When I do try to check the links, they often don't support what's stated. If I call out Gemini, it says something inane like "I don't actually use these links the way a human might --- I didn't misinterpret it because I didn't even read it!"
Recently I was searching for information about fig trees (the fruit). It kept telling me things that didn't make sense, luckily the links revealed it was answering about fiddle leaf fig trees (the houseplant). Unfortunately that was not a correctable problem --- it kept happening even after I pointed out the mistake. So I'm really not too sure what the purpose of the system is. You still have to read through the references in their entirety (if not paywalled!), and with skepticism. And then if you find a problem, AI isn't going to get back on track.
There's the new pelican bench mark
People just want to pay their bills and don't want to use up their limited cognition beyond what's minimum.
For example, I work for 3 different employers. They don't know and I don't care if they find out. I use AI exclusively. I don't have Kubernetes experience but I am managing their infrastructure without issues. I will just read whatever LLMs give me and repeat it back. It has never failed me and there's no need to learn something that AI will do much better job at.
Sometimes I'm not even at my desk, I go to the gym and exercise instead which is a better use of time only answering on Slack when they ping me.