>> Recognizable repetition can be abstracted
    > ... which is the whole idea behind training, isn't it?

The comment I was answering specifically dismissed LLM's inability to answer same question with same... answer as unimportant. My point is that this ability is crucial to software engineering - answers to similar problems should be as similar as possible.

Also, I bet that LLM's are not trained to abstract. In my experience they lately are trained to engage users in pointless dialogue as long as possible.