Search papers, labs, and topics across Lattice.
1
0
2
34
MLLMs struggle with visual prompts, showing a 17.8-point accuracy drop when questions are embedded in images, revealing a critical semantic gap in multimodal reasoning.
