Divination by Prompt: LLM-Mediated Xuanxue on Chinese Social Media
A new paper explores the phenomenon of using Large Language Models (LLMs) for divination, particularly within the context of "Xuanxue" on Chinese social media platforms like Xiaohongshu. The study, which analyzed over 23,000 posts and conducted interviews, found that users primarily turn to LLMs for guidance on relationships, careers, and gaming, often refining prompts collaboratively. While users report positive experiences, attributing accuracy to personal fit and confirmation bias, professional diviners dismiss LLMs as lacking genuine spiritual power. The research suggests LLM-mediated divination offers scalability and co-production, altering traditional notions of divinatory authority. AI
IMPACT Reveals a new cultural application of LLMs, highlighting user-driven prompt engineering and evolving concepts of AI authority.