How we measured AI writing across arXiv, and where the measurement breaks

AI & Machine Learning(unslop.run)view on HackerNews
AI-generated writingarXiv papersmachine learningdetectorfalse-positive rate

Author: dopamine_daddy

Date: 7/20/2026

Article Summary:
Researchers measured the prevalence of AI-generated writing in arXiv papers using a detector calibrated to identify genuine human writing at a 0.4% false-positive rate.