Scaling Wargaming for Global Catastrophic Risks with AI — LessWrong