Reproduce “Frontier models need only one text to unmask its human or AI author” 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
Atrey Desai, Deniz Böloni-Turgut, Ty Brassington, Minjoon Choi, Jenna Russell, Rishanth Rajendhran, Bradley Emi, Hila Gonen and 4 more
The authors test whether language models can identify writers from a single text, without candidate lists or web access. They ask models to name authors across peer reviews and other text domains, and find accuracy varies with model familiarity. Claude Fable 5.2 correctly identifies 56% of 398 private reviews, but those reviews came from only six contributors.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.