@arstechnica This study's methodology sounds flawed. They asked random teachers, who gave random prompts, and the researchers compiled these into prompts they thought are representative.

Good LLM work can't be done like that. Ever heard of "garbage in, garbage out"? A prompt needs to be thoughtfully constructed, and the results iterate until the desired outcome is achieved.

Simple example: the study says AI lesson plans did not touch on higher order Bloom's concepts. Well, this expectation is wrong. Who expected them to do so, when the prompt says nothing about this requirement?

That's not good research, I am sorry to say.