Apple study maps when LLMs act human-like — and when developers should dial it back

Apple researchers analyzed 21,000+ conversations to map when LLMs show human-like behaviors like emotional language or boundary-setting, and how controllable that is.