LLMs do not most of the things that you say (like radiology readings, birth defects...) Those are language models. It's on the name.

What you're thinking are generally NN, trained to categorize those things in particular. You can give those tasks to ChatGPT and will hallucinate an answer that somebody who doesn't know would feel correct.

The fact that everything is labeled AI makes it so that people like you greatly overestimate what ChatGPT does to idiots.