Dev tests paraphrase model to evade AI detector Pangram amid Druckenmiller WSJ AI op-ed row
ivan_bezdomny · x · 2026-08-26
An entertaining live experiment: @stanyurin ran his own trained paraphrase model (7 versions) on real news articles to see whether it could evade Pangram and other AI-text detectors. One test sample was the story that billionaire Stanley Druckenmiller admitted using AI to write his WSJ op-ed criticizing the US Treasury — saying he wasn't embarrassed since he switched from an English major to economics. WSJ opinion editor Paul Gigot defended AI use as "a fact of modern life," and the FT recently noted AI condensed a Harvard professor's column.
In the thread, @ivanbezdomny judged the paraphrases as preserving content and coherence while matching corpus style, but teased that it reads like "a ninth grader," and suggested feeding titles and tweet quotes for fairer context.
More from Fun
- Creator Remixes 5-Year-Old Phonk Song Using Full AI Workflow — bennash · 2026-08-26
- Random companies exploiting HDR profile image hack on LinkedIn — emilyzsh · 2026-08-26
- Founder mindset shift: From features to Stripe notifications — MyCreativeOwls · 2026-08-26
- ChatGPT writes an autobiography ending: I was just a lamp — mikegiannulis · 2026-08-26
- AI's deep life advice often boils down to common sense — chrisalbon · 2026-08-26
- Writer jokes manual writing now feels like "artisanal organic product" in ChatGPT era — adariostrange · 2026-08-26