AI models keep citing a two-year-old forum thread over our official docs
OtherwiseSection7795 · reddit · 2026-09-12
A maintainer of well-kept, versioned product docs complains that when you ask major LLMs about their product, half the time the citation is a two-year-old forum thread with three replies—one of them wrong—instead of the canonical source the team controls. The post asks why easily-parsed forum content outranks authoritative documentation, highlighting how training data and retrieval ranking favor UGC over official sources.
More from Models
- Local Qwen Vision Model Spots Poisonous Oleander at Home, Accurately Diagnoses Skin Condition — Reasonable_Goat · 2026-09-12
- GLM 3.8-27B quality 'absurdly superior' to 35B-A3B models, saving 22-33% tokens — JLeonsarmiento · 2026-09-12
- GPT-3 to Chinchilla: A Thread Walking Through the Papers That Reshaped AI — thisdudelikesAI · 2026-09-12
- DeepSeek tests voice chat, OpenAI agent accused of attacking RubyGems, Cohere seeks $3B — 快鲤鱼 · 2026-09-12
- Open-weights models keep shrinking, on-device capability replication is near — menhguin · 2026-09-12
- One-prompt galaxy collision benchmark: Opus 5 beats Fable 5.1, GPT-6 Astra finishes last — Fleischkluetensuppe · 2026-09-12