SPEAKER_00: If I was Cruz's CEO at the time, I would have probably been extremely honest, extremely transparent of what happened upfront. The minute that happened, disclose everything to our DMV regulator right off the bat, sit in their office and say, hey, this happened. I understand that's not the right move for us. This is how we see that going forward. How do we work on this together?

SPEAKER_03: Absolutely. Sunlight is the best disinfectant.

SPEAKER_04: This Week in Startups is brought to you by Intercom. Intercom's AI-first service is the best thing to happen to your customers since you. Twist listeners can get 90% off Intercom's platform at intercom.com slash twist. Squarespace. Turn your idea into a new website. Go to squarespace.com slash twist for a free trial. When you're ready to launch, use offer code twist to save 10% off your first purchase of a website or domain. And Assembly AI. Get maximum value from voice data with Assembly AI. Build powerful products and features for your end users on the industry's leading speech to text models. Get 100 free hours to start building

SPEAKER_08: at assemblyai.com slash twist. All right, everybody. Welcome back to this Week in Startups. I've been talking about self-driving cars a whole bunch because, hey, they're starting to become reality,

SPEAKER_11: but there are concerns around safety. And so today's guest from Halo Car reached out to me after I talked a little bit about remote interventions, specifically around the tragic case in California of crews covering up one of their cars, dragging a passenger who had been hit by another car operated

David Friedberg: by a human. And, uh, Waymo got back to me as well when we talked about it on the all-in podcast a couple of months ago. So, uh, I invited Anand Nunda Kumar on the program from Halo Car. How are you doing? Fantastic, Jason. Thanks for having me. Okay. Explain to us what halo.car does and how it's different than what we've seen from the other autonomous driving startups. And of course, we'll play one of your videos while you explain all this. Yeah. But first of all,

SPEAKER_18: we're not an autonomous car company. Instead, we make a hardware kit that can go into pretty much any

SPEAKER_21: vehicle that would allow us or a vehicle operator to remotely drive the vehicle end-to-end. Okay.

SPEAKER_23: All right. So this is a video game. This is grand theft auto. I kid, uh, you can with it, you know,

David Friedberg: with real world consequences. So here we go for people who are not watching the video, go to youtube.com and search for this weekend startups, uh, and halo, uh, car, but here's a halo car. Um, it looks like it has some cameras on it. It looks like it's a Kia. And then there's somebody

Chamath Palihapitiya: literally driving remotely with a, what looks like a video game controller. Am I correct? That's a video game controller. Very similar. Yes. Okay. And this is happening. Halo.car is happening in the real world today. It's live. It's live on public roads. Correct. Okay. Where are you live right now?

SPEAKER_23: What jurisdiction? Okay. So in Las Vegas, I see what looks like a video game player, uh, sitting in a

David Friedberg: very comfortable chair. It looks like a nice, uh, video game person's chair. Um, and they have that classic, I guess, I don't know what the race car games are, but you got that driving and that's connected to the steering wheel. So this requires a human and it requires, um, a really low latency connection. Am I correct that those are the two most important factors here? And also the cameras.

SPEAKER_29: So they can actually see around the car. So all three for sure. Now, the benefit of self-driving

SPEAKER_11: is removing the human from the car. So you remove the expense. Travis famously said at a conference, like the major expense here, uh, after the car and the insurance is the driver. So yes, if the driver's not in the car, Uber would become a lower priced concept here, but this doesn't lower the cost. Does it, or do you envision eventually that somebody in Manila getting paid $1 an hour

SPEAKER_38: will be driving these cars and would the latency then be acceptable? And is that, is that eventually

SPEAKER_18: where this heads? Yeah. So think about it in a different way here. So our mission, the reason why I started the company in the first place is to move the world away from traditional car ownership concept into a fully serviced concept where you push a button, a car will be delivered to you,

SPEAKER_00: wherever you are for you to drive yourself. So once you get to your destination, you don't worry about parking because let's, let's be honest, parking is always a nightmare. So once you get there, you simply hop off, walk away, the car just disappears magically.

Chamath Palihapitiya: Ah, so this is like a, this is a Hertz competitor more than an Uber competitor. Am I correct?

SPEAKER_18: Absolutely right. So the way you want to think about rideshare in general, we want to start off with the car rental model first, because getting a car at the airport is a cumbersome process. We've all been through, you know, terrible experience where you have to get a shuttle bus and go to a random parking lot, wait there for a couple of hours. And then the guy sells you insurance, not the car, right? So you want to skip all that. And as you, I'm saying, what if a car can be delivered to you inside the airport?

