Aligning brains into a shared space improves their alignment with LLMs
6 hours ago
- #language-models
- #neuroscience
- #brain-mapping
- Large language models can predict neural activity during natural language processing.
- A shared response model enhances encoding performance by 37% (from r = 0.188 to r = 0.257).
- Greatest improvements occur in brain areas specialized for language comprehension, like the superior temporal gyrus and inferior frontal gyrus.
- The study uses electrocorticography data from eight participants listening to the same podcast.
- The shared space allows denoising of individual brain responses and better generalization across individuals.