OpenAI trains a GPT-4-based model, CriticGPT, to catch errors in ChatGPT code output

According to BlockBeats, on June 28, OpenAI said that it had trained a GPT-4-based model called CriticGPT to capture errors in ChatGPT code output. OpenAI began integrating models similar to CriticGPT into their RLHF to provide explicit AI assistance to OpenAI's trainers. OpenAI plans to further expand the application scale of RLHF on GPT-4 and put it into practice.