Informed oversight through an entropy-maximization objective — AI Alignment Forum