Lookback Lens: Detecting and Mitigating Contextual Hallucinations in Large Language Models Using Only Attention Maps

TL;DR

This paper introduces Lookback Lens, a simple attention-based method to detect and reduce contextual hallucinations in large language models, improving answer accuracy without extensive retraining.

Contribution

It proposes a novel, efficient hallucination detection method using attention ratios that transfer across models and tasks, and demonstrates its effectiveness in reducing hallucinations.

Findings

Linear classifier using attention ratios matches complex detectors in accuracy.

Lookback Lens transfers across models without retraining.

Decoding guided by the detector reduces hallucinations by 9.6% in summarization.

Abstract

When asked to summarize articles or answer questions given a passage, large language models (LLMs) can hallucinate details and respond with unsubstantiated answers that are inaccurate with respect to the input context. This paper describes a simple approach for detecting such contextual hallucinations. We hypothesize that contextual hallucinations are related to the extent to which an LLM attends to information in the provided context versus its own generations. Based on this intuition, we propose a simple hallucination detection model whose input features are given by the ratio of attention weights on the context versus newly generated tokens (for each attention head). We find that a linear classifier based on these lookback ratio features is as effective as a richer detector that utilizes the entire hidden states of an LLM or a text-based entailment model. The lookback ratio-based detector -- Lookback Lens -- is found to transfer across tasks and even models, allowing a detector that is trained on a 7B model to be applied (without retraining) to a larger 13B model. We further apply this detector to mitigate contextual hallucinations, and find that a simple classifier-guided decoding approach is able to reduce the amount of hallucination, for example by 9.6% in the XSum summarization task.