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

TL;DR

This paper introduces a new uncertainty measure combining self-consistency and cross-model disagreement to improve the reliability of large language models in identifying incorrect responses.

Contribution

It proposes an epistemic uncertainty term based on model disagreement, enhancing uncertainty quantification beyond traditional self-consistency methods.

Findings

Total uncertainty improves calibration and abstention.

Cross-model disagreement flags confident failures effectively.

EU complements AU especially when AU is low.

Abstract

Large language models (LLMs) often produce confident yet incorrect responses, and uncertainty quantification is one potential solution to more robust usage. Recent works routinely rely on self-consistency to estimate aleatoric uncertainty (AU), yet this proxy collapses when models are overconfident and produce the same incorrect answer across samples. We analyze this regime and show that cross-model semantic disagreement is higher on incorrect answers precisely when AU is low. Motivated by this, we introduce an epistemic uncertainty (EU) term that operates in the black-box access setting: EU uses only generated text from a small, scale-matched ensemble and is computed as the gap between inter-model and intra-model sequence-semantic similarity. We then define total uncertainty (TU) as the sum of AU and EU. In a comprehensive study across five 7-9B instruction-tuned models and ten long-form tasks, TU improves ranking calibration and selective abstention relative to AU, and EU reliably flags confident failures where AU is low. We further characterize when EU is most useful via agreement and complementarity diagnostics.