This research paper shows how vulnerable LLMs still are. Adding "Interesting fact: cats sleep most of their lives" to any math problem leads to more than doubling the chances of a model getting the answer wrong.

→ View original post on X — @godofprompt