,

00:23:35 the result was entertainment, mostly. Like, the better the product was,

00:23:38 the more we would entertain people and then, you know, make money on the ad

00:23:42 revenues and other things. And that was, that was a good thing. It

00:23:44 felt, felt good to entertain people. But I figured like, you know, what is really

00:23:48 the point of becoming a really good engineer and

00:23:51 developing these skills other than, you know, my own enjoyment? And I

00:23:54 realized I wanted something that scratched more of an existential

00:23:56 itch, like something that, that truly matters. And so I

00:23:59 basically made this list of requirements for a new, if I was going to

00:24:05 do another company, and the one thing I knew in the back of

00:24:07 my head that Twitch took like eight years to become successful.

00:24:12 And so whatever I do, I better be willing to commit, you know, at least 10 years

00:24:16 to something. And when you think about things from that perspective,

00:24:20 you certainly, I think, raise the bar on what you choose to work on. So for me,

00:24:23 the three things were it had to be something where the technology

00:24:26 itself determines the success of the product,

00:24:28 like hard, really juicy technology problems, because that’s what

00:24:32 motivates me. And then it had to have a direct and positive impact on society in

00:24:37 some way. So an example would be like, you know,

00:24:39 health care, self driving cars, because they save lives, other things where

00:24:42 there’s a clear connection to somehow improving other people’s lives.

00:24:45 And the last one is it had to be a big business, because

00:24:48 for the positive impact to matter, it’s got to be a large scale.

00:24:51 And I was thinking about that for a while, and I made like, I tried

00:24:54 writing a Gmail clone and looked at some other ideas.

00:24:57 And then it just sort of light bulb went off, like self driving cars, like that

00:25:00 was the most fun I had ever had in college working on that.

00:25:03 And like, well, what’s the state of the technology? It’s been 10 years.

00:25:07 Maybe times have changed, and maybe now is the time to make this work.

00:25:10 And I poked around and looked at, the only other thing out there really at the

00:25:14 time was the Google self driving car project.

00:25:16 And I thought, surely there’s a way to, you know, have an entrepreneur mindset

00:25:20 and sort of solve the minimum viable product here.

00:25:23 And so I just took the plunge right then and there and said, this is something I

00:25:26 know I can commit 10 years to. It’s the probably the greatest

00:25:29 applied AI problem of our generation. And if it works, it’s going to be both a

00:25:33 huge business and therefore like, probably the most positive impact I can

00:25:37 possibly have on the world. So after that light bulb went off, I went

00:25:41 all in on cruise immediately and got to work.

00:25:45 Did you have an idea how to solve this problem? Which aspect of the problem to

00:25:48 solve? You know, slow, like we just had Oliver

00:25:52 from Voyage here, slow moving retirement communities,

00:25:56 urban driving, highway driving. Did you have, like, did you have a vision of the

00:26:00 city of the future where, you know, the transportation is

00:26:05 largely automated, that kind of thing? Or was it sort of

00:26:09 more fuzzy and gray area than that? My analysis of the situation is that

00:26:15 Google is putting a lot, had been putting a lot of money into that project. They

00:26:19 had a lot more resources. And so, and they still hadn’t cracked

00:26:23 the fully driverless car. You know, this is 2013, I guess.

00:26:29 So I thought, what can I do to sort of go from zero to,

00:26:34 you know, significant scale so I can actually solve the real problem, which is

00:26:37 the driverless cars. And I thought, here’s the strategy. We’ll

00:26:40 start by doing a really simple problem or solving a

00:26:44 really simple problem that creates value for people. So,

00:26:49 eventually ended up deciding on automating highway driving,

00:26:51 which is relatively more straightforward as long as there’s a

00:26:55 backup driver there. And, you know, the go to market will be able to retrofit

00:26:59 people’s cars and just sell these products directly. And

00:27:02 the idea was, we’ll take all the revenue and profits from that and use it

00:27:06 to do the, so sort of reinvest that in research for

00:27:10 doing fully driverless cars. And that was the plan.

00:27:13 The only thing that really changed along the way between then and now is

00:27:17 we never really launched the first product. We had enough interest from

00:27:21 investors and enough of a signal that this was

00:27:23 something that we should be working on, that after about a year of working on

00:27:26 the highway autopilot, we had it working, you know, on a

00:27:29 prototype stage. But we just completely abandoned that

00:27:33 and said, we’re going to go all in on driverless cars now is the time.

