SquintBench

New 'Squint Text' Challenge Identified for Multimodal LLMs

Researchers have identified a new challenge for multimodal large language models (MLLMs) where they fail to read text embedded in an image's coarse structure, a phenomenon termed 'Squint Text.' To address this, they dev…