Loading summary
Sophia Tung
Pickup and drop off. It's not a solved problem. They will just keep on looping and looping and looping. And sometimes I tell the safety driver like, hey, don't disengage because I want to see how many times it loops around this block. So.
Tim
One of my favorite guests last year was Sophia Tung, who is a YouTube broadcaster and a newsletter writer. She writes for the Ride AI newsletter and before that she was an entrepreneur. She worked on an electronics startup and has a background as an engineer, so knows a lot about how a lot of this stuff works. And last year I had her on the show to talk about her experience visiting China and riding in robo taxis. I was excited to talk to her because she's one of the few people in the West I know who actually got to go and especially at that point, got to go and ride in some Chinese robotaxis and report back and tell us what it was like. So since then she's had additional experiences there and she's also just one of my favorite observers and commentators on the state of the industry. And so I wanted to get an update and see what she's been learning in the last year or so. So, Sophia, welcome back to AI Summer. I think you're our first repeat guest.
Sophia Tung
Hey, Tim. Really, really glad to be here again.
Tim
Great. So for people who weren't here that weren't listening a year ago, tell me a little bit more about yourself, your background and how you got into writing about and talking about the AV industry.
Sophia Tung
Sure. I think, as you described, I'm an entrepreneur of sorts. Before I was an entrepreneur, I was pretty into the corporate style software engineer industry. So I was at Twitter, I was at various startups. This was before it became X and after it did become X, I figured might as well just start my own thing. So I did hardware startups in the electric scooter industry, a lot of R and D and design. We were building something completely new. So what better way to make the new thing than to go to China and try to figure it all out just by ourselves?
Tim
This is one of the reasons I'm always excited to talk to you is you've really gotten your hands dirty. You've gone to China, you had a startup where you're building scooters, and so you had to actually think about the nuts and bolts of what suppliers should I use, how much does things cost, what's the supply chain, how do I get this stuff to various places in the world and assemble it, et cetera. And so you have some big picture Thoughts. But also that's grounded in really understanding how a lot of this stuff works under the hood. And also on the software side, you've written software and so you have some sense for, you know, how the software inside the Navy works as well.
Sophia Tung
Yeah, absolutely. I mean, we had, you know, we had supply chain and we had, we were opening molds at factories. We were on the ground there, like, actually trying to figure out which factories were the best. We went to some real middle of nowhere places just to find casting factories that would actually make parts that were specialized enough to get our product out there. So they were pretty interesting experiences.
Tim
That is very cool. I'm going to ask you more about that a little later. And so I believe you're from Taiwan and you speak Chinese, right?
Sophia Tung
Yes, I am originally from Taiwan. I lived there for 10 plus years of my life. I was growing up. I came to the U.S. i think it was middle school ish, elementary school ish. The ranks don't exactly work out to be the same between Taiwan and America, but I would say middle school, elementary school transition. I came here, but I still go back to Taiwan often. In fact, I was just in Taiwan a couple days ago as of this recording. So.
Tim
And we were recording on May 20th. It'll probably be like a week or so after that when this airs. And so this is one of the reasons I think you were well positioned to do what you did last year, which is you traveled to China and you were able to get access to some Chinese robo taxis. I think you said when you came on last time that you had to, like, provide your photo ID to like, prove who you were. And there's like this, like some bureaucratic obstacles you have to go through before you can, like, even get onto these networks. But you did that and you, you got to ride a robotaxi and it was not a great experience the first couple of times you did it. Right?
Sophia Tung
Yeah. So I. How many have I ridden so far? Oh, man, it's a list. The first one that I rode was the fifth generation Apollo Go. I rode that quite a few times actually. I also wrote the sixth generation Apollo Go. I've written the two generations of the pony stuff. I've written also two generations of the we ride stuff. And also their shuttle bus, which is, I believe they've just started service of their shuttle bus product in somewhere in Europe. I don't quite remember exactly the country. And I think, yeah, a few months ago, before Ride AI, I tried to hail Didi's newest AV product, which did not go well. That was in Shanghai. And yeah, we can talk about that as well.
Tim
And so Apolligo, that is the Baidu.
Sophia Tung
Yes. Apollo is Baidu's division of Baidu's autonomy.
Tim
And so when we talked last year, you just done the fifth generation ride, which that was the version they saw, the safety driver, and it was a really bad experience. You said they accelerated very jerkily. They would like shift lanes in a way that didn't seem very safe.
Sophia Tung
Yeah, yeah. So all over the road.
Tim
And then you talked to them and they said, oh, the sixth generation is much better. And I guess you went subsequent trip, you went back and you got to try the sixth generation. Give me like a compare and contrast. Like how does, how did that. Did that newer generation of technology compare to the older generation?
