One use of LLMs that I haven't seen mentioned before is to use them as a sounding board for your own ideas. By discussing your concept with an LLM, you can gain fresh perspectives through its generated responses.
In this context, the LLM's actual comprehension is irrelevant. The purpose lies in its ability to spark new thought processes by prompting you with unexpected framings or questions.
Definitely recommend trying this trick next time you're writing something.
If your machine is beefy enough, running a local model works pretty well and there are a lot to choose from now tuned for different use cases. I've had good luck with using gpt4all as a no hassle app for running this stuff on my machine.