Comment by tpm
10 days ago
It's always possible to deny the relevancy of a comparison based on some quality of the compared data. The autonomous car pilot trials will be by their very nature restriced to some locations, with specific weather patterns, etc., so even after the mileage will be 1000x of the current one there will be still options.
At which point will the comparison be considered relevant?
When the confidence interval is narrow enough. For the human events in 1 trillion, we get 95% interval around [1997229, 2002773]. That barely changes the observed ratio. For 9 events of Tesla robotaxis, we get [4.11537, 17.08480]. That means "the real ratio is maybe half, maybe twice the measured one - we can't tell". The September was so out of usual distribution, that if you checked just October and November, Tesla would have a much better rate than humans.
I'm not saying the relevant agencies shouldn't be checking every crash. Just that "Tesla’s own Robotaxi data confirms crash rate 3x worse than humans even with monitor" is completely unjustified. But "crash at rate maybe the same as human and maybe 18x - we're 99% sure" doesn't get clicks.