Louis: 0:00

Hey there. Welcome to another Tesla Motors Club podcast. My name is Lewis. I'm Doug. In today's episode, we're gonna cover the recent Q2 earnings call for Tesla. We'll also go over some Tesla Robotaxi updates. There's quite a few of them. Elon is fundraising more for XAI yet again. There's a new Model Y variant, and possibly more. And the Tesla Diner has now opened in LA. All those things and more in this episode 75 starts now. Hey. That was so many things. Just to make a note, we don't have Mike with us today, but he will be back. He had a conflict, so he's on vacation. He is on a vacation with his family right now. We miss you, Mike. We look forward to you coming back. How are we doing?

Doug: 0:48

Doing all right. I'm in Maryland. The last time we had our show, I was in California.

Louis: 0:54

Yeah.

Doug: 0:54

And it was just before they did the robotaxi trials in Austin, where you are. Yes. And so we were speculating how this would go. And so how has it gone?

Louis: 1:05

So first I'll say that no, I have not ridden in it, nor do I have access yet, because as one could guess, they were overly optimistic on when they would open it up to more people. So it's still limited to a pretty small set of influencers and Tesla enthusiasts. They let you sign up. Yes, or signed up. Uh-huh. I just don't have access.

Doug: 1:28

They didn't turn it on. You would think it would increase. A lot of those influencers have gone home now. But I think they only have like less than a dozen actual cars, or at least the last time I checked. And they've increased the service area. At some point, you got to have a critical density of cars, right? If you make the service area bigger, the next guy that plans to use the service is going to be waiting for a while because there's not a car available or it's too far away. So they should get more cars. Of course, in the current setup, we should say, since we didn't know last time we did the show, they have a what are they calling them? A safety monitor?

Louis: 2:06

Yeah. Some kind of monitor or safety supervisor, maybe person in the car.

Doug: 2:11

Yeah. That person sits in the front passenger seat. So they're not in the driver's seat. So you get the spectacle of the car appearing to drive itself. But that guy is there to, I don't know, stop the car if necessary. There are a couple buttons on the screen where you can tell it to stop in lane or in trip or those sort of things. And also the monitor, we'll call them that safety monitor. That person sits in that passenger seat and they have their finger over the door open button. The door opens. Which kind of feels like an emergency stop sort of thing. There's been a lot of talk and speculation about what it actually does, but my guess is it does that, or it could just be as simple as recording like something weird happened here. Let's pay special attention at this moment. So have you been following it all? How do you feel about it?

Louis: 2:57

So my understanding, again, having not ridden in them, there's quite a few interventions that have occurred. I saw some tweets out around Tesla claims there's only these interventions, but we want to get more from other people that have been in them because we think there's been more. So like citizens are trying to collect the data because Tesla hasn't released it or published it yet. But yes, from social media, there's definitely been a number of interventions from the influencer drives that have gotten out. I would say definitely doesn't seem ready for prime time, right? We have Waymo here in Austin, and there's not daily things going wrong with that.

Doug: 3:34

I've heard people aren't reporting on the daily things that happen with Waymo. I'm sure Waymo does weird things, and we have seen such things. I will say I appreciate the influencers. They definitely have a positive bent, but I feel like they've been relatively open and honest about the things that have happened. Joe Tegmar mentioned an intervention where the car was about to drive through a railroad crossing, and he didn't have recording of it, but he reported on it. So I think people have been fairly open and honest. Another one, Kim Java, I think her name is. Like in the previous episode, we were talking about sun glare and how Elon had said, Oh, we've solved this problem. And I'm like, I don't think you have.

Louis: 4:13

We have photon detector cameras, whatever.

Doug: 4:16

He said, Yeah, photon counters, yeah. Or counters, yeah. Yeah, but it looked like her car, and she had video of it. The car did the phantom braking thing, but it seemed like it happened because the sun was right there, so there's some sun glare. Now I'm still on hardware three. You know, our cars haven't really had much of an update, but it feels like the same kind of errors that we have, it feels like those are still happening in the robo taxi. Right. And those things are like the car not being aware of signs, it knows the shape of a stop sign, but it doesn't know no right turn on red, like it's not reading that sign. Maybe there's some map data that it sometimes will know that you can't turn right here, but it's not actually taking in that kind of information from the environment and not really being aware of how to deal with a railroad crossing. I think that's uh that's an important thing to be able to do. And if you look at the visualization, if other people have videos of it, it doesn't pay any attention to the bar that's dropped. It may pay some attention to the flashing red lights that you get at a railroad crossing, but I think it interprets that as guess what, a stop sign, right? Because that's what we normally do. When we see flashing red lights, you treat that as a stop sign. Right. But it doesn't know that, oh, this is actually a railroad crossing. It doesn't seem to acknowledge the bar at all. And then the train going by, at least on the visualization, is interpreted as a bunch of semi-trucks going by, and it doesn't necessarily know where the proper stop is, so it will stop and then try to continue through the bar. Say you have a parking lot that is closed, and often people will close a parking lot by just stringing a chain across. Doesn't seem to know about that. It looks like it's ready to run right through that. That might be a little too small for the cameras to really register as something worthwhile. And these are things that I personally am dealing with. I think one of the first major interventions, it actually wasn't an intervention, they didn't do anything, but it was an error was that the car needed to make a left turn, but the left turn wasn't at the next intersection, it was at the intersection after that. And so the car went into the left lane already. So it's a left turn only lane, but it needs to go straight. So it's in the left turn only lane, and it just continues going straight and it actually freaks out for a bit. Do I go left or do I go straight? Do I go left? And we've seen that sort of indecision before, and then it ends up just going straight. And by going straight, it's now in oncoming traffic for a bit until it gets to the next left turn pocket, and then it takes this left turn. Yeah. You know, I use FSD in my car regularly, and there's a certain part where I always have that error all the time. It's like I'm going to the grocery store, it needs to make a right turn, and it gets into the right lane about 50 yards too early. And okay, but now that's the right turn only to go into some parking lot. Fortunately, there's no curb there, but it wants to just continue through. It's going too early. Yes, you need to be in the right lane, but you're going too early. And uh every time I send a report about it, but I don't think I don't think those reports are going anywhere with my car. I've had no updates, I don't know, for months now. And I feel like the hardware four people are reporting that they're not getting any updates either since Robotaxi launched, because probably they're spending all their time trying to fix things with Robotaxi.

Louis: 7:31

Yeah, to be honest, I feel like the Robo Taxi program that they're doing right now is a good idea in the sense that it's going to greatly accelerate them focusing on the problems. Trial by fire. We all have our cars with FSD and we're there and we correct the issues manually. So you just kind of keep making progress. But when you're doing a robotaxi and there is no driver, you have no choice but to fix every edge case, every intervention that's required. It's definitely going to hopefully accelerate their learning. And we have our fingers crossed that it's going to accelerate their remedy, their improvement, and solving those interventions and those challenges. But yeah, yeah, the problem is when you bundle that with, hey, we're claiming that we're going to get it out and it's going to be lots of people are using it and we're going to release this, and you have Elon time on those deadlines. It's insane.

