Hallucination as an Anomaly: Dynamic Intervention via Probabilistic Circuits

TL;DR

This paper introduces PCNET, a probabilistic circuit-based method for detecting and correcting hallucinations in large language models by analyzing the latent space geometry, improving factual accuracy without degrading correct outputs.

Contribution

It presents a novel approach using probabilistic circuits to identify hallucinations as geometric anomalies, enabling targeted interventions during decoding.

Findings

Achieves up to 99% AUROC in hallucination detection across multiple benchmarks.

Outperforms state-of-the-art baselines on TruthfulQA with higher factuality scores.

Reduces mean corruption rate to 53.7%, preserving 79.3% of correct generations.

Abstract

One of the most critical challenges in Large Language Models is their tendency to hallucinate, i.e., produce factually incorrect responses. Existing approaches show promising results in terms of hallucination correction, but still suffer from a main limitation: they apply corrections indiscriminately to every token, corrupting also the originally correct generations. To overcome this drawback, we propose PCNET, a Probabilistic Circuit trained as a tractable density estimator over the LLM residual stream. The method detects hallucinations as geometric anomalies on the factual manifold, which is done via exact Negative Log-Likelihood computation, hence without the need for sampling, external verifiers, or weight modifications, as in existing techniques. To demonstrate its effectiveness, we exploit PCNET as a dynamic gate that distinguishes hallucinated from factual hidden states at each decoding step. This triggers our second main contribution, PC-LDCD (Probabilistic Circuit Latent Density Contrastive Decoding), only when the latent geometry deviates from factual regions, while leaving correct generations untouched. Across four LLMs, ranging from 1B to 8B models, and four benchmarks covering conversational reasoning, knowledge-intensive QA, reading comprehension, and truthfulness, PCNET achieves near-perfect hallucination detection across CoQA, SQuAD v2.0, and TriviaQA, with AUROC reaching up to 99%. Moreover, PC-LDCD obtains the highest True+Info, MC2, and MC3 scores on TruthfulQA in three out of four models, in comparison with state-of-the-art baselines, while reducing the mean corruption rate to 53.7% and achieving a preservation rate of 79.3%. Our proposed method is publicly available on GitHub.