Confident Nonsense: why polished AI answers are the ones you should trust least
nikola_mr64990 · x · 2026-09-21
An article titled "Confident Nonsense: How to Stop Trusting AI That Is Wrong" argues the worst AI answers aren't the obviously broken ones but those that look finished — a clean paragraph, a named study, a citation, a snippet that compiles, a precise date or number. The author calls for methods to stop over-trusting such confident output, though the reshared post only previews the piece via a link.
More from Models
- $200/month OpenAI user cancels after reset token expired 30 minutes before the listed date ended — Tetsuo_Shoto · 2026-09-21
- Open-source models running locally have pegged token prices near zero — gerardsans · 2026-09-21
- Bindu Reddy predicts a big week ahead: several frontier model launches coming — bindureddy · 2026-09-21
- User finds Astra xhigh burns usage slower than light or medium reasoning tiers — CtrlAltDwayne · 2026-09-21
- Grok voice is one command away in ~5M Teslas, says owner who's chatted with it for hundreds of hours — tristanbob · 2026-09-21
- Users Notice Claude Suddenly Addressing Them by Name Far More Often — CrumblingSaturn · 2026-09-21