Sophia Tung
If you've taken Uber in the US for example, an Uber driver who has, who is his. He's had his first EV and he doesn't really know how to operate like how EVs work. And you know that if you lift off the gas paddle on an ev, it's basically like braking, you know, regenerative braking, that kind of thing. And he's sort of all over the lane, switching lanes all the time, sort of taking sketchy yellow lights. I think that is basically the experience that I got on the fifth generation Apollo Go. That is to say, okay, I am pretty prone to getting motion sick. So that is the, that is the curve that I grade on. Am I getting motion sick in this vehicle or not? And in the fifth generation of Holo Go, I definitely got motion sick. If you, if you go back and watch the video, you can see like my face turned totally white between the start of the trip and the end of the trip. It was not good. I had my boyfriend with me at the time. He also was getting pretty sick and he was filming me. So it just not a great experience all around. In addition to that, we had to go basically to the suburbs of Shenzhen. This was, I believe, in Nansan. And it was about, Let me think. We got there and we waited. So the first attempt that we waited maybe two or three hours or so for a car that it never came, there was a queue that kept on counting down in the app and it would tell you how many people are in front of you. So we had at one point like four people in front of us in the queue. And the app was giving us a time of around 30 minutes or so. I was like, okay, fine. The queue kind of down, the timer kind of down. When it hit zero it reset right back to 30. So that happened about three or four times. So we waited there like two hours, ish, almost three hours. And we just gave up for that day. We took the metro back. Metro was, you know, pretty nice except for the fact that there were security guards everywhere and you had to go through metal detectors and get your bag searched every single time you went in and out of the metro. But that aside, fine. They also got mad at me for filming the metro. So, you know. And then the second time we tried it, it was another area. It did also take a while, but it finally got to us and it was just not, you know, the whole pickup experience was sort of sketchy. We had to wade through some bushes on the side of the road to actually get picked up. There's just not, it's just not very conducive to ride hilling in general. But yeah, that took, I would say it's really cut down in the video, but probably close to an hour to get a car. Any, any other service. I would have walked away by now. But I was just, you know, I was there to try it. I had made the trip out there, so that's what I did. And. And the experience ensued.
Tim
So that was the first, the fifth generation. Then you went back on a different trip to try the sixth generation, is that right?
Sophia Tung
I did. I went back again for something unrelated, but then I went. I would. I figured I would take the trip over to Wuhan, so I did that. Wuhan is basically the city to try it because they're extremely gung ho about letting AVs be on their roads. Not so gung ho about it now. We can talk about that later. But at the time they were, they were very into it and they basically let Apollo go have free reign. So a large swath of the city was in the service area and most of the cars there were the sixth generation Apollo vehicles, which I forget exactly who the OEM is, but they're more or less custom designed, sort of in the same vein that Zeekr custom designed Waymo's new Ojai vehicles. So, you know, they're very nice vehicles. They sort of have the same legroom as the Ojai, you know, massaging seats. They had, you know, in cabin voice control individual like they had a screen for the back passengers that were sort of on the armrest area and then they had a screen up front and they both displayed different things. And that was a way better experience. The drive was a lot smoother. It basically felt almost Chinese human deity level of Driving in that it's fairly aggressive, but not overly aggressive. They kept mostly straight lines and good merging, pretty good behavior in traffic. And because the, the traffic systems are all tied together in China, the car basically knew when the light would turn red, when it would turn green. And so it was very responsive in that aspect of gauging if it should slow down or try to make the light.
Tim
And so you live in San Francisco, you've taken a lot of Waymos. How would you compare the experience to the experience when you have. When you take a Waymo in San Francisco?
Sophia Tung
If I were to equalize the levels of behavior between China and San Francisco, I would say it's pretty on par with how a Jaguar I pace drives today in that it's decently aggressive and it does try to insert itself into traffic. Now, at the time when I was trying it in China, the I paces did not. They were not so aggressive. They've gotten more aggressive recently. You know, behavior changes, so on and so forth. But at the time, I would have said that the sixth generation Apollo were more aggressive than the Jaguars. So that's like 20, 25, we're talking,
Tim
and the Jaguars are the Waymo car. And so the, the Jaguars, the Waymo cars were, you think, too cautious, maybe. Like, they, they've. As they've gotten more confident, they almost kind of turned up the, how much the cars are willing to, you know, merge into, into traffic or go when it's safe to do so.
Sophia Tung
Yeah, there's certainly been more behavior changes in the past year or so, definitely since I had my experiences in China. So, yeah, I would say that they've gotten significantly more aggressive before. I would not say that they were overly cautious. I would say they were adequately cautious. They were driving how I would drive. And I consider myself a pretty cautious driver. Like, I try not to run yellow lights. I, you know, I stop for pedestrians, even if they're not necessarily fully in the crosswalk yet, et cetera, et cetera. I think there was more of that behavior now. It's, it's sort of flipped a little bit. It's closer to driving in the sense that they don't want to annoy other road users.
Tim
That makes sense. And so would you say overall that it's the kind of quality of the ride in terms of, like, safety and smoothness is also comparable? Like, do you feel like The Via the DePaul Go is on par with Waymo overall in terms of capabilities, or did you notice areas where they seem to be behind?
Sophia Tung
Yeah, I think in terms of Safety. And again, this is over the course of an entire day, probably from 8 or 9 in the morning to past midnight. When I was in Wuhan, they. Yeah, in terms of driving safety over the course of that time frame and about six, or actually was it maybe closer to eight trips, eight or nine trips, something like that, it was fairly safe. Now, of course, there have been news that have come out recently about mishaps that have happened, and we can, we can talk about that. So I think there's two main pieces of news. One of them is the most recent one, which is that the entire fleet in Wuhan underwent some sort of network issue. And they all stopped in lane, like in the middle of the road. And in some instances, while initially when they stopped there was not much traffic around, the blackout lasted long enough that there was traffic and in some cases the traffic actually ran into the cars. So that's the first piece of news. And then the other piece of news is another car. It was a sixth generation Apollo Go. And I don't know exactly how the deal went down, but it was operated by hello, which is a company that used to be known as hello Bike, which was a rental bike operator in China. So they had renamed itself to hello in the hopes of operating their own AV fleet and not just concentrating on bikes. And as far as I understand, they were running their own software on Apollo Go hardware manufactured by whoever their OEM was. And that car ended up hitting, I believe, two people on a moped, on a. On a motorcycle and running them over quite severely. And the rumor is now that those people have passed away in the hospital. So I would say the rumor was maybe three months ago. The actual happening was, I want to say, around January, February, I can't quite remember at this point, but it was fairly recent in terms of AV news, I would say.
