OpenAI has launched the CriticGPT model to capture errors in ChatGPT code outputs

ChainCatcher news, according to Caixin, OpenAI has launched a new model based on GPT-4------CriticGPT, designed to capture errors in ChatGPT's code output.

The role of CriticGPT is akin to allowing people to use GPT-4 to find errors in GPT-4. This model can provide critical comments on ChatGPT's response results, thereby helping AI trainers identify errors discovered during the "Reinforcement Learning from Human Feedback (RLHF)" process, offering clear assistance to AI trainers.

Research has found that when trainers use CriticGPT to review ChatGPT's output code, their performance is 60% higher than those who do not receive assistance. Moreover, with the help of CriticGPT, their corrections are more comprehensive than when they work alone.