How the TESLA BOT will change absolutely everything! | Robert Scoble

Foreign.

Let's get started.

Hey, everyone.

So I'm Tesla Herbert, and today we have Robert Scoble with us, and we're going to be talking about a topic that I'm very excited about, Tesla Bot. I know it's a topic that many of you are excited about as well. And we're very fortunate that we have one of the few people in this world who has his pulse not only on the bleeding edge developments of bots, but also spatial computing and just technology in general. And as we'll hear, these three disciplines are all kind of merging together in accelerated speed, and they're required for bot development. So when Robert was at Microsoft in 2003, I was a big fan, even way back then. And he became, as the Economist put it, a minor celebrity for geeks around the world. I think you're no longer just a minor celebrity, but it's still for geeks, right?

Yeah. So if somebody recognizes me in the street, it's almost always a geek, even though it's always somebody who watches Leo Laporte and, you know, is nerding out every night, you know, calling in our Twitter spaces. Crazy.

So Robert, many of us know him for his blog, Scobalizer. He's an author of four books. He's best known for his interviews, his early peaks. He somehow gets himself in all these scoops for the most exciting startups in the world. He's one of the most foremost experts in spatial computing. So that's virtual reality, augmented reality, mixed reality. What's that?

Any. I. I say it's any computing that you move through or your robot. The robots.

Okay, well, we're gonna dig deep. That's why I'm so excited to have you here.

And you're gonna have a virtual being soon. It's running around your floors.

That's it. That's not just a Roomba, right?

Oh, whatever you want it to be. This is going to be about to really flip. This is a big, big, big wave of.

That's why I'm excited about this. I'm going to do a whole series on bots and bot experts. And you're my first to come in because just a week ago I saw that you published a YouTube channel, a YouTube video that you had an exclusive look inside giant AI robotics. Love that video. People should check out. So welcome, Robert.

Thank you.

Thank you.

The guy who deserves the credit for that is Adrian Kahler, who built the first. The computer vision system for the first Stanford autonomous vehicle that won the DARPA Grand Challenge. Right. What year was that in the 80s, in the 80s. He's been doing computer vision a long time. He wrote the book that they still use at computer science labs all over the world to teach computer vision. He started the core software team at Magic Leap, which is an augmented reality company which is just got one best of show at AWE two weeks ago. And now he's building a robot company with the latest AI possible. So yeah, he's a real interesting. There's not many people who very intelligent.

I love that video. He was very, very sharp and like you said mentioned though his background was actually magically or even like way back then and then now computer vision, Magic Leap and now for cars, I magically and then now bots.

And he says they're all tied, right? He says there's a reason that he's built robots, autonomous cars and augmented reality glasses. Right. Because they're all tied together by the technology and inside the, inside the brain.

So what's your opinion? Because you are obviously your spatial computing, that's been your area for the last, I don't know, six years or something like that. What's your thought about that?

Yeah, so there's been three eras of AI so far, right? So Siri was the first app that consumers saw that had AI in it. And Siri, everybody knows the flaws of Siri, right? It doesn't understand my son, so it's not accessible. You can talk to us, Apple, we'll give you the voice data so you can fix that, right? In AI2, it doesn't learn very fast. And I had dinner with the guy ran Surrey six years ago at Apple and he said we already know it doesn't learn fast enough because Google is learning faster than we are. They knew this six years ago. And I said, why do you know that? He goes, well, we instrumented both us and Google and we're watching how fast they're learning and they're learning at a faster rate, which means they're going to win everything because exponents go really nuts and take over everything, right? Yeah, yeah. And so yeah, when you start talking with these guys and start realizing where the problems are, the first version of AI was this very inflexible. Siri actually stood out. When you uttered something to it. It spit out 50 lines of code that were hand coded by Apple or by Siri Corporation back then. And it wasn't very flexible. For instance, my favorite question I keep asking these guys to fix is how many people are checked in at the Half Moon Bay Ritz Carlton on Foursquare, Foursquare has an answer to that question. Actually there is an answer. You can go to Foursquare and see how many people are checked in right now and be nine or one or two. And it has an API. But Siri, because it's hand coded, doesn't hook up to the API and doesn't answer the question and it fails over to go big and give you a stupid answer. Right. And so it's really frustrating because it keeps doing this failover all the time and it doesn't give you good answers. Google gives you way better answers because they have all the data from all the searches ever made by humans and they have all sorts of technology to make that useful. So I have a Google in my Sonus. It's always this.

Yeah, so that's the first.

Oh, it's too low.

So that's the first. You said the first stage of AI?

Yeah. The second stage is it's starting to learn faster. So you saw a whole range of companies like Waymo and Cruise and Zoox is even a little bit later stage. But you see a progression. If you saw a map of when these things were started, you see a progression. And the middle stage is sort of this in between where it's starting to learn faster, it's a little bit more flexible, but it still doesn't fix all the problems. Right. And now we're in a stage of AI where the AI people brag, I can drive down a street I've never been down before. Right. Which is the Tesla system and the way to make that system work. Okay, so Waymo and Zooks Cruz, they're all up in San Francisco. They don't leave San Francisco, they don't go across Golden Gate Bridge. That's why I go out to the Golden Gate Bridge and watch like I count cars out there. I just sit up there and watch how many cars. There's a Tesla every 10 seconds going across the Golden Gate Bridge, which is really mind blowing. And everybody else has a geofence. It can't go to the Golden Gate across the Golden Gate Bridge. Right, right. Because their AI doesn't learn fast enough. They have to go down the street and scan it, the neighborhood and scan it and do a lot of work to make the the map all work all memorized for device. Right. The Tesla one like doesn't assume that at all. It's trying to figure out how to navigate down the street without ever have been down the street like a human being can. Right. I've driven all over America and I can drive on any I can even drive backwards in Cork Ireland Lane on the wrong side of the street. Right. Which is very hard to cognitively fight. Right. Because you learn to drive on the right so hard. But even there you learn to drive down a street you'd never been down pretty quick. And that was Tesla's philosophy, is we should be able to learn what a human is seeing and react just like a human being.

Yeah. So that's coming true.

Yeah.

Elon says that Tesla has the real world AI, the ability to recognize people and cars and there's even like a big dog and a small dog, stop signs, trees. You've done a lot of work.

Let me talk about it. I was just at that Muir woods with my family and I turn on full self driving because I have the Beta now. So I'm using it for everything. And I'm coming around a street that curves and at the end of the curve is a stop van with its blinkers on and there's a kid barfing next to the van. You know, obviously the parents had to get him out of the van as fast as possible. Stopped in a bad space. My Tesla is going 30 miles an hour, comes around the turn and goes, pulls around it and goes right into it. It does a good job doing that. Holy shit. That's exactly what I would have done. But it did it so fast and so smoothly and oh man, never.

So definitely real world AI can recognize items when you spent work with the mission, mixed reality, AR reality. Can it do that now is any company can do that. Didn't you one time say that you can put, put your iPhone with this lidar, point it against a fridge, open it up, it'll recognize a Coke can. Are they able to do real world AI too using a phone and AR glasses?

It's not far away, I mean from both companies. I mean Elon might push. If we were having a conversation with Elon right now, he might go, oh no, I'm not doing AR glasses. I think that's a stupid idea. Right, right. Going to do the glass. That's how entertainment's going to be done. And he cares about how entertainment's going to be done because he has all the Tesla.

Pause for a second. You think that Elon Tesla is going to do an air glasses? Because I actually believe the same thing.

He'S hiring up to do them.

The reason I've been saying that for a while is that the reason why FSD Beta is able to succeed so well for so long is because they have millions and millions of cars. With millions of billions of cars, video, visual data to be able to feed into the neural networks. How if you're creating a bot, don't you need as many videos as possible, Whether it's a factory or the home? How are you going to do that? And so I thought they're either going to have air glasses and people walk.

