gwern comments on Estimate Stability