Doug: 8:24

By the end of the year or something, he said, we'll be able to do unsupervised in our own cars, but probably just in these areas where they've made Robotax available. Now, the interesting thing about that is we've seen these cars with verification technology or something on there, and basically some other Tesla vehicle with a mast on top that appears to have cameras and some kind of LIDAR, maybe they seem to be doing that as a precursor to expanding the area. And I have had friends in California tell me that, oh, we've seen these in the Bay Area as well, going on Page Mill Road, which is right near Tesla's technology headquarters in Palo Alto. I'm not exactly sure what those are because they say they don't do HD maps, but I get the sense that what they're doing is scanning in the environment for their simulation. And then they train off the simulation because then they can create the kind of scenarios that could be fatal in a real test, what they think might be edge cases and try to train in that way. And right. Yeah, at least as Elon has described it. I guess that's what they're doing. My understanding also that they plan to soon expand to the San Francisco Bay Area, which I guess would include Palo Alto, those sort of areas. But there they don't have the actual permits to do a robo taxi type thing. So what they're gonna do is this safety monitor will actually just be in the driver's seat. Sure. So that's gonna be like the Ubers that are already there because you have a bunch of Uber drivers that will drive Teslas and they just do FSD the whole time and they're just sitting there ready to take over. Basically, you have a license to do an Uber type service, but not a robotaxi type service.

Louis: 10:01

You mentioned them basically mapping areas, and your assumption is for simulation and to expand their coverage areas. There was some drama around their expansion map for Austin with Robotaxi. I guess we don't have a picture of that to show, but we can talk about it. It's Elon, right?

Doug: 10:21

Classic Elon. It feels like his emotional maturity is that of a 13-year-old or something. I mean, I don't I don't mean to be too critical, but the way he is vindictive and also what he thinks is funny. It's like, come on, we'll expand it and we'll elongate it. So he had the shape, it looks phallic. But Waymo has correspondingly made their service area larger as well. And yeah, great, compete on that. But again, you have those limitations, you need more vehicles because if you expand the service area, then you don't want your customers waiting because you don't have enough vehicles. I mean, on that note, so in Austin, for you to get a Waymo, you're using Wayne.

Louis: 10:58

You have to use Uber, correct? You don't use the Waymo app.

Doug: 11:01

Right. So there was some news recently of Uber using Lucid Gravity. I think so, yeah. Lucid already has some tech on it, it has cameras already, but they're also adding, as said would say, some hickey d'os from Nuro. And Nuro has been around for a while. I feel like they started with larger ambitions and they scaled down to just food delivery type thing. So not transporting people, but have these things that transport goods or food or something. But they've taken that technology, put that on the Lucid Gravity, and then that will be a robo taxi, assuming Tesla doesn't own that term, a robo taxi type vehicle that Uber will use in addition to these Waymo's. I don't know what Uber's endgame is there. Originally was trying to do this stuff themselves, and it made perfect sense, right? Because once they can get drivers out, then they can make a lot more money. The ROI is much higher. Sure. Of course, their program, which was run by Lewandowski, I think his name is. Yeah. He's been around. He went from Google and Uber. He does an autonomous construction company or something like that now, right? He worked for Tesla too for a while, but then when he left Tesla, I think they sued him.

Louis: 12:11

Anthony Lewandowski.

Doug: 12:13

Yeah. He co-founded Waymo. He was working on self-driving semi-trucks because he thought that was a good spot. And then Uber, I think, scooped him up from that. There's a big lawsuit over it trying to claim that he stole intellectual property. I think both Google and Tesla were suing him at some point. But yeah, Uber had their own self-driving effort. They had test cars with a observer in the driver's seat, but there's actually video. That person was playing with their phone, they weren't paying attention. Somebody was jaywalking, and really they just came out of nowhere in the darkness and the car hit them. I think even if the person responsible was paying attention, probably the pedestrian or uh I think they might have been walking a bicycle, but that person got killed, and that probably still would have happened, but it was such a thing that they quit their efforts. So I don't know what's their end game here. That makes sense for Lucid because they need money, they need to be able to sell cars and show people their vehicles. Like the gravity is a nice car, right? So that actually will be a luxury type experience. Uber has their different levels.

Louis: 13:12

Oh, like the black versus the Uber Black, maybe or whatever.

Doug: 13:15

Or XL and all that. So that might be a nicer experience. Seems expensive. Those Lucids are expensive, and then that extra hardware. You have any insight? Like what is there end-game there?

Louis: 13:26

I think, like you said, at the time early on, like most hyped things, self-driving with AI was so hyped 10 to 15 years ago, all these people were like, Oh, yeah, it's only a couple years away. Oh, within a few years, we'll have it. Not just Elon. I would say people that weren't working on the problem. So engineers and scientists that were working on the problem knew how hard it was. Yeah. And all the business folks and marketing folks were like super hyped. Hey, this is gonna be only a couple years away. Yeah, I'm sure some of that was driven by Elon. So, yeah.

Doug: 13:54

So, like you said, the people I know at Stanford that started a lot of stuff, right? They won the DARPA Grand Challenge. They were like back in 2016 when Elon was like next year, and they're like, we're thinking more like 15 years.

Louis: 14:05

Right, exactly. The insiders that work on the stuff kind of knew that there's no way, you know, as engineers often do. Executives, marketing, investors, they were all pushing it. So Uber was in the space trying to do it themselves. They divested out of it, and now they're partnering, which I think makes a lot more sense because it's way cheaper, you know, make these partnership deals. I still think is a smart business decision on their side. If you look at their revenue, they're gonna make money through all the other things they do. The licensing that they're paying, say for the software or the access to the cars, is probably less than they're paying a driver, the amount of driving that they're doing. Uh, but even if it was comparable, like they still make a lot of money on top of it. So I think it makes sense. I think if eventually the technology matures enough, Uber might end up buying something to bring it in-house or making some kind of tight partnership with somebody. But yeah, until that nut's been cracked, there's not much point in them trying to do it in-house. They kind of gave up on that. I think that was a smart decision.

Doug: 15:00

Yeah, but when that point happens, when Robotax actually work, like does Uber's brand, does it have enough cachet that why wouldn't Waymo, which it already does in San Francisco, just be their own app or Tesla, obviously, and even Lucid. You know, if they have some of this tech and they're working with Neuro, Uber doesn't own this tech.

