Reproduce “MixDetect: Word-Level Localization and Quantification of AI Editing” 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
Hongrui Bao, Yubing Ren, Zhendong Pan, Fang Fang, Shi Wang, Yanan Cao
MixDetect predicts whether each non-punctuation word in edited text was changed and, conditional on editing, how substantial the change was. It trains on source-edited alignments, then makes predictions without the source and aggregates them into scope, intensity, and overall magnitude. On EditLens, Edit-Only reaches 0.885 AUROC against alignment-derived labels; these predictions are not definitive authorship evidence.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.