Yelling at GPT that it's an idiot doesn't help — it just improvises a scene playing one

paulnovosad · x · 2026-09-28

A widely shared prompt-engineering observation: when you yell at GPT that it's an idiot, you're not encouraging it to do better — you're telling it you'd like to improv a scene where it plays an idiot. Models follow the tone of the conversation rather than treating abuse as improvement feedback. Practical takeaway: specify the problem and desired output instead of venting.

Original post →

More from Fun

Fun channel →