Illusions of Confidence? Diagnosing LLM Truthfulness via Neighborhood Consistency

TL;DR

This paper introduces Neighbor-Consistency Belief (NCB), a structural measure to evaluate and improve the robustness of LLMs' factual beliefs under contextual perturbations, addressing limitations of point-wise confidence metrics.

Contribution

It proposes NCB as a new belief robustness metric, a stress-testing protocol for LLMs, and a Structure-Aware Training method to enhance belief stability.

Findings

High-NCB data resists contextual interference better.

NCB correlates with response stability across models.

SAT reduces knowledge brittleness by approximately 30%.

Abstract

As Large Language Models (LLMs) are increasingly deployed in real-world settings, correctness alone is insufficient. Reliable deployment requires maintaining truthful beliefs under contextual perturbations. Existing evaluations largely rely on point-wise confidence like Self-Consistency, which can mask brittle belief. We show that even facts answered with perfect self-consistency can rapidly collapse under mild contextual interference. To address this gap, we propose Neighbor-Consistency Belief (NCB), a structural measure of belief robustness that evaluates response coherence across a conceptual neighborhood. To validate the efficiency of NCB, we introduce a new cognitive stress-testing protocol that probes outputs stability under contextual interference. Experiments across multiple LLMs show that the performance of high-NCB data is relatively more resistant to interference. Finally, we present Structure-Aware Training (SAT), which optimizes context-invariant belief structure and reduces long-tail knowledge brittleness by approximately 30%. Code is available at https://github.com/zjunlp/belief.