Predictability of Decisions and the Diagonal Method — LessWrong