FSD ver. 12 is a massive improvement over ver. 11, but not something I'd want to trust my life on. The vision system alone is reason not to trust it (it's just

FSD ver. 12 is a massive improvement over ver. 11, but not something I'd want to trust my life on. The vision system alone is reason not to trust it (it's just a "multiple choice" image classifier). The actual driving in ver. 12 (now neural net vs C++) looks much smoother. The continuing problem is that at the end of the day these are just rule-based systems with zero intelligence. They'll work fine 99.9% of the time, and then do something life-threateningly inappropriate in the 0.1% of cases they were not trained to handle. The only way to make these things as safe as a human driver is to base them on human-level AGI, which is still a long ways off in the future. I don't know if Musk is planning to add his Grok "AI" into the mix, but having a hallucinating chatbot as backup system or driver is not my idea of safety.