They may eventually be useful in this space. But for now, they are more work than they’re worth and completely discredited for proper fact-based research. And the teachers my kid has had who used it for testing resulted in completely wrong answers that the teacher didn’t bother to check.

Yes that is the teacher’s fault, but so is using it to generate a test in the first place.

I will die on this hill. LLMs of any kind right now are not something that should be trifled with in a critical thinking-based curriculum. In time, perhaps. But not yet, not when LLMs are so easily manipulated (whether trained on public data or private). The various implementations haven’t earned credible trust despite CEOs drooling over them.