There is some utility, and frankly it won't go away so kids should eventually be taught its use and its failures.
These LLM models can be very useful for organizing large data sets. I know I've saved hours of time by feeding it very specific instructions to create some data classes and subsequent symbology for a couple of hundred points.
However, it never would have worked if I didn't already have the knowledge on how the software I was using to display my data worked and a strong knowledge of my data. It only worked because I was able to give it hyper specific instructions on how to process my data. Even then I still had to proof and make some tweeks, but ultimately a time saver.
This is what kids need to be taught. The first assignment given in high school allowing the use of AI should task students with running a selection of prompts to produce an incorrect statement or answer. Then students should be instructed to "trick" the AI into giving a wrong answer themselves.