Pangram v4 flags pre-2020 thousand-line texts as 100% AI-generated

birchlse · x · 2026-09-23

Antoine Tilloy points out that several thousand-line texts written before 2020 — long before modern LLMs existed — are flagged as "100% AI" by AI detector Pangram v4. The false positives underscore how unreliable such detectors are for long texts and the risk of false accusations.

Original post →

More from Fun

Fun channel →