> LLMs tend to stay very generic, to have a lower chance of being "wrong".

It depends entirely on how you prompt them.