French AI-generated novel axed from Goncourt longlist; writers demand AI text watermarking
zetalyrae · x · 2026-10-20
The anonymous account behind "Balance ton Claude" explains why it exposed Thélyson Orélien's prize-winning novel C'était ça ou mourir as largely AI-generated — it was favorite for the Goncourt and translated into 20 languages before Pangram v.4 (claimed 0.0041% false positive rate) flagged every page at or near 100% AI; the Académie Goncourt removed it from the longlist.
Key arguments from the Persuasion piece:
- The issue isn't that AI writing is bad — it's already good enough to pass the Turing test and win literary prizes, and will keep improving
- Detectors will become fallible (Pangram already emits false negatives; people fine-tune open models to evade it), so confidence in spotting LLM prose is misplaced
- AI text can be produced at near-zero cost without limit; even 0.001% passing as human would crowd out human writing and discourage writers
- Therefore AI literary writing must be clearly marked — otherwise reader demand for human-written text cannot be expressed and won't be supplied
More from Fun
- Satirical 'Fable 5.5' leak thread: the safest model is no model at all — patience_cave · 2026-10-20
- 116 AI agents make a music video in 44 hours using 2.4B tokens — tetsuoai · 2026-10-20
- Clairvoyance turns a C++ strategy game into a pixel-perfect web app for rapid idea testing — draginol · 2026-10-20
- Dormant Roland SC-55 project revived after 20 years, built entirely with Codex — OpenAIDevs · 2026-10-20
- Dev scratches vinyl records in the browser using Web Audio API — jh3yy · 2026-10-20
- Pangram CEO: we could build human-like AI writing, but we won't — TuhinChakr · 2026-10-20