Around and they already digitized their entire factory floor, a huge floor, sorry, Volkswagen, and they're not on many factory floors have been digitized and turned into a virtual machine for training purposes and for safety and for reorganizing the machine when you need to reorganize it and figure out how to do that efficiently. And they're looking for efficiencies on the workers too. How do you get parts closer to them quicker so that you can make supply chains work more efficiently? So they're. Yeah, they're all digital. They're getting digitized by vjooqic.

So, okay, so let's go back to bots again. So you went and visited this giant AI. Tell me about giant AI, what they're doing. Would you consider one of the leaders in the space?

Yeah. In terms of AI, they're ahead of Tesla right now if you look at it the wrong way. And they know that. Right. So they started this company because robots before this were inflexible, had to be made very precise because they were inflexible so that they moved their hands or their grippers to the exact right location on a machine. Right. Or on a thing that is being, you know, milled or whatever. And those robots were very expensive and you couldn't be adapted to new jobs very quickly. Right. Once you trained it to do one thing, you expected it to live there for 30 years building the same thing over and over and over and over again. The new strategy is to come up with consumer level parts like your motors, your arm, your torso is all built out of a 3D printer. So low cost industry open small things that are lightweight that can do a job. And the problem with using cheap equipment to make the price come down is the robot arm will overshoot. Sometimes like a job bottle or, you know, a can or something, or a screw, it'll go too far. The, the magic of having modern AI is it recovers from mistakes very quickly, figures out the arm went too far and pulls it back just like a human being does. Right. We overshoot. If we pick up this iPhone, my hand sometimes overshoots the iPhone. And we humans figure out how to, you know, without even really thinking about it, just Smoothly get it back on path, and pick up your iPhone. This new AI does that, and it also comes with a benefit. If you're a worker and you have this robot, you can teach it a new job in, let's call it six times. You show it how to fold six times, and then it can fold laundry. And it's doing it visually by seeing everything you saw. It's using eyes that are seeing in neural radiance fields, and it's using the latest computer vision techniques to see everything and be able to fold it up. So this, this new AI lets them use cheaper parts to get the price down, but also makes it easier for a human to train. And so you're going to see it used for a lot of random interesting jobs that a robot needs to do an hour a week or something like that.

So they have the ability to. You said neural radiance vision. So that. Is that also the same thing as I can see a Coke can, I can see a water ball, and I know it is a water ball. And I see it's your hand when you. When in that video, you shook its hand. Of course, in that case, there was a human behind it to try to train it to do something.

But yeah. All right, so the Radiant, I mean, it's using two cameras that see 2D, and the AI is converting that to math. Right. Radiance fields. Because it sees in radiance fields, it navigates the fingers in radiance fields. That's how it perceives the world after it gets converted to these radiance fields. I've seen autonomous cars with two cameras that are building radiance fields in front of the car. And I've seen other devices that build them pretty quick. Now, the Nvidia one right now is taking a little longer, but it's improving at a pretty quick rate. The thing to watch is the exponents. It's not. Where is it today? Because, like, the Apple headset's not gonna. You're not gonna have it in stores for another year. So the question is, how much did it improve in the past year? And it's exponential. It's going crazy. The people who build a radiance field say a year ago it was impossible to do all sorts of stuff, and today they're, like, just chewing through problems and making it faster and faster and faster. And next year it's going to be faster and faster and faster. Yet there's a lot. There was 47 or 49 papers at the latest computer vision conference, and that's. That was the hot topic of the conference. That was.

So do you, do you think that Elon and Tesla is ahead in real world AI? He claims that you need to have, you know, a supercomputer, you need to have neural networks, you have to have millions of data. Yeah, my friend, there are other companies that are going to be able to.

Do this as well. They're all going to try, but. But my friend builds data centers for Microsoft and others up in Washington. And he said there's a data center he's been in which is just packed with Tesla AI, 100 computers, a huge data center, a good part of the 500,000 square or. No, no, he said this one data center, two something million square feet is huge. And a huge portion of that is Tesla's data center. Right. That's a lot of machines to throw at a big ass job. Right. And that's the simulator. We call it dojo. Right? Yeah. It creates a real time simulator of the world. In other words, Elon can see all sorts of different things about the world in this simulator because it's watching all patterns. So here's one. My Home Depot burned down the other day. Two months ago, the whole Home Depot.

Burned down the whole time. Yeah.

Did burn down. How many Teslas were in the lot? Three, with eight cameras going. How many cameras do you need to make a neural Radiance Field? 3, but not 8 cameras. You can make Unity claims you only need one camera, by the way, to make a radius. So. And you can actually move around a 3D space that it builds off the 2D camera a little bit. If you get two cameras, you get more 3D. If you get three cameras, you get more. And in a football stadium, they're putting 10 around the stadium.

Sure.

So anyway, so imagine you're in the simulator. You call up Elon, hey, Elon, the Home Depot's burning down. He can watch it in real time in the simulator or near real time, you know, a minute later maybe.

I mean, how hard is it going.

To be for that before the AI stitches it all together and puts it in the simulator and you get to watch the Home Depot burning down?

I mean, yeah, so right now all of their videos are from cars and vision systems. The real world AI is learning how to ride a roadway. Now we're talking about transferring that to a bot. I mean, people are saying, hey, the bot is a home, it's a factory, it's, it's everything in this world. It's this trees, it's the stairs. How transferable is that?

A lot. Because the street, as a Tesla drives down the street, it images about. It shows you on the screen about 200 meter, 200 yards around, around the car. It's seeing everything. Right. So if it needs an image of the front of your house, no problem. Right. And can it figure out, oh, this person has stairs up to a second apartment door. Right. Is, are they a Tesla owner? If they are, they're carrying their phone and the phone is always reporting location as you go upstairs. It has a barometer in the phone that I'll even.

Oh, interesting. Even the app, of course.

Right. So you started thinking. I call this Israeli thinking because the Israelis taught me how to. How they think about this stuff. Right. And. And no, they just chunk down what is the thing going to do and how do we get the sensor reading for that? Yeah, there's a lot of things that can be done.

Okay, so before we go there, let's, let's go back again. I want to, Yeah, I want to hear about your. If you've heard about the other robotic companies. Right. So I saw Amica is this face that can copy very, very. You know, it's the one that everybody saw on viral video. We saw.

I don't believe that's really an unencumbered AI. It was too. It's too advanced for today's technology. They can puppeteer that kind of thing pretty easily and. Right. In fact, art, our device, you notice that one AT worker can jump into it, into the giant AI device.

And Yes, I saw that in VR.

And he's seeing what the robot sees and he controls its moving of the hands. Right. Or the moving of the head too. Right. So as he moves the head left and right, the head is moving left and right. So it matches human experience. It's designed to work at a workspace where a human works. Right. So it has to work sort of like a human move the eyes left and right. So it track, you know, is the some task going by and it needs to be able to move its hands and grasp things and move.

This is amica you're talking about. Giant AI. Giant AI, Giant AI. All that. Yeah, yeah.

There's a variety of. I mean, I've seen so many different robots, hundreds of robots that do various different tasks. I mean, we already have roombas in our house. Right. That's a robot. And you can get a Chinese one that goes on the windows, cleans the windows. You just keep going. Right. And in manufacturing lines, they, they have everything from a huge robot that spins a car around right. And Puts it on a different line. If you go into the Tesla factory, they have one that spins the car around in a cage, right? And yeah, you can watch it, but you don't get close to it because you'll get crushed. That's such a thing, right?

We've seen Boston Dynamics show off their Atlas robot that is doing parkour as jumping. It's flipping, it's opening a door, it's running through the fields. We've seen. And then there's a dog version we've seen in the Japan Olympics. We saw this robot that was on a wheels but in a basketball court and was able to take shots from half court and get it in perfectly.

