Mechanisms of Prompt-Induced Hallucination in Vision-Language Models

TL;DR

This paper investigates how vision-language models hallucinate in response to prompts, identifying specific attention heads responsible for this behavior and demonstrating that their ablation reduces hallucinations significantly.

Contribution

It uncovers the internal attention mechanisms behind prompt-induced hallucinations and shows that targeted ablation of certain heads mitigates this issue without retraining.

Findings

A small set of attention heads are responsible for prompt copying behavior.

Ablating these heads reduces hallucinations by at least 40%.

Ablation increases model correction towards visual evidence.

Abstract

Large vision-language models (VLMs) are highly capable, yet often hallucinate by favoring textual prompts over visual evidence. We study this failure mode in a controlled object-counting setting, where the prompt overstates the number of objects in the image (e.g., asking a model to describe four waterlilies when only three are present). At low object counts, models often correct the overestimation, but as the number of objects increases, they increasingly conform to the prompt regardless of the discrepancy. Through mechanistic analysis of three VLMs, we identify a small set of attention heads whose ablation substantially reduces prompt-induced hallucinations (PIH) by at least 40% without additional training. Across models, PIH-heads mediate prompt copying in model-specific ways. We characterize these differences and show that PIH ablation increases correction toward visual evidence. Our findings offer insights into the internal mechanisms driving prompt-induced hallucinations, revealing model-specific differences in how these behaviors are implemented.