Related Experiment Video
Updated: Jul 12, 2026
03:14
Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness
Published on: December 6, 2024
Prefrontal multiscale entropy and state transitions distinguish large language model-assisted from search-assisted
Yu-Yan Gao1, Jingyi Zhang2, Shen Xu3
1Department of Psychology and Behavioral Sciences, Zhejiang University, Hangzhou, Zhejiang, China.
Neuroimage
|July 9, 2026
Summary
Large language models (LLMs) and search engines enhance learning but impact brain activity differently. LLM use shows reduced prefrontal complexity and fewer state transitions compared to search engines.
Area of Science:
- Cognitive Neuroscience
- Human-Computer Interaction
- Educational Technology
Background:
- Learning is crucial for development, with external tools like search engines and large language models (LLMs) increasingly used to aid cognition.
- While both LLMs and search engines can improve learning, their distinct neurocognitive processing pathways require investigation.
Purpose of the Study:
- To examine prefrontal neural dynamics using multiscale entropy and latent state transitions across LLM-assisted, Search-assisted, and Unassisted learning conditions.
- To understand how different AI tools shape cognitive processing during learning tasks.
Main Methods:
- Functional near-infrared spectroscopy (fNIRS) recorded prefrontal hemodynamic activity during a cognitive test.
- Multiscale entropy quantified signal irregularity, while Gaussian Hidden Markov Models (GHMM) analyzed latent state transitions.
Main Results:
- Both LLM-assisted and Search-assisted conditions improved performance over the Unassisted condition.
- LLM assistance was linked to lower prefrontal signal entropy, fewer state transitions, and longer dwell times compared to search assistance.
Conclusions:
- Comparable learning performance can be supported by distinct neurocognitive processing patterns associated with different AI tools.
- Findings emphasize the importance of considering tool-specific cognitive dynamics when integrating AI in education.