This research introduces a novel framework that analyzes large language models using methods from human brain studies, revealing how LLMs process information in networks rather than isolated neurons.

• Identified functional networks within LLMs that work together for specific language tasks
• Discovered hub neurons that play central roles in coordinating information flow
• Demonstrated how networks, rather than individual neurons, enable complex language capabilities
• Established direct parallels between human brain organization and LLM architecture

This interdisciplinary approach bridges neuroscience and AI, potentially enabling more brain-inspired AI architectures and offering new insights into both artificial and biological intelligence systems.

Brain-Inspired Exploration of Functional Networks and Key Neurons in Large Language Models