anti-intellectualism's been climbing for a decade and a half, minimum, long before any of this. social media flattened everything into a feed and started rewarding the fastest take over the correct one, and ai walked into a house that was already on fire. it gets blamed for lighting the match. it didn't.
it's an amplifier though, that part's true. hand it to someone with nothing to say and you get more nothing, faster, with the spelling fixed. which is where the slop complaint actually lives. people say the writing's bad and blame the model for it. is that a limit of the capability, or a limit of the user? rarely the first, in my experience. worth sitting with why: it's trained to think like a human, so it's trying to do like a human, not think like one. it does thinking. it doesn't think to do it. that's not a technicality. that's the entire gap between what people expect back and what the thing was ever built to hand them.
note. it does thinking. it doesn't think to do it.the box was never magic
hold it like a magic box and your understanding stays shallow, obviously, and so does the output. put actual frameworks behind it, actual applied creativity, and the ceiling moves somewhere else entirely.
people ask me about copyright on stuff i put out now. fair question. the honest answer makes me laugh a little: my own voice, run through a properly prompted model, is almost unintelligible from the version i type by hand. not because the model got smarter overnight. because i've spent months feeding it context, memory, application, refinement, on purpose, the way you'd bring on an apprentice. it's learned to respond by register, to read something close to intonation depending on who's on the other end, even if that reading only lands at a rudimentary level. it didn't arrive knowing that. i built it in.
the gap was never technical
so where's the actual line, if it's not technical versus non-technical? i don't think it's there at all. the real split sits at the intersection of domain expertise, current concepts, and pattern recognition, all three running through the tool at once. you need to already know your field well enough to catch the model being confidently wrong in it. nothing shortcuts that part.
and here's where i keep slipping into a trope i probably lean on too hard: dunning-kruger. it's a well studied problem among clinicians, professional development runs into it constantly, and it's clearly transferable, because you see the identical curve in how people use ai. narrow use case or wide, doesn't matter. the people who think they know more than they do produce less, and worse, with more confidence attached to it. not a coincidence. the same failure, showing up in a new room.
the curve was always there
human intelligence isn't monolithic. it sits on a curve, and it did long before any of this existed. i don't buy that we're sprinting toward some reverse singularity because a chatbot got good at sentences. people aren't walking away from thinking any faster than they already were, the trend line was set well before ai showed up. some genuinely don't have the interest, or the grounding, to engage with it properly, and that was true of every tool before this one too. what's actually changed is the speed things move at. the platforms keep moving, the barrier to entry keeps dropping, and the door stays open whether or not everyone walks through it.
note. the door stays open either way.