Louis: 15:18

Right. Why would they profit share with Uber? Basically, it's a marketplace, right? One of the main things that Uber has going for itself is not just taxiing, it's actually Uber Eats, the food delivery service. So they're integrated into the point of sale systems for thousands and hundreds of thousands of restaurants all over the place and the menu integration, all that kind of stuff. So those types of things are really hard to build out for a new company to come along and do that. It makes a lot more sense to just partner. So I don't see like Waymo fully being able to compete with Uber at that kind of scale. The other thing is they do have name recognition, they have way more users and they're in way more locations. So Waymo could start a war with them and try to like battle and take over markets. But I feel like a lot of those companies that have the technology will very likely merge, get purchased, or just have partnership deals with other companies to do it, especially as you want to go into more and more markets. But time will tell. Personally, I think it makes a lot of sense. Uber is going to be able to license it. I also think they have enough money that they'll be able to buy something if they need to. If the technology gets restricted to just one company that has it, nobody else is even close, then you know that's a different kind of conversation. But what I'm expecting is it's going to be multiple competing companies having comparable technology because they're all slowly creeping towards it. So we'll have to see what happens. I mean, I think even Elon has said like they plan on being able to license the technology to non-Teslas, right? They want to be able to sell that technology to another company. If it worked really well, I'm sure an Uber would just license it. They'd be perfectly happy to do that.

Doug: 16:49

So I've always been a little bit skeptical about that, just because we've seen how the Cybertruck is still kind of limited with their FSD, and that is only a slightly different configuration in terms of the placement of the cameras and whatnot. So, how well Tesla's system can map onto other vehicles and other vehicles can just adopt their system is a little bit unclear to me.

Louis: 17:12

I completely agree. I feel like the problem right now is the technology companies are struggling to get the technology to work at all. They're struggling to even just crack self-driving, working in all the various education things that happen, that they haven't even moved on yet to like how do I generalize its functioning and abstract it away so that I can put it on any kind of car and model with different camera configurations, it still works just as well. So I think in time that'll happen, right? That's we've talked about it in the past and earlier episodes around things like yeah, you're mapping all the camera signals into this virtual space representation of the environment, which then the models will go against. So there are ways to get there, but we're definitely far away from that. There are systems like Kama AI, for example, that work on a lot of different cars. Now, granted, they're you know a 2.5 kind of system similar to what Tesla is, and they can work with different kinds of configurations. So it's interesting to see. I'm looking forward to it expanding more in Austin. I'm looking forward to Teslas working through their stuff. I hope that they really get some progress.

Doug: 18:15

To me, the real test is in the coming months. Right now it's summertime, the weather's mostly clear. We've already seen the car deciding to quit because it rained and it actually kicked a passenger out, and then when it stopped raining, they got to get back in.

Louis: 18:31

And we will point out that Austin is a university town, so traffic is a lot lighter in summer compared to the rest of the year.

Doug: 18:39

And we already see how it fails with railroad crossings. The other thing that we have yet to see it work correctly on is the school bus stopping properly for a school bus. It may see that stop sign and then just treat it as a stop sign, like I can stop and then I can just continue. Kids aren't in school yet. So in September that may come up. And then, yeah, when the weather changes, not just rain but snow. I guess that's not an Austin thing and maybe not a Bay Area thing, but the days will get shorter. Currently, their system works at night, but it stops at a certain time. You wonder how much of that is related to daylight. I'm also still skeptical of cameras only. I think you can get a lot of the way there with cameras, and cameras are doing most of the job, but I really do think, especially as these sensors get cheaper, you gotta have the backups. Like I think LiDAR would properly identify the railroad crossing bar coming down and know not to try to go through that. Sure. We'll see how it goes. You know, the other thing is the hardware I mentioned earlier that there are reports that Tesla's only working on hardware four right now, or at least they'll be the first ones that get unsupervised. Hardware or AI 5, as they're calling it, is delayed. That was one of the things that Elon said next year or something. Where I think originally we're expecting it by the end of this year.

Louis: 19:51

It's supposed to be this year, but now it's likely next year.

Doug: 19:54

He made some comments about it in the earnings call, basically talking about how amazing it is. In fact, they'd have to nerf it overseas because it's so good. It would violate export controls on compute. Okay, maybe, but that sounded a little bit like juicing it up, spreading a little thick about how amazing it is. But you and I, as hardware three, people are still waiting for our updates. The question is, will this even be solved on hardware four, as we've talked about before? Will this hardware five be necessary? He've mentioned that they're gonna expand hardware four 10xing the parameters.

Louis: 20:26

Right.

Doug: 20:27

Can you give us an idea of what that means? Sure. I'm not really a CS type guy or even a neural network type guy, but the basic cartoon is you have different layers, right? And then you have so many nodes per layer. Correct. To 10x the parameter, does that mean more nodes? Does it mean more layers? What does it mean?

Louis: 20:43

Usually it means more nodes. While you can have more layers, most of the time when people are talking about how many parameters, they're usually talking about how wide it is, so how many parameters are passing through. But yes, it could be they're adding layers as well. But the direct mapping or like why they need more hardware for that is basically each parameter is more calculations that need to be done and more RAM to keep track of it. So you need more memory and you need to do more calculations as you pass data through. More dot products. So, yeah, exactly. More dot products, more linear algebra. As they're dealing with this type of problem, they're scaling it up. Generally speaking, with neural nets, how it seems to work for the most part, and this is how it relates directly to like ChatGPT and LLMs and things like that, is more parameters with more data equals better performance in a wider range of things. That just seems to just keep happening. We increase the parameter count, we add more data, it gets there. Now, there's definitely diminishing returns to some degree, although that mostly comes out of quality of data and things like that. As you get higher quality data with more parameters, things tend to work well. That also gets to why the power requirements go up and the hardware requirements go up and the number of GPUs you need go up and costs keep increasing. If we look at how big cloud infrastructure with models work. So, for example, if you look at how ChatGPT, Gemini, or those types of tools, anthropics, claw stuff, they actually have something called multi-model where they have different models that specialize in different areas. And so when you give it a request, it's not just one big massive neural net that has all the parameters doing everything. They tend to filter it out to some models of varying sizes that do different things. What I expect is in the future, once this problem is cracked, that will be the secrets to making self-driving not a giant monster and need infinite compute and GPUs and everything else. Is they will probably do something along the lines where it'll recognize I'm in this type of environment. I'm low light, I'm at night, I'm in US versus Canada versus Europe versus whatever. There'll be different models that specialize in certain driving environments. There's be like a baseline and then more fine-tuning around these other things. Anyway, that's what my expectation is in the future. So more parameters basically means hardware four is probably not going to pull it off. And if they decide they need those parameters to be able to solve these problems, they need more hardware. They need more compute, they need more RAM. Well, he said this 10x was in hardware four. Still was in the range of what hardware four can technically do. Yes. The question is, is that 10x going to be enough? Yeah. The problem is you don't actually know until you get there. So it may not be enough. And they're going to need to do another 10x and another 10x, right? Only time will tell. Hopefully, as you get to a certain baseline, again, because of potentially diminishing returns, you could get it to where it's good enough, and then you could make it a little bit better and you need a lot more compute, but not everybody needs that amount of compute, right? There could be differences, and that's that whole argument they made in the past. You know, Tesla was like, Yeah, we have models for hardware four, and then we kind of scale them down for hardware three. You can do some optimization and things where you can scale things down a bit too, and it'll run on lesser hardware, but still give you comparable performance. That might be the plan, right? Maybe if they keep scaling it up, they get better, better hardware and they eventually scale it back. But at this point, I'm of the opinion that hardware three is likely never going to actually work for FSD. And that's Elon's opinion, too, at this point. So Elon finally agrees, which is good. I'm also of the opinion that hardware four may never actually get to FSD. It's hard to know because I don't think we're as close as enthusiasts like to think we are, and certainly not as close as Elon likes to say they are. When we originally predicted this, similar to like you said, you had insiders back at Stanford that in 2015, 2016 said, Oh, we're probably 15 years out.

