@tbortelsTom Bortels I bookmarked this because it is a VERY accurate summary. I was revisiting the bookmark because I had a long "argument" with the ChatGPT non-pro model that illustrated all of this.. here's a segment
To be fair - it's right. We project intent on a process with no real understanding - it's just a parrot.
But useful occasionally. It's a turbo button for boilerplate and when they throw a giant ball of someone else's uncommented code at your head. Just never trust it, but I don't trust my own code either.
I use the janky lying robot (my new term for AI) to do code assist - it is, like most tools, useful but has a sharp edge or two even when used correctly. The first sentence in my global AGENTS.md included in every prompt is:
"Do not apologize. We are writing code, and code fails during development, that's how the process works."
And - well, it's still a janky lying robot, but this brings down the obnoxiousness level significantly.
If you have a fediverse account, you can quote this note from your own instance. Search https://infosec.exchange/users/tbortels/statuses/115577765636961643 on your instance and quote it. (Note that quoting is not supported in Mastodon.)