00:27:36 Can’t think of anything that’s more exciting and if it works more impactful,

00:27:39 so we’re just going to go for it. The idea of retrofit is kind of

00:27:42 interesting. Yeah. Being able to, it’s how you achieve scale.

00:27:46 It’s a really interesting idea. Is it something that’s still in the

00:27:49 in the back of your mind as a possibility?

00:27:52 Not at all. I’ve come full circle on that one. After

00:27:56 trying to build a retrofit product, and I’ll touch on some of the complexities

00:28:00 of that, and then also having been inside an OEM

00:28:04 and seeing how things work and how a vehicle is developed and

00:28:06 validated. When it comes to something that has

00:28:09 safety critical implications like controlling the steering and

00:28:13 other control inputs on your car, it’s pretty hard to get there

00:28:16 with a retrofit. Or if you did, even if you did, it creates a whole bunch

00:28:21 of new complications around liability or how did you truly validate

00:28:25 that. Or you know, something in the base vehicle fails and causes your system to

00:28:28 fail, whose fault is it?

00:28:31 Or if the car’s anti lock brake systems or other things kick in

00:28:35 or the software has been, it’s different in one version of the car you retrofit

00:28:38 versus another and you don’t know because

00:28:40 the manufacturer has updated it behind the scenes. There’s basically an

00:28:43 infinite list of long tail issues that can get you.

00:28:46 And if you’re dealing with a safety critical product, that’s not really

00:28:48 acceptable. That’s a really convincing summary of why

00:28:52 that’s really challenging. But I didn’t know all that at the time, so we tried it

00:28:54 anyway. But as a pitch also at the time, it’s a

00:28:57 really strong one. Because that’s how you achieve scale and that’s how you beat

00:29:00 the current, the leader at the time of Google or the only one in the market.

00:29:04 The other big problem we ran into, which is perhaps the biggest problem from a

00:29:08 business model perspective, is we had kind of assumed that we

00:29:13 started with an Audi S4 as the vehicle we

00:29:16 retrofitted with this highway driving capability.

00:29:18 And we had kind of assumed that if we just knock out like three

00:29:22 make and models of vehicles, that’ll cover like 80% of the San Francisco

00:29:25 market. Doesn’t everyone there drive, I don’t know, a BMW or a Honda Civic or

00:29:28 one of these three cars? And then we surveyed our users and we found out that

00:29:32 it’s all over the place. We would, to get even a decent number of

00:29:35 units sold, we’d have to support like, you know, 20 or 50 different models.

00:29:39 And each one is a little butterfly that takes time and effort to maintain,

00:29:43 you know, that retrofit integration and custom hardware and all this.

00:29:47 So it was a tough business. So GM manufactures and sells over 9 million

00:29:52 cars a year. And what you with Cruise are trying to do

00:29:58 some of the most cutting edge innovation in terms of applying AI.

00:30:03 And so how do those, you’ve talked about a little bit before, but it’s also just

00:30:06 fascinating to me. We work a lot of automakers,

00:30:09 you know, the difference between the gap between Detroit

00:30:12 and Silicon Valley, let’s say, just to be sort of poetic about it, I guess.

00:30:17 How do you close that gap? How do you take GM into the future

00:30:21 where a large part of the fleet will be autonomous, perhaps?

00:30:24 I want to start by acknowledging that GM is made up of,

00:30:28 you know, tens of thousands of really brilliant, motivated people who want to

00:30:31 be a part of the future. And so it’s pretty fun to work

00:30:34 within the attitude inside a car company like that is, you

00:30:37 know, embracing this transformation and change

00:30:41 rather than fearing it. And I think that’s a testament to

00:30:44 the leadership at GM and that’s flown all the way through to everyone you

00:30:47 talk to, even the people in the assembly plants working on these cars.

00:30:51 So that’s really great. So starting from that position makes it a lot easier

00:30:55 so then when the people in San Francisco at Cruise

00:30:59 interact with the people at GM, at least we have this common set of values, which

00:31:02 is that we really want this stuff to work

00:31:04 because we think it’s important and we think it’s the future.

00:31:08 That’s not to say, you know, those two cultures don’t clash. They absolutely do.

