On 2022-09-30, Tesla held its 2022 AI Day in Palo Alto, California. This is billed primarily as a recruiting event for engineers, intended to show the company’s position and ambitions in artificial intellgence, computer design and manufacturing, and robotics.
This is a three hour and 23 minute video, and will take some time to digest—I have not finished it myself. Here is a summary from CNBC, “Elon Musk shows off humanoid robot prototype at Tesla AI Day”.
Musk estimates the Optimus robot should be ready to go on sale in “three to five years” (this may be in “Elon time”) at a price “much less than a car—maybe less than US$ 20,000”.
I wonder how much time and money Musk could save if his “AI experts” had lossless compression as a benchmark?
The NeRF paper is from Aug. 2020:
“Our method requires only 5 MB for the network weights (a relative compression of 3000× compared to LLFF), which is even less memory than the input images alone for a single scene from any of our datasets.”
“We’re also very excited about the recent breakthrough in Neural Radiance Field, or NeRF…”
And even NeRF isn’t really where things should be.
The hardware architecture Musk has locked his company into would be very different as it would focus on sparsity. It could easily cost a lot of lives to shift to sparsity if, indeed, the regulatory and civil infrastructure emerging around self-driving even permits it.