Search papers, labs, and topics across Lattice.
University at Buffalo
1
0
3
7
MMLMs can be made 99% safer against harmful multimodal inputs without sacrificing utility, thanks to a novel calibration approach.
