A model flagging potentially harmful user content is being tuned. Missing genuinely harmful content is far more damaging than reviewing something safe unnecessarily. Which metric should be prioritised and what follows?

Written from the published competencies and our own syllabus, written from the published curricula of BlueDot Impact, the Center for AI Safety, DeepMind and Stanford. Not actual exam questions, and not affiliated with or endorsed by Product Digest.