The Hidden Life of Tokens: Reducing Hallucination of Large Vision-Language Models via Visual Information Steering