On “first critical tries” in AI alignment — AI Alignment Forum