[blubber]

Is this a problem for which the (human) solution is well documented an known and was learned during the training phase? Or is it a novel problem?

I personally think anthropomorphizing LLMs is a bad idea.