This paper presents a novel framework for analyzing and comparing the operational principles of large language models (LLMs) with biological cognitive processes. Despite the LLM's complex structure and numerous parameters, we explore its intermodular interactions and functional characteristics using a network-based approach. Specifically, we reveal that LLM modules exhibit patterns similar to distributed yet interconnected cognitive structures observed in the brains of birds and small mammals. This distinction from biological systems highlights the importance of dynamic interregional interactions and neural plasticity in LLM skill acquisition. This analysis enhances the interpretability of LLMs and suggests that leveraging distributed learning dynamics as an effective fine-tuning strategy is crucial.
