If you know what you’re doing, you can achieve near-human-quality judgements of literacy performances using classical NLP methods. LLMs pretty much get us over the line.
Since students themself use AI to write essays, those grading AI will be evaluating the most average writing anyway.
We've now come full circle, just losing our ability to write in the process.
The hero's journey: end up back where you started, having gained something in the process.
AI's journey: end up back where you started, having lost something in the process.
The AI gains more training data. It's not exactly the Odyssey, but still.
The AI is inbreeding. Model collapse is imminent.
Exactly. It would be amazing if the same LLM were used for both generation and grading, and could see what tweaks the human (student) made before turning it in. That would be very useful training data.