Detecting Token-Level Hallucinations Using Variance Signals: A Reference-Free Approach
This provides a lightweight, interpretable diagnostic tool for analyzing generative reliability in LLMs, addressing a critical issue for users and developers, though it is incremental as it builds on existing variance-based methods.
The paper tackles the problem of hallucinations in Large Language Models by introducing a reference-free, token-level detection framework using variance in token log-probabilities across stochastic generations, showing it reliably highlights instability and correlates with hallucination patterns on unanswerable questions from SQuAD v2 across models like GPT-Neo 125M, Falcon 1B, and Mistral 7B.
Large Language Models (LLMs) have demonstrated impressive generative capabilities across diverse tasks but remain susceptible to hallucinations, confidently generated yet factually incorrect outputs. We introduce a reference-free, token-level hallucination detection framework that leverages the variance in token log-probabilities across multiple stochastic generations. Unlike prior methods that require ground-truth references or sentence-level verification, our approach is model-agnostic, interpretable, and suited for real-time or post-hoc analysis. We evaluate our method on unanswerable question prompts from the SQuAD v2 dataset and benchmark across three autoregressive models of varying scales: GPT-Neo 125M, Falcon 1B, and Mistral 7B. Through both quantitative metrics and visual diagnostics, we show that token-level variance reliably highlights instability in model outputs and correlates with hallucination patterns. Our framework is lightweight, reproducible, and adaptable to multiple domains, offering a valuable diagnostic tool for analyzing generative reliability in LLMs.