A study analyzing over 12,750 papers submitted to arXiv between 2021 and 2026 found that approximately 32% of new submissions read as machine-generated. The research employed a detector calibrated to minimize false positives on pre-ChatGPT era scientific writing, establishing a baseline of 0.4%. Computer science papers showed the highest rate of AI-written content at around 65%, while mathematics papers had a lower rate, though this may be due to the detector's limitations with specialized notation and structure. AI
IMPACT Indicates a significant shift in academic publishing, potentially impacting peer review and the authenticity of research.
RANK_REASON Research paper analyzing AI content in academic submissions.
Read on Mastodon — fosstodon.org →
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →