Let's agree to disagree.

What I see is that while LLMs can do real tasks, they often produce overengineered unmaintainable slop (plenty of examples where code can be reduced 10x to do the same). I hope this is not a base to continue training LLMs on.