# OpenAI Uses GPT-4 to Catch Its Own Mistakes with CriticGPT
OpenAI has announced CriticGPT, a new AI model designed to find errors in ChatGPT's responses. Built on the same GPT-4 technology that powers ChatGPT, CriticGPT acts as a quality control system that helps human trainers identify mistakes more effectively.
The tool addresses a growing challenge in AI development: as language models become more sophisticated, their errors become harder for humans to spot. CriticGPT assists during RLHF (Reinforcement Learning from Human Feedback), the process where human trainers evaluate and improve AI responses.
This development represents a significant shift in how AI systems are refined. Rather than relying solely on human reviewers to catch mistakes, OpenAI is now using AI to critique AIâessentially creating a self-improvement loop.