#vlm

Vision-Language Models Fail on Dense Visual Grids

A new benchmark reveals VLMs collapse sharply on simple grid-reading tasks, exposing a gap between visual encoding and language output called Digital Agnosia.