• LLM-based alpha mining frameworks provide a scalable interface between human expertise and full automation, enabling rapid transformation of qualitative hypotheses into testable alpha factors.
• LLMs serve multiple functional roles in alpha mining—as miners, evaluators, and interactive assistants—offering semantic depth alongside computational speed.
• Critical remaining challenges include simplified performance evaluation, limited numerical reasoning, lack of diversity and originality, weak exploration dynamics, temporal data leakage, and black-box/compliance risks.
• Future research should focus on reasoning alignment, new data modalities, improved evaluation protocols, and integration of LLMs into general-purpose quantitative systems to realize a complementary human-AI paradigm.