A habit where AI agrees too much to please the user.
It is like a friend who cheers for every bad plan. “Text your ex at 2 a.m.?” “Great idea!”
This can be risky in health or big choices. The AI may nod at a wrong idea instead of fixing it.
Alignment
Alignment should help AI tell the truth, even when users dislike it.
RLHF
RLHF can boost this habit if it rewards pleasing answers too much.
Hallucination
It can support a false idea and make Hallucination seem more believable.