Finding GPT-4’s mistakes with GPT-4

In three linesOpenAI introduces CriticGPT, a GPT-4-based model that writes critiques of ChatGPT responses. The tool assists human trainers in spotting mistakes during RLHF (reinforcement learning from human feedback) training.

Summary generated by Claude — human-verified