(How) Do MLLMs Report Bistable Images Like Humans?

Title:(How) Do MLLMs Report Bistable Images Like Humans?

Abstract:Bistable images such as the duck-rabbit are classic stimuli in which one image supports multiple mutually incompatible interpretations, typically reported one at a time in humans. We ask whether multimodal large language models (MLLMs) show similar report behavior and what internal computations support it. Using the LLaVA family, we study two tractable dimensions: modulability, whether reports can be biased by bottom-up visual cues and top-down linguistic priors, and exclusivity, whether responses commit to a single interpretation. We test both on the canonical duck-rabbit and on synthetic Visual Anagrams to mitigate memorization confounds. Behaviorally, both visual and linguistic manipulations systematically shift reports in human-consistent ways, while responses remain predominantly exclusive. Mechanistically, these effects arise from competing image-token representations, distinct pathways for bottom-up and top-down modulation, and a link between exclusive reporting and object-count encoding. Code and data are available at this https URL.

Accepted to EMNLP 2026 (Main Conference). 31 pages, 46 figures, 2 tables

Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)

arXiv:2609.13254 [cs.CV]

(or arXiv:2609.13254v1 [cs.CV] for this version)

https://doi.org/10.48550/arXiv.2609.13254 Focus to learn more arXiv-issued DOI via DataCite

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.