That the tech has evolved to be better actually is an assumption. The novel data problem hasn’t been meaningfully addressed really at all so mostly we assume that progress has been made… but it’s not meaningful progress. The promises being made for future capability is mostly pretty stale hype that hasn’t changed year to year with a lot of the targets remaining unchanged. We are getting more data on where specifically and how it’s failing, which is something, but overall it appears to be a plateau of non-linear progress with different updates being sometimes less safe than newer ones.

That actually safe self driving cars might be decades away however is antithetical to the hype run marketing campaigns that are working overtime to put up smoke and mirrors around the issue.