logoalt Hacker News

bitmasher9yesterday at 11:56 PM0 repliesview on HN

I often have the expensive models give relatively simple inaccurate answers, even when they cite sources that directly contradict them. The error rate is lower, but you can’t have confidence with llm answers.