They're good at math.

What's that?

They're good at math. It's just, you know, it's sine wave or something. I, you know, I don't even remember.

It's one of the side level. That's obviously a lot of, you know, it's, it's dexterity and so forth. And then there's Dennis Hong who has seven different robots out there. So of all the robots you've seen, is there anything that, that is even coming close to?

I'd say an autonomous car is a robot.

That's what Elon's been saying.

The Tesla software is the most stunning thing I've ever seen. And we haven't even been in the simulator. I'm talking about what it shows on the screens in the Teslas. It's, it's stunning when you think about what is it doing and how is it predicting everybody's next move? It's like, whoa, that's a lot of.

Software, a lot of all these robots you've seen. Is there anything, anybody that's coming close to what they're attempting to do with the Tesla bot or what they're proposing it's going to be?

No, because they don't have the data.

The real world AI, right?

Huge data center with a huge number of machines to collect the data. My Tesla is sending Elon two and a half gigs a day and so is yours. And so we're all uploading such mass little strings data into this AI model that he's a, I mean, look at Dali. Dali ingested 225 million images into its AI and now it hallucinates images back to you. You just tell it anything and it'll send you back an image. Right.

So, okay, so the, the real competitors then is somebody not quite as big.

As Tesla's, but also big ass data center.

So yeah, I was Going to say the real competitors would be Google or Apple. Is Apple going to get into robotics if they can do AR glasses recognizing real world homes services.

Hey Apple. Hey Siri, can you deliver a pizza tonight for my family? Oh, sure. It'll be there in 39 minutes. And here's the, here's your pizza now going in the oven. The pizza guy has glasses on that are imaging the oven and can tag your pizza as it goes through the oven. Right.

But you're, you're pretty close to all these, you know, especially Apple. You've been closer to them. Do you have any hints at all that they may go into robotics or.

They'Re building a car, right? They, they building a car with a bunch of lidar, solid state lidars on the top of the car. Absolutely. Apple's building stuff and it is watching Tesla and going, shit, we're fucked. So they, when companies that have a lot of money say that, then they spend a lot of money to get back in the game, you know. And do they.

So you said earlier about Google being in, in, in the AI bit about it.

What? Nobody is driving a car 10 every 10, 10 seconds across the go bridge.

Only goes back to again, that's a car. And that. But how does that then translate into real world AI for homes and factories? We're making a big leap. But I believe again, it's a self learning system.

My friend AJ went to Carnegie Mellon as an AI pioneer. He's one of only two people I know who's built an auto tagging AI.

Okay.

And he explained to me why it's so important. Elon just laid off a bunch of workers that do the tagging labeling. That's right. AJ Jenega says auto taggers get everybody to go on vacations. Once you get an auto tagger, it starts tagging everything and it's so good.

You don't need to.

Yeah, it does it all fast. And so if it sees something new that it's never seen, how many micro milliseconds does it take to go to Google and figure out, you know, how does this work? You know, what is it? And is there any YouTube videos about it? Oh, let's chew through all of those in a few milliseconds. Right? I mean it's not quite that good, but it's close, you know, and so it knows all data. This is why Google works so good. Because it's seen every single search we've ever done and it has an answer now for everything that's been done. When it hits something new The Google AI is not flexible. It gives you a stupid answer, and sometimes it gives you wrong answers, and you have to know that it's a wrong answer. But, yeah, that's another problem. So far. Most of the things that actually matter aren't wrong. Right. So where you find errors is in, like, if you ask Google who was the first James Bond, it assumes you're asking about the movie, because that's probably what most people are asking about. But there was a play before the movie, and there was a different James Bond. I don't even remember his name. But my. My friend who builds AI told me about this problem, and it's still there. It still doesn't work.

Sure, sure.

It's a little bug that we all enjoy.

Okay, well, this is good. So we had. We spent a lot of time talking about the AI and understanding that a little bit better and where the different companies are. And then I want to talk about the humanoid form. And there's. Do you know Dennis Hong? He's a UCLA professor. Uh, he runs. He's the founder of Romella Robotics Robotics and Mechanism laboratory. So he's kind of famous. He's been doing it for a decade or longer. So he had seven kinds of robots, and they're all variations from a spider to jumping ones and others. And then he just did a TED Talk a couple weeks ago, and I was able to just watch it in preparation for this call, and he said, you know, what's next? He's got seven new robots. But what was interesting about what he was talking about was he found it very, very difficult to have a humanoid form, to stay balanced and to not fall. And so as much as he wants to create a human form, he always reverts back to spider with six legs or three legs or he created a humanoid form, but instead of it being bipedal like we walk, it actually is kind of like it has. It kind of does this kind of thing. Um, and so. And then, and then it's so funny because he goes, it's so hard.

That created a new giant legs, right? They don't have legs on giant.

They don't even have legs. Giant. There's no legs. Yeah. They just saying this is going to be hands.

Reason. Because the math you need, the training data, you need to walk and not fall. Is it? Very.

So, so then what Hong did is Dennis Hong. What they did was they actually created another kind of robot which doesn't look like a robot at all. It's a balloon from. Filled with helium, and it has two kind of almost like rope, like legs, so that it can always stay. It's just a way to keep it so that it doesn't fall now, but then you have that, and then you're comparing that to the Boston Dynamics. Big giant heavy metal things that can jump and flip and all that. And they seem like you can't flip them over. In fact, you've seen videos where the graduate student would kick it and it fall and get back, back up again.

Or something like that. It's all AI and.

Yeah. So why, why can't, why, what's going on? Why is Dennis Hong saying that they can't make a robotic, bipedal human or humanoid robot that can walk?

It's a tough problem. Our human brains are doing a lot of math, trying to figure out, you know, how to, how to navigate ourselves and keep us from phones. I once met the autopilot team at Tesla at a supercharger. It's funny. Met five of them, I don't know, 50 people who work in autopilot engineering. And they said, we have a philosophy. If your human can do it, we can do it. Yes, yes. And so I start with that philosophy, like, how is this possible to make this thing walk? Well, they have more data about how things walk. I stood, I parked in front of my kids school, and my Tesla was showing all the humans. Yes. I saw that around the car. I'm like, oh, interesting. All the data to train all the robots on how to walk because, oh.

My gosh, I never thought of that.

Right. He has a lot of human samples of people walking around the car in front of the car, next to the car. Right. And all of that is being studied. Now. I know a guy who has patent on bone, you know, structures and vascular structures to do all sorts of identity stuff. This is. Studying the human body is a pretty deep science now. I mean, we have iPhone apps you put on a basketball court and it watches you play and it shows how many. Yes. You made. Right? Yeah. It's like, okay, so you start thinking about what do they got? Oh, yeah, they got a, a few million computers in a data center that are all optimized just to do math for AI. And they have a lot of data of people walking around, dogs walking around, all sorts of kids, old people. Right. They can, they can study each one of those data sets in mass.

To be a devil's advocate, couldn't you just pull up YouTube, have Google search everything and say, show me all the ones where humans are walking. Just use that as a simulator and just work on that, yes.

But now, do you have what the sidewalk out front of my house looks like and the staircase and the depth.

And the height and the mud?

Because it has to go through mud sometimes. It has to go in the rain sometimes. Right. If it's actually going to enable what I am expecting it to enable, why does Elon want a robot? He's going to have an autonomous car soon. And if you have an autonomous car, transportation per hour goes from $60 retail to about 10. Therefore, a delivery service that charges $10 for pizza goes down to 2. And they'll take a profit. So 5. Right. It'll reduce the price of a lot of things. And if I'm going to deliver pizza, I need to move the pizza from my pizza oven at Round Table into the vehicle. Right. To send it away and get it to go to somebody's house. And then how do you get the pizza from the vehicle to the front door? At minimum. Just that. Just that's the rule.

