People are experiencing mental health crises and physical self-harm due to their interactions with Large Language Models (LLMs)