Anybody who has tried to shortcut themselves into a report on something using an LLM, and was then asked to defend the plans contained within it knows that writing is thinking. And if you outsource the writing, you do less thinking and with less thinking there is less understanding. Your mental model is less complete, less comprehensive.
I wouldn't call it "cognitive decline", more "a less deep understanding of the subject".
Try solving bugs from your vibe coded projects... It's pain, you haven't learned anything while you build something. And as a result you don't fully grasp how your creation works.
LLM are tools, but also shortcuts, and humans learn by doing ¯\_(ツ)_/¯
This is pretty obvious to me after using LLMs for various tasks over the past years.
On the bright side, if you are forced to write AI code, at least reviewing PRs of AI generated slop gives your brain an exercise, albeit a frustrating one.
I'm sure they are, but maybe they just need some guidance. I was fortunate to learn this by myself, but when you just start out, it feels like magic. Only later do you realize you have also sacrificed something.
I wouldn't call it "cognitive decline", more "a less deep understanding of the subject".
Try solving bugs from your vibe coded projects... It's pain, you haven't learned anything while you build something. And as a result you don't fully grasp how your creation works.
LLM are tools, but also shortcuts, and humans learn by doing ¯\_(ツ)_/¯
This is pretty obvious to me after using LLMs for various tasks over the past years.