Cross-Modal Noncommutativity Affects Multimodal LLM Judgments

Research shows that the order of conflicting evidence significantly alters multimodal LLM outputs, a phenomenon termed cross-modal noncommutativity. In vision and speech tasks, placing perceptual evidence after conflicting text consistently shifts model answers toward the latter, revealing a positional bias in evidence integration.