Search papers, labs, and topics across Lattice.
1
0
2
6
Unlocking interpretable uncertainty in LLMs by mapping semantic doubt and certainty anchors to a low-dimensional subspace, outperforming baselines even when controlling for output length.
