1 paper
Kailin Jiang, Yuntao Du, Yukai Ding +7
Large Multimodal Models (LMMs) store vast amounts of pretrained knowledge but struggle to remain aligned with real-world updates, making it difficult to avoid capability degradatio…