LLMs cannot find reasoning errors, but can correct them

LLMs cannot find reasoning errors, but can correct them. 142 comments, 239 points on Hacker News.