Is it your. Is it your guess then, that basically we're saying that the first use case for the Tesla bot, he's been saying factory. Now you're saying maybe door.

Putting an air conditioning unit in there, putting some wiring harnesses in there. And that's another job a robot can do. And it's hard to do that with today's industrial robots. Right? Yeah. So you could. And if he has this kind of AI that learns quick, like in less than six tries, it's not going to be hard to teach that robot how to put something in over and over and over and over again. Then they take over the job. Yeah, yeah.

So factory first. And then you said you think that maybe they'll be used for delivering the pizza from the car or something like that.

It's totally about the car because it enables a new kind of transportation service that is pizza laundry, food laundry. All the things that are around your city that need to be delivered. When you get that cybertrike, it can deliver Sheetrock. Right?

Okay, this. This is interesting. Yeah, that's right. Okay, a lot of people said this, but I just. I don't think so.

Depot and pick out the Sheetrock. You're gonna pick that out in augmented reality glasses and you're. Oh, yeah, that's what I need. That right there.

And what I thought it would be personally, is the semi. So after the factory, they'll help with the semi, because when you're driving a semi, it's autonomous. That's great. But then there's so many things that the driver does Outside of that. And then they would take over that. And then that sort of, then it allows them to do it.

Totally makes sense what you said. That's having a workforce, a robot workforce that comes with the semi, jumps off the back and you know, organizes all your boxes onto lines. Yeah, it's already happening. Anyways, I mean, what I loved about what I've heard of, I met a guy at a game at the game developer conference and we started talking about his job. He's like, yeah, yeah, I build these kinds of systems for warehouses. I work for a supply chain company. I'm like, oh, cool. Because. Yeah. And this was two years ago, he said, and we just turned on our first cold storage automated. A completely automated factory, I. E. Lettuce comes in one door on a truck and robots take it off, redistribute it online. Everything's automated, no human beings. And I'm like. And I just saw that on one of the news journals today, like, oh, this fully automated warehouses. Yeah, we're two years, the news is two years behind what's actually going on.

Wow, that is absolutely cool.

So why do we need a Tesla robot on the semi and such? Well, there's lots of places that don't have that kind of infrastructure. Right. And you know, you're 7 11, you're dropping some milk off too, or. Right. You see these semi trailers been driving around the neighborhood delivering, you know, different products to different stores. Right. So having, so having a few robots back makes total sense that it would get out and you know, stock your milk and stock your Diet Cokes.

What giant AI said was that. So Boston Dynamics, it's like steel. It's really thick and it needed that, that width and strength. But then John A. Said, you know, you can reduce the price. Heavy pack military plastic. Yeah, they were heavy pack notes. That's, that's who their market is probably in Boston Dynamics, but for, for construction giant A.I. they're, they said, you know, if you make it light, you, it's easier to move, you won't harm anybody, you can work side by side with workers. And that seems to be what Tesla bot is designed to be. You know, some sort of plastic, not too heavy.

If you bump into it, it's not going to hurt you. It'll react like if you, if you, if you tackle it, it'll be interesting to see what does, what does it. I mean this is the kind of if Elon's going to build a robot, you know, it's going to do stuff like, hey, go for a bomb so.

One of the reasons why I reached out to you, because you just tweeted out something that I really loved a week ago or whatever it was, and you said, you know, Tesla bought in a Day on September 3rd. Because I've been asking the same question. So when they announced that Tesla bot was going to be possibly at the AI Day, they first said AI Day was going to be August 19th. And then he comes out and says, oh, sorry, guys, we need till September 30. Everybody kept saying, oh, that's because they're delayed, they're not there yet. I actually thought of as positive. I think what happened was they showed him this incredible bot. He got really excited and he said, okay, we're going to do it on August 19th. Get it done, guys.

And.

But then he goes, you know what? I needed to do this to show that it's not just a bot that can jump and dance. That's not enough. I need it to show its intelligence, because that's what we are. And so what could be cool in your tweet? You threw out some ideas that I thought, yeah, those are good ideas. So what do you think in Tesla AI Day, when he shows off the robot, what does it need to do to prove that this is real world AI? This is not just all the other bots out there that can do fancy stuff that are coded in the Boston Dynamics guy can slip and turn and then they can walk out the door. But when you look closely, you'll see QR codes in every door. So it's just, you know, robotically opening a door. You need to show that Tesla bot is able to maneuver based on changing environments.

What's your ideas and what does Elon know about the real world? Every.

I still don't agree with you there. I don't think so, because he doesn't have videos in the house.

Hear me out. There's four on my street. Okay. And they drive up my street a lot. So. Yeah. How many times has the front of my house been imaged by a Tesla or Google for that matter? Google and Apple have also imaged the front of my house. And it's 2cm per volumetric pixel or a box in the street. And it. It'll do some pretty amazing things. I mean, the eyes are going to know where in 3D space the head is that. That's why it won't stumble, because it always knows what it's looking out at. And there are little strings going to the voxels that are around it, and it knows where those voxels Are. And it knows what those voxels are. Is that a stop sign? Is that a light pole? Is that a telephone pole? Is that a car? Cars don't move very fast, but they do move once in a while. Right. The AI treats those different and shows them different on the screens than a telephone pole, which never moves. Telephone pole is good for looking at voxels. Right. Because it never moves and those voxels are there.

So I agree with you. Then delivery from a car to the front door can be done because they've got that video for sure.

And he can do it while dancing. Yeah.

So is that an example of something you might see on AI Day 2, where he'll go, here's the bot, and he's gonna open up. He's gonna. You said it's gonna drive in through the Tesla, whatever. Why does that have to happen? Opens the door, comes up, picks a package, and delivers the thing.

What he's about to do. Yeah. Is turn on the Robotaxi network. Right. I mean, it's still four years away, but that's not very long. The iPhone is 15 years old. Right. I remember that. So of course I do. To sell Robo Taxis to us, he needs to get us to understand the utility of what happens when transportation goes autonomous. The wholesale cost of Transportation goes from $18 an hour to 13 $30 an hour wholesale. And, you know, so it all sorts of weird shit's gonna happen because of that, because you. You're gonna use transportation more than you do today, but you're gonna use it not to take a ride somewhere, but to go pick up your laundry or go pick up your Sheetrock eventually, when the cybertruck comes out, or pick up your food or pick up dinner. Right. Or pick up your weed. Right? They deliver all sorts of stuff to your house now, Right? All sorts of different services are delivering stuff. He's going after all that because he understands how the human city works. He understands how many people are standing in line at a nightclub in San Francisco. So does GM and Waymo, but they only operate in San Francisco and Phoenix and Carnegie Mellon. Yeah.

Okay, so you think that demo is going to be delivery from the car into the front door or something has.

To show off why automatic driving matters beyond just saving lives. Because if he does that, you start thinking about, oh, I better reor. My brother who owns a bar in Virginia is already thinking about this. It's like, oh, what's coming from my bar? Is it more business, less business? How do I get the business right? This is all going to come through the Tesla network. He's putting in Tesla chargers for that reason. So Tesla owners get used to going and plugging in and teaching the AI in the cloud about this bar and going, hey, this is a cool place for Tesla owners. Right?

You just gave me an idea. There's a camera probably watching you plug in. Maybe instead of that snake thing it just.

There's Chinese microphones on the walls over there. So there's probably a camera. I mean I'm talking to the camera. It stares at me all day long.

Right.

And stares at my living room and watches the TV while I'm watching tv. I mean it's right behind me, right.

