Aligning LLM with Human Neural Pathways: Enhancing Cognitive Synergy

ACL ARR 2025 May Submission2527 Authors

Abstract: Recent studies have highlighted the similarities between human neural pathways and the operational mechanisms of Large Language Models (LLMs). While the Solo Performance Prompting (SPP) exhibits cognitive synergy akin to human collaboration in LLMs like GPT-4, it faces limitations in generalizability and defining the conditions for cognitive synergy emergence. To address these issues, we introduce Brain Performance Prompting (BPP), an innovative framework inspired by human neural pathways. BPP dynamically activates task-specific brain region personas to LLMs, enhancing self-collaboration and advancing cognitive synergy beyond SPP. Moreover, BPP offers new insights into the emergence of cognitive synergy by revealing its partial presence in smaller models, including GPT-4o-mini, Qwen-2.5-7B-Instruct, and Llama-3.1-8B-Instruct. Our experiments demonstrate that BPP significantly outperforms SPP and other approaches across knowledge-intensive and reasoning-intensive tasks on GPT-4o. These findings suggest that drawing inspiration from human brain information processing principles can play a crucial role in optimizing LLM performance. Our code will be made publicly available upon acceptance.

Paper Type: Long

Research Area: NLP Applications

Research Area Keywords: NLP Applications, prompt engineering

Contribution Types: Model analysis & interpretability, NLP engineering experiment, Approaches to low-resource settings

Languages Studied: English

Submission Number: 2527