CRITIC-GPT, a model based on GPT-4, is used to critique and detect errors in responses from ChatGPT. It helps AI trainers identify inaccuracies and improve the quality of RLHF (Reinforcement Learning from Human Feedback). By integrating CRITIC-GPT, trainers can detect more issues and provide more comprehensive feedback compared to working without assistance. The model also reduces the number of “hallucinated” errors that AI alone generates.
Read more at: OpenAI – Finding GPT-4’s Mistakes with GPT-4