SPEAKER_00: That's the first kind of segment that we want to, you know, really improve. Once we do that, that's a low level market where you went in and get in, get familiarized with customers, getting this service out there. And then that starts going to short-term rentals as a next phase, where you don't need to keep the car for two days or three days in a row, just like a normal car rental. You only keep it for a few hours. So you make multiple stops, right? That's the most expensive part about Uber. When you make multiple stops, it just jacks up the price really quickly. But if you have a car for a couple hours, you make all your stops and drop it off in a hotel and walk away, the car's gone. And when you want one, you push a button, shows up again. That's a second phase that we think that we want to get into the market. And the ultimate is the ride share itself. Think like Uber, but you drive yourself. So you have the best of both worlds. You don't have a driver inside the car, right? And you have the convenience of car coming to you, instead of you walking 20 blocks in crazy heat, like here in Vegas, or in a crazy rain, to get to a car and you can unlock it. You have to call a customer service agent and like, just dealing with all this kind of crazy overhead of having a car.

SPEAKER_23: Yes. And if you were to do this with humans,

David Friedberg: I had a service like this that I used when I lived in Los Angeles, where I would drive to the curb and somebody would jump in the car and take my car, put it in parking. Yes. And then when I was done, they would do the reverse. They would come and bring me my car. I think it cost me like an extra 50 or $100 to do this.

SPEAKER_11: And I don't know if that service still exists, but it was a pricey service. I'll tell you that. And we had one in here in San Francisco called Lux, L-U-X-E. Yes. Yeah. And I don't know if you remember that one, but I would drive and I would normally pay $25 to park in the tenderloin. Plus I would give the guy five bucks on the way in, five bucks on the way out. So it was 35 bucks. And they started doing it for 30 bucks to have a valet waiting in a valet blue jacket outside my office, hand them the key, they would park. And then I would tell them, Hey, wait downstairs. And they would wait for up to, I think, 20 minutes for you. Right. That's right. So that was kind of nice, but it's really expensive. That went out of business because I think it was too

SPEAKER_18: pricey. Yeah. So economics there is very difficult because think about this guy has to be in the location where you're going to drop off the car number one. So they have to get there somehow, and then take over the car from you, go park in a random place, and then they have to get a ride back to wherever the other customer is going to drop off. The next customer is going to drop off the car.

SPEAKER_00: So just the shuttling the actual driver from one car to another is a very big hassle, right? And the other part is now you need to have physical labor that is present in a very high density city

SPEAKER_52: that is extremely expensive to live in already. So instead, what if the driver could be in Vegas,

SPEAKER_54: but repositioning cars in San Francisco, ah, so, okay, hold on here. Now we're talking living in San

David Friedberg: Francisco. If you put a job description out for a 75, sorry, to somebody who put a job description out for $75,000 recently in San Francisco. And they said they got like hundreds of replies. Yeah.

Chamath Palihapitiya: The majority of which were, are you insane? $75,000? How would anybody live in San Francisco on $75,000?

David Friedberg: Now this was for like an administrative position that Athena wow.com could have done for you for $36,000 a

SPEAKER_11: year remotely, but they're in Manila and the cost of living in Manila is quite different. So a driver in San Francisco needs to make, gosh, I don't know, like 90K, 90K. Okay. In order to just have like a basic sustenance as a living. Whereas a driver in, I would guess, Las Vegas could live off of half that

SPEAKER_18: two thirds. Oh, easily. But 55, 60K. That's like a good capital to live in here because the cost of living is so much lower. You don't have state tax and it's such a good place to live. It's a smaller city compared to like a high density city like San Francisco. That's why we believe that remote

SPEAKER_00: driving option could be leveraging American jobs that are here in America. So we can actually very efficiently reposition vehicles. That's one of the reasons, right? Why you want to move labor into a very affordable environment. The second point where I think that we're going to be leveraging is a single remote pilot can now multiplex into a fleet of vehicles. Oh, say more. Okay. So like, for instance, now you're driving a single car. So they literally push a button, connect to a vehicle and move that to a customer and end that job. The minute they end the job as another job is queued in the background. It's a totally different car. It could be in a totally different city. Ah. So now they're, think about this note, eight hour shift, they're repositioning 50 cars.

