I do use it quite often in my work. I just downloaded an Excel worksheet with all standard mailtexts (I work at a company offering courses), about 500x3. I gave it a list of criteria they should follow, and made it find those that didn’t. This worked pretty well. And it can work pretty well so long as you’re in control and you don’t take the result as truth.

That’s beside the obvious privacy issues, obviously. I hardly ever use LLMs outside of work (though when I do, I like to run models locally).