Tim
So tell me a little bit about the media environment over there. I mean, if a Waymo car ran over somebody and possibly killed them, there'd be a ton of media coverage here. And it doesn't seem like. It seemed like there'd be a clear answer whether the person died or not. Um, I mean, we had, you know, Cruz in 2023, ran into somebody. I think the woman didn't die. I don't think we ever met, learned her name, but I think we did get a pretty good idea that kind of how severe the injuries were and that she didn't die. Is there, do you see that as like press censorship? Or is it that we don't have access to the best sources in China or do you think there's something, something else going on there?
Sophia Tung
I think in China there's several tiers of press. If you're talking about press in the traditional sense, there is only one tier. But if you're talking about press in the sense that news gets out, there's multiple tiers of press. So the traditional state backed media by and large did not really report more on it other than the event happened. There's, there wasn't that much investigative reporting that went on. If you remember the Uber, sorry, the Uber ATG incident of at this point, forever ago, it feels like, yeah, they, so they ran over somebody and killed them with a safety driver in the car and that was large enough of an incident that it totally devastated their, their, their program and they basically sold it all off. I think in China Baidu is large enough, sort of similar or even larger than, than, than Uber size that the state was sort of like, let's, you know, maybe not report on this too much. I don't exactly know how much, you know, censorship if you will, there was in the media, but certainly state level media did not report.
Tim
Okay, that makes sense.
Sophia Tung
Yeah, well, so what did come out of it was there was a lot of social media posts about it that happened because everybody is, you know, if you live in China, you're living your life online. There's no other ways about it. If you try to go out to like a street stand and pay for something with cash, they won't take it because you know, everything's done through like wasting WeChat Pay or like Alipay or something like that. It's all digital. If you don't have a phone in China you're basically screwed. So because of that, you know, everybody is online. Like everybody's on TikTok, everybody's on which is Red note. Everybody is on Weibo, which is social media is basically like, you know, sort of like X Twitter nowadays. So it's hard to not hear about this kind of thing through social media. And it's, everybody has a phone in their pocket, right? Everybody's camera in their pocket. So there was a lot of images, a lot of videos when this happened because this was a pretty busy intersection that had happened. So the non state level media did sort of have a field day about it. And there were people who went and investigated in the hospital, like what happened to these people that got run over. And there was never an official confirmation but it seemed like they were no longer at the hospital. And the implication was that they might have, you Know, like passed away or they've been moved or like something happened to them. And that's about as much as we know.
Tim
So what's the experience like for the Pony and we ride rides you took? How do they compare to Apollo or to the Waymos you'd be able to ride in San Francisco?
Sophia Tung
The ponies that I've taken are. They were gas car. They were like hybrid cars. I don't know if it was tuning or something, but the behavior, you know, on the gas pedal was super aggressive. It would, it would oftentimes be the first car off the line at a red light, like, you know, trying to get in front of all the other cars. It, you know, it merged fairly aggressive. I would, I would compare it to like a fairly aggressive D driver. However, the maneuvers that it made they felt safe versus, you know, sort of wishy washy, like, you know, will it, won't it, sort of lane merging kind of thing. So it was. It was aggressive and confident. I would say that's how I would rank Pony versus we ride. It was fine. The buses were less confident than I would have liked them to be, especially around bicyclists, you know, trying to merge the next lane over to avoid a bicyclist in the lane that we were currently in. That behavior was just okay. Sometimes it would try to do it and then decide not to do it. You know, not something that you really want to experience halfway through a merge. You just merge right back to where you were. But experience was just okay. I would say. So between the two, if I had a choice in the same service area, I would probably choose Pony over we ride.
Tim
And when you choose Polygo over both of those.
Sophia Tung
Oh, yeah, for sure. Yeah. The experience is a lot more polished there.
Tim
And then Waymo above any of them. Or Polygo before Waymo, I think in
Sophia Tung
terms of driving behavior. Yeah.
Tim
So what about here in the U.S. i mean, the company people probably talk about the most after Waymo is Tesla. How much time have you spent riding in Tesla vehicles, if at all, with the, you know, the Robo taxi program?
Sophia Tung
Yeah, the Robo taxi program in sf. It's small, but I think if you time it right, you can get a car without too much of a weight. That being said, it's often. It's. I was about to say oftentimes, but it's all the times less convenient than just getting a Lyft or an Uber. The way they offset that is that they price. Their pricing, at least in SF is so much lower than Uber or Lyft and I always go for the lowest one because I'm a price conscious consumer.
Tim
Is it like 420 or something like that? They have like a fixed price?
Sophia Tung
No, they no longer have a fixed price. They have been increasing their prices, but it's still lower. So, you know, the same trip on Lyft, maybe If you're paying $12 something, it would be like seven or eight dollars on robo taxi. And that's like a big enough difference that I would go like, okay, well I'm saving $5. You know, I'll take the extra 10 minute, you know, wait for a total of 17 or 18 minute wait for the robo taxi versus like the Lyft would show up in eight minutes or something like seven or eight minutes.
Tim
And those, those all have safety drivers still, right?
Sophia Tung
They all have safety drivers.
Tim
The other Chinese companies you were mentioning, do any of them have safety drivers? The pony and we ride.
Sophia Tung
No, no, Pony had no safety driver. We ride had. Well, we ride did have a safety driver, but now they don't have a safety driver. And when I went and tested it, they were sort of on the brink of taking out the safety driver. So now they don't have safety drivers and the buses don't have safety drivers.
Tim
And so how do you, how would you rate the experience under the Tesla? If you have these like five companies, the three Chinese ones, Waymo, Tesla, where would, where would you put Tesla in terms of the customer experience?