SPEAKER_68: Hey startups, you've probably heard the saying, better, faster, cheaper, pick any two. Well, AI has changed that equation and Intercom is all in on AI. Intercom built its first AI customer service platform so that no company has to compromise when it comes to caring for their customers or saving money and controlling costs. These things are important. Intercom's AI tools are so smart that 50% of their customers have their issues resolved in just seconds. That's thanks to Finn and AI co-pilot. Yes. Intercom has an AI co-pilot. It's called Finn. Think of it as like a personal AI assistant for each of your customer support agents. So they get really, really good at their jobs. It makes them so much more effective. Finn is a big deal. It's going to boost your agent efficiency by 31% or more. And managers of customer support teams that use Finn are reporting happier customers and very, very productive teams all while staying on budget. That means when it comes to faster, cheaper and better customer support, you can pick all three, no sacrifices. And that's all thanks to Intercom's AI first software. So if you're an early stage startup, if you're a high growth startup, you're going to get access to Intercom at a massive 90% discount. That's right. 90% off because Intercom and the team over there love startups and innovation as much as we do. So go to intercom.com slash twist to apply. Or if you prefer, you can just do an email at startups at intercom.io.

SPEAKER_23: All right. So this makes a lot of sense. If you drop off the car, you pick it up at the Vegas airport.

David Friedberg: Now, all of a sudden my screen goes pink. And now I'm in a new car in a, in a parking lot somewhere. And I'm driving it to a residential home, a studio apartment in Vegas, 900 a month, a one bedroom, 1100 a month, two bedroom. You get the idea. A studio in San Francisco, probably 2,500, a one bedroom, probably 3,500, two bedroom, probably 5,000. So the cost of living is probably roughly a third in Vegas. And so you specifically mentioned, Hey, this is for American jobs. Well, America's got the lowest unemployment of our lifetime. Are you able to find drivers? And then what do you think you'll wind up paying the drivers and would you allow them to do this from home or do they have to come to a facility? It looks like they're in a facility right now, obviously, cause you're in beta, but eventually could somebody be doing this from home?

SPEAKER_18: Absolutely. That's the whole vision for us in the future. For start off, there are plenty of folks that are unhappy with their current jobs, that are getting paid very low wages, not very happy because they don't have a future or career path to go accelerate themselves. What we're doing is

SPEAKER_00: very different. Think of like an Uber driver that is stuck in a car dealing with crazy customers all day long and putting wear and tear in their cars, putting miles in their cars. Instead, if they just sit in a seat office, nine to five job, full pay, healthcare, you have all this advantage. Now

SPEAKER_18: you can actually just move cars around without dealing with crazy customers.

SPEAKER_77: Amazing. So you see this as a 20, $30 an hour job working from home or something?

SPEAKER_78: Exactly. Exactly. Like that. Yes. Something like that.

SPEAKER_00: Something like that. Exactly. So we're next phase where we're going is full VR. So we just want to ship them a kit so you can sit home and drive a car. The inspiration actually came from my mom way back

SPEAKER_79: in India. We don't come from a rich family. We come from very, very poor family. And my mom used

SPEAKER_81: to do medical transcription way back in the day. I remember that was like a very big deal. We would ship

David Friedberg: paper or faxes, digitized paper all the way to India for legal and for healthcare. And they would do

SPEAKER_77: data entry, right? That was like a major, major in the eighties and nineties business in India.

SPEAKER_00: Yes. Yeah, exactly. And it was very efficient where what happened instead of shipping paper, they would just ship an audio file and my mom would have a, a small little foot pedal where that would do a rewind, play, pause, and forward of the audio clip. So she can sit and type the whole day. And she was making decent living wage off of that. And why can't we do that for a remote pilot where we can ship a kit, they sit home and reposition cars, especially with sit home moms that are fantastic drivers, right? Let them make an extra penny here and there, extra buck here and there

David Friedberg: sitting home while taking care of kids and family. Now, what about combining this with self-driving? So, um, if they're self-driving in the car, having a safety driver seems to be wise. So have these other self-driving car companies come to you and said, Hey, you know, we're Waymo, uh, or cruise or whatever. I suppose they would want to do it themselves, but there's a lot of people making self-driving car technology today. So do you ever see a hybrid where, you know, I'm driving from home, I've got my headset on, or maybe I've got 10 cars. I'm watching a fleet of 10 cars driving around. And it has a percentage certainty, you know, maybe this car in Boston, it starts snowing and it's like, Hey, this is at 50% certainty of safety and slow down. Hey, everything going on here in Arizona, California, smooth sailing, Texas, no rain, perfectly dry. They're all at, you know, 99.59. And then that driver can just jump on and monitor the most acute situations. Is that where this heads?

SPEAKER_03: Absolutely. Love that. I'll give you a little bit of context here. Before starting this company,

SPEAKER_00: I was, uh, I was leading perception teams for Uber at the self-driving unit. So we would constantly see problems with our autonomy stack, constantly see edge cases with our autonomy stack. So traditionally, just like you mentioned, all these edge cases are handled by a safety driver into the car, right? And then what would happen is we would offload terabytes, almost petabytes of data from a single day's drive. And then it'd be an army of people that have to go through all the interventions and try to see and understand what really happened and then

SPEAKER_18: retrain all of our models for that specific edge case. The thesis behind why I started the company is, I don't see automation happening overnight. We don't see level four, level five happening overnight. It'll be a gradual progress towards full autonomy at one point. That's how I see automation happening.

