Power-seeking for successive choices — LessWrong