Generalizing POWER to multi-agent games — AI Alignment Forum