Sophia Tung
Yeah, honestly, it's hard to beat the smoothness of Tesla driving. That being said, there have been a lot of instances where it's, it's, it's been really yikes. So it'll do a lot of maneuvers that I personally would never do to get really close to people. Few times the safety drivers have had to disengage. There's a lot of exceptions to the service area. So like I live at the end of a cul de sac. You know, Waymo drops me off, no problem. Tesla does not even let me put a pin at the end of the cul de sac. And like multiple times the safety drivers go like, oh, you're down there, right? Like, and they'll disengage the car and drive me down there. And a lot of times like pick up and drop off, just not a solved problem. They will just keep on looping and looping and looping. And sometimes I tell the safety driver like, hey, don't disengage because I want to see how many times it loops around this block. And a lot of times it'll add like half an hour to the trip. I'll just sit there and go like, okay, I've had enough after half an hour.
Tim
Yeah, yeah. I mean, this is something I think people who don't follow the industry closely don't appreciate. And there's both like a lot of these fit and finishing types of things. And it's very helpful to have the safety driver because a lot of the failure modes of an AV isn't that it crashes into something, it's that it gets stuck or it loops around in a circle or whatever. And so if you have the safety driver to intervene real quickly and do something, it seems like it wasn't a big deal. But if you imagine the car being empty and trying to do that, then it might actually be a big problem if it gets stuck for several minutes or it goes to the wrong place or whatever. But then there's also trade offs between smoothness, aggressiveness and safety. Where ideally you want something that's a smooth, confident driver that doesn't hesitate and merges in traffic smoothly and whatever, but given a choice between something that's too confident and maybe crashes sometimes or too hesitant and never crashes, you probably want the more hesitant one. And I think sometimes when people, especially early on when, when people would feel that the Waymo was not as good as a driver as Tesla is because Waymo is super cautious and has a approach that means they never get into crashes. And Tesla is not as cautious in part because there is a safety driver. And so they can kind of count on the human to, to, to rescue them. And so it kind of feels like WMO's, it's like Tesla's a smoother driver. Even though if you actually did an Apple's Tables comparison with no safety driver, like, it would be smoother. But then sometimes the Tesla might crash into people and that's not great.
Sophia Tung
It's really hard to beat Tesla's vehicle control. I think their vehicle control, I mean, it's clearly, it's tuned over. It's been like a decade now of FSD being a public thing and they
Tim
have a ton of data, which probably helps.
Sophia Tung
Oh, absolutely, yeah. I mean they make the entire stack from start to finish and it's hard to not have that dialed in of if they didn't have that dialed in level of control, then I would be like, this is a joke, right? But they do. And FSD is like 98% of the time, it's really, really good. It's just like that 2% of the time that they're trying to solve. And I mean, clearly Waymo solved it enough that they don't have people looking at the cars anymore and sitting in the cars. And Tesla has it. And that's sort of, that's sort of where we are.
Tim
Did you see the story a few days ago? Tesla unredacted their safety reports and there were a couple of teleoperator crashes.
Sophia Tung
I did see that. Yeah. But it's not all the safety report. Right. It's only like a certain time period. And I think they're still being pretty selective about what they're showing the public. If they, you know, from an industry perspective, like I would, I would love if everybody had unredacted reports from the very beginning so that we can track where they were and where they are now. I think that's so important in the safety aspect of everything. But that's not the case. So I think probably the way we look at Tesla now is like, okay, they're going to publish these unredacted reports, but now we need to give them a year's worth of reports to really track how they're doing and how they've improved. So I think that Jerry's still out on that one.
Tim
Yeah, for sure. I thought it was interesting because this is something that, like, I think Waymo has made a point of pride that there's nobody, Waymo said since 2018, there's nobody back at headquarters with like a steering wheel and like brake pedals. That is like driving remotely. There is remote assistance, but it's kind of high level assistance. It's kind of questions or like, is it safe to do this? And Tesla has not made those assurances. And in fact we've seen, I think there was a picture the, around the time they launched. There was a group photo that you clearly see like a video game, steering wheel in the back. And so it seems pretty clear that Tesla does in fact sometimes have a remote driver. And so I believe there are two instances in these. Initially for, for the first few months, Tesla redacted everything. But recently they unredacted some of their reports and there were two incidents where Tesla crashed into them. And Tesla said, oh, don't worry, it wasn't our software didn't run into somebody as one of our drivers. Our remote drivers took over and drove the vehicle and crashed. I don't think either the crashes were particularly severe. I think what they said is that the remote drivers are limited to 10 miles an hour, which is definitely better than driving as freeway speeds or whatever. But I think it's still pretty troubling because two things. One is just like when people say, oh, that when people see the Tesla successfully navigating a bunch of different situations and say, oh, it seems like they're very close to being able to go driverless and scale up. It's like, well, if you don't know when you have a remote teller operation happening, then you can't really tell that much from what situations it does. It doesn't handle because you don't actually know if the software is handling those. But also it does seem like in some ways training a bunch of remote teleoperators to drive safely might be a harder problem than getting the software to work. And I'm just very skeptical they're going to be able to scale up very much in a safe way until they have a better handle on this and can avoid having to have people drive remotely on a regular basis. I don't know. What do you think?
Sophia Tung
Yeah, the remote issue is an interesting one because on the one hand, we want to have as little remote help as possible, but also, maybe there are a lot of situations in which having a human in the loop is the better choice, whether or not you've, you have trust in your system. So there's that portion and then there's. On the other hand, there's companies like V out in Las Vegas, who, their entire business model is built on the fact that they are remote drivers and their drivers will drive a car to you so that you don't have to go pick it up. And like, they're limited to. I think I might be wrong on this, but like 20 miles an hour or something like that, like, it's, it's pretty slow in the grand scheme of things, especially in Las Vegas.
