New writing model Deft hits beta; 86% of outputs pass as human on Pangram
TuhinChakr · x · 2026-08-20
AI writing lab Deft, cofounded by rosmine and @jmrphy, launched a public beta for "better writing" — its launch announcement was written by the model itself. The team says 86% of user queries now come back fully human on Pangram, and advises adding more prompt detail or switching styles for better results; it works best for analysis, creative writing and rewrites, less for marketing copy. A tester found that total nonsense output was still flagged as AI, sparking jokes about detector reliability.
More from Fun
- Sick vibes perfectly captured by one dazed cat — michellechen · 2026-10-16
- Researcher complains AI-written jargon replies now flood over half of his X mentions — teortaxesTex · 2026-10-16
- Musk's 2019 tweet resurfaces: new Roadster would 'actually do something like this' — XFreeze · 2026-10-16
- Publishing lists of people you want to flatter is a surprisingly effective growth hack — signulll · 2026-10-16
- AI researcher Rosie Campbell catches the Elizabeth Holmes stage play, jokes about cosplaying her — RosieCampbell · 2026-10-16
- Yacine: vibe-coded repos with 3,330 agent-written tests are all useless — yacineMTB · 2026-10-16