Complementing Self-Consistency with Cross-Model Disagreement for Uncertainty Quantification

Kimia Hamidieh, Veronika Thost, Walter Gerych, Mikhail Yurochkin, Marzyeh Ghassemi · 2026

Read Complementing Self-Consistency with Cross-Model Disagreement for Uncertainty Quantification by Kimia Hamidieh, Veronika Thost, Walter Gerych, Mikhail Yurochkin, Marzyeh Ghassemi (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-17112.txt.

Description

Hamidieh and four colleagues improve uncertainty scores for language models that sound sure while wrong. Self-consistency, the usual proxy for chance noise, collapses when an overconfident model repeats the same wrong answer. Their epistemic term compares answers across a small ensemble of similar-scale models, measuring the gap between inter-model and intra-model semantic similarity. Their sum improves ranking and selective silence.

Catalog record

CREATORKimia Hamidieh, Veronika Thost, Walter Gerych, Mikhail Yurochkin, Marzyeh Ghassemi
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE69091
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.17112

Browse the full library · About the Tower of Babel