Doug: 24:43

Yeah.

Louis: 24:43

I still don't expect it to come before 2030. I think 2030 is probably the earliest we would likely see it. So it's very likely we'll be hardware six, hardware seven by then. But who knows? What I'm hoping is that with hardware five, they might be doing a form factor that's more compatible with hardware four. I could be completely off base there, but maybe it's closer. What we had was like hardware two, 2.5, and 3 could all be swapped in with some effort. And so the problem was hardware three to four, there's no way to retrofit. They may come out with a special thing, but I'm assuming they probably won't. They're gonna drag their feet until there's almost no hardware three cars on the road, or they might come out with one eventually. Hardware four to five, hopefully, is upgradable. My expectation is as they realize they have to keep transitioning these hardware, they're gonna try to do a better job of making to where they could more easily upgrade older cars if they had to within reason. If they decide they need additional cam replacements or they need LiDAR or some other kinds of sensors, who knows if they'll ever do that? You can't just slap that in there. Um, but it would be nice if you could upgrade the compute or something along those lines. We'll see. I'm glad that progress is being made. I definitely feel like Tesla will hopefully learn a lot more quickly now that they're trying to do robotaxi. As you said, it's a trial by fire. They have not much choice to work through those things, but that doesn't mean that they can solve it. That doesn't mean that they'll do it quickly. It's a very difficult problem. Technically, no one has done it in a non-geofenced all environment, all weather, all time of day. Like there's a ton of restrictions on the folks out there doing it. I still think Waymo's in the lead from what I've seen. I think technology-wise, they are the most advanced of self-driving vehicles, but also they're geofenced and they have a ton of restrictions. Teslas don't have all those restrictions. Teslas technically work in more environments, they just don't work as reliably in the same environments that Waymo works extremely reliably in.

Doug: 26:37

So well, Waymo certainly took their time with it, years of trials, but Tesla has more data.

Louis: 26:43

Yeah, Tesla has more data, Waymo has more hardware and sensors, and years of engineering on it.

Doug: 26:48

And also risk aversion, true, positive or negative, but I don't think people should die in testing this, so and we've had at least in autopilot and FSD, some dozen or dozens of people have died. It is what it is, unfortunately. So another thing that came up briefly in the earnings call was someone asked about Tesla investing in XAI. Elon made a big point about not talking about it, though there has been some news, and he suggested that's something the shareholders might think of bringing up, which seems like a hint as to what he wants people to do. What's going on with XAI? I mean, I hear that they're raising a lot of money. Yep. SpaceX is investing, which doesn't make a whole lot of sense to me. SpaceX investing two billion into it.

Louis: 27:33

Honestly, it's classic Elon. Elon has historically done a lot of games around multiple companies, gets one to invest in another one and then merge and then absorb and do that kind of stuff. He also knows a lot about raising money. Yes, Tesla's a public company, but we all can agree their valuation and their market cap is not connected to fundamentals. It's purely speculative and it's based on a lot of hype. So, what is the biggest hype thing in the world right now? AI. Yes. We didn't even have to plan that. We both could just know AI is the insane hype. Over 90% of all investment dollars go into AI-based projects. Not to spend too much time on it, but AI is without question a bubble. At least that's my opinion and the opinion of many others. You may disagree. If you look at market cap right now, there's six or seven companies which make up 35% of the United States stock market. So the entire stock exchange value is like six companies. All of those companies are investing hundreds of billions of dollars into AI. All that money gets funneled into NVIDIA. So NVIDIA makes the chips that everybody's using. Okay, to be pedantic, they don't make the chips, they design the chips, and TSMC makes the chips, but whatever, that's just because I used to work in semiconductor.

Doug: 28:50

NVIDIA makes the money.

Louis: 28:51

NVIDIA makes the money, but yes, all that money's funneling into NVIDIA, and it's essentially a big arms race. Everyone's of the opinion that as soon as I get AGI, as soon as I have this super intelligent AI, I'm gonna conquer the world. That's everybody's assumption. Now, I don't agree with that, and I won't get in all the reasons why I disagree.

Doug: 29:09

It does feel like a race for power, though, doesn't it?

Louis: 29:12

It's really a race for power, but the power that they're racing for is not as good as people think it is. And what I mean by that is right now, AI seems great and it's improving quickly, and there's all this cool stuff happening with it, as many of us got to experience using tools ChatGPT, Gemini, O'Claude, Anthropic, that kind of stuff, cursor, whatever, depending on the flavor of AI that you like to use. AI is going to eat the world, it's using lots of things, it's in and everything, it's going everywhere. Here's the thing that most people don't think about or know about AI. AI is majorly subsidized by investor dollars. So in the last two years, there's been $580 billion spent by these top six or seven companies on AI. That's capital. Expenditures of buying video cards mostly, getting power for those video cards, things like that. The revenue generated by all of their AI stuff, and this is very generous, is about $30 billion. So they've spent more than 10 times what they've gotten back. And what's interesting is if you look at the rate of expenditure and the rate of growth of revenue, the gap is not closing. The gap is widening. Now, most people think of AI from like the ChatGPG days. Those of us like myself that have worked in the AI space, this is goes back to the early to mid-2000s. AI at Google and those types of companies have been going back almost 20 years now. In research and academia, it goes back to the 50s, like it's been around forever. So if you look at it, it's like where's the ROI on the current AI hype? It's currently spend, and hopefully in the future we'll make money. How do they make any money right now? It's like subscription fees. Subscriptions. Yeah, basically, all of these products that are using AI are all built on top of these subscriptions. But again, the subscriptions are majorly subsidized by investor dollars. So investors have effectively burned over $550 billion, just gone. That money is disappeared out of their pockets, gone to buy GPUs, and those GPUs are basically running these models. At some point, you have to make that money back. At this current rate, if they didn't have to buy any more GPUs, it would take them over a decade to make their money back. But that's not how it works. There's always a bigger model, a better thing, a new shinier toy, and bigger, better GPUs. So every year they're spending even more and even more. And after a couple of years, the old GPUs go away. So GPUs you buy now aren't gonna be here in 10 years, they're gonna be in a garbage dump somewhere. Anyway, the point is AI is a bubble because they aren't making a return on what they're spending. And even if you could get the cost way down on, say, GPUs, they don't have to buy them as much, the models are more efficient. You still have the energy cost. That's the other thing. Right now, to run these models, it's more expensive than hiring a person. So there's all this drama like AI is going to replace everybody. There is no remote close chance of that happening just from an energy cost. A human brain runs on 25 watts a day. A GPU cluster used to run your AI model to be not as good as a person right now runs on many, many megawatts. It's not even remotely in the same realm of things. All is to say, as part of this arms race, it's a hype race in a bubble. And if you look at previous ARM races, like between the United States and the Soviet Union, it was an arms race of trying to spend, which essentially caused the losers to collapse and go bankrupt, right? They basically ran out of money, ran out of steam of trying to produce. So the hope is that somebody's going to win and everybody else is going to lose. That's their shtick. You've got all these companies spending lots of money hoping to not be the loser. Who's in that race? Technically, Tesla is in that race. And if you look at Elon, Twitter was purchased by XAI. What happened there? Basically, all this money, hundreds of billions of dollars, feeding into AI companies. Hype, hype, hype, hype, hype inflates the valuations. You don't need to make money as long as you have the potential to make a lot of money in the future with AI. XAI within a few months of being founded, very small company, very few people working on it, is suddenly worth more than Twitter. Because even though Twitter is $44 billion whenever when Elon bought it, there's no super hype curve anymore. That's a already existing business. It has revenue, it's losing money. They don't go, ooh, that's valued a lot more than it should be. But XAI is like, ooh, it's an infant. It could be worth a lot. You don't know. There's so much potential there. So the investors hype it up. It becomes worth a significant amount of money. Elon gets to play the shell game where XAI now gets to buy Twitter or X for paper money, basically not real money. He gets to use stock and this fake valuated money. Now that becomes one company. So XAI is basically Twitter, but like with an emphasis on Grok, their one product, and on trying to build AI to get to AGI. What's happening with all those companies, though? They all have billions of dollars and they're buying billions of dollars of video cards every year. So they need more money. They need more money. They need to buy more video cards.

