Study Finds Natural-language Training, Not Code, Shapes Brain-aligned Model Geometry
A new arXiv paper reports that code-trained language models can achieve positive brain scores, and examines whether abstract structure from programming languages or natural-language training drives this brain-aligned geometry.
Read the full story at arXiv cs.CL ↗
Timeline · 1 report
- 2026-11-09 05:00 · arXiv cs.CL
Language, Not Code Abstraction: Natural-language Training Shapes Brain-aligned Geometry