AI-written critiques help humans spot flaws in summaries (openai.com)

Summary:

• AI models trained to write critiques can help humans identify more flaws in summaries
• Larger models are better at self-critiquing, showing promise for AI-assisted human supervision