logoalt Hacker News

supern0vatoday at 5:25 PM0 repliesview on HN

Exactly. LLMs regress to the mean. If you want don't want average, you need to feed the context to push it away from the mean.