So in the last an interview that Lex Fridman did with Elon about the bot, Elon said there's two things that we can do that no one else is able to do. So he said real world AI, which we talked about. And the other one is manufacturing. He's able to manufacture this. So any thoughts or ideas about that? Have you seen other robots and do they onesie twosies?

I visited China and understood how factories work over there. Seagate used to sponsor my show and took me over there factory. The guy who runs design at Harman Kardon, which Samsung bought and for billions of dollars they make head units for Toyota and stuff like that. He said when I moved my design team from around different places around the world to Shenzhen, China, they started shipping product faster, winning awards and making sales. They made a lot more money per designer. And he said the reason for that is if they have a problem with the plastic in some parters something along with. Or if they just have a weird ass idea. I want to make a camera that is waterproof. They just need to walk across the street with the plans and go, hey, can you guys build such a thing? Is that possible? And they go, well yeah, but you have to change that one piece that, that's hard to manufacture right now, change that to maybe a thicker piece and we can make that. And in half a day they're to work on a new product, right? Yeah. That iteration cycle is very fast because they're close together. It's a real problem for America because if we want to design products, I have to, you know, use augmented reality. That's the only way we're going to compete with China. Because if we want to have that same iteration cycle with the factories in China, which is what you're going to do, you're going to make a lot of things in China still. You need to have that kind of, don't get on a plane kind of attitude. You have to fix it right now and get, get back to work on this thing you're building.

Well, so Tesla is their factories.

Yeah. They have the whole, the whole universe mapped.

The whole universe.

Their AI knows every thing. Serious. And it's next level. It's like, what's the retail store sales of this store versus this store.

How would they. How would, how would Apple or Tesla know this?

Ingesting of data from places.

So you think there's. So they have supercomputers.

Tesla. What did Elon ask me when I laid down my money for my Model 3? He said, Send me your driver's license. I saw an Israeli company that takes a picture of that one picture even sent over a cheap ass camera on your phone and it validates whether it's authentic or not. Knows that it actually came from a real government id. Yes. My credit card. Both sides. So he knows security code on the back. Right. And insurance card. Both sides, so he knows everything about my insurance agent and whether I have insurance or not. Those three things let you do a lot of things in finance.

Oh my gosh.

And you think, yes, I saw Israel AI several years ago that ingested every everything. Wow. Okay.

That's interesting. Your mind is.

Yeah. And so the robots, you know, you start thinking about how do you design a robot to not stumble? Oh, shit. He has a lot of data. Nobody else does.

Wow.

He has all the data.

Shocked how quickly.

But the eyes are getting good enough to see, you know, feature points around you. And the computer is thinking this way. Insta360, a camera company came to my house and showed me their camera. It works basically the same way.

I saw you do that. Yes. Yeah.

The AI in the, in the inside the cameras looking for features. Features like edges or horizons. Right. And it's moving the ball of pixels as you move the camera left and right. So it's steady cam, you know, it's. It's moving. This dude. I mean, this is why I was like, come on. This dude lands things on barges from space.

Right? Right, right. So they must be able to be able to. They can land that rocket on the barge at that speed, at that height, a robot floor is nothing.

Right.

That's expectation.

Come on. But they have simulators where they can do. Where they, their AI can learn from other things over and over, real fast.

Got you. Okay. You just, just, you just made me go crazy here. Okay.

Auto tagging. Because you want to know, oh, that pedestrian's a child. Right? That one's an Old person, that one. Right. It's auto tagging everything in the scene around it and it knows what it's looking at. Look at the Apple one, the Apple plan room plan at wwdc, they shift a scanning system. That scan recognizes things your room and recognize all sorts of stuff and distances.

All just from now you can move the foreground out because it knows exactly what one object versus another object.

I'm wearing this shirt of the good. This guy did a scan. He was. He watched WWDC with me and then he scanned my house and he had a fashion show where people are walking on my floors through my house. Right? Yeah. So if he can make a virtual thing walk through my house, robot Tesla can make a robot thing walk through my house because it sees the world the same way. Yeah. He was telling me, yeah, soon you're going to have performers walk down the hall into your space. And.

That'S so cool. Okay, so let's, let's go back to, you know, your, your life is.

The hard part then is hooking up the AI with the motors. But that too is something that is trainable, learnable. Like, didn't they get engineering specs for every motor? Don't they have an idea of how inflex flexible or inflexible the arm is, how many millimeters it overshoots things every single time. Don't they have that? And doesn't the math keep, you know, getting the hand better and better and better every time?

Yeah, yeah.

So if I had a simulator, I wouldn't have to do all those things those people were doing in the video. They would just do it, right? Virtually.

This virtually.

Instead of doing it, let it go to town and figure out how to run a machine from scratch.

Yeah, yeah. So he said, Elon said that humans are doing this.

Right. I couldn't figure out how to put gas in my rental Toyota because I got a Toyota for a couple of weeks because my Tesla was in the shop, because somebody hit me, got a new bumper. And I couldn't figure out, my wife has a RAV4 and I know where the button is on her car. But I couldn't figure it out. I was like clueless. I was like, so what do I do? I pull out Google Assistant, Talk to Google assistant. There's a YouTube video that shows you how to do it. So you're telling me that I can't ingest all that and go, oh, we have a 3D thing now that shows you how to do that task? Yeah.

Okay.

Yeah. How much further should we Go. I mean, this data center is fairly big. How much more data can we ingest? A lot.

Yeah, no, you're right. I mean, that's what he said. He goes, in order to do real world AI, you, you have to show me you have a supercomputer, you have billion miles or videos that you could train it. And then you have to have the neural networks and the software. If you don't have any of those or you're short on that, you're not going to.

I saw a paper written by an innovator at the Computer Vision conference. It figured out how to go over the obstacles from one picture.

Oh, okay. Just showing a picture.

If AI can learn from one picture how to go over things like a humanoid form.

Well. Right. Tesla's the expertise in heights and depth of things. It needs to know the curb.

You can go through Yosemite in the wintertime or the summertime or in sunset or in sunrise or in the middle of night. Night. Right. Or in rain or in fog or tornado. We can simulate a tornado in this thing. Come on, let's go.

You're so bullish. It's awesome.

Then it can teach a robot how to walk in the snow. That's in the 3D scene that the camera's. Because I drove in the snow in my Tesla and it was capturing the snow. Yeah.

Not exactly. Yeah.

The simulator piece of the snow might have been off my car. Another piece of the snow was off another car. That's what a computer does. It takes little pieces of things and makes new surfaces. And his material scientists are using this to do material science. Yeah.

So, yeah, their expertise with Tesla's expertise, when you show them the video, when you show them a video of something, they're able to quickly figure, okay, I don't know what this item is. And then they try to figure out what it is and they label it. And then before they used to have people, humans would. He said that we had like 100,000 people would label things and then the system would see if it was right. Now they move those people to the end. They would label something and the people were there just to confirm if it's correct. And so when they let go some people, it's because it's so accurate already that they have. They need less people to actually confirm it's right. And so what you've got there is something that you show a brand new scene hasn't seen, let's say it's never seen a balloon before. Once you show them. You know, I think Karpathi said in one of his AI days, he goes.

On your iPhone, you can get an app called Chuch C H O O C H with the standard 2D camera. It knows 200,000 different items on your iPhone. You aim it at a Coke can, it says, that's a Coke. You aim it at a Toyota.

And if you have a supercomputer, it can do that. Yeah. So, yeah, Karpathy said that. Let's say you have a car that has debris coming out of it. In just four hours, we can get 10,000 videos of different cars in different scenarios with debris. And then the system figures out that that's a car with the debris coming out of it and figured out, put.

A mudslide in the simulator, put an avalanche in the simulator.

So if you throw a bot in a home, it'll just figure it out on its own eventually, once it has enough videos of that, and then it'll just self learn and figure it out. When one robot figures it out, the rest know already right away that that fridge is open this way. That fridge is open this way.

