Pendulums, Policy-Level Decisionmaking, Saving State — AI Alignment Forum