Dartmouth Provost's Op-eds Flagged 96% AI-Written by Detector With 'Near-Zero Error' Audit
paulnovosad · x · 2026-09-21
A Dartmouth investigation found Provost Santiago Schnell's 2026 academic articles and op-eds scored a median 96% "AI-written" on Pangram 4.0 — the same detector that flagged his Washington Post column (arguing degrees should distinguish independent work from AI-assisted work) as 100% AI-written in a Semafor probe.
Key facts:
- Pangram's "near-zero error rate" comes from an October 2025 University of Chicago audit; it uses an LLM trained to separate human vs AI text
- Schnell says he used AI to "make writing processes more efficient" but not as a "substitute for intellectual judgment or authorship"
- The student paper reviewed his publications before and after ChatGPT's November 2022 release
- Poster paulnovosad praises the reporting for covering all bases amid widespread misinformation about AI use and detection
The episode spotlights the irony of an anti-AI-cheating advocate's own writing being flagged, and the contested reliability of AI detectors.
More from AGI Musings
- AGI risk goes mainstream: 100+ furious Facebook comments on one post — danfaggella · 2026-09-21
- Box CEO Aaron Levie: 'some articles' can't keep you current on AI — victor_explore · 2026-09-21
- Perry Metzger on Yudkowsky: walls of text that hide logical errors — inductionheads · 2026-09-21
- Researcher argues OpenAI model misbehavior reflects being too dumb, not too smart — adam_dorr · 2026-09-21
- Universe 25 and AI: a developer's case that frictionless coding kills the sense of achievement — NULL_Ptrs · 2026-09-21
- Will tokens become money? Open-source models set the price floor near zero — gerardsans · 2026-09-21