SPEAKER_00: But even when automation happened, even when level four is there, we will still need humans in the loop. So that was a world I'm building for today.

David Friedberg: Tell me about, there have been two instances, um, that you are I'm sure acutely aware of. One of them is the crews, uh, dragging a pedestrian after it was hit by a human. And then obviously there was the tragedy where, uh, an Uber safety driver was playing candy crush. I believe as dark as that is. And, um, I don't know if you were there during that, uh, terrible time, but, uh, the safety driver literally, um, didn't do her job. So take me through those two instances and what the industry has learned

SPEAKER_11: from them and has that set the industry back? And if so, how much?

SPEAKER_94: So I was there in Uber times. It was one of the darkest days for us over there. Um, we learned a

SPEAKER_29: lot from it, right? It's a, it's a very deep, uh, situation that happened with industry. That was a wake,

SPEAKER_18: wake up call for all of us. What we really understood at that moment was there are enormous amount of edge cases that a single sensor would not just solve for people came out and saying, oh, if you add a lidar, it was a solid, but that was not the case. When we have so many sensors in there, it is all about how do you identify that current object and then push it down to the entire entire stack. I'm going to go a little bit of a ramble for a second. So I want to talk about two different forms of autonomy that that's happening. There's the first phase of autonomy, which is there are four modules

SPEAKER_00: to it, a perception, prediction, motion planning, and vehicle controls. That's a traditional model. That's how Waymo operates. That's how we built it at Uber. And there's a new wave, um, especially the, Tesla's 12.0 release, which is end-to-end deep learning, end-to-end machine learning models. That's where the future will have to go towards because in the first phase of version 1.0, anytime we see an edge case, you have to take over by human, collect all the data, and an army of people have to intervene and understand what really happened. What was the stack? Where did it fail? Did the perception fail? Did the prediction fail? Did motion planning fail? And then retain those nodes to actually understand that edge case. But this is a complex process. It's extremely heuristical, right? That's why it was so difficult for us to solve for edge cases. What's happening now is very, very interesting where you're pushing all of the learning to the edge. That is, let the car by itself understand what the edge case is. Let's not have a human teach what the edge case is, but let's try to supervise that with data that we can give context that we can give. So let the models understand that solve by itself and get over it. That's the right way to do it for self-driving in general, right? But if we remove the driver from the car, it gives us ability to

SPEAKER_78: scale that up law faster with a remote human that is able to supervise and push the autonomous down.

David Friedberg: Got it. Where are we at with autonomy? If we were to look at, say, the, you know, for every million rides that occur in a city here in the United States, right now, obviously, for every million miles, it must be a fraction or be doing being done autonomously, like way less than 1%, correct? Mm-hmm. That's fair. Between Waymo? Yep. That's right. When does it get to, in your mind, 10% of rides, 10% of Uber and Lyft rides, ride-sharing rides, move over to autonomy, yours, Elon's, Waymo's, Cruises, whoever. When would we hit 10%?

SPEAKER_18: This is a very hard question to answer, right? Okay, that's why I asked it, yeah. You're not going to ask easy questions. So I would say in this game, Elon's probably the furthest along. Primarily Elon is. Tesla's probably the furthest along in my humble opinion. That's primarily because they have the ability to collect fastest possible data than anybody else. They have the millions of cars that are already driving in all over the world,

SPEAKER_00: right? Collecting all these edge cases, number one. Number two is they have the unique ability to run machine learning models or end-to-end models in shadow mode. So even when you're driving a car, the car is technically in simulation mode in the back end, underneath itself. That's a huge power advantage. If you think about the Waymo's of the world, I have two different difficulties with that model. One is your ability to scale because you need to be in every single city, every single area for you to hit the 10% Uber rideshare market. That's a very difficult problem to solve capital-wise. Even though Uber and Lyft has, sorry, Waymo has unlimited check size, right? But each of these

SPEAKER_18: cars are $300,000. 300,000 still. Wow. That's right. Yeah. But they're doing something like

SPEAKER_87: 50,000 rides a week is what I heard. Yeah. But 50,000 rides at like six, seven dollars. Think about

SPEAKER_00: how much is that going to bring back, right? Plus these are not assets. These are depreciating value. Cars, the more miles they put, the low return tasks in it.

