Toward More Reliable Artificial Intelligence: Reducing Hallucinations in Vision-Language Models