Search papers, labs, and topics across Lattice.
1
0
2
MLLMs often start with the right visual predictions but override them for text, and we can reverse this bias with a simple intervention that boosts accuracy by nearly 10%.
