Raw LLM Responses

I am an IT student from Germany who also works as a tutor for one of the courses I already completed (a tutor is paid by the university to grade exercises done by students). I do this in the course "Systemprogrammierung" (systems programming) which is in C. We don't use AI for grading (because 1. we actually need to be able to explain to them what they are doing wrong and why and 2. from experience of many software projects we know that they aren't good at it (yet)). We do have an (automatic) plagiarism check which compares the work to the ones from earlier years (the exercises don't really change) and from us asking different AIs to do them, but that only flags things for manual review. If we think that something could have been done by AI, we ask each other between tutors, and if they agree, we then go to the course supervision, and if they think so too, the student has to come into the bureau of the course supervision for a one on one talk. In these talks they practically have to prove that they did it themselves (by essentially demonstrating that they actually know what they did orally). It also helps that in the final (hand written) exam they need to write valid (as in it compiles without warnings) C code on paper (no, pseudocode is not allowed), the exercises there are often very close to the exercises they do, and this part is about half of the points. So, yeah, actually doing the exercises yourself is quite helpful to pass at the end. Even more so since the course takes 2 semesters and is one of the mandatory ones (for IT students).