HalluSAE models LLM generation as a trajectory in a potential energy landscape using sparse autoencoders to locate hallucination transition zones and attributes errors to specific sparse features for improved detection.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
HalluSAE: Detecting Hallucinations in Large Language Models via Sparse Auto-Encoders
HalluSAE models LLM generation as a trajectory in a potential energy landscape using sparse autoencoders to locate hallucination transition zones and attributes errors to specific sparse features for improved detection.