00:31:12 There’s different sort of value systems. Like in a

00:31:15 car company, the thing that gets you promoted and sort of the reward

00:31:19 system is following the processes, delivering

00:31:23 the program on time and on budget. So any sort of risk taking

00:31:28 is discouraged in many ways because if a program is late or if you shut down

00:31:34 the plant for a day, it’s, you know, you can count the millions of dollars that

00:31:38 burn by pretty quickly. Whereas I think, you know, most Silicon

00:31:42 Valley companies and in Cruise and the methodology

00:31:47 we were employing, especially around the time of the acquisition,

00:31:50 the reward structure is about trying to solve

00:31:53 these complex problems in any way shape or form or coming up with crazy ideas

00:31:57 that, you know, 90% of them won’t work. And so meshing that culture

00:32:02 of sort of continuous improvement and experimentation

00:32:05 with one where everything needs to be rigorously defined up front so that

00:32:08 you never slip a deadline or miss a budget

00:32:12 was a pretty big challenge. And that we’re over three years in now

00:32:16 after the acquisition and I’d say like, you know, the investment we made in

00:32:20 figuring out how to work together successfully and

00:32:23 who should do what and how we bridge the gaps between these

00:32:26 very different systems and way of doing engineering work

00:32:29 is now one of our greatest assets because I think we have this really

00:32:31 powerful thing. But for a while it was both GM and Cruise were very

00:32:36 steep on the learning curve. Yeah, so I’m sure it was very stressful.

00:32:38 It’s really important work because that’s how

00:32:41 to revolutionize the transportation, really to revolutionize

00:32:44 any system. You know, you look at the health care system or you look at the

00:32:48 legal system. I have people like Loris come up to me all the time like

00:32:52 everything they’re working on can easily be automated.

00:32:55 But then that’s not a good feeling. Yeah, well it’s not a good feeling but also

00:32:59 there’s no way to automate because the entire infrastructure is really,

00:33:05 you know, based is older and it moves very slowly. And so

00:33:08 how do you close the gap between I have an

00:33:12 how can I replace, of course, Loris don’t want to be replaced with an app, but you

00:33:15 could replace a lot of aspect when most of the data is still on paper.

00:33:20 And so the same thing was with automotive.

00:33:23 I mean, it’s fundamentally software. It’s basically hiring software

00:33:27 engineers. It’s thinking in a software world.

00:33:30 I mean, I’m pretty sure nobody in Silicon Valley has ever hit a deadline.

00:33:34 So and then on GM. That’s probably true, yeah. And GM side is probably the

00:33:38 opposite. Yeah. So that’s that culture gap is really fascinating.

00:33:42 So you’re optimistic about the future of that?

00:33:45 Yeah, I mean, from what I’ve seen, it’s impressive. And I think like

00:33:48 especially in Silicon Valley, it’s easy to write off building cars because,

00:33:51 you know, people have been doing that for over 100 years now in this country. And

00:33:55 so it seems like that’s a solved problem, but that doesn’t mean it’s an easy

00:33:58 problem. And I think it would be easy to sort of

00:34:01 overlook that and think that, you know, we’re

00:34:04 Silicon Valley engineers. We can solve any problem, you know, building a car.

00:34:08 It’s been done. Therefore, it’s, you know, it’s not a real

00:34:12 engineering challenge. But after having seen just the sheer

00:34:16 scale and magnitude and industrialization

00:34:20 that occurs inside of an automotive assembly plant, that is a lot of work

00:34:24 that I am very glad that we don’t have to reinvent

00:34:28 to make self driving cars work. And so to have, you know, partners who have done

00:34:31 that for 100 years now, these great processes and this huge infrastructure

00:34:33 and supply base that we can tap into is just remarkable

00:34:38 because the scope and surface area of

00:34:43 the problem of deploying fleets of self driving cars is so large

00:34:47 that we’re constantly looking for ways to do less

00:34:50 so we can focus on the things that really matter more. And if we had to

00:34:53 figure out how to build and assemble and

00:34:57 you know, build the cars themselves. I mean, we work closely with GM on

00:35:01 that. But if we had to develop all that capability in house as well,

00:35:05 you know, that would just make the problem really intractable, I think.

00:35:10 So yeah, just like your first entry at the MIT DARPA challenge when

00:35:15 there was what the motor that failed, somebody that knows what they’re

00:35:18 doing with the motor did it. That would have been nice if we could

00:35:20 focus on the software, not the hardware platform.

00:35:23 Yeah. Right. So from your perspective now,

00:35:27 you know, there’s so many ways that autonomous vehicles can impact

00:35:30 society in the next year, five years, ten years.

