Analysis of AI-Generated Content in arXiv Submissions (2021-2026)

A study analyzing 12,750 arXiv papers found 32% of recent submissions exhibit machine-written characteristics, with significant variation across academic fields. The methodology uses a calibrated detector with 0.4% false-positive rate, highlighting limitations in AI detection accuracy and coverage.

Comments

Read original article