Sparse Autoencoders Learn Monosemantic Features in Vision-Language Models Paper • 2504.02821 • Published Apr 3 • 10