Do VLMs Possess Human-Like Geometrical Intuition?

Demonstrated through experiments that Large Multimodal Models do not possess the same geometric intuitions innate in humans. Their inability to identify simple geometric properties that are trivial for humans, when contrasted against strong VQA benchmark performance, reveals that their inductive biases differ from those underlying human cognition.