Lyapunov Probes for Hallucination Detection in Large Foundation Models

TL;DR

This paper introduces Lyapunov Probes, a novel method for detecting hallucinations in large language models by analyzing the stability of their representations through dynamical systems theory, leading to more reliable hallucination detection.

Contribution

The paper proposes Lyapunov Probes, a new stability-based approach for hallucination detection in LLMs and MLLMs, leveraging derivative constraints and systematic perturbation analysis.

Findings

Lyapunov Probes outperform existing baselines in hallucination detection.

The method effectively identifies unstable, hallucination-prone regions.

Experiments show consistent improvements across diverse datasets and models.

Abstract

We address hallucination detection in Large Language Models (LLMs) and Multimodal Large Language Models (MLLMs) by framing the problem through the lens of dynamical systems stability theory. Rather than treating hallucination as a straightforward classification task, we conceptualize (M)LLMs as dynamical systems, where factual knowledge is represented by stable equilibrium points within the representation space. Our main insight is that hallucinations tend to arise at the boundaries of knowledge-transition regions separating stable and unstable zones. To capture this phenomenon, we propose Lyapunov Probes: lightweight networks trained with derivative-based stability constraints that enforce a monotonic decay in confidence under input perturbations. By performing systematic perturbation analysis and applying a two-stage training process, these probes reliably distinguish between stable factual regions and unstable, hallucination-prone regions. Experiments on diverse datasets and models demonstrate consistent improvements over existing baselines.