Just MLPs: Efficient Visual State Reconstruction for Multimodal Language Models Paper • 2609.34972 • Published 4 days ago • 28