Search papers, labs, and topics across Lattice.
1
0
2
Task-asymmetric miscalibration in on-device language models leaves developers blind to when their outputs can be trusted, with confident-wrong outputs indistinguishable from confident-correct ones.
