This research demonstrates how explainable AI techniques applied to large language models can predict actual brain activity patterns during language processing.

• Uses attribution methods to quantify word influence in LLM predictions
• Applies these explanations to successfully predict fMRI data from human participants
• Establishes stronger links between computational models and biological language processing
• Suggests new ways to understand both LLMs and human cognition

This work advances medical neuroscience by providing novel methods to analyze brain imaging data, potentially improving our understanding of language disorders and creating more brain-like AI systems.

Explanations of Large Language Models Explain Language Representations in the Brain