There was a time when many within the AI community thought that every problem could be solved using neural nets, tuning and lots of training. Elon truly believe

There was a time when many within the AI community thought that every problem could be solved using neural nets, tuning and lots of training. Elon truly believed them, and believed himself that if you trained a large enough network on trip recordings and/or sensor data, the network would magically come up with correct driving inputs. It seemed to work well enough on closed courses, so everybody, including Elon, thought that within a year's time we'd have figured out the kinks. That's when he made the mistake of promising FSD. Turned out, what worked in one city/country didn't work in another. And then we had inconsistent road markings, missing lane dividers, lanes going into walls, pigeons... So they had to get back to more 'conventional' machine vision combined with some neural to recognize objects. It turned out driving is a hard problem that humans happen to be very, very good at. And here we are, many years later, and the problem isn't close to being solved - and not for lack of trying. IMHO finding a solution to self-driving isn't so far from finding a way to scale intelligence enough to reach AGI. I'm sure we'll get to self-driving before we get to AGI, but solving self-driving will be a major stepping stone towards AGI. Hence Elon is now aiming square at it: https://www.nextbigfuture.com/2022/01/elon-musk-must-lead-new-tesla-development-programs.html https://www.nextbigfuture.com/2022/01/elon-musk-must-lead-ne... *edit for typos.