AI and the World's Hard Problems · Life sciences · arXiv · 2026-10-19
What Shapes Brain Alignment of Large Language Models during Dialogue? Insights from Language Production and Comprehension
The study examines how large language model (LLM) representations align with brain activity recorded by fMRI during dialogue, covering both language production and comprehension. Alignment was assessed across LLM layers and brain regions, along with its link to the intrinsic dimension of LLM representations and the contributions of several linguistic and communicative features. The results show distinct alignment patterns for production and comprehension, with strong associations with intrinsic dimension in the brain regions linked to each mode.