Research map / Large language models / Language, safety and interpretability
When Language Models Lose Their Mind: The Consequences of Brain Misalignment
Gabriele Merlin, Mariya Toneva
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 |
|---|---|---|---|
| train the model to be brain misaligned instead of brain preserving | BERT/GPT2/Llama · Holmes linguistic competence (200+ tasks) | average win rate | Hurt |
| brain misalignment on GPT2 trained on Harry Potter | GPT2 · Holmes linguistic competence (Harry Potter) | overall win: p-value = 0.055 | Mixed |
| brain misalignment across linguistic subfields (discourse, morphology, reasoning, semantics, syntax) | Brain Misaligned model · Holmes subfields | win rate | Hurt |
| brain-tune (increase brain alignment) instead of brain preserving | Brain Preserving model · Holmes linguistic competence | average win rate | Improved |
| brain-tune compared with original pretrained model | Pretrained model · Holmes linguistic competence | win rate | Mixed |
| brain tuning on semantics and syntax subfields | Brain Preserving model · Holmes subfields | win rate | Improved |
Closest papers by meaning
- Brain-tuning Improves Generalizability and Efficiency of Brain Alignment in Speech Models (NeurIPS 2025 · similarity 0.91)
- Scaling and context steer LLMs along the same computational path as the human brain (NeurIPS 2025 · similarity 0.90)
- Improving Semantic Understanding in Speech Language Models via Brain-tuning (ICLR 2025 · similarity 0.90)
- Far from the Shallow: Brain-Predictive Reasoning Embedding through Residual Disentanglement (NeurIPS 2025 · similarity 0.88)
- Divergences between Language Models and Human Brains (NeurIPS 2024 · 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.
