User proposes a bet: one misclassified snippet in a 1,000-page book wins
ctjlewis · x · 2026-11-16
User @ctjlewis is in a dispute over a bet about text safety classifiers: the other side claims any snippet from any safe text would be classified correctly. He counters that in a 1,000-page book, a "rare failure case" could just be one specific 750-word section misclassified while the remaining 99.9% is fine — which would still win him the bet, and complaining about it would be welching.
More from Fun
- PyBot: 1,000 agents wrote a 600k-line rule-based chatbot that matches SmolLM — lvwerra · 2026-11-16
- One Day ChatGPT May Tire of Writing Substack Posts — and the World Will Lose It — martyamark · 2026-11-16
- 1,000 agents, one week, 630k lines: HF builds a weights-free Python chatbot — _lewtun · 2026-11-16
- Reading a 1960s kids book makes dev sad: nobody has a boat anymore, they're all doing AI — granawkins · 2026-11-16
- OpenAI FP bounty unsolved not because impossible, but stakes too high — ctjlewis · 2026-11-16
- Xbox Teams Up with DJ Khaled for GTA 6 'Doctor's Notes' Stunt — tomwarren · 2026-11-16