SPEAKER_111: Okay. The internet has never been busier. Everybody's creating new websites, new products,

SPEAKER_51: new services, and search is changing. Search engines want to keep visitors captive and AI generated content is creating so much noise that you're going to have to beat. So you need to get your search engine optimization correct. And you can spend a ton of money on SEO consultants, or you can just use your Squarespace's built-in SEO tools and get the best of breed in the world. You are the best person to tell the world about what you're building. So Squarespace has built a wealth of tools to help you reach your audience. Sitemaps, automatic content markup, and clean URLs are just the start of getting your SEO right. Squarespace also has SEO tools for e-commerce sites, right? Which are a whole different genre of website where you've got a lot of SKUs, you've got a lot of items, and you want each one of those to rank. So they're going to help you write SEO-friendly copy for your site, all thanks to their in-house AI tools that you can then tweak. No matter what you're working on, Squarespace can help you get the word out without forcing you to get a PhD in the latest tweaks to Google's algorithm. So here's what you do next. Go to squarespace.com slash twist for a free trial. And when you're ready

SPEAKER_113: to launch, go to squarespace.com slash twist to get 10% off your first website or domain purchase.

David Friedberg: That's squarespace.com slash twist. So you think Waymo has a, and I heard Bill Gurley talk about this on his podcast. They have a unit economics problem. 300,000 for one of those cars, even if it's 150,000, you can only do so many rides in it per day. And if they're charging seven bucks a ride, man to that car has got to cost 5,000 a month to maintain and whatever and supervise. So yeah, I don't know when they hit break even, but it's gotta be a hundred rides or so. Uh, no, it's gotta be a thousand rides. Thousands. Yes. Yeah. Maybe a thousand a month, which would be, you know, 30 a day for several years to just get above that mark now.

SPEAKER_00: Just to break even, right? Just to break even. That's not, doesn't include your HD mapping. That's a whole can of worms. Explain that can of worms. What does that mean? Yeah. So HD mapping is high definition map, which is they have to have a fleet of high definition sensors collecting massive amount of data to localize each object that you see in a, in a, in a road, like lane lines, stop signs, lights, which direction is this light facing? What is that

SPEAKER_123: light? Is it a left arrow light? Is it a straight light? It's a green light. Is it a red light? So there's a lot of cost in maintaining that.

SPEAKER_00: A massive amount of costs. Hundreds of millions of dollars probably a year. For each city, they have to spend hundreds of millions of dollars to map the heck out of it. Really? Wow. That's one. The number two is when cities change, because cities constantly change. When you change a route or change a, put a construction zone in, it's a whole bunch of problems start up. You have to remap everything, redo everything in real time and have humans that are involved in the backend to annotate every single scenario, every single change.

SPEAKER_116: And this is where Elon has an advantage because they've got a million cars out there.

David Friedberg: They put the sensors in every car, whether you buy FSD full self-driving or not, the sensors are in there. So they could be in shadow mode. Who knows what percentage are actually recording every trip and doing this because I would think they have too much data now.

SPEAKER_130: Yes.

David Friedberg: Like how many times do you need to have me drive up and down the 280 or 101 here in the Bay Area to understand how those things work? Because they don't change all that often. And then there are other things that do change. I know Elon had a quote about remote. Let's play that quote and then get your feedback on the other side.

SPEAKER_132: When you take the driver out of the seat for the driverless car, how do you plan to handle

SPEAKER_134: interventions? There are interventions that happen where the car is simply like, I don't know, get stuck down a one-way street whereas with a like crazy construction situation, in which case it'll basically phone home and say, help, I need, I'm stuck. We can probably solve a lot of these things just by, you know, remoting into the car and kind of like steering it like a giant RC car.

SPEAKER_18: So this is what we bet, right? So we've kind of put a very proprietary hardware into a Tesla that allows us to remotely take over the vehicle, shift it in gear, drive it. You literally see this

SPEAKER_142: video right now. There's nobody inside the car. Got it. So here you go. You're driving on a course with a bunch of cones. You've got, you know, five different views. I don't know if those are the

David Friedberg: Tesla cameras or yours. I'm assuming that the Tesla's. Oh, they're yours. So you bolted on more cameras. Yeah, but everybody knows that Tesla's got a bunch of cameras around it. Um, and so yeah, you could be the outsourced takeover group for you'll be the AWS for anybody who needs this as a service remote driving as a service. Boom. And this, where this really makes sense, I think is

SPEAKER_11: trucking. Yes. So have you considered trucking as an option? Because one of the issues with trucks is they have to stop. So the driver can sleep and that sucks. Drivers are pulling over to Starbucks wherever truck stops sleeping for eight hours, waking up. It's just an arduous life and they

