Synthetic Error Injection Fails to Elicit Self-Correction In Language Models

Open in new window