
Is it a duck, or is it a rabbit? (How) do multimodal large language models (MLLMs) report bistable images? Terrific new work led by @SussexCentre PhD student @rtakatsky - to be presented at #EMNLP2026. Paper and explainer video below 👇🏽
Visual cues or expectations can flip what we see in bistable images—yet we typically see only one interpretation at a time. 🦆🐇 Do MLLMs show human-like reports? If so, how? Excited to share our #EMNLP2026 Main paper: “(How) Do MLLMs Report Bistable Images Like Humans?” (1/3)


