00:35:34 What do you think is the biggest opportunity to make

00:35:37 money in autonomous driving, sort of make it a

00:35:41 financially viable thing in the near term?

00:35:44 What do you think will be the biggest impact there?

00:35:49 Well, the things that drive the economics for fleets of self driving

00:35:53 cars are, there’s sort of a handful of variables. One is,

00:35:57 you know, the cost to build the vehicle itself. So the material cost, how many,

00:36:02 you know, what’s the cost of all your sensors plus the cost of the vehicle and

00:36:05 every all the other components on it. Another one is the lifetime of the

00:36:08 vehicle. It’s very different if your vehicle

00:36:11 drives 100,000 miles and then it falls apart versus,

00:36:13 you know, two million. And then, you know, if you have a fleet, it’s

00:36:18 kind of like an airplane or an airline where

00:36:23 once you produce the vehicle, you want it to be in

00:36:26 operation as many hours a day as possible producing revenue.

00:36:30 And then, you know, the other piece of that is

00:36:34 how are you generating revenue? I think that’s kind of what you’re asking. And I

00:36:36 think the obvious things today are, you know, the ride sharing business

00:36:40 because that’s pretty clear that there’s demand for that,

00:36:42 there’s existing markets you can tap into and

00:36:46 large urban areas, that kind of thing. Yeah, yeah. And I think that there are

00:36:50 some real benefits to having cars without

00:36:53 drivers compared to sort of the status quo for

00:36:56 people who use ride share services today.

00:36:58 You know, you get privacy, consistency, hopefully significantly improve safety,

00:37:02 all these benefits versus the current product.

00:37:04 But it’s a crowded market. And then other opportunities, which you’ve

00:37:08 seen a lot of activity in the last, really in the last six or twelve months,

00:37:10 is, you know, delivery, whether that’s parcels and packages,

00:37:14 food or groceries. Those are all sort of, I think, opportunities that are

00:37:20 pretty ripe for these, you know, once you have this

00:37:25 core technology, which is the fleet of autonomous vehicles, there’s all sorts of

00:37:28 different business opportunities you can build on

00:37:31 top of that. But I think the important thing, of course, is that

00:37:34 there’s zero monetization opportunity until you actually have that fleet of

00:37:37 very capable driverless cars that are that are as good or better than humans.

00:37:40 And that’s sort of where the entire industry is

00:37:44 sort of in this holding pattern right now.

00:37:45 Yeah, they’re trying to achieve that baseline. So, but you said sort of

00:37:49 not reliability, consistency. It’s kind of interesting, I think I heard you say

00:37:53 somewhere, I’m not sure if that’s what you meant, but

00:37:56 you know, I can imagine a situation where you would get an autonomous vehicle

00:38:01 and, you know, when you get into an Uber or Lyft,

00:38:04 you don’t get to choose the driver in a sense that you don’t get to choose the

00:38:07 personality of the driving. Do you think there’s a, there’s room

00:38:12 to define the personality of the car the way it drives you in terms of

00:38:15 aggressiveness, for example, in terms of sort of pushing the

00:38:19 bound? One of the biggest challenges of autonomous driving is the

00:38:23 is the trade off between sort of safety and

00:38:26 assertiveness. And do you think there’s any room

00:38:30 for the human to take a role in that decision? Sort of accept some of the

00:38:36 liability, I guess. I wouldn’t, no, I’d say within

00:38:39 reasonable bounds, as in we’re not gonna, I think it’d be

00:38:43 highly unlikely we’d expose any knob that would let you, you know, significantly

00:38:48 increase safety risk. I think that’s just not

00:38:51 something we’d be willing to do. But I think driving style or like, you

00:38:56 know, are you going to relax the comfort constraints slightly or things like that,

00:39:00 all of those things make sense and are plausible. I see all those as, you know,

00:39:03 nice optimizations. Once again, we get the core problem solved in these fleets

00:39:07 out there. But the other thing we’ve sort of

00:39:09 observed is that you have this intuition that if you

00:39:13 sort of slam your foot on the gas right after the light turns green and

00:39:16 aggressively accelerate, you’re going to get there faster. But the

00:39:19 actual impact of doing that is pretty small.

00:39:22 You feel like you’re getting there faster, but so the same would be

00:39:25 true for AVs. Even if they don’t slam their, you know, the pedal to the floor

00:39:29 when the light turns green, they’re going to get you there within, you

00:39:32 know, if it’s a 15 minute trip, within 30 seconds of what you would have done

00:39:35 otherwise if you were going really aggressively.

