ryan_greenblatt comments on Chain of Thought Monitorability: A New and Fragile Opportunity for AI Safety