Multimodal LLM Alignment with Human Theory of Mind and fMRI
October 19, 2026
A study evaluates MLLMs' ability to predict social and affective content in cartoons, comparing diffusion-based and autoregressive models against human Theory of Mind (ToM) ratings. Model ratings showed significant correlation with regional fMRI responses, particularly for affective valence.
HOW THIS AFFECTS YOU
●
researcherYou can leverage MLLM affective ratings as proxies for human neural responses in social cognition research.