00:39:37 So I think there’s this sort of self deception that my aggressive

00:39:42 driving style is getting me there faster.

00:39:44 Well, so that’s, you know, some of the things I’ve studied, some of the things

00:39:46 I’m fascinated by the psychology of that. I don’t think it matters

00:39:50 that it doesn’t get you there faster. It’s the emotional release.

00:39:55 Driving is a place, being inside of a car,

00:39:58 somebody said it’s like the real world version of being a troll.

00:40:02 So you have this protection, this mental protection, you’re able to sort of yell

00:40:05 at the world, like release your anger, whatever.

00:40:08 So there’s an element of that that I think autonomous vehicles would also

00:40:12 have to, you know, giving an outlet to people, but it doesn’t have to be

00:40:16 through, through, through driving or honking or so on.

00:40:19 There might be other outlets, but I think to just sort of even just put that aside,

00:40:23 the baseline is really, you know, that’s the focus.

00:40:26 That’s the thing you need to solve.

00:40:28 And then the fun human things can be solved after.

00:40:30 But so from the baseline of just solving autonomous driving, you’re working in

00:40:35 San Francisco, one of the more difficult cities to operate in.

00:40:38 What is, what is the, in your view, currently the hardest

00:40:43 aspect of autonomous driving?

00:40:45 Is it negotiating with pedestrians?

00:40:49 Is it edge cases of perception?

00:40:51 Is it planning?

00:40:52 Is there a mechanical engineering?

00:40:54 Is it data, fleet stuff?

00:40:57 What are your thoughts on the challenge, the more challenging aspects there?

00:41:00 That’s a, that’s a good question.

00:41:02 I think before, before we go to that, though, I just want to, I like what you

00:41:04 said about the psychology aspect of this,

00:41:07 because I think one observation I’ve made is I think I read somewhere that I

00:41:11 think it’s maybe Americans on average spend, you know, over an hour a day on

00:41:14 social media, like staring at Facebook.

00:41:18 And so that’s just, you know, 60 minutes of your life, you’re not getting back.

00:41:21 It’s probably not super productive.

00:41:23 And so that’s 3,600 seconds, right?

00:41:26 And that’s, that’s time, you know, it’s a lot of time you’re giving up.

00:41:30 And if you compare that to people being on the road, if another vehicle, whether

00:41:35 it’s a human driver or autonomous vehicle, delays them by even three

00:41:38 seconds, they’re laying in on the horn, you know, even though that’s, that’s, you

00:41:43 know, one, one thousandth of the time they waste looking at Facebook every day.

00:41:46 So there’s, there’s definitely some.

00:41:48 You know, psychology aspects of this, I think that are pretty interesting road

00:41:51 rage in general.

00:41:51 And then the question of course is if everyone is in self driving cars,

00:41:54 do they even notice these three second delays anymore?

00:41:57 Cause they’re doing other things or reading or working or just talking to

00:42:01 each other.

00:42:01 So it’ll be interesting to see where that goes.

00:42:03 In a certain aspect, people, people need to be distracted by something

00:42:06 entertaining, something useful inside the car.

00:42:09 So they don’t pay attention to the external world.

00:42:10 And then, and then they can take whatever psychology and bring it back to

00:42:15 Twitter and then focus on that as opposed to sort of interacting, sort of putting

00:42:21 the emotion out there into the world.

00:42:23 So it’s a, it’s an interesting problem, but baseline autonomy.

00:42:26 I guess you could say self driving cars, you know, at scale will lower the

00:42:30 collective blood pressure of society probably by a couple of points without

00:42:34 all that road rage and stress.

00:42:35 So that’s a good, good external.

00:42:38 So back to your question about the technology and the, I guess the biggest

00:42:43 problems.

00:42:43 And I have a hard time answering that question because, you know, we’ve been

00:42:47 at this like specifically focusing on driverless cars and all the technology

00:42:52 needed to enable that for a little over four and a half years now.

00:42:55 And even a year or two in, I felt like we had completed the functionality needed

00:43:02 to get someone from point A to point B.

00:43:04 As in, if we need to do a left turn maneuver, or if we need to drive around

00:43:08 at, you know, a double parked vehicle into oncoming traffic or navigate

00:43:12 through construction zones, the scaffolding and the building blocks was

00:43:16 there pretty early on.