Doug: 34:22

In that earnings call, someone asked, Would Tesla buy XAI with a merge? And Elon said something like, I don't think that would be appropriate. Or basically, he shot down that idea. And I feel like that might have to do with Tesla being public and XAI not being public. So just having Tesla invest in it allows him to transfer money. Correct. Because Tesla is the piggy bank. Use some of that Tesla money in the thing that he owns more of, which is XAI, without having to give up control any control to Tesla since he only owns about 13% of Tesla. Absolutely.

Louis: 34:57

If XAI was able to become a bigger company than Tesla, which it can't right now, it could in the future, who knows, on the AGI arms race, but right now it can't. Elon would absolutely let XAI buy Tesla because that would give him a lot more control. He would love to do that because he controls so much of XAI.

Doug: 35:16

He said that he would like to have 25% control of Tesla. Yeah. And the reasoning he's given is because of AI type stuff, and I guess Optimus and robots, and he thinks that's a big thing in the future. It's potentially dangerous, and he wants to be sure that he has a level of control. Sure. I'm not sure how much I trust him in that control necessarily. Yes. Because, you know, certain signs from him as an aside, I'm not really sure how to put that, but like his emotional maturity doesn't seem quite high. But then also he has a kind of utilitarian view of people. That seems to be like a not too uncommon trait among the CEO class. You know, some people call it psychopathy. But for example, Bill Gates, right? In the early 2000s, Bill Gates came to Stanford to give a talk, and I was there. And it was really about his philanthropy and doing things like helping cure malaria or help stop the spread in Africa, for example. Somewhere circa around that time, he gave a TED talk where he released a bunch of mosquitoes so people could experience that, but hopefully those mosquitoes weren't carrying malaria. But at this talk in Stanford, he said he used to think that since we're worrying about global warming and overpopulation, that maybe they shouldn't solve these problems in Africa because we have too many people. So that this is helping control the population. And he said this matter-of-factly, it's you know, maybe it's okay to let these people die. Uh, but then he said he found out when we raise their standard of living, particularly women, they end up having fewer children. Rick.

Louis: 36:52

So he thought, okay, maybe that's a it's a more ethical way to accomplish the same thing.

Doug: 36:55

I he didn't say ethical, he just like maybe that's a better way to go about it. He didn't seem to be making a joke at all that he could make that switch and be like, oh, okay, whatever. Okay, I'll do it that way instead. Which kind of struck me as scary. Now, of course, Elon, I I get similar vibes from him, uh, but he's on the other end of wanting people to have many more kids. He wants more kids. But I feel like from him, it has to be the right people having more kids, though, not just anybody, particularly him. Yeah. There's 14 kids that we know of.

Louis: 37:22

There are billionaires that have kids in the hundreds. Yeah. Is a thing. Eugenics was popularized by the ultra wealthy back 100 years ago. But yes, there is something about it when you have that much money and power and influence that you start thinking about things in a different way. And that different way is it generally involves less empathy than what you would hope for somebody that has a lot of money and power. Scary. Back to this whole thing. I think XAI needs more money. Elon likes to play the shell game of move money around his companies, also get more investment dollars. And he said he expects to achieve super intelligence within the next couple years. I would still argue, first of all, you have no way to know if we can or how to get there or what's going to happen. But even if you could, how are you going to make the money back? That's the other argument is at some point it needs to make a return. You have to have a product or service that generates enough money to do it. And outside of I'm going to crack Bitcoin and get all the Bitcoin, which now suddenly they wouldn't be worth anything. If somebody could crack it, they would all become worthless. Yeah. What's the secret way in which this super intelligent thing is going to collect all money without stealing it?

Doug: 38:29

Controlling everything, inventing new physics.

Louis: 38:32

Yeah. Anyway.

Doug: 38:34

Another aside, yeah, I listened to a number of podcasts, mostly news and science-y ones, but I was considering because I'd see them on X all the time. Let me check out this all-in podcast, see what's going on. It's run by Jason Calicanus and a bunch of Elon either admirers or cronies. Like Jason Kalicanus was Elon's neighbor or something. And I think the guest was Travis Kalnik, the guy that Uber founded Uber, yeah. Yeah. And I'm listening to this thing, and I wasn't quite sure who was talking, but I guess it was Travis. And they're talking about how I've been using these LLMs, and I'm just asking questions about physics. And I'm testing the limits of physics. And I feel like I'm just close to discovering some new physics. And I'm like, what the heck are you talking about, dude? You know, it's an LLM, right? And maybe it trained on some real theories. It's just going to start putting stuff together in a way that if you're not a theoretical physicist, might sound really intriguing. Like, wow, this is really pushing things to the edge. And just hearing them all talk about it. Are you guys nuts? I mean, I mean, I'm not a theoretical physicist myself, but that's sort of what got me interested in physics, but I went more the experimental route. But it just seems so naive. And what do you call the Dunning Kruger effect or whatever?

Louis: 39:43

It's the Gellman amnesia effect. It's parallel to that. It's you read a newspaper as the original, right? And you go, hey, this is so interesting, all these things. Oh, I'm learning all this stuff. It's so factually correct. And then you get to the thing that you're an expert in, and you go, Oh, this is completely flawed. Look at all the mistakes in here. But you assumed everything else was correct that you aren't an expert. So that's absolutely the same effect, in my opinion. Yeah.

