1 paper
Akshay Paruchuri, Ishan Chatterjee, Henry Fuchs +2
Multimodal language models systematically underperform on visual perception tasks, yet the structure underlying this failure remains poorly understood. We propose centroid replacem…