Their assumptions are just one way of reaching the 275 million mile mark needed to hit that 95% confidence interval. If we're looking at actual numbers, Waymo's recently boasted of hitting the 7 million mile mark over several years of testing so if anything, Rand's assumption of 275M miles in 12.5 years for a test programme looks to be erring on the generous side. Sure, if Chevrolet decided to use their entire production run of Bolts to test autonomous vehicle performance you'd get 275 million miles pretty quickly, but they might struggle to get consumer buy in and regulator approval for selling fully autonomous driving capability to 30,000 consumers unless and until they can produce evidence that their autonomous mode accident rate is lower than human drivers...

And there's a much bigger problem, in that the software being tested is continually evolving, and modifications which result in improvement in general driving conditions don't necessarily tends towards stable (or monotonically reducing) numbers of rare accident-inducing bugs.