And the AI videos are going the same way as full self driving:
1) looking like Tesla is easily two year, probably more behind everyone else
2) the others are seeing real SOTA performance ... and are not planning products because they think it won't work, or at least not yet
I must say ... really reminds me of the Tesla autopilot situation.
And I'd add 3) the really impressive robots, ie. the ones based on Boston Dynamics, are not based on ML algorithms. They are augmented by AI, not running actual AI algorithms in the control loop. The founder was an electrical engineering professor who moved into a CS direction (you know the sort of person who insists not just writing control loops in realtime, in assembly, but actually develops custom hardware for those algorithms. And I don't mean FPGAs or DSPs, I mean actual circuits)
So the entire approach of Tesla (and a lot of other startups) could be very wrong, and could very well be 5 theoretical breakthroughs removed from being feasible.
100%. We are seeing the bet of hard coded functions vs AI learned functions play out for us over the next year or two. Waymo has special case code for so much of their stack. Tesla removes special case code and replaces it with inference. Same with the Boston Dynamics vs Tesla robots. Tesla is making a bet that custom case code isn't going to scale nearly as fast as AI inference scales. Good news is that if one feels strongly one can place bets with one's money! If one doesn't feel strongly you can just comment on HN or Reddit or just watch.
It’s really hard to do sarcasm online in a way which is clear, still funny, and doesn’t normalize beliefs you oppose or make it easy for people to dismiss you with the “both sides” fallacy. It’s been a staple of internet humor for decades but I now think that was a mistake.
