3 ms·

I agree with you that comparing the rates of accidents is what's important. It's especially important to think about what's in the denominator of these rate me

I agree with you that comparing the rates of accidents is what's important. It's especially important to think about what's in the denominator of these rate measurements. I think the right denominator is "per km driven". (Accidents per hour of driving is OK but penalises faster cars -- people usually need to move a fixed distance, from A to B, and a faster car travels that fixed distance in less time.) The numerator (fatal accident count) for Tesla FSD may indeed be just 1 (I didn't see this claimed in the article, but it seems the data is somewhat public, so presumably if there were more fatalities, they would have reported them), but the Tesla FSD denominator might be quite low too, which might make its rate comparable to or worse than the human one. Of course, we'd need a much greater sample size for any confidence, so it would make more sense to compare non-fatal accidents, which are more common.

paulryanrogers 1y ago
Also note that over the air updates mean a Tesla may behave differently every time you get inside one. It could get better or worse or just different. If the human behind the wheel has to compensate, they'll have to be extra vigilant despite the non-incident times lulling them into false sense of security. So do you run the numbers per FSD version? Do you insist they must be certified before they make changes? People are unpredictable too, yet we have a mental model of their/our behavior and our own personal limits. When I get in the car I have confidence in how I'll perform, and without FSD, will remember to constantly maintain control. And a 'flaw' in one person's wetware is limited to them. No one can push out a fleet update causing humans to start targeting pedestrians instead of avoiding them.