CLR's recent work on multi-agent systems — AI Alignment Forum