Reproduce “Large Language Model Turnover Undermines Screening for Artificial Intelligence-Assisted Scientific Writing” 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
Kazuki Nakajima, Takayuki Mizuno
The authors test whether AI-writing detectors remain reliable as language-model versions change, using 4,000 pre-ChatGPT PNAS abstracts rewritten by 23 versions from OpenAI, Meta, and Alibaba. A RoBERTa-base classifier is evaluated at a 1% false-positive rate as new versions arrive. Past-version training catches only 26.6% of GPT-5 rewrites and 3.8% of Muse-Glimmer rewrites, showing abrupt failures in this English-abstract setting.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.