Hot take, but the amount that people try to go and make an LLM be less sycophantic and still have it be sycophantic in round-about ways is astonishing. Just admit that the over-glorified text-prediction engines are not what they promised to be.

There is no “reasoning”, there is no “understanding”.

EDIT: s/test/text