I think the problem of tainted training data is basically insurmountable at this point. LLMs get progressively worse the more of their own information you feed back into them, and the internet is getting more LLM content every single day. Maybe the next revolution in LLMs is pruning the training data to make it more pure? Problem with that is it would take a lot of human input, like more than you can do with Amazon's Mechanical Turk service.

I can barely search for anything these days. Unless I'm just looking for something extremely straightforward like business hours of some store or something, like... I'm at a point now where I will start to type out a question in a search bar, and I'll just stop in the middle of it and think, "What's the point? This question is more complex than where's the nearest McDonald's so I'm not gonna get anything useful."

Hallucinations aka rampant is real. With that said, it is great for finding the word for something I can only sort of describe and to just screw around with. I already don't pay for art. Who cares if I ask a robot to draw Steve Harvey jumping off of the top rop onto a rudo scoobie doo.