Zero accidents thus far has demonstrated that human-supervised FSD Beta is at least
500x safer than the average human, conservatively assuming the average beta tester has done only 1k miles on Beta since most testers were added recently.
If you want to estimate 5k miles per tester thus far, the safety is then at least
2500x safer than the average human.
I need to say “at least”, because:
- With zero crashes we can only have a one-sided confidence interval for the true probability of collisions
- E.g. Collision probability of 1 in a billion miles and in 1 in a trillion miles both would likely have shown zero accidents in a 100M mile sample.
- FSD Beta has been significantly improving over the time the sample was collected, so we’re estimating a moving target which moves because of the same process that generates the measurement data.
If each one of the 100k FSD Beta drivers has averaged at least 2 miles on Beta and if Beta were truly more dangerous than human driving, we’d expect to have seen a crash by now. If it’s soooo dangerous, where are all the injuries and deaths?
This is hard, indisputable evidence that these safety criticisms have no basis in reality and whoever publishes them either has no idea how to do basic statistics or has a malicious agenda.
Unfortunately, there are many people who don’t know basic statistics (or at least forget to think about them when confronted with emotionally-triggering anecdotes) and the smart malicious people are well aware of this fact.