So when you turn on the giant robot, the software starts running and it figures out it's a robot. When you start it up, the software is designed to start up in different things. The giant software, when it boots up, it figures out by the sensors that are around the machine that just booted it that it's a robot. If it starts up inside an autonomous car, it can figure out, oh, I have seven cameras now to play with, or seven cameras and a lidar to play with. I'm an autonomous car.

Yeah, car.

Again, that's AI thinking. It's like, how do we make this thing learn so fast that it just figures out what it is as it's building up?

Oh, that's so cool. Briefly, I want to get back to.

You on all situations, but that's their thinking. It's like, how do you make this thing flexible? And then someday you're going to have virtual robots too that are just going to be pals of yours.

Hey, Tesla, what's a virtual robot?

A virtual being that in your glasses you can see you put on your glasses next to your Tesla robot standing next to you, and that thing's going to be pretty smart. Like, hey, Tesla robot, you know, where should we go for dinner tonight? Yeah.

And Elon said that.

You ready for some sushi, babe? Yes, I, I'll call it.

Elon said that the, the robots will probably have personalities in the future.

There's a based on your personality gaping hole in The Tesla car. Everything else is automatic. The windshield wipers, the mirror. Everything's nice. Except when you try to talk to the car.

That's right.

You know, that ain't on purpose. I mean, that's on purpose. They're holding out for a system that they're going to bring out where you can talk to everything in the Tesla environment. A virtual robot, a car. Yeah. Augmented reality, glasses, the car itself. Yeah. And it's going to learn fast, so it's going to keep getting more and more things it does.

So, yeah, that's what I was going to talk about. It needs to be voice. Right. You need to be able to talk to it. You need. I think he already said that.

Right.

Go get me a beer.

I mean, here's one. You're driving 80 miles an hour through central California. You just left the supercharger. Right. And you're hungry all of a sudden. Hey, Tesla, can you tell me when's the next in and out? It knows where you're heading. It knows where all the in and outs are. It's seen that before with other Teslas that have gone through the drive, though, even three minutes ago, right? Yeah, yeah. So it's able to say. Yeah, there's one ahead. There's a line that's about 10 minutes right now, and you can decide if you're going to stop at that line or go a little further. Oh, there's no line at that one. And the AI will be able to talk. The little virtual thing standing here is going to be able to talk to me like that all day long. Like, the weather looks nice. You should go outside.

I can't wait till that happens. But, yes, you should definitely be able to talk to your car, at the very least in the course.

I didn't think about that. Maybe they shipped a virtual robot long before they ship the physical one to get us used to. We're going to talk to a robot and it's going to do.

Interesting. So the only way for that to happen is they have to have AR glasses, Right?

No, that phone. I'm going to Apple.

Well, you're thinking Tesla's not going to create a phone.

Yeah, no, but they have on the phone. They could add an augmented reality thing to the phone that lets you image a virtual robot in your house and talk to it.

Right? I doubt that, but. Okay.

That'S way easier than building a robot because that can right into the app.

I mean. Yeah, it's not as valuable.

That's okay.

So, Robert, I just want to spend some time I want to spend some time getting to know you a bit. So tell me a little bit your story how you started and then you. There's so many things that you've said that you're like. It's amazing the things that you've been the first to see or first to do. Like I heard you this morning on the space talking about the CERN being giving a private tour of CERN in Switzerland. Can you share a few ideas, a few of these examples that you've done that it's just mind blowing.

Yeah. I was the first child laborer for Apple computer at 13.

What do you mean by that?

My mom built Apple II motherboards in our house and had a whole assembly line and we were not quite slaves. We got paid a dollar a board to stuff the boards with. I don't know. You're kidding.

You're not kidding me.

This is the original hundreds of Apple II motherboards, right? Oh my God. I just was thinking about that because it's like, you know, Apple's iPhone is.

Does Steve Jobs know this? Did he. Did he know this?

Was.

Did was. Okay. That's so cool.

They. They ran out of factory line. They got so much demand for the Apple II that they couldn't make them fast enough in their factories. So they hired a woman, Hildy Licht in the Valley who hired housewives to build things under contract. She still is in business today, probably because she had the Apple account for a while.

That's so cool.

Their Apple hired you guys. You must do good work. In fact W told me those motherboards had fewer failures than the ones they made in the factory because people like my mom were very careful with them. And we washed them in our dishwasher. The flux off the solder. I learned all I learned how to solder on Apple II motherboards. Right. And I got a job later at HP running their manufacturing. That's early days. I met was the guy who designed the motherboards. And in 1989 later, you know, 12 years later and I saw him at West Valley Community College. He was parking in the same spot every morning and his car had the plates was. So I knew who it was what is was doing at community college. You know, started Apple Computer. You should be going to Stanford or Berkeley, right? You know, they let you in for free. When you start a company like that. Everybody wants you to start to come to their college because it's a big deal. But here he's going to community college and I stalked him and studied with him every morning for a Semester. And that led to a friendship for the entire life. But I was first the Tesla. I was the first to get a ride in the first Tesla with Elon. Elon gave me a ride before he gave his best friend a ride. How did this happen?

How did this happen? I thought it was that BC guy.

By accident because I was out to dinner with Jason Calacanis, who's.

Yes, Jason is his best friend, right? Yeah.

Or was at the time. I don't know if they're still best buzz, but they're pretty good. And he was like. We were having dinner at this restaurant. He's like, I think Elon just got his new car. You know. Should we call Elon? Of course we should call Elon. I mean, keep in mind, back then, nobody knew who he was.

Really. Yes.

I mean, nerds like us probably knew. Okay. He's a crazy guy. He started PayPal and had a. Has a. Some sort of car fetish. Something. He's doing something with cars. That's what I. That's what I thought of as Tesla that night.

I was like at the time, yeah.

I want to see a new car. Who doesn't want to see a new car? They have this dude bring out his new car, right? And we ended up drag racing it. And Elon had a yellow Corvette and the Tesla just wiped away. At least you know, what we were doing, which was stop sign racing in Santa Monica. And we would both step on the accelerators as good as we could, and they. And the Tesla was just completely eviscerated. The gas power and Elon was giving me. I was live streaming on a Nokia phone, so to 2,000 people or something. Gambit wrote about it still, you know, it's a story.

Yes.

Anyways, we. He. He gave me the first sales pitch of Tesla Corporation. Why it mattered, right? Why electric motors and vehicles matters. It's cheaper, it's safer, and it beats you. It beats the other guys. Like stop station stop sign racing, which is.

His work is fast.

He appeals to the child in all in me at least. And many, many other people mad who are Tesla fans, right? Because he's like, oh, he flies. He. He's. He's doing rockets that land on barges. Okay, that's just Next level. That's cool.

Pretty cool.

We've seen rockets all of our lives, but rockets that land on mar. That's like next level. Yeah. And. And we all want to be an astronaut, right? A lot of us want to. Yeah, we do have good eyes or something like that. And he. He's A car guy and he's doing all this other stuff and so he just appeals to the child in us, you know? Yeah. Wants to do weird shit, you know, and change the world and help people and stuff like that. Or put him out of work. I'm still trying to figure out what which. Is Elon nice or mean? I'm not sure.

Yeah, we'll see.

Some people say he's mean. Jury's still out there. Yeah, but the robots haven't killed me yet. And they had chances, millions of chances to kill me every day when I'm driving my Tesla because it's a foot, right?

Sure, sure.

65 miles an hour on curbs with a bus on the other side. And it does it perfectly every single time so far.

What other news? First things did you do?

Yeah, it does make sense.

The first things are.

