Fable 5.1 bio-safeguards trigger on harmless letter-counting, making it 'unusable'
maksym_andr · x · 2026-09-18
In an ongoing evaluation thread, maksymandr shows that simply counting letters in random word sequences triggers Fable 5.1's bio safeguards, causing refusals on a harmless task. The same thread found Fable consumes 3x+ more tokens than GPT-6-Astra on 1280-character passages while performing worse.
More from Models
- Mystery "Stealth Union Alpha" Model on OpenRouter Baffles Redditors — Iory1998 · 2026-09-18
- Typesafe AI launches general-purpose steerable low-latency classifier — andreisavu · 2026-09-18
- Cactus Releases Needle 3: An 8-29MB Foundation Model Running 4k tokens/s on a Raspberry Pi 5 — airesearch12 · 2026-09-18
- Classifier scores 100 YouTuber videos for sales intent in 12s at ~$0.02 — eptwts · 2026-09-18
- Dan Shipper gets early vibe check of secret new LLM from InstructGPT author Diogo — danshipper · 2026-09-18
- Six or Seven Chinese AI Labs Reportedly Racing to 10-40T Parameter Models — troll_khan · 2026-09-18