Just MLPs: Efficient Visual State Reconstruction for Multimodal Language Models Paper • 2609.34972 • Published 1 day ago • 14