Bayesian Model Testing Comparisons — LessWrong