I think this is a reasonable standard to hold, otherwise, like many before have said...send me the prompt. It's actually more interesting/better I know a coworker is struggling to communicate about something.
Likewise, I often used literary flourish and pleasantries like that above article about email decompression; I’m from the south so I think structured formality comes with territory.
I do think using LLM to turn notes and bullets into narratives should be considered no different than rendering CSV text into an excel format, just making it more digestible by recipient.
I will generate an LLM output that organizes scattered information and thoughts, resulting in 1.25x the text. I then read and edit it, generate executive summaries, and send it.
It saves effort for me in organizing, formatting and summarizing, and the LLM is producing more structure than content.
I often send out the LLM version, but still check if it contains the original thoughts correctly.
It's not a bad way to extend your vocabulary & catch spelling mistakes
Please don’t do this. You probably aren’t aware of how bad this can land. It’s not just about containing your original thoughts, it’s about the verbosity, repetitiveness, and absurdity of it all.
Grammarly is a much better tool for these kinds of purposes, and it actually guides and teaches you to improve your writing along the way.
A Google search didn’t reveal anything specific other than them using famous author names for expert review.
“keylogger mode” is optional, and to my understanding, you always see a visual indicator in the text area.
it doesn’t take every input as far as I know, and security firms don’t consider it a threat.
but point taken, it’s not for people with privacy concerns.
Regarding "honeypot" -- that's also what a honeypot is. They provide a service you want, then collect data. We have to take their word that they're only using this data to train their AI (which, btw, they are upfront about -- they log everything and feed it into their training. it's in their TOS).
Eg FBI putting up fake “buy drugs online” sites and logging your info once you place the (fake) order.
Tell it that you want a succinct professional email and it will do that. Give it examples of your own writing and it will match that style. If there's something you don't like, tell it to rewrite the part differently.
Theses are literally the things language models are best at.
This is not what the parent I replied to indicated, nor what people usually do.
I keep the LLM close to the original text content wise, but feedback was/is fantastic
The problem is that you lose your voice and adopt one that your audience knows all too well (and knows it isn’t yours). It makes your audience feel like you aren’t listening to them (even though you are!), because they feel like they’re talking to an LLM.
So there's a clear separation, a reply from me which I stand by and then some interesting chatbot stuff if you're into that.