Search papers, labs, and topics across Lattice.
1
0
3
MLLMs still fumble at visual tool use, struggling to compose even basic OpenCV operations into effective plans, as revealed by a new benchmark where the best model only scores 51%.