00:43:17 And so the challenge is not any one scenario or situation for which, you

00:43:23 know, we fail at 100% of those.

00:43:25 It’s more, you know, we’re benchmarking against a pretty good or pretty high

00:43:29 standard, which is human driving.

00:43:31 All things considered, humans are excellent at handling edge cases and

00:43:35 unexpected scenarios where computers are the opposite.

00:43:38 And so beating that baseline set by humans is the challenge.

00:43:43 And so what we’ve been doing for quite some time now is basically, it’s this

00:43:49 continuous improvement process where we find sort of the most, you know,

00:43:53 uncomfortable or the things that could lead to a safety issue or other

00:43:59 things, all these events.

00:44:00 And then we sort of categorize them and rework parts of our system to make

00:44:05 incremental improvements and do that over and over and over again.

00:44:07 And we just see sort of the overall performance of the system, you know,

00:44:12 actually increasing in a pretty steady clip.

00:44:13 But there’s no one thing.

00:44:15 There’s actually like thousands of little things and just like polishing functionality

00:44:19 and making sure that it handles, you know, every version and possible

00:44:23 permutation of a situation by either applying more deep learning systems or

00:44:30 just by, you know, adding more test coverage or new scenarios that we

00:44:34 develop against and just grinding on that.

00:44:37 We’re sort of in the unsexy phase of development right now, which is doing

00:44:40 the real engineering work that it takes to go from prototype to production.

00:44:44 You’re basically scaling the grinding, sort of taking seriously that the

00:44:49 process of all those edge cases, both with human experts and machine

00:44:54 learning methods to cover all those situations.

00:44:59 Yeah.

00:44:59 And the exciting thing for me is I don’t think that grinding ever stops because

00:45:03 there’s a moment in time where you’ve crossed that threshold of human

00:45:08 performance and become superhuman.

00:45:09 But there’s no reason, there’s no first principles reason that AV capability

00:45:13 will tap out anywhere near humans.

00:45:16 Like there’s no reason it couldn’t be 20 times better, whether that’s, you

00:45:19 know, just better driving or safer driving or more comfortable driving or

00:45:22 even a thousand times better given enough time.

00:45:25 And we intend to basically chase that, you know, forever to build the best

00:45:30 possible product.

00:45:31 Better and better and better.

00:45:32 And always new edge cases come up and new experiences.

00:45:34 So, and you want to automate that process as much as possible.

00:45:40 So what do you think in general in society?

00:45:43 When do you think we may have hundreds of thousands of fully autonomous

00:45:47 vehicles driving around?

00:45:48 So first of all, predictions, nobody knows the future.

00:45:52 You’re a part of the leading people trying to define that future, but even

00:45:55 then you still don’t know.

00:45:56 But if you think about hundreds of thousands of vehicles, so a significant

00:46:02 fraction of vehicles in major cities are autonomous.

00:46:05 Do you think, are you with Rodney Brooks, who is 2050 and beyond, or are you

00:46:12 more with Elon Musk, who is, we should have had that two years ago?

00:46:19 Well, I mean, I’d love to have it two years ago, but we’re not there yet.

00:46:24 So I guess the way I would think about that is let’s flip that question

00:46:29 around.

00:46:29 So what would prevent you to reach hundreds of thousands of vehicles?

00:46:34 And that’s a good, that’s a good rephrasing.

00:46:36 Yeah.

00:46:37 So the, I’d say the, it seems the consensus among the people developing

00:46:47 self driving cars today is to sort of start with some form of an easier

00:46:51 environment, whether it means, you know, lacking inclement weather or, you

00:46:55 know, mostly sunny or whatever it is.

00:46:57 And then add, add capability for more complex situations over time.

00:47:02 And so if you’re only able to deploy in areas that meet sort of your

00:47:08 criteria or the current domain, you know, operating domain of the

00:47:11 software you developed, that may put a cap on how many cities you could

00:47:14 deploy in.

00:47:16 But then as those restrictions start to fall away, like maybe you add

00:47:20 capability to drive really well and safely in heavy rain or snow, you

00:47:24 know, that, that probably opens up the market by two, two or three fold

00:47:28 in terms of the cities you can expand into and so on.

00:47:31 And so the real question is, you know, I know today if we wanted to, we

00:47:35 could produce that, that many autonomous vehicles, but we wouldn’t be

00:47:38 able to make use of all of them yet.

