Bookmark · 28 June 2024, 12:28 OpenAI’s new “CriticGPT” model is trained to criticize GPT-4 outputs | Ars Technica mltryme