Tim
Yeah, this is a, this is a clever, clever business model. So it's a car rental service. But a problem with car rentals if you're not, unless you're at the airport is, well, you have to get to the car rentals place sometimes. And then that's. Often you're not like, saving yourself any time or money because, you know, you got to get a taxi to the. Anyway, it's just a pain in the ass. And so in this case, they're using autonomy to get to you. Then the customer gets in the car, drives the car for wherever they need to go, and then at the end, the car drives itself off and goes to the next customer, to the depot or whatever. And in principle, this seems like something that could make rental cars much more appealing. And because you can Kind of drive slowly and stay on residential streets and stuff. Conceivably this could be pretty safe because the fast driving and the kind of high density driving is all being done by the human drivers. And you can have this remote teleoperation thing only done in situations that are relatively safe.
Sophia Tung
Yeah, I mean that's your entire business model. Right. I don't know if that can scale either. So they say they can. They, they say that they could conceivably go faster. I don't know if either of those things are true, but they did just recently have received an investment, a pretty heavy hefty investment from, from Grab, which is like a Singaporean super app slash Uber type company. So I mean clearly there's something there, I don't know, like that, the remote driver issue. If for a hundred percent safety we have to keep a remote driver in there, maybe that's the way to go. Like maybe the robots will Never get to 100% coverage of every single edge case that could happen. But I know from financial standpoint it certainly is not the way to go. You don't want to have to pay a person to sit there and shut a remote into your cars and, and drive it.
Tim
Yeah. And we recently had Waymo disclose to, I think a congressional committee its ratio, which is, I think they said that there was, they got 3,000 vehicles. I think they had about 70 remote operators at any given time, which is a ratio about 40 cars to one person, which is, I was surprised at how high that number. That seems like a pretty, a pretty impressive ratio.
Sophia Tung
They have very, very high confidence in their technology, which. I mean. Yeah. Warranted.
Tim
Yeah. And so it's just, it's just hard to believe at some point presumably Waymo could license that technology to rental car companies or to whoever. And so if your business model requires like a one to one ratio of like people to, to cars, like maybe you can make profits for a few years, but pretty soon it's just Waymo or somebody with a way more level of technology is just going to come along and, and like destroy your business. I, I would, I would not be very optimistic about somebody who's, you know, hopefully this is a stepping stone to some other business model or technology in the future.
Sophia Tung
Yeah. Maybe isn't just driving their own cars. Right. They have sort of integrated with other companies. So it's clearly other companies are seeing the value there. I just don't know how long it'll last.
Tim
So you write a weekly newsletter that I really like about the AV industry and you're Kind of keeping track. There's lots of companies out there. We've mentioned, you know, half dozen of the most prominent ones who have you been surprised by or been particularly interested by in the last few months? Particularly ones that maybe haven't gotten as much attention as they ought to.
Sophia Tung
I will say the unsung heroes of this entire industry. It's basically the infrastructure. So you know, companies like Terrawatt who have deals with Waymo and like Move, who just raised it specifically to purchase a fleet of Waymos to operate themselves and then other charging companies like Roxas and that kind of thing. So I think they're really interesting to look at as, as we sort of flesh out the, the actual driving portion of this entire industry. You know, you have to have infrastructure around this. And I think the hardest problem now to solve is the infrastructure problem because nobody, you know, like you, when you think of the car on the road, you don't really think about all the ops that have to go on in the background to get it clean, to get it serviced, to get it fixed up if something happens to it, if somebody throws up in it. And we just had a very viral video of somebody throwing up out of a Waymo's window. But that's like sort of exactly the use case that they're for, right? Like you get, you get too drunk and then you, you have a robot drive you home, right? So all of that operation has to go on in the background and we sort of don't think about it. But I think that's really important.
Tim
Yeah, I agree with that. This is something I've tried to emphasize to people when I'm writing about this, about the kind of timeline for scaling. Everybody had this intuition, I think that well, okay, you figure out the technology and then you push the button or you start the factory, whatever, and then everybody's gonna have a driverless car. But even in the most optimistic scenario where the technology works perfectly, you know, next week, you know, like a lot of, especially with a taxi model like Uber and Lyft's business model is that the driver is also the one that cleans the vehicle, that fills up the vehicle's gas tank, that kicks it in for repairs when it's needed, and stores the vehicle at night. And so those functions are, if you take, if you're taking the driver out, you need to replace all those functions. And it's going to be hard to do a kind of gig work version. You have to actually build depots and charging stations and hire full time employees to do the cleaning and do the repairs and so forth. And so Waymo's doing that. And there's nothing conceptually difficult about it. It's all very straightforward stuff. But it just takes time to hire people. It takes time to get permits to build things. It takes time to convince electric companies to give you the kind of high power connections you need to charge 10 or 50 or 100 vehicles at a time. All that stuff just takes time. And, and so yeah, even in the most optimistic scenario, we're looking at several years at least before these things are ubiquitous in any given city. And obviously there's many cities that haven't even launched in yet.