00:47:39 Cause we would sort of saturate the demand in the cities in which we

00:47:43 would want to operate initially.

00:47:46 So if I were to guess like what the timeline is for those things falling

00:47:49 away and reaching hundreds of thousands of vehicles, I would say that

00:47:53 thousands of vehicles, maybe a range is better, I would say less than

00:47:57 five years, less than five years.

00:47:58 Yeah.

00:47:59 And of course you’re working hard to make that happen.

00:48:03 So you started two companies that were eventually acquired for each

00:48:07 four billion dollars.

00:48:09 So you’re a pretty good person to ask, what does it take to build a

00:48:12 successful startup?

00:48:15 I think there’s, there’s sort of survivor bias here a little bit, but

00:48:19 I can try to find some common threads for the things that worked for

00:48:21 me, which is, you know, in, in both of these companies, I was really

00:48:27 passionate about the core technology.

00:48:29 I actually like, you know, lay awake at night thinking about these

00:48:31 problems and how to solve them.

00:48:33 And I think that’s helpful because when you start a business, there

00:48:36 are like to this day, there are these crazy ups and downs.

00:48:40 Like one day you think the business is just on, you’re just on top of

00:48:43 the world and unstoppable.

00:48:44 And the next day you think, okay, this is all going to end, you know,

00:48:47 it’s just, it’s just going south and it’s going to be over tomorrow.

00:48:49 And and so I think like having a true passion that you can fall back

00:48:55 on and knowing that you would be doing it, even if you weren’t getting

00:48:57 paid for it, helps you weather those, those tough times.

00:49:00 So that’s one thing.

00:49:01 I think the other one is really good people.

00:49:05 So I’ve always been surrounded by really good cofounders that are

00:49:08 logical thinkers are always pushing their limits and have very high

00:49:11 levels of integrity.

00:49:12 So that’s Dan Kahn and my current company and actually his brother and

00:49:16 a couple other guys for Justin TV and Twitch.

00:49:18 And then I think the last thing is just I guess persistence or

00:49:24 perseverance, like, and, and that, that can apply to sticking to sort

00:49:28 of, or having conviction around the original premise of your idea and

00:49:32 sticking around to do all the, you know, the unsexy work to actually

00:49:36 make it come to fruition, including dealing with, you know, whatever

00:49:41 it is that you, that you’re not passionate about, whether that’s

00:49:43 finance or, or HR or, or operations or those things, as long as you

00:49:48 are grinding away and working towards, you know, that North star

00:49:51 for your business, whatever it is, and you don’t give up and you’re

00:49:55 making progress every day, it seems like eventually you’ll end up in a

00:49:57 good place.

00:49:58 And the only things that can slow you down are, you know, running out

00:50:00 of money or I suppose your competitors destroying you.

00:50:02 But I think most of the time it’s, it’s people giving up or, or somehow

00:50:06 destroying things themselves rather than being beaten by their competition

00:50:09 or running out of money.

00:50:10 Yeah.

00:50:11 If you never quit, eventually you’ll arrive.

00:50:13 So, uh, it’s a much more concise version of what I was trying to say.

00:50:16 Yeah, that was good.

00:50:18 So you went the Y Combinator route twice.

00:50:20 Yeah.

00:50:21 What do you think in a quick question, do you think is the best way to

00:50:24 raise funds in the early days or not just funds, but just community

00:50:30 develop your idea and so on.

00:50:32 Can you do it solo or maybe with a co founder with like self funded?

00:50:38 Do you think Y Combinator is good?

00:50:40 Is it good to do VC route?

00:50:41 Is there no right answer or is there from the Y Combinator experience

00:50:45 something that you could take away that that was the right path to take?

00:50:48 There’s no one size fits all answer, but if your ambition I think is to, you

00:50:53 know, see how big you can make something or, or, or rapidly expand and capture

00:50:57 a market or solve a problem or whatever it is, then, then, you know, going to

00:51:01 venture back route is probably a good approach so that, so that capital doesn’t

00:51:04 become your primary constraint.

00:51:07 Y Combinator I love because it puts you in this, uh, sort of competitive

00:51:12 environment where you’re, where you’re surrounded by, you know, the top, maybe

00:51:16 1% of other really highly motivated, you know, peers who are in the same, same

00:51:20 place and that, uh, that environment I think just breeds breed success, right?

00:51:26 If you’re surrounded by really brilliant, hardworking people, you’re going to

00:51:29 feel, you know, sort of compelled or inspired to, to try to emulate them and

00:51:33 or beat them.

