Empirical evaluation on MATH500 shows no single unsupervised score is best for hallucination mitigation; effectiveness depends on the paired decoder and model capability, with self-verification performing well in most cases.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Score $\times$ Decoder: A Unified View of Unsupervised Inference-Time Scaling for Hallucination Mitigation
Empirical evaluation on MATH500 shows no single unsupervised score is best for hallucination mitigation; effectiveness depends on the paired decoder and model capability, with self-verification performing well in most cases.