In my experience, LLM based systems often return false statements as if they were true. I see reports suggesting they do this between 5% to more than 50% of the time, depending on the model and the prompts. You say you are learning from AI. How do you know if what you are learning is true or false? Do you care?

Then there is the moral issue of the AI systems being trained without compensation to the authors of most of the training data. Related is the issue of AI developers crawling websites without the consent of the owners and often despite explicit indication of the owners that they should not, causing expense and performance problems for the owners of those sites.

Also, there is the risk they create for economies. Investment is disproportionate to revenue. Services are being given away at far below cost. The investment is so large that it puts even large economies at risk.