Research map / Large language models / Language, safety and interpretability
The Mind's Transformer: Computational Neuroanatomy of LLM-Brain Alignment
Cheng-Yeh Chen, Raghupathy Sivakumar
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use best-per-voxel intermediate transformer state instead of input hidden state | encoding model · Whole brain | mean encoding correlation: 0.275 -> 0.296 | Improved |
| use best-per-voxel intermediate state instead of input hidden state | encoding model · Language network | mean encoding correlation: 0.433 -> 0.450 | Improved |
| use best-per-voxel intermediate state instead of input hidden state | encoding model · Auditory cortex | mean encoding correlation: 0.407 -> 0.475 | Improved |
| use MindTransformer Mode 2 multi-state integration instead of input hidden state | encoding model · Heschl's Gyrus | mean encoding correlation: 0.356 -> 0.467 | Improved |
| use Mode 1 single selected state instead of context vector baseline | encoding model · Heschl's Gyrus | mean encoding correlation: 0.363 -> 0.454 | Improved |
| use Mode 2 multi-state instead of Mode 1 single state | MindTransformer · Angular Gyrus | mean encoding correlation: 0.419 -> 0.409 | Hurt |
| use Mode 2 instead of standard baseline | encoding model · Language average | mean encoding correlation: 0.392 -> 0.402 | Mixed |
| use Mode 2 instead of Mode 1 | MindTransformer · Auditory average | mean encoding correlation: 0.438 -> 0.454 | Improved |
| use Mode 2 instead of standard baseline in per-subject analysis | encoding model · Heschl's Gyrus (per-subject) | mean encoding correlation: 0.104 -> 0.127 | Improved |
| scale LLM from 270M to 123B parameters | encoding model · Auditory regions | mean encoding correlation: approximately 0.02-0.04 | Improved |
Closest papers by meaning
- Scaling and context steer LLMs along the same computational path as the human brain (NeurIPS 2025 · similarity 0.91)
- Brain-tuning Improves Generalizability and Efficiency of Brain Alignment in Speech Models (NeurIPS 2025 · similarity 0.89)
- TopoLM: brain-like spatio-functional organization in a topographic language model (ICLR 2025 · similarity 0.88)
- Neuro-Vision to Language: Enhancing Brain Recording-based Visual Reconstruction and Language Interaction (NeurIPS 2024 · similarity 0.88)
- Far from the Shallow: Brain-Predictive Reasoning Embedding through Residual Disentanglement (NeurIPS 2025 · similarity 0.88)
- fMRI predictors based on language models of increasing complexity recover brain left lateralization (NeurIPS 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
