Applications of Large Language Models in Chinese Traditional Numerology: An AI-Driven Empirical Study

This study focuses on the combined analysis of the Zi Ping Ba Zi and Zi Wei Dou Shu dual systems, comparing the practical performance of three mainstream LLMs—Microsoft Copilot, OpenAI ChatGPT, and Google Gemini. It found that Google Gemini had the best overall performance in aspects such as chart calculation accuracy and academic logical reasoning.

Original Author Information:

• Original Author/Maintainer: goldmanau
• Source Platform: GitHub
• Original Title: AI-Logic-Driven-Study-on-LLM-Applications-in-Chinese-Metaphysics
• Original Link: https://github.com/goldmanau/AI-Logic-Driven-Study-on-LLM-Applications-in-Chinese-Metaphysics
• Publication Date: 2026-06-01

With the explosion of artificial intelligence technology, LLMs have penetrated the field of complex text analysis, and traditional numerology has become a new application scenario. This study chose the combined analysis of the Zi Ping Ba Zi and Zi Wei Dou Shu dual systems:

• Zi Ping Ba Zi: Based on the combination of birth Gan Zhi (Heavenly Stems and Earthly Branches), it deduces through the generation and restriction of the Five Elements and the relationships of the Ten Gods, emphasizing the integration of Qi-Li (energy principles) and Xiang-Shu (symbolism and numerology);
• Zi Wei Dou Shu: Centered on the Purple Star (Zi Wei), it interprets through the interactions between stars and palaces; The combined use of the two is an advanced technique for traditional numerologists, testing the AI's ability to understand and map the two symbolic systems.

The test subjects are three mainstream LLMs: Microsoft Copilot, OpenAI ChatGPT, and Google Gemini. The evaluation dimensions include:

1. Chart Calculation Accuracy: Calculating the accuracy of the Ba Zi four pillars and the star distribution in the Zi Wei numerology chart;
2. Error Correction Speed: Efficiency in correcting errors for edge cases (e.g., birth at Zi Shi [11 PM-1 AM], solar term transition days);
3. Dynamic Deduction of Liu Ri/Liu Yue: Cross-deduction ability across four time dimensions (Da Yun [10-year luck cycle], Liu Nian [annual luck], Liu Yue [monthly luck], Liu Ri [daily luck]);
4. Academic Logical Reasoning: Classic references, coherence of reasoning, and practical value of conclusions when explaining the meanings of star combinations.

The tests found that Google Gemini had the best overall performance:

• Chart Calculation Stability: Low error rate in lunar calendar conversion and solar term calculation, strong robustness for boundary cases;
• Academic Logic: Rich classic references, coherent reasoning, close to the style of professional numerologists;
• Dynamic Deduction: Low error rate in Liu Yue (monthly luck) and Liu Ri (daily luck), providing actionable time window suggestions;
• Error Correction Response: Better correction speed and quality than competitors, with stronger metacognitive ability.

Google Gemini's comprehensive performance in numerology analysis is significantly better than Copilot and ChatGPT. Deep insights: Highly structured traditional knowledge (such as numerology symbols and rules) is naturally compatible with LLMs, but AI may have 'technically correct but culturally biased' issues when dealing with the specificity of cultural contexts, which requires supervision and calibration by human experts.

AI numerology tools should be used as references rather than prophecies, and ethical boundaries need to be paid attention to. In the future, multimodal AI combined with astronomical calendar databases and historical numerology case libraries can improve deduction accuracy, but 'humans' are always the core of numerology—AI assists in calculation and logical sorting, while the meaning of life is independently written by humans.