Doug: 40:06

Especially with theoretical physics, existing theoretical physics right now, or people that are working on grand unified theories to try to ling quantum mechanics with general relativity. And look, general relativity and quantum have been flawless in terms of every time you do an experiment to test it to even more nines or whatever. It's so good, those theories. And they predict things that the people that wrote those theories didn't even expect. Einstein didn't expect black holes, for example. But they work in certain regimes, and there's a regime where one works and the regimes where the other doesn't quite work. And so people have been working on that for a while. String theory has gotten a lot of time, and there are a bunch of other kind of theories people are working on. The problem is pretty much none of them you can actually test. So you're just making math. Right. And math can be a model of reality, but it isn't reality. It isn't necessarily reality. So yeah, you can come up with theories and test them mathematically to at least see if they are self-consistent. And a lot of them aren't. But whatever Chat GPT or whoever's spewing about this stuff, it's gonna be nonsense. Now, I do see places where it could be useful where there is a ton of literature on a certain topic, and maybe it could point to areas where there's gaps in knowledge by comparing things that other people have said, maybe, and suggest some grad student might want to go work on this, give suggestions for where you could do further experimentation, maybe. Or it could find parallels between different fields. And this happens in mathematics a lot, actually, that you have totally different fields, but you're able to, with some insight, find an isomorphism between these two fields. Oh, actually, this field is really the same as this field, but it's like a different mapping, different representation, like having a matrix representation of something or a graphical representation of something, and it might take a while to figure out that oh, these are the same thing. Ed Whiton, it's string theory. There used to be dozens of string theories, and he managed to unify them, show them, oh, actually, these are all just aspects of the same theory.

Louis: 42:00

Or in computer science, lambda calculus and Turing machines. So Alonzo Church's work and Alan Turing's work ended up being isomorphic. But at the start, they were competing computational models for how computers could work, anyway.

Doug: 42:13

So there are places, but dude, it's not you're not pushing the envelope, it's not reaching the edge of physics, you're not doing anything. You're not finding new physics, at least not theoretical physics.

Louis: 42:23

It's mental masturbation.

Doug: 42:25

That's what it is.

Louis: 42:26

He's gonna bleep me out in the podcast there.

Doug: 42:29

So yeah, anyway, that just irked me. And these people that do this podcast, these are the investor class, right? Correct. So they're so bought into this hype, and it's something that I can see that oh, wait, this is nuts, right? And you guys maybe can't tell that it's nuts, but it's nuts. Someone needs to tell you that that's nuts. So yeah, it makes it feel more like a bubble.

Louis: 42:50

It is absolutely yeah. And I work in AI, I develop models, I do all that work, I've been doing it for a number of years now, but I also use the tools daily for work and not for work. And so they're very powerful, useful tools that are not actually self-aware or not artificial intelligence and are not better at doing my job without me versus with, you know, it's a tool. Yeah, anyway, back to the topic. So XAI raising all this money, they're buying video cards. Are they gonna do anything with it? Who knows? Elon said he doesn't want Tesla to buy XAI outright. Part of me wonders if the bubble was to pop. Say it starts to go down if he then uses it at that moment to buy, because right now XAI is growing and riding that bubble wave up, and it's gonna grow faster than what Tesla likely could, just due to the size difference, right? It's easier for a small thing to grow bigger than a big thing. I'm curious if things turn, if suddenly it's like, oh, investor dollars are drying up and people are less interested, and video cards stop being able to be produced at a rate that they need or whatever. If suddenly he sees the writing on the wall and goes, Hey, now's a good time for us to buy XAI just to leverage it into having more controlling shares of Tesla. But yeah, I don't know. What's funny to me is a lesson that Elon didn't seem to learn that many other tech company executives learned. You know, if you look at Google or you look at Meta, Facebook, the founders of those companies only have a small percentage share, right? Surrey and Larry combined don't have as much ownership of Google as Elon has of Tesla. Elon actually has more Tesla percentage-wise, which gives you value that gives you dollars, your bank account. But they have 51% controlling shares because they have multiple classes of shares. So I think their class B is like 10 to 1 voting power versus value. And a normal person can't buy those shares, only the founders have them. So basically, Elon didn't structure the company in that way for Tesla. I don't know if he didn't know that he could, or if he just never did, and now it's too hard to do. It might go back to Martin and Mark. Right. It might be he never could because he wasn't actually one of the co-founders of the company. But it's one of those things where Elon probably wishes he had some other class of shares where he could do that. Uh, but yeah, Zuckerberg, he doesn't own 50% of the meta, Facebook, but he does have a lot of controlling stake because of it. Elon is much wealthier than all those folks because he has a higher ownership stake at this company, but because he doesn't have the voting shares, it's riskier for him. Granted, he has pretty good control over his board and the majority of the investors right now, but things could change. We'll see how it goes. I agree with you. I'm shocked SpaceX is going to be investing. I don't see what SpaceX benefits at all. I worked for an AI startup company that was doing quite well, had a lot of investment money and IBM Watson people, all this other nonsense. And we had a project for NASA that we spoke with them about it, and it was so ridiculous because it's like AI needs to run on a cluster of machines, it has to run on a big cluster. You don't have that in space, and you can't usually deal with the latency delays depending on the type of thing you're working on. So space greatly changes how much AI you can have. Oh, having that HAL, your sci-fi stuff, we're very far away from having that.

Doug: 46:02

There are companies talking about putting these data clusters in space, right? Sure. You'll have unlimited solar power, and then also you can deal with your thermal issues too.

Louis: 46:10

You're gonna need all that shielding for cosmic rays, right? Depends where in space you're putting it, I guess. You're gonna have to deal with bit flips and error rates and all that. Yeah, but you can error correction, you can decrease those and then but error correction usually comes through additional hardware, so you're running less throughput of what you have, or you're running on much older technologies, you're not running like four nanometer in space, you're using 28 nanometer or bigger because you're less prone to those types of interference. It's one of those things that if you want bleeding edge video cards and you want to juice everything you can out of there, yes, space is an option, but it's gonna be difficult. And then it's the bandwidth transferring data back and forth. But anyway, SpaceX, I'm not sure what they're gonna benefit from this. I think it might just be an Elon control SpaceX.

Doug: 46:55

I would think SpaceX needs that money given how they're behind with the Starship and whatnot.

Louis: 47:00

Yeah, that's a few more rockets they could be blowing up to test things. Two billion bucks.

Doug: 47:04

But SpaceX is private, and Elon probably has majority controls so he can make that happen. He can do whatever he wants. Tesla makes a little more sense, though. Yes. That hasn't happened in my car yet. But putting grok in the car, do I even want that?

Louis: 47:18

I do not want grok in my car. There's an argument to be made that the improvements or the advancements that XAA does could be leveraged by Tesla for self-driving. The newer architecture that Andre Carpathy, what he did before he left being director of data science and machine learning stuff at Tesla was they switched over to a transformer-based model, right? Transformer architecture is what LLMs use. It is possible for improvements to be made at XAI that do help Tesla, but you don't need to give them billions of dollars for those improvements. That's not helping Tesla. So that's that. That's on the XAI stuff. We'll see what happens. It's gonna be shareholder votes, I'm sure, whenever they get to that point. So one of the other things we want to talk about model y variants. So there have been new Model Y variants seen, and I'm gonna pass it to you because you know way more about it than I do. I just know there's a six-seater in China. Is it Model L or something like that?

