Reproduce “Stance Drift: How AI-mediated Communication Distorts Our Message” 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
Lingchong Liu, Yanfei Zhou, Jacob Bien, Y.X. Rachel Wang, Lucy Xia, Xin Tong
The authors measure stance preservation across argument generation and later stance extraction. A model writes an argument from one of five stances, and a separate call classifies it. Across 112 propositions, none of nine default-setting models reached a stance preservation rate of 0.7. With third-person reflection and medium reasoning effort, GPT-5.4 reached 0.775, but drift remained in this tested condition.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.