§ feed · storyline

Finding GPT-4’s mistakes with GPT-4

CriticGPT, a GPT-4-based model, writes critiques of ChatGPT outputs to help human trainers identify errors during reinforcement learning from human feedback.

Jun 27 · 12:00:00 · primary fetch1 sourceupdated Jun 27 · 12:00:00

CriticGPT, a model based on GPT-4, writes critiques of ChatGPT responses to help human trainers spot mistakes during RLHF

read full article on openai.com ↗

§ sources1 publication · timeline below

openai.comFinding GPT-4’s mistakes with GPT-4primary12:00:00