Non-axiomatic math reasoning with naive Bayes — LessWrong