CS240 AI Cheating Retrospective
A university’s introductory C programming course has become a flashpoint over how to handle large‑scale use of AI tools like ChatGPT in student homework. Commenters dissect a professor’s strict “no AI, work alone” policy and his retrospective on flagging nearly half the class for suspected AI-generated code, with many criticizing the coercive tone of warning letters and the chilling effect on collaboration, while others defend the need to protect academic standards and the value of learning fundamentals without automation. The exchange widens into proposals for alternatives such as oral exams, in‑person proctored coding, and redesigned assignments, and raises broader questions about whether universities should ban AI, integrate it into curricula, or shift entirely to assessment methods that can’t be outsourced to machines.
Reactions to the Anti-AI Policy and Enforcement
- Many see the policy (no AI on assignments, strong policing of similarity) as rigid and out of touch, arguing that if ~50% of students break a rule, the course design may be misaligned with reality.
- Others defend the ban on AI in early courses: using LLMs is likened to using a calculator while learning basic arithmetic or a forklift in a gym; it bypasses the learning objective.
- Some commenters emphasize fairness: widespread cheating harms honest students competing for grades, majors, and later jobs; strict enforcement is seen as necessary.
Form Letter, Coercion, and Due Process
- The notification form threatening dean referral, course F, or zeroes if students “honestly” admit AI use is widely viewed as coercive and structured to force confessions.
- Critics say if the instructor is confident in the evidence, they should assign penalties directly instead of using a pressure-filled “admission” process.
- Others note the retrospective acknowledges mistakes and partial apology, but some feel the tone remains highly defensive and self-justifying.
Collaboration vs “Work Alone”
- The “do not work with other students / don’t look at others’ files” language is described as anti-collaborative and contrary to how humans learn.
- Supporters argue that at this level, implementation details are simple and collaboration too easily turns into copying; TAs are positioned as the safe help channel.
- There’s concern that policies treating “inadvertent exposure” as a violation create anxiety about even talking to peers or TAs and can select for more socially connected students.
Assessment in the LLM Era
- Several propose shifting to in-person proctored exams, oral exams, or live code walkthroughs to test understanding directly, accepting that take-home work will be AI-assisted.
- Others suggest project-based grading with in-person questioning or on-the-spot modifications to distinguish genuine understanding from “vibe coding” with AI.
- Scalability of oral/interactive assessment for large classes (e.g., 600 students) is noted as a major constraint.
Course Design and Use of Language Features
- Some students would naturally explore features like
malloc,sizeof, or other C constructs and resent being implicitly suspected of cheating for doing so. - The instructor’s perspective (as described in the thread) is that such features often correlate with copied solutions and can distract from the targeted learning goals; they were discouraged, not banned, and followed up with brief conversations rather than automatic accusations.
Wider Reflections on Academia and Cheating
- Opinions range from “zero tolerance, throw cheaters out” to “cheating is symptomatic of misaligned incentives and overly vocational degrees.”
- There is debate over whether universities should embrace AI as a core tool versus treating it as a threat, and whether CS programs should focus more on AI-era skills like effective LLM use.
- Some see large class sizes, bureaucratic metrics, and a research-first culture as underlying drivers of both poor teaching and adversarial integrity policies.