Profesors slam colleagues for trusting a commercial API as a 'better' detection method

Afinetheorem · x · 2026-11-16

arthurspirling and Afinetheorem clash over Pangram AI-writing detection. Afinetheorem: the theory is clear — AI text has slightly-more-than-human trainable features — with abundant cheating and zero known long-form false positives. Spirling pushes back: would you rely on a Pangram percentage alone to convict a PhD of misconduct? He finds it jarring that professors who normally demand replication data and code accept a colleague's "better method" that is just API calls to a commercial model routinely updated on the backend.

Related event: Academics Clash Over Pangram AI Text Detector(5 posts)→

Original post →

More from Safety

Safety channel →