Yeah, but I was first to see Flipboard, Cloudera, Instagram, all sorts of things. Wow, really? Yeah. Siri was launched in my son's bedroom. I had the first video of the Google self driving car on YouTube. Still is the first. And so I've been all over the world into R and D labs and other things. Matayo was one 11 years ago. They showed me monsters on the sides of skyscrapers with a computer and that became Apple's augmented reality headset. That's 11 years I've been waiting for this. See my son, he's giving me a nice back. It's 11 years.

Amazing, right?

So I've been tracking all these companies for a long time, all these startups that I like talking to who are starting two person companies doing something weird and they get sucked into Apple and disappear. And that's happened over and over and over again. I'm like, someday this company is going to ship something.

It's going to ship. Yeah, they're picking the right one.

And it's been in a lot of interesting things and companies and technologies and doing stuff like the M1 chip. It's really amazing. Chip. Here's one. I tried to talk Bill Gates into spent. I worked for Microsoft as a strategist.

Yes, he did.

So I tried to talk him into spending $300 million to do a variety of things and buy companies like Automatic, which today own 17% of the web.

Oh, automatic, yeah. Yeah, they did look at put a.

Bid in and I bought it. Did he? No, he didn't even give me $300 million. It cost him a lot, but he doesn't care. He's still making a lot. He's a billionaire and he always will be. But he turned me down for $300 million. And I had a pretty good case of why he should spend that money. This M1 I'm talking to you on, 16.8% of it is neural network. It's not being used right now. That alone, that piece of the chip alone costs multiple billions of dollars. So who talked Tim Cook into spending multiple billions of dollars on a processing capability that really is not turned on yet? Okay, cool. Steve Jobs. There's only one human being who can get Tim cook to spend $3 billion, and it's Steve Jobs. Steve Jobs told him long ago what to do to disrupt tv, disrupt everything.

Yeah, yeah, yeah.

They start buying augmented reality companies right after that and patenting things way back, even before Matayo got bought. Right? Yeah. So it's. Yeah, it's fun.

You've been excited for it. It's coming.

So I'm just a little pollinator. I see this startup do this thing. I see this robot company do this thing, and I see this guy do or this woman doing this thing. And I started put together. Oh, there's big change on the way.

You're like that forest gump of technology.

Every place you've ever been good at showing up at weird places first. I can't deny it. After having, you know, Bill Gates's robot lab, I got it in there first. Right? Yeah.

But it's so cool that you also did that with Tesla. And now you're. You can see the passion you have and the excitement you have with the Tesla bottom. You just. We just know it's going to change the world. Right? Have you. Have you given some thought a bit of all the things that are going to change? How significant.

Ice cream away from my son right now. Right.

A robot nanny. What would we call these robots? Anyways, I know that Elon stumbled with the word owner.

Hey, Tesla one. Do you have a brother? I'd like to hang out with your.

Brother, but what are you. What are you gonna be called together?

Right. Three Way games. We need another robot in the house to do something. Yeah.

I don't know. It's.

You're gonna rent them also. They're gonna come with. With the transportation network. You're gonna call them. You know, hey, I need. I need a pizza delivered, and it's going to arrive at the front door and it's going to go, hey, for $2 we can come in and set up your table and put out your napkins and all that. Another $2, we can clean your toilets and, you know, and another $2. We can.

Is there any way that you can weasel your way in so that you can get a first view of this bot? Can I join you? Is there a way that you can.

Find your friend watching AI Day right there? I. I don't. I don't get invited anymore to these things. So.

Yeah. Yeah.

I'm going to be watching on TV just like you are and going, okay, that's really interesting. The AI is really good at that. Not really good at that. Or has merge the batteries. How much power does it have? They're good. They could put on a good show. And it's going to be a show, right? Yeah. With a robot that looks real. I mean, like, you saw a giant. Giant could climb into the robot with a VR headset. Is there somebody.

Yeah.

Helping it? I think.

I hope they don't do that, though. That's cheating.

I hope they don't either. I think Tesla should say. This is Tesla AI. There is no human being you're interacting with. It's talking to you. It's playing jump rope with you. It's doing this.

Yeah.

Stuff in front of you because. Milan, get off that. Put the ice cream away, please. Okay. You know, could. Could it do very crazy things? It should like it. They should demonstrate it. You know, going through an obstacle course on the way to your front door. Like.

Yeah, no, you. You threw an idea, which is it has to have something random. Right. So you have to either throw a dice or something random happens, and then it. Things change because of that, and then they go out. Does something. Because it can't just be pre. You know, you watch it do that. It's got to be something random and changed it, Right?

I sure hope so. Yeah. We're going to be watching close.

What is this thing doing and where did it go? So let's do some predictions here.

What does it know about the street? Whoa, wait a second. It did something freaky there. How did it do that? Right, and we're going to be thinking about how the AI is seeing various things in the street. Like, does a bicycle come down the demo path and it jumps out of the way like a human does and lands on its fate and then comes back. It better do stuff like that because it. That's the expectation. Right.

That it doesn't have to do it by September, though, does it? I mean, it can. It can kind of like.

No. That's the whole point. If you can build. If you. The simulator. It's doing it in the simulator. So, okay, now, do they have the ability to manufacture A robot that can do the similar. That's hard. Adrian warned me about that. He said we can do a lot of things in Simulator 2. It's hard to put them in a machine that moves and has engines and sensors and has it all calibrated together. It's hard to do that. But, you know, they build cars. Come on. Don't they have a lab where they can build things? Yes, they build seats, they build everything.

Yeah.

Don't they have the ability to put, you know, all sorts of weird stuff onto weird things? Yes.

Actuators. Yeah, he's already said that. It's not going to look like the bot that we saw in a mannequin. It's going to be different. It's going to be probably bigger, heavier. It's got to have that. Right. What do you. What do you expect?

Where's the battery?

Is it the. Is it the button?

How long can this thing walk? Right, that'll be this thing today in the battery configuration. It has. Can walk, you know. Yeah. 20 miles or something.

4,680 batteries. A structure.

You need those answers, like, to fit it into your mental model. Oh, we're in trouble.

What's really cool about Tesla is whenever they show a prototype, it is very close to what they're actually going to do in production versus other companies with this fancy prototype.

Do they do a head fake and make us all think they're doing a physical robot for. Or do they have something on your phone that you look at?

It's not going to be a virtual. You keep. You're on that. It's not going to be. No, they're a manufacturer. They're a manufacturer and they're. Okay, we'll see. You want to make a bet?

If you can put a 3D robot on your 2D phone and you're talking to it and getting good answers, and it's doing things on your floor in augmented reality. Right. Like, you would see what the robot would do. Now you're getting.

The only reason I agree with you there is they need millions and millions of videos of people's homes and then to be able to recognize it. So they may do something like that, like a game, just so they can get all the videos.

Yeah. Would you like to play a game, Dave? All sorts of weird stuff throughout your house.

Would you like to play games?

Take it on a walk and watch it block. Watch it on your phone. Walking around, you'll realize they have the simulator working and you're seeing the robot walking around. Then you start believing it's going to work when they physically bring a robot to you with an autonomous car with a pizza in it.

Oh, I. I see what you're saying. That's what I mean. It can't just be a phone. It's got to be glasses. Right? Like I'm opening the do.

Do not underestimate your phone. It's supercomputer with nice sensors on both sides.

But you have to hold it.

I'm gonna get a beer following it down the street.

You're holding it Anyways.

I hold my phone all the time looking at my phone while I'm walking down the street. Right. So am I talking to my Tesla robot while I'm walking down the street and it's walking along with me? Hi, Tesla.

I see.

Right.

You have a wild imagination. We'll see.

Amazing that there's a thing walking alongside you, talking to you all day long. Do you forget that it's not a physical robot.

Do you want to buy one Eventually? Because it's. It's so exciting. And you got it.

