Why context matters in VQA and Reasoning: Semantic interventions for VLM input modalities

Kenza Amara, Lukas Klein, Carsten Lüth +3

The various limitations of Generative AI, such as hallucinations and model failures, have made it crucial to understand the role of different modalities in Visual Language Model (V…