One-step hypothetical preferences — LessWrong