Since Day 1 I've been on here arguing that these stats, first Autopilot and now FSD, are full of selection bias and other confounding factors that make them not at all apples to apples.It's just over 6x longer, but it's a different scale so it's not apples to apples comparable.
The top bar is FSD + Humans
The bottom bar is Humans
There is no bar for FSD on it's own. That might be between the two or it might be worse than humans, we just don't know.
But this is not really the way Tesla is portraying the stats in the tweet
I also don't know what these stats mean exactly in terms of what was operating... On the 4Q22 earnings call it was stated that FSD Beta has driven around 100million miles outside of highways. If we are to take the tweeted metrics in the way most people likely take them, that means there have been somewhere around 31 FSD Beta accidents that have been serious enough to deploy airbags and I find that difficult to imagine in the context of urban driving -- I've heard of no such accidents outside of highways, and you'd think at least some City Streets accidents would be all over the media and in O'Dowd's tweet history.
Makes me suspect these FSD Beta miles driven per accident are not distinguishing between the Autosteer on City Streets stack and the Autopilot highway stack, and there's nothing on the graph to suggest it does. When you have FSD Beta it is all-encompassing and flips between the stacks based on road type, so it's not like you activate Autopilot vs FSD Beta. You activate FSD Beta, and it decides which stack is active.
I don't know why they'd portray Beta this way, but the numbers don't make sense to me if we're just talking about City Streets.