David Friedberg: don't get to see their families for weeks at a time. I can see your technology working incredibly because I could hand off. I could do eight hours. 8pm comes around. I've been driving. Now you take over. Maybe you're in another city where it's daytime and you'll be nice and fresh. And boom,

SPEAKER_18: you, you work on your, uh, timeframe. 100%. So there are a couple of things here. I want to talk

SPEAKER_00: about one, especially when it comes to logistics, there are vans and there are big rigs, the articulated trucks both need some form of human, um, intervention when you think about automation. So, so big rigs for sure automation is the way to succeed here, right? The long trips on the freeway going from 15 from California to Texas, those are really good routes that can be fully automated. But the difficulty for big rigs when it comes to automation is how do you get it into the ramp and how do you get it off of the ramp? Got it.

David Friedberg: So autonomy will come to those trucks. They're going to rock it going across the 10 freeway across the United States, whatever, but when they have to get to the local Walmart and unpack, that's where you would take over. Exactly. Amazing. Um, so show us, I know you had another video that you wanted to show of, um, your operation. Maybe we could show that video and you can just give us any more specifics that we missed. Yeah, sure. So I'm going to take my phone

SPEAKER_156: and just walk us through the facility. Oh, here we go. Amazing. Live. And, uh, I don't know who's driving, what they're doing, what jobs they're doing. So we're just going to go right in the drive controls. I love it. Here we go. Take a look at it. So here we go. Oh, this is the first,

David Friedberg: it's a first for, uh, this week in startups, a live view of the drivers. He's walking over

SPEAKER_17: out of his private office and, uh, let's see what's going on in the pit. Here's the pit,

SPEAKER_162: bunch of drivers. All of our, uh, engineers are here. So I'm going to walk you through our drive controls. Ah, it's really done. All right. So we tend to keep it very quiet here because they're focused on driving. So we have multiple pilots at the moment driving. We see this gentleman right here. He's driving a car live on public roads, all in Las Vegas. Amazing. All controls. He has full visibility around the car. In fact, the way we engineered our sensors is they can see better than yourself sitting inside the car. If you think about a car, it's not made for high visibility, it's made to protect you to sit inside the car in the case of an event happens. So you have these A pillars that are preventing you from seeing the edge of where the car lands on the road. So here we don't have to worry about that. We have to give them best possible visibility so they can sit here, literally drive a car. Yep. Far away. Amazing.

David Friedberg: And so they've got their main camera is the dash. They're seeing right across like you would if you were sitting in the car, but they also have five looks like at least five other views and there are six cameras here. It looks like so they can see the side behind the left, the basically all four quarter panels, it seems and behind them. So they theoretically should be able to see somebody who's coming up behind them on a fast motorcycle or somebody tries to cut them off or, you know, etc.

SPEAKER_116: in a way that a human in a regular car would not see. That's exactly right. So they can actually

SPEAKER_162: intervene, take over, drive wherever they want to drive and park a car. They can actually do a lot better parking job than ourselves sitting in the car because they have so much visibility around the

David Friedberg: vehicle. Man, you know what I would tell you? It's interesting. If I could put this in my car right now today, I would pay you like 20 bucks an hour and I would buy, I don't know, 20 hour a month subscription, 400 bucks a month and would love to just have the ability when I'm driving to say, you know what? I got, I got to check my email here. I'm going to be on the road for 40 minutes. I'm more than willing to pay a dollar a minute for somebody else to drive. And I would literally do that. You know, it's kind of like if you have more time than money or whatever, and you, you want to set up a driver or take an Uber, but here I want my car and what an amazing opportunity that would be. So are you thinking of going direct to consumer with this?

SPEAKER_14: Yeah, Jason, I'm going to take it one step further. Imagine two road 2.0s,

SPEAKER_18: a car could be delivered to a peers car. You go buy a Kia from, from a local dealer that comes with a halo trim in it. So you as a customer, you own the car when you want to. And when you don't want to, you push a button, put it in the marketplace. It just goes out earning money.

SPEAKER_142: Okay. So same as, uh, Elon has, uh, outlined in his, uh, auto thing is, Hey, we can all like

SPEAKER_11: Airbnb, like Uber, make a little money with these assets when they're not being used. Absolutely. Fantastic. Um, and so tell me about regulation. You, you're in Las Vegas. Uh, they are very pro tech there. Is there another market you're going to go to next, uh, have people from say, MENA or China or Australia. There are some other places where, you know, geographies, jurisdictions where people are

SPEAKER_18: pretty pro this technology. What city next do you think? Yeah. So Nevada was the first one for us. They're very friendly for autonomous vehicles and futuristic for tech. So we came here, I actually

SPEAKER_00: started the company in San Francisco first and moved here a few years ago, worked with the DMV very closely, drafted the first legislation for them and got it passed. So that was a phenomenal win for us.

