Reproduce “The Chameleon's Limit: Investigating Persona Collapse and Homogenization in Large Language Models” 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
Yunze Xiao, Vivienne J. Zhang, Chenghao Yang, Weihao Xuan, Jen-tse Huang
The authors evaluate persona collapse by comparing simulated populations with human responses across coverage, uniformity, and complexity. On the BFI-44, none of eight models matches the US human reference on both coverage and complexity; Qwen3-32B’s high persona fidelity (ρ = 0.94) accompanies an average High-versus-Low trait gap of Cohen’s d = 13.7. The findings apply to tested models and tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.