On principle, my stance is that we write for people to read, not score. If we have a systemic problem of too many students per teacher, we should work on that. In the meantime, anyone studying autonomously of course could benefit from AI-generated feedback, especially in early language learning.