Writer composes with every AI stylistic marker — Pangram still scores it 100% human
TuhinChakr · x · 2026-10-23
Responding to recent criticism of AI detector Pangram, the author ran a controlled test: he wrote a paragraph himself but deliberately packed it with every known AI stylistic marker — em-dashes, series of three, "it's not X, it's Y" constructions, rhetorical questions, setup phrases like "here's where it gets complicated," and clichés like landscape, delve, and realm — while keeping it "too short to classify."
Pangram classified the paragraph as 100% human. He argues this debunks the common "I just write like AI" excuse and that the backlash against Pangram is a campaign to inject doubt and discredit AI detection. A booster adds that even AI-sounding human writing won't trigger false positives.
More from Apps
- Ex-judge Bridget Mary McCormack joins Clio to build AI for courts, not just lawyers — EJWalters · 2026-10-23
- Varformer: multimodal transformer prioritizes drug targets from population-scale variation — anshulkundaje · 2026-10-23
- Hermes Agent hits 395 real user stories; one IT staffer runs 80% of a 165-person company's work — NousResearch · 2026-10-23
- An Arduino learning app built with Opus 5.5 breaks hardware into tiny circuit steps — FinanceYF5 · 2026-10-23
- Microsoft Research's Flint-interactive turns AI-generated charts into clickable dashboards — aigclink · 2026-10-23
- After weeks with always-on agents, the killer use case isn't errands — it's surfacing a class action — talkaboutdesign · 2026-10-23