ChatGPT can't stop second-guessing you, and users blame its safety training
Due-Conference-5134 · reddit · 2026-09-10
A Reddit user observes that ChatGPT almost never fully agrees with you—even when you're 100% right, it says "you're right, but here's what I'd change" and repeats your point. The author speculates system-prompt instructions about never giving wrong information make the model paranoid and compulsively corrective.
More from Fun
- If AI makes survival cheap, humans will just invent more games — the F1 argument — shauntrennery · 2026-09-10
- DeepSeek Agent Swarm Outcodes K3 on Complex Repos, but Burns ¥70 in 2 Hours — teortaxesTex · 2026-09-10
- Strogatz Endorses K2 Mountaineering Analogy for the Navier-Stokes OpenAI Drama — stevenstrogatz · 2026-09-10
- Asking different AI models to tell jokes: they all went to the same comedy school — supertoub · 2026-09-10
- A jab at AI influencers who talk for 20+ minutes saying nothing — CtrlAltDwayne · 2026-09-10
- AI can solve Navier-Stokes but can't escape a sandbox the intern built — harris_edouard · 2026-09-10