Researchers from MIT, Google DeepMind, and Oxford Unveil Why Vision-Language Models Do Not Understand Negation and Proposes a Groundbreaking Solution