NeuroCogMap maps the internal representations of LLMs onto functional systems, mechanistically identifies failure patterns such as hallucinations and bias, and simultaneously improves prediction of human brain activity.