DigiNews

Tech Watch by Johan Denoyer

← Back to articles

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

Quality: 8/10 Relevance: 9/10

Summary

The article describes a methodology to measure AI-written text in arXiv papers using a detector calibrated to a 0.4% false-positive rate. It reports that around 30-40% of recent submissions show machine-like writing, with substantial variation by field and clear discussion of limitations.

🚀 Service construit par Johan Denoyer