at work, i can tell that people are not thinking and losing the ability to thinking hard.
I keep coming back to what LLMs "really" are. Recursive autocomplete, it keeps choosing the best fit for the next token. This really demystifies it for me. It also reinforces that is not "reasoning", so there's no hidden mental model.
its also wild how quickly and willingly many people handed this over. the apathy and desire for connivence runs extremely deep among my coworkers. we are at the point where people are not even running their local dev servers and clicking around to ensure their work is functionality correct, they just ingest the ticket, dump it into the model, commit the output, and say they're the most productive they've ever been.
the difference now is it is shamelessly encouraged.
one interesting topic is code reviews of AI code. code reviews have always been a bottleneck on the team and there's lots of trust and rubber stamping. I don't believe people are reading the code they ship.
I will say that it is a beautiful tool if you turn it on its head. if you don't quite understand something, it will never tire of answering your questions. It is a great time to need handholding.