cs.AI, cs.LG

From Flat Facts to Sharp Hallucinations: Detecting Stubborn Errors via Gradient Sensitivity

arXiv:2605.00939v1 Announce Type: cross
Abstract: Traditional hallucination detection fails on “Stubborn Hallucinations” — errors where LLMs are confidently wrong. We propose a geometric solution: Embedding-Perturbed Gradient Sensitivity (EPGS). We h…