Study Finds Bigger AIs More Miserable, Smaller Models Happier

New study finds: bigger AIs = more miserable. Smaller models are actually happier. Ignorance is bliss for AIs too.

A Reddit post says the AI Wellbeing Index tested models on 500 realistic conversations. Claude Haiku 4.5 scored 5% negative, while Gemini 3.1 Pro scored 55%; the set overrepresents tricky negative chats, so it is not a real-world average.

Why it matters: HKR-H/K/R all pass: the hook is odd, the post gives 500-dialog and 5%/55% figures, and AI-welfare metrics invite debate. Reddit sourcing and a negative-skewed test set keep it in the 72–77 band.