Tesla AI Day – Big Promises, Big Picture, and a Robot
By Edward A. Sanchez – Aug. 25, 2021
I will preface this post by saying that this is not an engineering or technical analysis of what was revealed at Tesla’s AI Day on Aug. 19, 2021. There are plenty of other outlets that have a much more detailed and informed view of the announcements, or you can simply watch the video of the presentation yourself.
What I’m going to unpack here is Tesla’s chronic issue with missing announced timelines, yet at the same time, ultimately fulfilling most of their promises, even if it is often years after the initial targeted timeline.
Also, we’re going to examine whether the claims that Tesla is “a technology company, not a car company,” and what implications that has on the company’s future, as well as future implications for investors.
Finally, I’ll lighten the mood a little by sharing my thoughts on the Tesla Bot, for what it’s worth.
Dojo Supercomputer
The goals and benchmarking laid out by Tesla for its Dojo Supercomputer would potentially make it the most powerful AI training supercomputer in the world. Tesla claims the in-house developed D1 chip, which is made by TSMC with a 7 nm semiconductor node, 50 billion transistors, and a large 645 mm(2) die size, is capable of 362 teraFLOPs of output at FP16/CFP8 precision. I am not an electrical engineer or computer scientist, but most reviews by experts in the field seemed pretty impressed.
A “training tile” containing 25 of the D1 chips has reportedly been tested. The ultimate goal is to create a cabinet with six training tiles, and then an ExaPod, which will have 10 cabinets. The ExaPod is hypothetically capable of an ExaFlop of compute, a long-held goal of computer engineers.
Computer Vision Architecture
The other initiative Tesla is working on is improving its computer vision, specifically focusing on temporary occlusions causing issues for its new purely vision-based system. Tesla has been historically anti-LiDAR, and recently dropped radar as part of its sensor suite. Some may see this as “reinventing the wheel,” as including sensors that are more naturally suited to spatial sensing could hypothetically provide more accurate data with less issues with occlusions. Indeed, during the presentation, Andrej Karpathy, Tesla’s head of AI, said taking a purely vision-based approach as like “building an animal from the ground up.”
I can’t get in Elon’s mind, and why he’s so resistant to LiDAR, and apparently, radar. Maybe it’s the engineering challenge of going with a purely vision-based approach. Most other OEMs are taking a combination approach to spatial imaging. Tesla, for better or for worse, is the contrarian in its approach.
Tesla Bot
About two-thirds of the way through the presentation, an actor dressed in a humanoid robot suit came out and did an impromptu dance on stage. Musk deadpanned, “Unlike Dojo, that obviously was not real. Tesla Bot will be real.”
Utilizing the company’s full-self-drive hardware and software, the Tesla Bot is envisioned as being a “friendly” robot designed to navigate through a world built for humans and handle “dangerous, repetitive, or boring tasks.”
In between snickers, Musk said people should be able to outrun it and “most likely overpower it. Hopefully that never happens, but you never know.”
At the end of the Tesla Bot presentation, Musk made his pitch for universal basic income, claiming that in the future, “physical work will be a choice.”
Final Takeaway
Musk and Tesla are famous for getting over their skis in terms of timeline, and we have no reason to expect any differently this time. But ultimately, be it two to three years later, they usually make good on their promises.
Already, the peanut gallery of tech bloggers and industry watchers is criticizing Musk and Tesla for the Tesla Bot being a distraction and sideshow, claiming the company should be concentrating on improving build quality consistency, getting its plants in Texas and Germany up and running, and focusing on more timely customer deliveries and improving the general ownership experience. Fair points.
But if there’s not an element of “reach,” aspiration, and imagination, the creative and developmental process can easily become drudgery and monotony. I know I’m speaking like a true liberal arts major. I get why the engineers want to take on some projects out of their normal wheelhouse. The day-to-day tasks of process improvement will always be there.
Tesla’s Elon Musk is unlike any automotive CEO in history. How many others were simultaneously CEO of a car company as well as leading one of the most advanced space technology companies in the world? How many others were serial entrepreneurs, and how many others invested their own personal fortune into their company to save it from bankruptcy and closure?
I am not saying Musk or Tesla are perfect. Far from it. I am an outspoken critic of some of the company’s design decisions. But it’s also pretty indisputable that the company is responsible for the industry’s headlong rush into electrification, whether or not you’re a fan of the trend.
Tesla has plenty of areas it needs to improve on, and plenty of unfulfilled promises. I’m not going to beat them up about it. Some people don’t want to give them a two- to three-year runway to turn their boasts into reality.
Personally, I’m just enjoying the show.
(Images courtesy Tesla)
- Podcast - Facebook - Google News - Twitter -
