CS240 AI Cheating Retrospective
5 hours ago
- Course CS 240 had a clear policy prohibiting AI/LLM use for assignments, repeatedly communicated via syllabus and lectures.
- The tool Argus used static analysis (not an LLM) to detect indicators of AI-generated code, with high accuracy across multiple semesters.
- Out of 584 identified students, 267 were flagged after manual review; 207 enrolled students received an email with a self-report form.
- The process faced backlash over coercion fears, timing near the drop deadline, and due process concerns, leading to university intervention.
- Resolution: retroactive findings were discarded, but Argus continued for remaining assignments; 74 of 144 students who dropped re-enrolled.
- Key takeaway: the instructor acknowledged the process should have been better—clearer communication, individual meetings, and earlier deployment.
- Data showed students flagged for AI use performed significantly worse on midterms (10–14% lower) compared to peers.