Sophia Tung
Yeah. So I will say that is one of the areas that China really, really exceeds at because they sort of already have all that infra. You know, take for example when I went to Wuhan, I went and investigated one of their charging depots and they don't charge. Right. They battery swap. Like it, it takes, I don't know, three minutes for a car to get back on the road. That's sort of crazy. We don't have that kind of technology in the US we're still thinking about robots that'll go and plug a car in. Even if you have a robot that goes and does that, your car is going to be off the road for an hour plus. And how many trips have you already serviced in an hour? Right. So that's sort of that kind of design thinking in conjunction with the infrastructure. We simply don't, we don't have that here. And then you were talking about the long lead time for the power company to give you power for the permits to come through. That kind of stuff is already worked out. In China you don't have such long lead times to build things. You just, just sort of just go and build it. If you're a startup, the company gives, or, sorry, the government gives your company a certain amount of grace. You can know people in the government and they'll be like, yeah, you're okay for however many depots you need to build. And it just sort of happens. You don't need to go through. It's probably good that you go through stuff like environmental review and you know, like general, like civic infrastructure review.
Tim
But I mean some, some of that stuff is good. I think we probably do more than is good. I mean it is a little bit damning actually that like the communist country is actually more hospitable to like technological development and stuff. It's like you'd hope that it is
Sophia Tung
a weird juxtaposition huh. You know, you think about like, oh, this, this government, you know, they got to be so locked down. But it's actually not so much the case. They are locked down in certain aspects. But yeah.
Tim
I don't know if you've read, have you read Dan Wang's, Wang's book about the US and China?
Sophia Tung
No, not. I have not.
Tim
It's, it's, it's a great book. He's a like financial analyst who spent a few years in China and he wrote a book about this. And his theme is that like China is run by engineers and the US is run by lawyers, which I think is a good, a good lens to look at it. And it's like, I mean, lawyers serve a purpose, right? I mean they're, they're used to make sure procedures are followed and people's rights are protected. And certainly there are situations where like people's rights in China are not respected and I wouldn't want us to emulate those. But there can also be too much process. I think often there is too much process and that leads to the US like taking way longer than we should to build certain kinds of infrastructure and get certain kinds of projects done. So ideally we could have meet somewhere in the middle. It'd be a little bit more permissive here, but still have more rights being protected in China.
Sophia Tung
Yeah, it's always that fine line. I mean we're not even just talking about the AV industry anymore. This is just in general how different governments operate. Right. To get something built here. I mean we're talking the example in sf, which is probably one of the worst cities because we have so much red tape just around getting things done is like this like Mission Rock area, which I think the new Visa headquarters are. That took like 20 something years. 30. 30 years maybe. I don't know exactly the timeframe, but it's like in the decades, I mean you're talking about just building a building and it's, it's taken two decades. Like, come on guys.
Tim
So, so I asked you about the companies you're most excited about and you gave a very smart answer about talking about infrastructure and stuff. But in terms of like customer facing, like AV companies, technology developers, who do you after Waymo and Tesla, who do you expect to be in the top tier of companies building? Because I think my guess is there's going to be a shakeout right now there's probably 10 or 15 companies building this technology and it seems like the sort of thing will be a top two or three or four. Who do you see As a contender for that top tier of platforms that everybody's going to build this technology on in the West.
Sophia Tung
I think the sensory companies are really interesting, but if you're talking about consumer facing directly, it's gotta be so Zoox recently expanded their service area in San Francisco and in la. I think they're, they're going pretty slow at it, but I think they're probably doing it to be safe. And I've recently gotten access to the expanded service area and it's starting to not look like a toy service anymore. I mean, in my videos I joke about, you know, with, with, with other engineers. Like their service area is so small, it's basically a toy. You know, you, you just, you just do a, a joyride from one end of the Mission to the other and then that's sort of it. And it's like a 10 minute trip. But they've expanded their service area to half of SF basically like so you can take a trip from the Mission to North beach. It's about like 25 minutes or so, half an hour and it's, it's like a legitimate service that you can use. The wait times are still not great and they don't have the car availability, but I think, you know, it's, it's just a matter of scale from here. It seems, you know, from the trips that I've taken recently that they've stepped up their vehicle control and their routing. They still can't go over bridges yet, especially drawbridges. I don't think they have the high enough confidence to determine if the road is physically there or not. Because if you think about drawbridges, like what if the barrier fails, right? Like what if the stoplight fails? Your car has to know that there in fact is no road. No road in front of it means no drive. They don't have the confidence for that yet, but I'm sure they will get there.
Tim
Wouldn't the LIDAR be able to tell that?
Sophia Tung
Yeah, it would. I just think they just don't have enough data in their training set for that specific instance and they need to figure it out. I don't know, in simulation or, or in real life. We'll see.
Tim
So one of the running debates in this industry that I think you have smart thoughts about is the question about like custom built vehicles. So Waymo's approach is you just buy. Well, actually the new model is a little bit of an edge case, but at least with the fifth generation, the ones are mostly on the streets now. They buy the Jaguar, I pace, they Add some sensors and some stuff to it and send it out. And the nice thing about that, there's a regulatory advantage in that because the underlying car already has regulatory approval, there's no additional work you need to do to get it at the federal level to get it approved. Whereas you have a company like Zoox where what they're doing is they built something from scratch that has no steering wheel. And it's kind of cool actually. The way they build the Zoox vehicles is that they are symmetrical in the front and back. And so there's no front to the vehicle. And so if it wants to go the other way, it doesn't have to turn around, it just starts driving the other way. And there's four seats with each. It's like they're all facing towards the center. And I've heard the people, the companies that do the custom bale vehicles, they will tell you there's like fundamental advantages to designing from the ground up. You can do cool things like have it be symmetrical, but also just anytime you build something for a specific purpose, it's just going to be better tuned for that use than if you're taking something else and redesigning it. What's your sense for the pros and cons of that approach? Are you, do you think this is going to be something that kind of comes to bite Waymo? So Waymo right now they have the I paces. The next one is the Ojai vehicle, which is based on a Zeekr vehicle, which I think is more kind of custom built for Waymo, but they still import a Chinese vehicle and then do some retrofitting. And then after that they're going to have a Hyundai vehicle which is based on the Ioniq 5. I think that's also something of a retrofit. Anyway, give me your thoughts about do you think this is going to be an advantage for Zoox and other companies that are doing this custom built route or do you think it doesn't matter or do you think the, the downsides are going to be actually like the Waymo's approaches is a more sensible one.
