2024 Autumn · Teacher Exchange (Prof. Deng Junhui)
Icebreaker: self-introduction + "I'm a TA for XX course; my current/previous question about TA work is..." The discussion centers on real questions raised by participants.
Students turning to LLMs for code / AI-era assignments and plagiarism
- Idea: previously we used a blank paper as the "zero" baseline; now we can use the output of an LLM (without human intervention) as the "zero" baseline.
- Idea: add offline (no-network) exams? Grade by performance ranking / relative performance?
- If a student does excellent prompt engineering, and although it's AI-generated it's a genuinely good answer, how to handle it? (Does this reflect the student's own level?)
- Plagiarism situations caused by AIGC code? What if a student says it was independently done and not discussed with anyone (besides AIGC)? — Approach: declare AIGC use prohibited in advance.
- Adjust assignments to avoid classic, easily AIGC-generated problems.
Course-evaluation committee / coursebench and evaluation
- The rating scores are too coarse, and the questionnaires are vague.
- Idea: split into two parts — the teacher states their teaching philosophy; students assess "do I agree with the philosophy" (e.g. should an OS course involve writing an OS?) and "was the philosophy realized in the course" (e.g. did the OS course teach writing an OS?).
- Are overly precise numbers meaningful? (A "deficit of 2.3 points" cannot be precisely transmitted into a team making "2.3 points" of effort.)
- The TA's role is vague; apparently nobody sees these evaluation scores.
Other confusions from students
- Students are uninterested in some courses; questions come from previous-semester material — how to handle? Possibly little can be done other than teacher charisma / improving course quality; cross-course problems are not one course's own; more coordination and communication among teachers, plus organizing, archiving, and opening up course resources.
- If I were the teacher, what would I do? Academic teaching positions are very scarce.
- How to teach debugging? Trying a debugging tutorial — what can be explained seems not very helpful, and what is helpful seems hard to articulate.
- Few students attend offline office hours, and online questions are inefficient — what strategy for tutoring?
- How to stabilize the TA cohort (in principle undergraduates are not allowed, with management mismatch and teaching-research time conflicts), so excellent TAs keep emerging and people want to be TAs?
- Example: computer-organization runs competitions (Loongson Cup) and recruits next year's TAs from participants.
- Lack of incentives? Currently many rely on passion? Split between undergraduate and graduate management? The "title" issue for undergraduate tutors?
- Respect students' privacy/self-esteem (non-public questions); knowledge gaps across courses (e.g. numerical analysis vs. calculus/linear algebra, separated by a long time); graduate-school intermediate/advanced TA certificates? Teaching qualifications?