Personally, I treat it as both a tool and source of entertainment.

Run an LLM locally, crank up the temperature, give it a very simple prompt on purpose and ... watch it go. xD

Or, feed it a lot of context, reduce the temperature and rerun the prompt a few times to get different versions/takes and use it like a tool. It can give you some great pointers - but it can also make you run into a wall if you aren't careful x)