How to solve deception and still fail. — AI Alignment Forum