[atleta]

I had similar thoughts when I read a comment from a (non-technical) project manager in a discussion a few months ago. He said something like he uses ChatGPT to summarize the long winded emails of their developer. (He was working for/owning some kind of agency where apparently software developers weren't in the focus, really.)

He also mentioned writing emails with ChatGPT (though I can't remember if these were for communicating with the devs, but it doesn't really matter). I remember thinking that well, you my friend are pretty ignorant and not someone I'd like to work with. Even unethical.

So someone puts in the work to inform you about important details of their work/thought process and then this guy just ignores the whole thing. Even if the developer in question is really sharing more details than needed the right thing to do is to actually manage them and teach them to share less and make their own decision. (Hard to tell from the outside, of course, and the manager in question might be wrong too.)

Now if someone also blows up their own emails with generated details, especially if it's toward the same person whose email they have compressed with the AI then besides robbing the time of others, they may successfully deceive others and themselves and pretend that they have the knowledge what they don't have. And that will hurt the decisions made by them and by others. Basically the shared reality that you want to create with communication will not be a shared reality but out-of-sync views and you'll end up with a lot more "oh, but when I said that I thought you meant X...."

[disqard]

You raise a good point, and it reminds me of a sort of "Prisoners Dilemma" (sender uses/avoids LLM-verbosification and/or recipient filters all emails through LLM-summarization).

The worst-case scenario is that human communication becomes completely routed through this LLM-in-the-middle.