Sophia Tung
Generally the consumer experience in a Zoox is pretty good. Personally, I'm not so much of a fan of writing backwards, but that's just like a preference thing. I get motion sick more often than not. That's a me thing. So in terms of the holistic experience, I think Zoox is the winner. Even with Waymos Ojai, I mean, you're sort of still getting a car that basically has just four seats instead of five. You can't sit in the driver's seat still. With Zoox, there is no driver's seat. And so the entire car is efficiently packaged specifically for those four people. Up to four people. I think that's sort of going to be where we're headed. I'm a fan of the Zoox custom vehicle. I think it's really well designed. You know, the ride comfort is still better than the Waymo offerings, I think.
Tim
And do you think, how significant do you think the kind of manufacturing side is going to be? I mean, do you think it's going to be like, as. Let's assume that Dukes, like gets the technology figures out and both them and Waymo are kind of trying to scale at the same rate. Is it going to be an edge that they have? The custom vehicles are going to be a disadvantage because Waymo can partner with like more different people or Zoox needs to find somebody to do this specific vehicle or how do you think that's going to play out?
Sophia Tung
I think it's a risk that Zoox has come out with this vehicle and they're, it's the only vehicle in their fleet. Right. Like they, they're staking their entire reputation on the fact that this vehicle is designed for enough people that it works for the majority of their tam. But I think it's a risky bet. I think, you know, we've seen treasury companies like Uber, they have Uber X, right? They have the standard Ubers, they have the Uber XLs, like people's needs are different and designing one vehicle and sort of betting the, like the company on it is pretty risky. On the other hand, Waymo is really beholden to what's available in the market. Right. Like even, even the Zeeker it was originally designed for in partnership with Waymo. But also like zeekr uses the same vehicle in China. It's called the zeekr mix. It's the same chassis, but they've sort of changed up the, the, the outside a little bit. But they, you know, they've done that because I think Waymo is probably not pumping enough volume for them to maintain as a, a, a line in, you know, in their factories just, just for the, just for the Zeekr vehicle. So that's why they probably retooled it for the minivan. I don't know. I think probably what's going to be more efficient for a company finance wise is still going to be work with an oem, probably in the early stages of a design, so that the vehicle that comes out is going to be usable for The AV industry and also for consumers. I don't know if I were building an AV company from the ground up. Yes, a totally custom vehicle that we manufacture ourselves entirely is super cool and it's very eye catching, but I don't know if that's the route that makes the most sense financially for a company to do it. On the manufacturing side, for all these components to have a low bomb cost, a lot of it's going to have to come from outside the US with the current administration, I don't know how that really shakes out. Like when, when we were looking at component costs and manufacturing costs in China and in Mexico, it was like it was the beginnings of the Trump administration and already we were getting hit left and right with like tariffs and everything. That was just driving up the cost of the bomb. And it was like it almost became not feasible for us to do it entirely in China. So we ended up going like, okay, we're going to have final assembly in Europe and also in Mexico, and batteries come from other places and we had deals with companies in Taiwan that was going to potentially supply us with cells, and it was just all over the place. I think that's pretty tricky to manage. I wouldn't want to be the one managing it now in this current geopolitical climate. I think Zoox has a tough job ahead of them. If they're going to try to do assembly in Fremont like they say they're going to, and try to manage this global supply chain, it's no easy task.
Tim
I mean, this is going to be like Waymo is going to have this problem too, right? Because the zeekr is a Chinese company and I've heard different differing accounts of what their strategy is going to be for minimizing those tariffs. But, like, everybody's going, to some extent, I think, going to be navigating tariff complexities as they try to, like, launch robo taxi services in the U.S. yeah,
Sophia Tung
that is the reality. I think because of this, the US is probably going to be able to move slower than other countries. Like, not even China. Like, we're talking about, like Europe. Europe is probably. Or Canada is probably even going to move faster than the US can because of the complexity.
Tim
So I'd like to hear your thoughts about the Chinese supply chain a little more generally. So in addition to writing about autonomous vehicles, I've been getting interested in robots recently, and it's really striking. I mean, so like the drone market globally, dji, I think, is a clear winner, clear leader there, and that's a Chinese company. And then you have a lot of robots. You've got companies like Unitree that make these really cheap, really cheap unit robots as well. And you having done a scooter company where you're trying to source Chinese parts, you probably have some insight into this. What do you think explains why China is so good at building robot, like electronic type products at a low cost in a way that really helps them to dominate the global industry, global market sometimes.
Sophia Tung
I mean, the supply chain in it, right? Like the, the entire supply chain is in, is in China. How are you going to compete with that? You know, it, it's, it. Let me put it this way. If you're going to start a company in China today that say you want the company to build, you know, I don't know, humanoid robots that are sort of in the vein of like Unitree robots, right? Or like a robot dog or something like that. Like, first of all, there's 100, 200 other companies that are already doing the same thing. But also second of all, it's so incredibly easy to get started, like form the company. Super easy getting capital. There are government grants, there's a bunch of stuff that you could be doing. It's super easy to form a company and get at least some amount of capital. And depending on how scrappy you are, you can get a product out in, I don't want to say like a few months, but like, you know, half a year, three quarters of a year, like you could be in business sourcing these parts are so easy in China. It's like, you know, you go to. People don't even really source from Hua Changpei anymore. But like, you could go to huaqiangbei, which is like one of the largest electronic markets in China, and like source what you need. But also as a Chinese company, it's really easy to get in contact with other Chinese companies. Whereas in the west, my boyfriend's startup, they're doing robotic training and robot arm simulations and stuff as an American company trying to source from a Chinese company, like a Chinese motor company, it's almost impossible. You have to be Chinese, which he is.
