04:00
2026-05-25
arxiv.org
large-language-models
Sparse Autoencoders Map Brain-LLM Alignment onto Cortical Semantic Topography
Researchers decomposed GPT-2 XL and Llama-3.1-8B into 16,000 to 32,000 interpretable features per layer using sparse autoencoders, finding that semantic features alone recovered 94% of peak brain-resp…