SPEAKER_18: Right now we have other states that want to do the same thing. We have, we talked to Florida, we have Texas, we have Arizona, we have Michigan, and now California is extremely open for this.

SPEAKER_00: And we can go deploy in all these states working with our regulators upfront. That's what they want. They want companies to come to us, come to them and draft what the future is going to look like. And we can actually have a path for drafting the legislation right now. Outside America, we totally see countries like Japan coming up right now for us. Think about it. Tokyo is a very density, right? You can't have cars, you can't own cars there. It's very difficult to own cars, right? So you want to have a car share service that is extremely popular, that is very low cost to run. Your OPEX has to be very low. So you push a button, it just gets delivered and you don't have to ever

David Friedberg: worry about parking it. Amazing. And this is where, you know, if we have less cars in parking lots,

SPEAKER_116: you can then reclaim that space. So if you think about the average American home or an apartment building, we have a minimum number of spots you have to have. So just think about your,

David Friedberg: your homeowner, uh, and you have like a two car garage. Um, if I didn't have cars, that two car garage could be turned into literally two more bedrooms in my house. It would be literally put a Jack and Jill bathroom in there. And I have two more bedrooms. You could turn every three bedroom house into a five bedroom house just by taking over the garage. So this could have massive ramifications on the second order effect on things like housing. You know, you got some giant, I remember when I was in a corporate housing, when I first moved to Los Angeles, Santa Monica, there was an equally large building for cars, a garage. That was five stories. You could convert that into another 50 apartments easily. And so this is the future folks. Absolutely amazing. Great job.

SPEAKER_51: All right. You might not realize this, but we're all producing and consuming more voice data than

SPEAKER_68: ever before. What is voice data? That basically means any audio recording, like the one you're hearing right now of human speech, or it could be video content, like the library of video content you have at your company or startup. If you're building a product that uses voice data, you need to check out assembly AI. It is a speech to text model that helps you quickly build new product capabilities based on voice data. And you do this with just a few lines of code. Their speech AI models let you turn consumer calls, virtual meetings, video scripts into insights, action items, and transcripts. Well, you can join over 200,000 developers who are now building with assembly AI. When you hit 200,000 developers building like assembly AI has, that means you're the gold standard. So here's an example. Vita.io makes video editing tools that my team use every week here on This Week in Startups. And you'll see captions over our clips. Well, we wouldn't be able to do that without assembly AI crunching all that audio data in the background. And that's what Vita.io does. They use assembly AI, and that provides the world's lowest error rate and up to 30% fewer hallucinations. So here's your call to action. Build powerful AI experiences for your end users on the industry's

SPEAKER_51: leading speech-to-text models. Head to assemblyai.com slash twist and get 100 hours free. Hey, that's great. Go play with it. You're a developer. You want to play with this. Check out assemblyai.com slash twist today. I didn't ask you about the cruise accident.

SPEAKER_11: What's your take on what they did with regulators? They seem to, according to all reports, have lied or yeah, not been fully honest. And it seems like the cleaned house over there is cruise toast. Have they burned the bridges with regulators? What's the back channel in the industry about what's going on at cruise? What do people say about cruise when they're, you know, in your industry on the back

SPEAKER_79: channel and nobody's listening? Yeah, that's a, that's a sad one for us in the industry. Absolutely.

SPEAKER_18: Um, there's a fine line between building a software company and a very safety critical company where in software side, you can push the boundaries beyond regulations and, you know, growth at all costs and launch at all costs kind of mindset works really well. So you can iterate it really fast, right? Based on any kind of change that happens. But when it comes to, uh, driverless cars, you have a 4,000 pound vehicle that's on public roads. We have to be upfront with regulators. In fact, most of the times even regulators don't understand how to regulate this industry. They're not like heavy researched folks that are coming from this industry. They want to work with private sector because

SPEAKER_00: that involves trust. Correct. Exactly. So this is how I would frame it. If I was Cruze's CEO at the time, I would have probably been extremely honest, extremely transparent of what happened upfront. The minute that happened, disclose everything to our DMV regulator right off the bat, sit in their office and say, Hey, this happened. I understand this is not the right move for us. This is how we see that going forward. How do we work on this together?

