Power-seeking for successive choices — AI Alignment Forum