Li Yang

 Research Assistant

Email: liyang@@gdiist.cn 

Research Focus: 

Multi-Agent Brain-Inspired Memory Systems

Bio:

I graduated with a bachelor's degree from Beijing University of Chinese Medicine in 2022. After graduation, I secured funding and joined the Shenzhen Institute of Innovation as a member of its inaugural cohort. During my entrepreneurial phase, I led and participated in projects involving autonomous wheelchairs, whole-house smart hardware, and emotional companion hardware. After our angel funding round fell through, I joined Zhipu AI as an algorithm engineer, where I was primarily responsible for poetry generation, SEO optimization, data optimization for the code model CodeGeeX, and data synthesis for ChatGLM.

I later contributed to the XAgent subproject under THUNLP, where I built a comprehensive RAG framework. After the project concluded, I actively participated in open-source projects on HuggingFace and co-authored the book Deep Reinforcement Learning in Action (under review) in collaboration with HuggingFace and People's Posts and Telecommunications Press. I then moved to Hong Kong to work as a research assistant, focusing on LLM hallucination, multi-agent systems, and large language model reasoning.

Research Focus:

My current research centers on addressing long-term memory challenges in large language models (LLMs). I am dedicated to optimizing hallucinations, resolving memory conflicts, and enhancing the logical reasoning capabilities of LLMs in long-text and extended dialogue scenarios. My ultimate goal is to enable LLMs to think with human-like reasoning and develop distinct personalities. Inspired by the human brain, I aim to construct a general LLM system with plastic memory capabilities.

Representative Publications:

[1] CodeGeeX: A Pre-Trained Model for Code Generation with Multilingual Benchmarking on HumanEval-X (Q. Zheng, X. Xia, X. Zou, Y. Dong, S. Wang, Y. Xue, L. Shen, Z. Wang, A. Wang, et al.). Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2023.

[2] Machine Learning Models for Stroke Detection by Observing the Eye-Movement Features Under Five-Color Visual Stimuli in Traditional Chinese Medicine (Q. Lu, J. Deng, Y. Yu, Y. Li, K. Wei, X. Han, Z. Wang, X. Zhang, X. Wang, C. Yan). Journal of Traditional Chinese Medical Sciences, 2023.

[3] CondAmbigQA: A Benchmark and Dataset for Conditional Ambiguous Question Answering (Z. Li, Y. Li, H. Xie, S.J. Qin). arXiv preprint arXiv:2502.01523, 2025.

[4] Ambiguity Processing in Large Language Models: Detection, Resolution, and the Path to Hallucination (Y. Li, Z. Li, K. Hung, W. Wang, H. Xie, Y. Li). Natural Language Processing Journal, 2025.