Last night, Tesla CEO Elon Musk spoke at the virtual World AI Conference 2020 in Shanghai. He shared just how close Tesla is to reaching “Level 5” autonomy. He also said that Tesla China will get a chance to create original designs and engineering in the future.
Tesla China & Autopilot
Elon Musk said that in China, Tesla’s Autopilot worked “reasonably well,” and that Tesla is building up its engineering team in China. “If you are interested in working at Tesla China as an engineer, we would love to have you work there. That will be great.”
One thing that Elon wanted to really emphasize was that Tesla is going to be doing original engineering in China. It won’t be converting things from American designs into Chinese, but will crate actual, original Chinese designs.
Level 5 Autonomy
“I’m extremely confident that Level 5 or essentially complete autonomy will happen, and I think will happen very quickly. I think at Tesla, I feel like we are very close to Level 5 autonomy. I think — I remain confident that we will have the basic functionality for Level 5 autonomy complete this year,” he said.
Elon reiterated his April 22, 2019 Autonomy Day remarks:
➡️ Level 5 autonomy is possible
➡️ Tesla is getting close to achieving it
➡️ He’s absolutely confident it can be accomplished with the sensor hardware in all Tesla vehicles
Meaning HW 2.0 or later (built since Oct 2016) pic.twitter.com/uETEicgYHb
— James Stephenson (@ICannot_Enough) July 9, 2020
Here are @elonmusk’s comments on FSD:
— autonomy is coming, and it will happen very quickly
— Tesla is very close, feature complete by the end of 2020
— no fundamental challenges remaining for level 5
— simulation doesn’t cut it. you need real data pic.twitter.com/63hcqdJZy5
— Whole Mars Catalog (@WholeMarsBlog) July 9, 2020
Elon Musk also said that he thought there were no fundamental challenges remaining for Level 5 autonomy and dove into the small problems Tesla has encountered.
“There are many small problems. And then there’s the challenge of solving all those small problems and then putting the whole system together and just keep addressing the long tail of problems. So you’ll find that you’re able to handle the vast majority of situations. But then there will be something very odd. And then you have to have the system figure out a train to deal with the very odd situations. This is why you need a kind of real-world situation. Nothing is more complex and weird than the real world. Any simulation we create is necessarily a subset of the complexity of the real world.”
Elon Musk’s words are similar to his recent Autopilot rewrite update that he shared on Twitter earlier this month. In that update, he shared that it was going well and that a lot of functionality would be ready to release in 2–4 months, but that it still needs to be proven safe for owners to use.
Going well. Team is kicking ass & it’s an honor to work with them. Pretty much everything had to be rewritten, including our labeling software, so that it’s fundamentally “3D” at every step from training through inference.
— Elon Musk (@elonmusk) July 2, 2020
A lot of functionality will happen all at once when we transition to the new software stack. Most likely, it will be releasable in 2 to 4 months. Then it’s a question of what functionality is proven safe enough to enable for owners.
— Elon Musk (@elonmusk) July 2, 2020
“So yes, I think there are no fundamental challenges remaining to Level 5 autonomy,” Musk said, explaining all the little details Tesla will still need to focus on to reach Level 5 autonomy. “So we are really deeply enmeshed in dealing with the tiny details of Level 5 autonomy. But I’m absolutely confident that this can be accomplished with the hardware that is in Teslas today and simply by making software improvements.”
Autopilot AI Chips, Dojo, Vector Space
“In developing AI chips for Autopilot, what we found was that there was no system on the market that was capable of doing inference within a reasonable cost or power budget. So, if we had gone with conventional GPUs, CPUs, and that kind of thing, we would have needed several hundred watts and we would have needed to fill up the trunk with computers and GPUs and big cooling systems. It would have been costly and bulky and have taken up too much power, which is important for range for an electric car. So we developed our own AI chip, the Tesla Full Self-Driving computer with dual systems on chips with the eight-bit and accelerators for doing the dot products.”
Elon explained that AI consisted of doing many dot products. For those (such as myself) who know nothing about this, a dot product is the sum of the products of the corresponding entries of two sequences of numbers. Keeping that in mind, what Elon said next was aimed at highlighting how powerful the human brain truly is: that “effectively means that our brain must be doing a lot of dot products.”
Elon noted that Tesla still hasn’t fully explored the power of Tesla’s FSD computer. They’ve only turned on the second system on the chip “a few months ago.” Elon said that making full use of the FSD computer could take at least another year or so.
Tesla Dojo System
Elon spoke about Tesla’s Dojo system and said that it was a training system that is intended to be able to process fast amounts of video data to improve the training for the AI system. “The Dojo system — that’s like an FP16 training system and it is primarily constrained by heat and by communication between the chips.”
Tesla is developing new buses (subsystems used to connect computer components and transfer data, not transit buses) and sort of heat projection or cooling systems to help solve these challenges. “We are developing new buses and a new sort of heat projection or cooling systems that enable a very high operation computer that will be able to process video data effectively.
“How do we see the evolution of AI algorithms? I’m not sure how’s the best way to understand it, except what neural net seems to mostly do is take a massive amount of information from reality, primarily passive optical, and create a vector space, essentially compress a massive amount of photons into a vector space.”
He shared that earlier that very morning he was wondering, “Have you ever tried accessing the vector space in your mind? Like, we normally take reality just for granted in a kind of analog way. But you can actually access the vector space in your mind and understand what your mind is doing to take in all the world data.”
He explained that what we are actually doing is trying to remember the least amount of information possible. “So it’s taking a massive amount of information, filtering it down, and saying what is relevant. And then how do you create a vector space world that is a very tiny percentage of that original data. Based on that vector space representation, you make decisions.”
In essence, it’s a compression and decompression that is going on on a massive scale, which, Elon says, “Is kind of how physics is like.” Just after this part, there was some type of audio error or upload issue where the audio of what Elon was explaining became muted and the clip cut out to explain the Giga Shanghai updates. However, you can hear Elon musing about how we are all made from hydrogen and that the universe is sentient. (I believe that as well. I personally found that fascinating and would love to hear Elon talk more about that!!)
Giga Shanghai Updates
Elon shared that things at Giga Shanghai are going really well, and that he is incredibly proud of the Tesla team. “They’re doing an amazing job. I really can’t say enough good things. Thank you to the Tesla China team. And I look forward to visiting Giga Shanghai as soon as possible.”
“It’s really an impressive work that’s been done. I really can’t say enough good things. Thank you to the Tesla China team. We expect over time to use more AI and essential smarter software in our factory. But I think it will take a while to really employ AI effectively in a factory situation. You can think of a factory as a complex, cybernetic collective involving humans and the machine. This is actually how all companies are really.”