Imagine peering inside the 'mind' of an AI like ChatGPT and seeing a map of its cognitive functions — much like neuroscientists map human brain regions. A team of researchers led by Zhongxiang Sun has done just that with NeuroCogMap, a cognitive neuroscience-inspired framework that organizes internal features of large language models (LLMs) into functional parcels. Their paper, posted on arXiv in July 2026, shows that these parcels are stable, semantically coherent, and partly conserved across different models. Moreover, they found that major AI failures like hallucination, bias, refusal failure, and sycophancy correspond to distinct disruptions in these representational and behavioral-control systems, enabling mechanism-guided detection and intervention.
The Research
The study, published on arXiv with 14 authors from institutions including Chinese Academy of Sciences and Renmin University of China, analyzed internal representations from several LLMs. NeuroCogMap parcels the models' internal features into interpretable functional units. The parcels form a stable organization that is functionally linked to model outputs. The researchers identified that common LLM failures — hallucination, bias, refusal failure, and sycophancy — are tied to specific disruptions within this organization. Importantly, NeuroCogMap improved prediction of human cortical responses during naturalistic language comprehension, especially in higher-order association cortex areas. This suggests that the internal representations of LLMs share organizational principles with the human brain.
Why It Matters for Your Brain
This framework not only helps us understand and improve AI but also sheds light on human cognition. By revealing latent strategies in LLMs that refine classical models of human decision-making, NeuroCogMap offers a new tool for cognitive research. For you, this means that advances in AI neuroscience can directly inform our understanding of how your own brain organizes cognitive functions, potentially leading to better brain training methods and assessments.
What You Can Do
Stay curious about how your own mind works. Engage with brain-training exercises that challenge multiple cognitive domains, as the brain's functional organization is key to efficient learning. Understanding that cognitive functions are organized into coordinated systems can help you approach learning and problem-solving in a more holistic way.
Source: arXiv q-bio.NC
Curious about your own brain? Take our free adaptive IQ test or try 306 brain training levels.
