Tesla AI Day 2022 Rumors: Elon Musk's AI Plans, Tesla Bot and Driverless Cars

Speaker 1: Last year's Tesla AI day gave us a demo of full self-driving technology and a weird dancing robot. This year, we could see the first working prototype of the Tesla bot as well as huge news on Tesla's dojo, artificial intelligence, which is at the heart of its goals for full self-driving cars. You might think of Tesla as a car company, but they're actually one of the biggest names in AI and Tesla. AI day is a big [00:00:30] deal on this year's tech calendar. So let's break it down, what to expect, what we'll see, and whether you'll see any of this tech in the real world. Anytime soon, Speaker 1: The 2022 Tesla AI day was originally scheduled for August, but in June Elon Musk tweeted that the event was being pushed back to September 30 because the company may have an optimist prototype [00:01:00] working by then. He also added that this year's event would be quote epic. You might remember optimist from last year's AI day. It's the official name for the so-called Tesla bot, which was revealed at the end of the event in a slick looking video. 3d renders showed a white humanoid robot with articulated hands and a futuristic faceless head. But what we saw in person was well, a person where some kind lame [00:01:30] dance moves. Yeah, look, I've seen a lot of weird robots during my time in the tech world, all these ones, but this was up there. Musk admitted the Tesla bot wasn't real yet, but he said it was coming and that it was the next logical step for the company's AI tech. After all Musk has said, Tesla's cars are quote semi sentient robots on wheels. So why not make a robot on legs? Speaker 2: It's intended to [00:02:00] be friendly of course, and, uh, navigate through a world, uh, built for humans and, uh, eliminate dangerous, repetitive and boring tasks. We're setting it such that it is, um, at a mechanical level, at a physical level. Uh, you can run away from it. Um, <laugh> and, and most likely overpower it. <laugh> Speaker 1: We only got a few details on the Tesla bot. It's built from lightweight materials. It measures in at five feet, eight inches tall and [00:02:30] weighs 125 pounds. And that haunting black head it's filled with sensors and autopilot cameras to help the robots since the environment using Tesla's full self-driving computer. We've learned a little bit more about optimist since last year's AI day Musk says his goal is to have the bots in millions of households doing tasks like cooking, mowing lawns, and caring for the elderly. So what are we expecting to see from optimist at this year's AI day? Well, [00:03:00] for that, I asked my friend CNET colleague and total AI pro Steven Shanklin, Speaker 3: Elon Musk has said, he's hoping to show a prototype. So we might see something now that said it is really hard to make self-driving cars that can pilot roads. And I think it's even harder to make a robot that can pilot the, even more complicated, the infinite variety of people's houses or workplaces or factory floors. Speaker 1: So we might see a physical robot prototype and it could well [00:03:30] be the kind of thing that moves around or interacts with people or objects. But don't forget. Musk has only promised a prototype, a robot working in a very controlled environment is really different to having a fully functional robot ready to ship. Speaker 3: I still think it's not likely to be delivered for years, given the difficulty and the complexities. And in fact, even Elon Musk said, well, maybe by the end of the decade, you'll buy one of these as a present for your parents. Maybe if you're really, really rich, <laugh>, I'm, I'm kind [00:04:00] of skeptical. But so if, if you talk about the end of the decade, then you know, it's much more attainable. If you talk about something useful this year or next year, forget it Speaker 1: Still. If there's an optimist demo, then I am here for it. And in a way it's kind of not so important what the robot looks like because what's happening under the hood could be way more impressive. And here's where we see that idea of the robot as an extension of full self-driving AI. And that brings [00:04:30] us to Tesla's dojo. Now, Tesla, isn't getting into the martial arts. You saw that robot's dance moves, right? It's the name of the artificial intelligence system that Tesla wants to use to understand the real world in order to train up its self-driving cars. Speaker 3: Dojo is Tesla's AI training infrastructure. There are two steps to using AI. The first is you have to train a system, you feed it a huge amount of real world data, and it learns to recognize patterns. And then from that, [00:05:00] you create a system that you actually run inside, something like a self-driving car or robot. So you train it. And then once you have that model trained, you actually package that into a product. And that's where the AI execution happens. So dojo is the, the learning part of the operation. That's essential. It's critical. It's really hard. You require extraordinary amounts of data to make it work. Speaker 1: Dojo is built on the back of Tesla's D one chips. The idea is that they're all linked together in groups, in Tesla's data centers to process huge [00:05:30] amounts of complex data. Tesla designed the D one chips to process real world video data. The footage that's continuously being collected by Tesla's cars. And that's how the AI learns about navigating the real world. But Tesla says dojo could also be used to tackle other AI jobs like language models could be an example. That's a big area in AI right now. And something that companies like meta, Google and Invidia are investing [00:06:00] a lot of money in teaching machines to understand human language commands and complex questions. If learning about the physical world is one part of the AI puzzle. Then understanding language is the other well, if Speaker 3: You're gonna be talking to a robot, it turns out that, you know, you're gonna be using language to tell the robot to pick something up for you. So I think, you know, that's an example of something where Tesla could be using that too, maybe to train its own language models. Speaker 1: But before Tesla can make robots that meet your every command, it [00:06:30] still has to nail self-driving in the real world. While we got a video demonstration of full self-driving at the 20, 20, 21 Tesla AI day, we're still a long way off self-driving Teslas hitting your local neighborhood. Remember there are different levels of autonomous driving from level one driver assistance, things like adaptive crews control all the way up to level five full self-driving, where you can read a book in the back seat, or there's no human in the car at all right now. Tesla's [00:07:00] on-road vehicles are at level two. The car is controlling the accelerator brake and steering wheel, but you still need your hands on the wheel. Tesla has been selling full self-driving for years, and it says every Tesla released since 2016 has the hardware necessary to pull it off, but it still hasn't delivered the capability to customers. So when will you be able to jump behind the wheel of a full self-driving Tesla car? Speaker 3: To me, it looks like actually delivering a true self-driving car [00:07:30] is pretty hard personally. I I'm impressed with Tesla's progress, but the more I look at the progress, the harder the problem looks to be driving is just moving from one weird corner case to another weird corner case something weird, an emergency vehicle parked in your lane, a pothole there's just infinite variety on the roads out there. That's why competitors like Waymo and crew are working on self-driving technology. That's very geographically limited to, you know, Phoenix or to San Francisco. Tesla [00:08:00] is trying to do self-driving cars everywhere, and that's a much harder problem. Speaker 1: Well, we'll find out all of Tesla's plans. Soon. Tesla AI day kicks off on September 30th and you can stay across all the coverage on scene at.