Scaling human oversight of AI systems for difficult-to-evaluate tasks (openai.com)

Addresses methods for scaling human feedback and oversight of AI systems performing tasks where quality is difficult to measure or evaluate automatically.