OpenAI’s new “CriticGPT” model is trained to criticize GPT-4 outputs https://arstechnica.com/?p=2034089