Visual Anchors Are Strong Information Aggregators For Multimodal Large Language Model | Read Paper on Bytez