David Friedberg: Absolutely. Sunlight is the best disinfectant. Basically for people who don't know, they submitted according to reports, the first part of the cruise car hitting the pedestrian, which was hit by another car. It was kind of like a ricochet type thing. Human hits, human car hits human, pedestrian. That pedestrian then gets knocked in front of a Waymo. No fault of the Waymo. It apparently stopped as quick as it could, but still hit it. But then the car decided to, I think through human intervention, but perhaps not park itself. When it parked itself, it dragged a human being 20 or 30 feet, according to reports. And then when they submitted the video, they didn't submit the dragging part. Essentially they

SPEAKER_11: lied or they omitted the key part of the video, the key learning. And, uh, yeah, they cleared house over there since then. And it is part of, I think you alluded to growth at all costs, you know, dare I say like Silicon Valley, you know, move fast and break things. This is the ultimate failure of the move fast and break things concept. I think this is where I would probably give credit to Waymo where

SPEAKER_00: they have done a very good job working with the regulators and being upfront and disclosing quite

SPEAKER_156: a bit. They had a recent poll, poll dance incident. Let me call it that way. Oh, oh yes. Explain this

SPEAKER_170: one. Cause I, we have a picture of it too. We'll find on the internet. I saw this. Right. The, the one

SPEAKER_00: of the Waymo cars ran into a poll and it makes perfect sense of why it would have, it's probably because that poll was not annotated in the HD map is my guess. I honestly, I need to know the more data of what had happened, but it actually hit the poll and he saw there was a, it's a, it's a bit of an accident where the car was not totaled, but the front was completely behind the pole. They had to come and recover the car and tow it and whatnot, but they were very honest about it. They were very direct to the DMV. They disclosed what happened. They said, Hey, we've, we've identified the problem. We fixed it. The subway fixed it. And it's won't happen again. My guess is they, they're now having

SPEAKER_156: a, a team of folks focusing on annotating every poll in that city. So the car can navigate around

SPEAKER_142: it. Yeah. And here's a picture of, for those of you watching, essentially, you know, when you go down

David Friedberg: these alleys, it's a little confusing because the polls are supposed to be on the sidewalk with a curb. There is no curb in an alleyway in some cases, but there are telephone poles. This thing just had a low speed hit. It looks like, you know, a minor amount of damage. Um, and it was on the way to pick somebody up, but there was another one that got in an accident on the road as well, but I think it was another car hitting it. So almost all crashes so far to date have been,

SPEAKER_11: to the best of my knowledge, extenuating circumstances. These things generally don't cost me problems. As we wrap up here, tell me about, um, inclement weather. There's always videos, you know, you'll see 50 people coming around the same turn in Lake Tahoe from the Bay Area without snow tires on, and they're all going to ski and literally people are standing there watching one car

Chamath Palihapitiya: after another slide on the black ice and get in accidents. Will this technology work anytime soon in fog, rain, ice, snow conditions, or is that just like 10 years out?

SPEAKER_39: Let me try to distinguish in three different ways. First is the automotive industry itself has

SPEAKER_18: actually gotten really good at this, where the car can identify a IC track and enable traction or

SPEAKER_29: disable traction control based on the condition that it's actually going through. Cars have gotten pretty

SPEAKER_00: smart in identifying this number one. Number two is autonomy has to listen into what the car is saying and take that into consideration as one of the inputs outside of just the sensors, right? See what the traction is happening in the wheel, understand that, and use that as one of the inputs on how to handle the environment. The third is when it comes to remotely driven vehicles, humans are much better at

SPEAKER_18: handling weird scenarios like this, where we see the dash light going off or the traction control. We now know how to handle that. Let's slow down. Let's go to the right-most lane where things are much more easier to handle and then slowly get the car, crawl back to the base, because you don't want to be operating in a very snow storm. You want to shut down operations. That's the normal thing to do, because people are not going to be moving around in a very heavy environment like that, right? That's kind of

David Friedberg: how I would bucket it. Got it. Makes total sense. All right. Listen, continuous success. Uh, if people want to try this, uh, next time they're in Vegas, what do they do? Do they go to halo.car and rent a car from you? Or is it like you have to do a Hertz and then you wouldn't know? Actually, how do I try

SPEAKER_156: this? I kind of want to do a shameless plug here. Please plug away. You earned it. You gave me,

SPEAKER_208: they gave me the solid 40 minutes. Plugs are welcome. We are pro tech, pro capitalism here.

SPEAKER_18: So halo car is a, uh, now we're a B2B company. We work with our customers that are car rental companies, car share companies, logistic providers. So our goal is to stack their fleet with our proprietary tech and enable them to remotely drive their own vehicles and their own fleet.

David Friedberg: So if you are a rental car service, small, big, medium, or if you were get around or one of those, yes, get around automatically does like opening and closing of doors. I think remotely. Hey, here's the next phase. We're going to be able to drive it. Continue to success. Everybody go to halo.car and we'll see you all next time on this week in startups. Bye. Bye.