[JoelSutherland]

Fatalities per 100 million miles driven is not a great metric for measuring the reliability of Autopilot.

That number is influenced not just by Autopilot avoiding accidents, but also by the overall safety of the car. Teslas are incredibly safe in accidents. The current 25% lead that Autopilot has over the average is probably not attributable to Autopilot at all, and instead to the overall safety of the vehicle.

* Fatality rates vary by make/model quite a bit (up to 8x). Some of this is surely attributable to the relative driving skill of purchasers, but most is probably the safety of the cars themselves. Given the spread seen in this IIHS report I would guess that Autopilot performs worse than the average human. http://www.iihs.org/iihs/topics/driver-death-rates

[studentrob]

It's not even a good figure for comparing fatalities. See my comment here [1]

Those 100 million miles include non-divided roads, motorcycle deaths, adverse weather conditions, and cheaper, older cars.

[1] https://news.ycombinator.com/item?id=12087603