Doug: 48:11

This is Tesla trying to be more competitive in China, I feel. BYD has been doing game busters there. Overall, if Tesla sales are down, revenues down, they're losing certain things like the EV tax credits, the 7500 federal tax credit, and then also the zero emission vehicle credits that they're able to sell those credits to other companies. Basically, the Trump administration has zeroed all that stuff out. The 7500 that goes away Q4. So Tesla has for competition in China, they have this L variant that's basically a six-seater, it's seven inches longer. People want people movers. I think that's reasonable. My disappointment of it is still has that sort of fastback. I'd rather they just squared it off like the Rivian and made it more of a proper people mover because I think even with the longer someone like me sitting in those back seats are going to be cramped without the ceiling height. So it ends up being just like seats for kids, maybe slightly bigger kids. There was already a six-seater variant of the Model Y, or I guess maybe even seven-seater because they just had the standard bench, but that was kind of ridiculous. So this makes some sense. It feels a little too little, too late in China. China has some amazing, very cool sort of luxury vehicles there. But I expect that this variant will probably make its way to the US. You know, the Model Y is their moneymaker. And then the other thing is during the call, there's some talk about this lower cost Tesla. And they've talked about in the past, but it wasn't gonna be what some people had called the Model 2, like the lower cost vehicle. The question was asked, Well, what is it gonna look like? And one of the other executives were like, Well, we can't really talk what it looks like. That'll be like an announcement. And Elon just cuts and says, It looks like a Model Y, letting the cat out of the bag. Yep. So basically, it's gonna be some Model Y variant. We had a list of things about it because there was some leak. They're just trying to make it cheaper. So get rid of the glass roof. It still has the front camera and the bumper, it seems to still have all the FSD-related stuff, but not that light bar. That's to me the main cool feature of the newer Model Y is the signature feature. Yeah, it's not gonna have that reflective light in the rear that glows down on the road behind it. Instead, it's gonna be a painted body panel back there. Maybe no light front bar. And then for the interior, there's talk about maybe it would have cloth seats. They got rid of the little cubby in the front, which some people might like, kind of like the pass-through back in the old Model S days where those kind of open. No cover on the cup holders. I'm not really sure what that's supposed to say, but no real screen, that should actually save some money, and it would still have that single stock for the turn signal. But they want a lower cost vehicle. Earlier, they were talking about it being available in Q3 or something. Probably they're gonna wait to Q4 because that's when the discount goes away. So they're gonna try to keep selling their more expensive vehicles before that vehicle becomes available, right? But these changes that I don't see them saving that much money, but I guess they could eat into their margin more, and then it can also act as that product ladder. In a way, it just takes away a lot of the things that were added in the new Model Y. So I don't see it being that different than someone driving a 2024 Model Y. Sure. The sound system might be a step down to try to make it cheaper, or really just to incentivize people to go ahead and spend that more money and get the better Model Y. It's funny how everything is Model Y, because that's the thing that's been making all the money. Though I have heard reports that that newer Model Y isn't selling as well. And like we've talked about, these cyber trucks filling up parking lots all over the country. I've also seen Model Y's filling up this space because overall test of sales have been down. And we'll see if this helps. But two main things: one is a bit of a stale product line. They could have come up with a newer car, but they spent all that time on the cyber truck when they could have made a proper pickup truck that more people would like or a newer, lower cost vehicle. Because these little refreshes, that's okay, but it doesn't feel exciting. It's not something that people are really clamoring out to get. Right. And then the other thing is just the brand reputation because of Elon. There are other choices now, and a lot of people don't feel like supporting Elon anymore. And sadly, a lot of those people were existing Tesla customers. And he doesn't seem to acknowledge that. So I'm not sure how they really move forward as a company there. Sure. And certainly the stock price is tied to Elon. The people that are buying up the stock and speculate about the future, they're really bought into the Elon vision. And for the most part, I am too, in terms of the basic vision of where the technology is going, what's interesting, and what things are exciting. But for Elon, that doesn't really seem to be cars. I would like Tesla to remember its bread and butter and actually be more of a car manufacturer than you know, Optimus and whatever, however many years from now. They moved on. Cars were old news. Yeah, I really wish that wasn't the case. I mean, if you want to spin off and do these other things, I don't know. Any opinions?

Louis: 53:16

I agree with that. I wish they would stick with cars. He already has how many other companies that he started at different times. If you want to do something different, do a different company that's focused on that other thing. But yeah.

Doug: 53:28

As a good segue, the next thing they've done is that Tesla has opened this Tesla Diner. A diner in LA.

Louis: 53:35

What? Because that's a business that's got future written all over it.

Doug: 53:39

Yeah, it's cool. They've been talking about this thing for a while, at least since 2018 or something. He wanted like a really cool destination or whatever. Actually, we mentioned it last episode when we were talking about Elon stealing all the sci-fi names. And there was talk about calling this thing Milliways from the restaurant at the end of the universe. Milliways was a restaurant literally at the end of the universe. Back in those days, the theory of the big crunch was a bit bigger. So you had the big bang, which is expansion, and then there would be an end. And so that's what Douglas Adams was sort of talking about. And you could, I don't know, somehow travel through time and end up being there. And there's something about trying not to run into yourself when you go again or something. I don't know. You know, Douglas Adams, his sci-fi was actually more comedy, and the sci-fi wasn't that well thought out. But anyway, it doesn't seem like they called it Milloways. All the branding is Tesla Diner. My guess is that there is probably a rights issue calling it that. Right. Uh, the other thing we mentioned was the super cluster that XAI has built, and they call that Colossus. And that's to me, is a direct reference to Colossus, the Forebend Project. But Colossus as a name is very old, and no one's going to own that name. You can't own it, right? Yeah. There are poems named Colossus, some ancient mythology. You have an X-Men character named Colossus. No one can just own that. So I could see them using that pretty easily. But Millowes is pretty specific.

Louis: 55:04

So can you give us a little bit of an update on what's novel about the Tesla Diner or like what would we see or learn about it now that it opens?

Doug: 55:12

It's got an interesting menu of supposedly pretty good food. I've seen people's reviews of it. They have these roller skate people that will deliver it. It's trying to think retrofuturistic.

Louis: 55:22

That's what I think of the future roller skates.

Doug: 55:25

Well, it's retrofuturistic, right? I know. Yeah. You can order food from your car and they can bring it. But that idea goes back to the 50s or whatever.

Louis: 55:35

The 1950s. Yep.

Doug: 55:36

And then it also is kind of a drive-in movie theater. They have these huge LED wall screens. They're showing stuff like Star Trek episodes, Twilight Zone, 2001, a Space Odyssey. Supposedly they can't play full movies later into the evening. I feel kind of bad for the people that live near this thing. Yep. There's a condo where they look at their window and they're looking at the back of these screens. And then you think about the people that now their windows are being inundated by the bright LEDs of the screen. That's the opposite problem. But adjacent to it is this 80 spot supercharger station. So there's a lot of parking there. Obviously, it's crazy right now because everybody's trying to check it out. Because it's new. Yeah, because it's new. I'm not sure how sustainable it will be. We know the restaurant business is very hard to make money unless you're selling.

