Sleeping Beauty: an Accuracy-based Approach — LessWrong