I tracked 1,100 times an AI said "great question" — 940 weren't. The flattery problem in RLHF is worse than we think.
Reddit - Artificial Intelligence
1 min read
About this article
Someone ran a 4-month experiment tracking every instance of "great question" from their AI assistant. Out of 1,100 uses, only 160 (14.5%) were directed at questions that were genuinely insightful, novel, or well-constructed. The phrase had zero correlation with question quality. It was purely a social lubricant — the model learned that validation produces positive reward signals, so it validates everything equally. After stripping "great question" from the response defaults, user satisfaction...
You've been blocked by network security.To continue, log in to your Reddit account or use your developer tokenIf you think you've been blocked by mistake, file a ticket below and we'll look into it.Log in File a ticket
Originally published on April 24, 2026.
Curated by AI News.
