Victor-Alexandru Padurean, Kaitlin Riegel, Alkis Gotovos, Jyotika Mahapatra, Ahana Ghosh, Adish Singla (2026) — arXiv:2607.05068 [cs.SE, cs.CY]
As generative AI becomes central to software development, CS education is shifting toward prompt-centered workflows where students describe intended behavior in natural language to elicit code. But professional practice demands careful review of GenAI output that may look correct yet harbor subtle faults — a challenge in CS1, where current models solve tasks correctly and dull students' incentive to inspect generated code. Padurean et al. (2026) investigate how prompt-centered programming activities can be designed to deliberately surface buggy GenAI code, forcing students to practice verification rather than blind trust. The work speaks directly to the over-reliance problem and the cognitive-offloading temptation of strong code models, and complements code-review-genai-cs1, which found oral code review preserves learning outcomes under rising AI use. By intentionally injecting faults, the approach reframes generative-ai in cs-education as a metacognitive training ground rather than an answer engine, reinforcing ai-literacy and metacognition. Implications for student-experience and formative-assessment design suggest that 'wrong-on-purpose' perturbations are a scalable way to build debugging habits in an era where reshaping-cs-education-genai is already underway.
Related Pages
- over-reliance — related
- generative-ai — related
- cs-education — related
- code-review-genai-cs1 — related
- ai-literacy — related
- metacognition — related
- student-experience — related
- reshaping-cs-education-genai — related
Citation
APA: Victor-Alexandru Padurean, Kaitlin Riegel, Alkis Gotovos, Jyotika Mahapatra, Ahana Ghosh, Adish Singla (2026). When AI Is Wrong on Purpose: How Students Respond to Buggy GenAI Code. arXiv:2607.05068. arXiv:2607.05068 [cs.SE, cs.CY].