I can remember when it seemed preposterous that a computer could beat a chess grandmaster. The problem was just too hard to solve. A computer could never beat a strong player. It was no match for human intuition, which is something we couldn't even understand, much less simulate. But after 10 years of research, IBM's Deep Blue beat reigning World Champion Gary Kasparaov, who is arguably the greatest player who ever lived.
Similarly, we are about 10 years into Tesla's autopilot/FSD efforts. And we don't even need to match the greatest driver who ever lived.
They are not comparable and winning at chess seemed impossible only for people outside CS. You can build a pretty strong chess algorithm quite early while learning to program (I did it as part of a homework in my second year of university), how it fared against a grandmaster was mostly a function of processing speed and iterating on those algorithms. There was also nowhere near the level of investment we are seeing now with autonomy, where billions of dollars are spent annually across the industry to get to a solution first.
Broadly, i also think it's a logical fallacy to take for granted something that has never been done before, just because we've been able to do other seemingly impossible things. Arguably there's an even bigger collection of things that always seemed to be on the cusp of being solved, but which ended up nowhere.
In the absence of anything else, my metric is the community based FSD tracker. The improvements are in the hundreds of percent, but still only a fraction or what's required.