Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster B in Poster Session B: Tuesday, August 4, 2:00 – 3:45 pm, Kimmel Center, Shorin & Rosenthal Rooms

Divergent scaling patterns in the auditory cortex vs. other brain regions reveal distinct sources of LLM-brain alignment

Taha Osama A Binhuraib¹, Anna A Ivanova¹; ¹Georgia Institute of Technology

Presenter: Taha Osama A Binhuraib

Large language model (LLM) features predict neural responses during story listening, and this predictivity is often assumed to improve with scaling. Here we show that the scaling patterns differ sharply across cortical systems. In naturalistic story-listening fMRI, encoding performance in the language network increased with representational dimensionality, training progression, model size, and number of training stories. Category-selective visual regions showed scaling profiles similar to the language network. In contrast, auditory cortex showed weak or non-monotonic gains. We further show that low-level features (word/phoneme rate) explain the majority of encoding model performance in the auditory cortex but not elsewhere. Our results demonstrate that successful LLM- based brain prediction can arise from different underlying signal features across cortical systems.

Topic Area: Auditory, Speech & Language Processing