New Framework Reduces AI Hallucinations in Multi-Model Vision Systems

Researchers introduced SCoOP, a training-free framework that combines multiple vision-language models while measuring system-level uncertainty to detect hallucinations. The method improved hallucination detection accuracy by 10-13% on benchmark tests with negligible computational overhead, advancing reliability in multimodal AI systems.