logoalt Hacker News

mysterydiptoday at 11:31 AM0 repliesview on HN

If LLM training data is human-written, and LLM output mimics that input, how could you not have false positives?