You named it Everything 2024. When the autonomous cars get regulated, that thing is going to be real. Coming and delivering pizza to you. And you can see how it would do it.

Yeah, Right.

And you can see the simulator all around you too. Do you have any clue how amazing the simulator?

That's pretty amazing. Yeah, yeah, yeah.

Here's. Here's one thing. I was in lane one, going 80 miles an hour, and something was on the road. A paint bucket. And human beings were going around the paint bucket randomly, you know, getting. Doing random things up ahead. And so I. I was in autopilot. I let the computer drive and I wanted to see what it would do. Didn't do much.

It.

Now it knows about pink buckets because of my car, because I grabbed the steering wheel to take control of it. It taught the AI to take the videos off the car at that moment and put it up in the learning system. And the learning system tagged the bucket and said, we haven't seen a bucket in lane one before. Might think about how we're going to handle that in the simulator. And then they can try it 100,000, hundred million times, both virtually as well as physical Human drivers driving a simulator down the street, Seeing a bucket in the road, figuring out what to do.

That's what Karparthi said. If you wanted to recognize buckets on a road, he just says it. And then four hours later, he's gonna have 10,000 videos of different buckles and buckets and they'll and then they'll teach it how to recognize it, label it, and then move on and.

And do the predictive math of where the car needs, how it's gonna. When it sees a buck in the road today, it does a hundred thousand pass instantly. A hundred thousand. And then it starts evaluating the pass while you're going a few feet. Right. At 80 miles an hour, you're only going a few feet. And it's doing this. And then it figures out, oh, that's the right path. That's the path that leads to the most success in our simulator. Right. So now if it needs to interact with a human being out on the street and it's doing jump rope, did it learn how to play jump rope the same way? And does it joke around with the kids who are playing the jump rope?

This is going to be too cool. I was excited.

That's the expectation. Right. So if it doesn't meet that expectation.

Yes, that's okay.

What do they show us in the simulator? It's a cover for it.

Well, it is. It is AI Days.

Like you're at AI Day, you're watching tv, you have no idea what is real and what's robot.

Sure.

Or, I'm sorry, what's robot and what's augmented. Yeah. Because the two can be made to look the same now.

Yeah, I've seen that.

Yeah. In the simulator. Right. And if you're in the simulator and if you're looking at it on a phone screen or looking at it home on a tv, it's going to be hard to tell what's. What's in the simulator and what's the real robot doing. And it might even be nice to us and say, this is the real robot. This is the simulator. Yeah. So we can see. What did the simulator learn? How was it taught? How does it work? What's that data set? Because it's big. And then what does the robot. How did we get the code, the instruction, the. The thinking, the brain, the software out of the simulator and into the robot? And then how much work did you have to do over there to fix, you know, things that didn't work? Right.

Yeah, I fully expect that they're gonna announce that the dojo is already live. There's. There's no way that they're going to do an AI day two and then say, oh, dojo is still six months away and you're away. They can't do that. What an embarrassment that would be built.

No, no. I have friends who have been inside the warehouse. It's built they have all the machines in data center. So it's built. So built.

I know that, but they. He said, he said that the day that his AI team decides to switch from the current supercomputers to the dojo, they will decide if it's better or not. And we'll need to find out if they turned it on and actually are using it. Live production, of course, you have a lot of optimism for.

It's the reason Giant isn't really going to matter much. Right? Yeah.

They don't have the intelligence.

Once you let a robot on your factory floor and it's seeing the world in 3D, it's seeing everything.

I think this is the. What you said earlier is actually the answer when I brought up the idea that Dennis Hong and the others, they created these bots but they can't manufacture to not stand trade. Well, what they're missing is the supercomputer, the AI to then videos like you said, even the photos of it. It'll feed it and then they'll learn. And if they can land a rocket, if they can get it, they can get a car to, you know, move on micrometers. They can, you know, not easily, but they should be able to teach a bot to walk. I don't know. We'll see.

It'll be very fun. That's what I'm hearing. They did. So we see in September. Okay. Or do they do something even more mind blowing? I don't know. Yeah, there's limits.

He's our new. He's our new Steve Jobs. We miss Steve Jobs so much.

And sort of show a piece of the physical robot and say, this is where we are with the physical, but let's show you the simulator. And it's so mind blowing to put it on your. To update the Tesla app and then get the AR simulator and talk to it all day long and it's walking around your house and doing things. Oh, that's going to be pretty interesting.

Well, you brought up some things that I haven't even thought of before and I'm a wild imagination.

Does it know what's on the roof? If I have a Tesla solar roof, does it show me that? Does it talk to me about that? Does it go, hey, your batteries are full now. Thanks for the solar power?

Or some people were saying that solar roof is another area that it's repettable, repeatable, dangerous and boring.

That they would probably use a bot.

To help with solar roof installations, you.

Know, now that would be. That would go across the line. I'm Expecting if it could climb up a ladder on the side of a house and start installing a Tesla solar panel. Yeah.

That's not September 30th.

Yeah, that's.

Now that's my new tears.

I'm expecting that.

No, you don't. Do.

You don't think it's ready for the solar panel.

No, no, no, no. Though. This is.

It's you in the simulator, though.

Yeah, we're.

We're.

You're right. We'll see. So the way. Normally it's like this, right? Things are slow, and then all of a sudden, boom. It can do something you did not expect, but at this point, you're expecting something a little too early.

Maybe if we go from one plus one is two, you're already. It's just small. Two plus two is four. That's exponential. Four plus four is eight. That's exponential. Yeah.

Four times four.

Sixteen. Yeah. By the end, you know, by the time it does that 30 times, you have millions, right?

Millions. Millions, yeah.

Yeah. These are changing real fast. He has millions now. He has, I think so to ingest two and a half gigs from every one of the Tesla cards, which is what's somewhere around 2 million cars now. And he's making how many in Fremont a month? Like 60, 50,000amonth. New cars just in Fremont. And he has Texas and China. China's beating Fremont every month. Right. They're making more and more cars over there. So. And Germany's coming online. Right. So he's gonna have nodes that are spreading out across the world, and he's going to be able to see all the data. Plus he has starlink over your head. You don't. There's a nose. Okay. If not, there's planet that has a camera. So, you know, he gets the. Yeah. The government showed me taking a single photo from a military satellite, right. And extruding it into Yosemite National Park. One photo from overhead. It was excluded into a 3D thing you could fly through in VR. And so it already has a lot of the world's data all everywhere. It's just better data if you're at the Golden Gate Bridge because there's a Tesla going across that every 10 seconds.

Tesla satellites can actually map faster and quicker than Google the people walking around with their backpacks.

By the way, did you see what Woz is doing in retirement?

No, that's that.

He just released an app that shows everything in space around the Earth in real time. He's one of three people who started a new company. I forget the company name now that I'm talking.

That's so cool.

Yeah, yeah, it's really cool. Alex feeling is the other partner. He's a goot. Alex Fieldman built a robot company too, so he's a real smart dude as well. Let me find this satellite photo. Yeah. Leo Labs. Click on Leo Labs and you can. And it shows the globe with everything flying around the globe in real time. If you watch it closely, it's showing you.

Okay. From the satellite view. Yeah.

Yeah. Really? So that's what Woz does on retirement. Yeah, retirement, exactly. You get the co founder of Apple bored on a beach somewhere and he starts doing like that with his friends. It's pretty cool.

All right, well, thank you so much for your time, Robert. Everyone please follow Robert on Twitter under his handles Globalizer. And of course you got to visit his YouTube channel at Roberts Global. Hopefully you learned something new. And if so, please subscribe and like, thank you everybody. Thank you again.

Robert.

This was amazing. I learned so many new things. Excited even I somebody with a dream or idea person. You've made me blow my mind a few times. So thank you again.