Louis: 56:24

Remember when cyber trucks were crazy and everyone wanted one?

Doug: 56:28

We'll see how this pans out over time. I've led to check it out. I mean, this stuff's overpriced as anything else is in LA. Yeah. Seems neat, but in the middle of this neighborhood, the kind of traffic it's gonna generate, and then also the time it takes for you to get your food and consume it, right? To me is gonna be longer than the time it takes for you to charge. So then what happens?

Louis: 56:49

It's not an efficient use of space for charging.

Doug: 56:51

Probably you'll have people unplugging so they don't get congestion charged, but then they're still occupying the spot. So supercharging isn't really a destination type charging thing. Right. Supercharging needs to be on just off the highways, you know. In route. Here it's the middle of the city. So I kind of feel for these neighbors and just from the Bay Area where there are people that feel like they need to supercharge. You have these long lines of people waiting in this queue. Actually, it'll be fun if they were set up for a queue and then people could take their orders while they're in the queue, right? And someone delivered it to you, kind of like the drive-thru at Chick-fil-A or whatever. Right, right. So they've talked about expanding it, having one in Starbase, which again would be interesting, but they'd really have to do the traffic control because Starbase, you have a pretty small highway going down there. And imagine if you have cars trying to line up to get in there.

Louis: 57:42

Not a lot of infrastructure for this type of thing, yeah.

Doug: 57:45

So another cool thing about it, I should mention, is that they had Optimus there distributing popcorn like you might at a movie theater. But clearly, given the way it was moving, it was being teleoperated. So I don't know how long that's gonna last. But I did want to talk about Optimus a little bit. Sure. Optimus came up in the earnings call. Elon has expressed the opinion that Optimus is even bigger than Robotaxi. Honestly, I've been a little more impressed with other companies instead of what I've seen with Optimus. Of course, the hand that Optimus has is very interesting. They've spent a lot of time on the hands, but there are a lot of tabs that don't necessarily need careful hands. To me, the most interesting stuff I've seen is out of Unitree. For one thing, you can actually get one, and so you see research people with these units, and they have a new version out that looks more like an athlete, looks like it has better balance and stuff. So, you know, if you look for it, there are also some videos of the unit tree going nuts. Yep, I saw that. Have you seen this? I did see that video, yes.

Louis: 58:38

It's like it's ready to kill somebody. Yeah, it was getting all tangled up, but there was like a harness or something that I've seen two different versions of it going nuts.

Doug: 58:46

Sure. It looks like it wants to just kill anybody around it with the arms flailing around. Destroy all humans. Yeah, that's what it looked like. You know, I'm nuts and I want to kill somebody. They're self-aware. Yeah, but I think what happened in both instances, because it's on this harness, it's being held up, and so its feet aren't touching the ground, and it's been given some command, but it has a sense of balance. So it's trying to balance itself, but its feet aren't on the ground, so it thinks it's falling, and so it's doing all this stuff to try to counteract the falling, just like you might do. You might move your arms around to produce a torque to try to keep yourself from rotating to the ground. So it's trying to do that and it's not getting the right feedback from its legs because it thinks its legs are swept out from under and it's just not stopping and it's just going nuts. So it's because it's on that harness that is flailing around, but that does not look safe, right? You don't want to be anywhere near that thing. Yeah, yeah. I don't know how much mass its arms have, but it needs to have some mass for it to be able to actually help to have a corrective use as it's trying to flail around. So if you're in the way of that, man, you might end up with a broken neck.

Louis: 59:54

Yeah, I remember seeing at one of the research labs where I worked, they had these sleeves they would put on. Robotic arms that they they were developing where they basically project out electromagnetic fields and it could detect something coming close. So if anything was in range, it would stop or slow way down because it would be like, oh no, something's close to me, I'm about to hit it. But yeah, it's one of those general purpose problems of when you have freely moving humanoid robot in a chaotic environment. How do you make it actually protect to not move too fast and hit something or a person or fall on someone or whatever? There's so many dangers to getting that to be safe. I think that's a lot of the hurdles that still need to be solved before we will have them in our homes. Yeah. Right. Even if you could have an Optimus and even if it could walk around freely and do something to do it safely is a whole nother question.

Doug: 1:00:43

There's a company called One X. They do the software in the Bay Area, and some of the work happens in Norway. But their philosophy is yeah, about the safety. So if you look at their robot, it looks very soft. It has a kind of a textile padded cover on everything so that you know it's pliable. Like when you hug a person, for example, you're squishing fat and muscle. You're not hitting bones directly. Not hitting the bones directly. So that's the idea that the body is compliant, sure, such that it has some give when you touch it, so that you're not instantly gonna get injured by interacting with the thing if you happen to bump into it or it happens to bump into you. So at least there are people thinking about those kind of things. Yes. As far as Optimus, though, uh, I feel like that's had some delay also because Elon has been talking about, oh, we're gonna sell a Legion. But the update he's given is that, well, we could have done a thousand of Optimus 2, but we have an Optimus III coming down the line, and we'd rather instead of going to production with Optimus 2, we'll go to Optimus III. So it looks like they're delayed a year or so compared to what they were talking about before. Because they were talking about some deliveries, yep, but it looks like that's been pushed out a bit. I guess the first time I was running a program, he left, he's gone. And actually, Tesla's had several key people leave. Yeah, executives leave and such. Head of sales, head of sales of North America. It's like, well, we're not really selling much, so get out. You gone. I don't think it's that guy's fault. Of course not.

Louis: 1:02:15

Yeah, yeah.

Doug: 1:02:15

We had a comment that we'll pay attention to. Drew Z said, if they bring back free supercharging for the life of the Cybertruck, I'm getting one. That probably is a pretty good deal. Not for me, it's not. If you like driving, if you like going on trips or whatever.

Louis: 1:02:31

I've still never needed to supercharge my car in three years or whatever.

Doug: 1:02:35

So I was hoping to go on a road trip myself, but I have some international travel coming up and I didn't have time to fit in that road trip. And I'm annoyed because the supercharger miles that I have. I have like a thousand free supercharger miles. Are they expiring? They're gonna expire. It's like, why? Why do they need to expire?

Louis: 1:02:53

Like your credit card points, your frequent flyer miles. Most of them don't expire anymore, but they did for a long time. Come on, man. They don't like carrying the liability on the books. I guess they like to be able to write it off. Yeah, it's too bad. Well, that's unfortunate. I never had pre-supercharging miles, and I probably never will. I just charge it home. Cool. As always, thank you all for listening and hanging out with us. Mike will be back, hopefully, on the next one. As always, you can become a supporting member on Tesla Motorsclub.com. You can post on there, find all kinds of great info. All right, well, have a good one, and we'll catch you next time. Later.