x
Reward uncertainty — LessWrong