Bayesian inference on 1st order logic — LessWrong