Making LLMs more accurate by using all of their layers | Not Hacker News!