Overthinking Causes Hallucination: Tracing Confounder Propagation in Vision Language Models
This research addresses the critical problem of hallucination in Vision Language Models, which impacts the reliability and trustworthiness of AI systems for all users.
Vision Language Models (VLMs) frequently hallucinate non-existent objects, and this paper demonstrates that this issue stems from a process called 'overthinking,' where models repeatedly revise object hypotheses across layers before committing to an incorrect answer. By introducing the Overthinking Score, the authors significantly improved hallucination detection, achieving 78.9% F1 on MSCOCO and 71.58% on AMBER.
Vision Language models (VLMs) often hallucinate non-existent objects. Detecting hallucination is analogous to detecting deception: a single final statement is insufficient, one must examine the underlying reasoning process. Yet existing detectors rely mostly on final-layer signals. Attention-based methods assume hallucinated tokens exhibit low attention, while entropy-based ones use final-step uncertainty. Our analysis reveals the opposite: hallucinated objects can exhibit peaked attention due to contextual priors; and models often express high confidence because intermediate layers have already converged to an incorrect hypothesis. We show that the key to hallucination detection lies within the model's thought process, not its final output. By probing decoder layers, we uncover a previously overlooked behavior, overthinking: models repeatedly revise object hypotheses across layers before committing to an incorrect answer. Once the model latches onto a confounded hypothesis, it can propagate through subsequent layers, ultimately causing hallucination. To capture this behavior, we introduce the Overthinking Score, a metric to measure how many competing hypotheses the model entertains and how unstable these hypotheses are across layers. This score significantly improves hallucination detection: 78.9% F1 on MSCOCO and 71.58% on AMBER.