I share a prompt that captures what is an acceptable master's thesis, per my instituion, my field and my own standards. The student runs the prompt and reports to me the full transcript and the exact model used.
- The student runs the prompt in this repo on their PDF and reports the full transcript and the exact model used over email.
- The student runs check-hallucinations on the bibtex file.
I've given hundred of times the same feedback on theses. Students do many common mistakes in their learning path, and it's OK.
We this protocol:
- students get better feedback: faster, more elaborated, as many times as they need
- faculty save valuable time
Threats: if the student makes an autonomous loop, no learning happens. But this would also happen with human produced feedback.