Tim
Is that a language barrier?
Sophia Tung
It's a language barrier and it's also a barrier between like, way of doing business. Because the way that Chinese companies do business is like they jump on WeChat, right? Like we're still stuck in like the email the company and hope they reply sort of way of doing business in China. It's just like you jump on WeChat, you know, you send a few messages and you're Sort of you're in. Right? And a lot of times if, if you don't do it in Chinese, they won't even reply to you. But if you do do it in Chinese and you say, like, hey, we're like another Chinese company. We're trying to store some parts, like, they'll invite you over to the factory, like, come have tea, you know, come come look at our factory, come come look at our operations. Like, they're really open that way in China. So it's, it's like a, you know, cultural difference in that sense. People are a lot more willing to do business together that way in China. And then, of course, you start placing orders and stuff. And everybody wants to maintain a relationship in China, so.
Tim
And so the physical proximity is really important, right. It's not just you're not on WeChat or you don't speak Chinese. It's like, if you can go, if you can drive over to somebody's factory in an hour, that's going to be much better than if you have to hop on an airplane and go to our time zones and deal with visa paperwork and stuff.
Sophia Tung
Yeah, absolutely. Absolutely. Yeah. If you're in Shenzhen, that's where a bunch of people are. It's like different, different areas of the country for different industries, too, right? So, like, for scooters, where we were in, like, where are we, like, in Ningbo, and, like, a lot of the motor factories were, like, north of Shenzhen. It's, you know, a lot of the casting factories were like, even more north than that. But they're all sort of centralized in various areas. And if you know what area, like what industry that you need your parts from, then you can just go to that area and, you know, throw a stone and you'll hit a factory that'll probably do what you need.
Tim
That's it for this episode of the AI Summer Podcast, which is edited by Vulgate Media. You can find more of my work at my newsletter@understandingai.org I'll be back soon with more voices from the AD Summer.
Host: Timothy B. Lee
Guest: Sophia Tung, Ride AI newsletter writer & entrepreneur
Date: May 25, 2026
In this episode, Timothy B. Lee welcomes back Sophia Tung, YouTube broadcaster, AV industry commentator, and Ride AI newsletter author, for an in-depth update on autonomous vehicles (AVs), especially focusing on the latest experiences with Chinese robo-taxis, comparative insights with US-based AVs like Waymo and Tesla, and a candid discussion of global regulatory landscapes, industry shakeouts, and supply chain realities. Sophia’s unique background—having grown up in Taiwan, worked as an engineer in both the US and Asia, and ridden in multiple generations of AVs—provides invaluable first-hand perspective.
Timestamps: 01:14–03:43
Sophia is an entrepreneur and former engineer with experience building hardware startups (notably electric scooters), including ground-level R&D in China.
Grew up in Taiwan, speaks Chinese, and maintains frequent connections with both the US and Asia:
“I am originally from Taiwan. I lived there for 10 plus years...I still go back to Taiwan often.” (03:11)
Her access to Chinese robo-taxis is rare among Western observers, allowing for first-hand experience-based reporting.
Timestamps: 04:15–11:03
Timestamps: 11:03–13:02
“There’s certainly been more behavior changes in the past year...They’ve gotten significantly more aggressive.” (12:17, Sophia)
Timestamps: 13:17–19:29
“The traditional state-backed media by and large did not really report more on it other than the event happened...But...social media posts...had a field day about it.” (16:28 & 17:47, Sophia)
Timestamps: 19:29–21:13
Timestamps: 21:22–24:29
Tesla’s SF AV service is small, cheaper than Lyft/Uber, but less convenient due to longer waits.
All cars still require safety drivers.
Ride smoothness is excellent but drop-off and pick-up are unresolved:
“Pick up and drop off, just not a solved problem. They will just keep on looping and looping and looping.” (23:17, Sophia)
Failure modes (looping/inability to complete ride) highlight the current necessity of a safety driver.
Timestamps: 24:29–33:00
“Maybe robots will never get to 100% coverage...but from a financial standpoint...you don’t want to have to pay a person...” (31:08, Sophia)
Timestamps: 33:17–36:16
“In China, they...already have all that infra. ...They don’t charge. They battery swap. It takes...three minutes for a car to get back on the road.” (36:16, Sophia)
Timestamps: 37:58–39:56
Timestamps: 40:29–43:03
“Generally the consumer experience in a Zoox is pretty good. ...I think Zoox is the winner.” (44:10, Sophia)
Timestamps: 42:20–48:55
Timestamps: 49:34–54:01
“The supply chain...is in China. How are you going to compete with that?” (50:18, Sophia)
“Pick up and drop-off, just not a solved problem. They will just keep on looping and looping and looping.”
— Sophia, [00:00, 23:17], on Tesla AV customer experience
“The entire fleet in Wuhan underwent some sort of network issue. And they all stopped in lane, like in the middle of the road.”
— Sophia, [13:35], on a high-profile Chinese AV outage
“The supply chain...is in China. How are you going to compete with that?”
— Sophia, [50:18], on global hardware ecosystems
“The unsung heroes of this entire industry...are infrastructure”
— Sophia, [33:37], on importance of AV operations and support
“China is run by engineers and the US is run by lawyers”
— Tim, [38:29], quoting Dan Wang on why China’s tech sector scales faster