This paper introduces Alpha-GPT, a novel human-AI interactive system for alpha mining in quantitative investment. The research addresses limitations of traditional methods such as hand-crafted factor synthesizing and algorithmic factor mining (e.g., genetic programming), which often struggle to effectively implement quant researchers' ideas. The authors propose a new paradigm that combines human expertise with artificial intelligence through interactive processes, specifically leveraging large language models with a specialized prompt engineering framework.

The paper develops Alpha-GPT as a practical implementation of this paradigm, creating a system that can heuristically 'understand' quant researchers' concepts and generate creative, insightful, and effective trading signals. Through multiple alpha mining experiments, the authors demonstrate that their approach outperforms traditional methods, offering significant advantages in quantitative investment research by bridging the gap between human intuition and algorithmic efficiency.