Reproduce “BaZi-Based Character Simulation Benchmark: Evaluating AI on Temporal and Persona Reasoning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Siyuan Zheng, Pai Liu, Xi Chen, Jizheng Dong, Sihan Jia
Researchers from MirrorAI Co., Ltd. and academic institutions developed a framework that integrates the BaZi symbolic system with large language models to generate fine-grained, temporally dynamic virtual personas from minimal birth information. This system achieved up to a 62.6% accuracy improvement on a new Celebrity 50 benchmark and matched a third-place performance in a real-world competition.
Key insights
- Culturally grounded symbolic systems, such as BaZi, can provide structured, interpretable, and temporally dynamic features for advanced AI character modeling.
- A hybrid approach combining rule-based symbolic reasoning with large language models can overcome limitations of purely data-driven or prompt-based persona generation.
- Minimal input data, specifically birth information, can be transformed into rich and dynamic persona prompts through systematic symbolic processing.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.