LLM Hypnosis: Exploiting User Feedback for Unauthorized Knowledge Injection to All Users