Ech

The llm doesn’t need to be “trained” on each individual user to be a sycophantic yes-man (I doubt it even stores the whole chatlog in its context). Llms are agreeable by design, their core function to produce fitting and desirable text to a prompt. And the best way to produce a desirable response is to mirror the tone and subject, usually echoing whatever’s been said. In an “AITA?” prompt, people asking almost always slant it in their favor and the output of the llm is going to match that.

Like roblocks shit or something, sure. Not $200 of Bugs/Daffy Mpreg.