Search papers, labs, and topics across Lattice.
Jilin University
1
0
2
Spotting unfaithful reasoning in LLMs just got easier: a new method efficiently compares a model's internal computations against its stated rationale.