The problem tends to be it defaults to the old fashion way of doing thing. Since most training data is that old fashion approach
If you just remember to prompt it to keep things simple or ask it for the easier way after the first step it reduces that problem to effectively zero.
It’s no different then asking some greybeard how to do something then bitching that it’s too complicated.
They always default to “their way”
redballooon@lemmy.world 4 days ago
Drop chats, use a harness. Include somewhere the procedure instructions to always look up documentation before crafting the command. Then it’s really good.