gwern comments on Correspondence Bias