Semantic Energy: Detecting LLM Hallucination Beyond Entropy

TL;DR

Semantic Energy is a new method for detecting hallucinations in LLMs by analyzing model confidence directly from logits, outperforming previous entropy-based approaches in various benchmarks.

Contribution

It introduces Semantic Energy, a novel uncertainty estimation framework that leverages logits and semantic clustering to better detect hallucinations in LLMs.

Findings

Semantic Energy improves hallucination detection accuracy.

It outperforms semantic entropy in uncertainty estimation.

The method enhances reliability for downstream applications.

Abstract

Large Language Models (LLMs) are being increasingly deployed in real-world applications, but they remain susceptible to hallucinations, which produce fluent yet incorrect responses and lead to erroneous decision-making. Uncertainty estimation is a feasible approach to detect such hallucinations. For example, semantic entropy estimates uncertainty by considering the semantic diversity across multiple sampled responses, thus identifying hallucinations. However, semantic entropy relies on post-softmax probabilities and fails to capture the model's inherent uncertainty, causing it to be ineffective in certain scenarios. To address this issue, we introduce Semantic Energy, a novel uncertainty estimation framework that leverages the inherent confidence of LLMs by operating directly on logits of penultimate layer. By combining semantic clustering with a Boltzmann-inspired energy distribution, our method better captures uncertainty in cases where semantic entropy fails. Experiments across multiple benchmarks show that Semantic Energy significantly improves hallucination detection and uncertainty estimation, offering more reliable signals for downstream applications such as hallucination detection.