Teachers still have to make AI math problems feel real

A new study found that while AI can generate math problems and catch basic errors, teachers and students still struggle with making the problems feel authentic and relevant to the real world. This means that even with AI assistance, human teachers must heavily edit the context of problems to make them useful for students.

Everyone assumed AI could handle the "real-world context" of educational problems, but it turns out that is where human teachers are still essential. This paper shows that AI is good at the mechanics of problem generation and catching mathematical errors. But it cannot yet create problems that feel genuinely relevant or "authentic" to students without significant human input.