This is wrong, just look at this comment here:

https://news.ycombinator.com/item?id=46222523

LLM can't grade reliably human text. It doesn't understand it.