New task, VICIS, evaluates vision-language models' ability to infer visual concepts from example images. Current VLMs struggle with this reasoning, performing poorly on the newly defined task. This research highlights a key gap in VLM visual understanding.
Opening Kapyn…