Research map / Large language models / Language, safety and interpretability
Scaling laws for language encoding models in fMRI
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| scale language model from OPT-125M to OPT-30B | fMRI semantic encoding model · fMRI encoding, 3 subjects | encoding performance (prediction correlation): ~4.4% per order of magnitude; ~15% overall | Helped |
| scale OPT beyond ~30B parameters (66B, 175B) | fMRI semantic encoding model · fMRI encoding, ~33,000 training points | encoding performance | No clear effect |
| increase fMRI training stories by 10x | OPT-125M encoding model · fMRI encoding | encoding performance: +122% per order of magnitude | Helped |
| scale Whisper encoder size | fMRI audio encoding model · fMRI encoding | encoding performance: +32.2% per order of magnitude | Helped |
| increase training data by 10x | Whisper Large encoding model · fMRI encoding | encoding performance: +144% per order of magnitude | Helped |
| use LLaMA instead of OPT language models | fMRI semantic encoding model · fMRI encoding | encoding performance: about 5% higher correlation | Helped |
| add Whisper layers via stacked regression to LLaMA-33B layer 18 (ablation) | semantic encoding model · fMRI encoding voxels | encoding performance | Mixed |
| scale Whisper size, within auditory cortex | fMRI audio encoding model · auditory cortex voxels | encoding performance | Helped |
| allow long contexts that reveal story position (no exclusion of first 100 s) (ablation) | encoding model evaluation · fMRI encoding | measured performance: inflates measured performance by up to 20% | Hurt |
Closest papers by meaning
- fMRI predictors based on language models of increasing complexity recover brain left lateralization (NeurIPS 2024 · similarity 0.92)
- The Brain's Bitter Lesson: Scaling Speech Decoding With Self-Supervised Learning (ICML 2025 · similarity 0.88)
- Scaling and context steer LLMs along the same computational path as the human brain (NeurIPS 2025 · similarity 0.88)
- OWLS: Scaling Laws for Multilingual Speech Recognition and Translation Models (ICML 2025 · similarity 0.88)
- BrainLM: A foundation model for brain activity recordings (ICLR 2024 · similarity 0.87)
- Neural Encoding and Decoding at Scale (ICML 2025 · similarity 0.87)
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.