CriticGPT

On the 27th local time, OpenAI announced that it had trained a model called CriticGPT based on GPT-4 to find errors in the output of the ChatGPT chatbot. It can write comments to highlight inaccuracies in the answers generated by ChatGPT. According to reports, CriticGPT is designed to assist human AI trainers in their work - using a method called "Reinforcement Learning from Human Feedback (Note: Reinforcement Learning from Human Feedback)" to improve the accuracy of the chatbot's output.