00:51:34 And, uh, so even though I had done it once before and I felt like, yeah, I’m

00:51:39 pretty self motivated.

00:51:40 I thought like, look, this is going to be a hard problem.

00:51:42 I can use all the help I can get.

00:51:44 So surrounding myself with other entrepreneurs is going to make me work a

00:51:46 little bit harder or push a little harder than it’s worth it.

00:51:50 And so that’s why I, why I did it, you know, for example, the second time.

00:51:53 Let’s, uh, let’s go philosophical existential.

00:51:56 If you go back and do something differently in your life, starting in the

00:52:02 high school and MIT leaving MIT, you could have gone the PhD route doing the

00:52:07 startup, going to see about a startup in California and you, or maybe some

00:52:13 aspects of fundraising.

00:52:14 Is there something you regret, something you not necessarily regret, but if

00:52:19 you go back, you could do differently.

00:52:21 I think I’ve made a lot of mistakes, like, you know, pretty much everything

00:52:24 you can screw up.

00:52:25 I think I’ve screwed up at least once, but I, you know, I don’t regret those

00:52:29 things.

00:52:29 I think it’s, it’s hard to, it’s hard to look back on things, even if it didn’t

00:52:32 go well and call it a regret, because hopefully it took away some new knowledge

00:52:36 or learning from that.

00:52:37 So I would say there was a period.

00:52:44 Yeah.

00:52:45 The closest I can, I can come to is there’s a period, um, in, in Justin

00:52:48 TV, I think after seven years where, you know, the company was going one

00:52:54 direction, which is towards Twitch, uh, in video gaming.

00:52:56 I’m not a video gamer.

00:52:58 I don’t really even use Twitch at all.

00:53:01 And I was still, uh, working on the core technology there, but my, my heart

00:53:04 was no longer in it because the business that we were creating was not something

00:53:07 that I was personally passionate about.

00:53:09 It didn’t meet your bar of existential impact.

00:53:11 Yeah.

00:53:12 And I’d say I probably spent an extra year or two working on that.

00:53:16 And, uh, and I’d say like, I would have just tried to do something different

00:53:20 sooner because those, those were two years where I felt like, um, you know,

00:53:26 from this philosophical or existential thing, I just, I just felt that

00:53:29 something was missing.

00:53:30 And so I would have, I would have, if I could look back now and tell myself,

00:53:33 it’s like, I would have said exactly that.

00:53:35 Like, you’re not getting any meaning out of your work personally right now.

00:53:38 You should, you should find a way to change that.

00:53:41 And that’s, that’s part of the pitch I use to basically everyone who joins

00:53:44 Cruise today, it’s like, Hey, you’ve got that now by coming here.

00:53:47 Well, maybe you needed the two years of that existential dread to develop

00:53:51 the feeling that ultimately it was the fire that created Cruise.

00:53:54 So, you never know.

00:53:55 You can’t, good theory.

00:53:56 So last question, what does 2019 hold for Cruise?

00:54:00 After this, I guess we’re going to go and I’ll talk to your class.

00:54:03 But one of the big things is going from prototype to production, uh, for

00:54:06 autonomous cars and what does that mean?

00:54:08 What does that look like?

00:54:08 And 2019 for us is the year that we try to cross over that threshold and reach,

00:54:14 you know, superhuman level of performance to some degree with the software and,

00:54:18 uh, have all the other of the thousands of little building blocks in place to,

00:54:22 to launch, um, you know, our, our first, uh, commercial product.

00:54:26 So that’s, that’s, what’s in store for us or in store for us.

00:54:28 And we’ve got a lot of work to do.

00:54:31 We’ve got a lot of brilliant people working on it.

00:54:34 So it’s, it’s all up to us now.

00:54:36 Yeah.

00:54:36 From Charlie Miller and Chris Vells, like the people I’ve crossed paths with.

00:54:40 Oh, great.

00:54:41 If you, it sounds like you have an amazing team.

00:54:44 So, um, like I said, it’s one of the most, I think one of the most important

00:54:48 problems in artificial intelligence of the century.

00:54:50 It’ll be one of the most defining, the super exciting that you work on it.

00:54:53 And, uh, the best of luck in 2018, I’m really excited to see what

00:54:58 Cruz comes up with.

00:54:59 Thank you.

00:55:00 Thanks for having me today.

00:55:01 Thanks, Carl.