Detecting AI Hallucinations in Finance: An Information-Theoretic Method Cuts Hallucination Rate by 92%

TL;DR

This paper introduces ECLIPSE, an information-theoretic framework that significantly reduces hallucinations in large language models for finance by detecting unsupported answers through entropy and evidence utilization analysis.

Contribution

ECLIPSE is a novel entropy-capacity method that effectively detects hallucinations in LLMs by combining entropy estimation and evidence utilization metrics, with proven convexity and stability.

Findings

ECLIPSE achieves ROC AUC of 0.89 and precision of 0.90 on financial QA dataset.

It outperforms entropy-only baseline with AUC of 0.50.

Effectiveness depends on calibrated token-level uncertainties, confirmed by ablation studies.

Abstract

Large language models (LLMs) produce fluent but unsupported answers - hallucinations - limiting safe deployment in high-stakes domains. We propose ECLIPSE, a framework that treats hallucination as a mismatch between a model's semantic entropy and the capacity of available evidence. We combine entropy estimation via multi-sample clustering with a novel perplexity decomposition that measures how models use retrieved evidence. We prove that under mild conditions, the resulting entropy-capacity objective is strictly convex with a unique stable optimum. We evaluate on a controlled financial question answering dataset with GPT-3.5-turbo (n=200 balanced samples with synthetic hallucinations), where ECLIPSE achieves ROC AUC of 0.89 and average precision of 0.90, substantially outperforming a semantic entropy-only baseline (AUC 0.50). A controlled ablation with Claude-3-Haiku, which lacks token-level log probabilities, shows AUC dropping to 0.59 with coefficient magnitudes decreasing by 95% - demonstrating that ECLIPSE is a logprob-native mechanism whose effectiveness depends on calibrated token-level uncertainties. The perplexity decomposition features exhibit the largest learned coefficients, confirming that evidence utilization is central to hallucination detection. We position this work as a controlled mechanism study; broader validation across domains and naturally occurring hallucinations remains future work.