CMU professor redesigns course assessments after AI agents ace every assignment
AustinZHenley · x · 2026-09-25
CMU professor Christian Kästner details how he rebuilt assessments for his ML in Production course (100-170 students) after AI agents could complete every assignment.
Key points:
- Back in 2021, before ChatGPT, GPT-3 already passed reading quizzes without reading the assigned papers, but he changed nothing then.
- Learning goals barely changed—engineering tradeoffs, risk mitigation, teamwork—since the course isn't about writing code.
- Core strategy: stop testing understanding via take-home work; shift to TA interactions, in-person exams, and video demos.
- Students are explicitly allowed to use AI anywhere without attribution, even where changes violate evidence-based pedagogy best practices.
- Intro or traditional software engineering courses would likely need far deeper learning-goal revisions.
More from AGI Musings
- Founder argues polymath strategy is negative EV in today's risk-punishing system — felpix_ · 2026-09-25
- Are We Overestimating AI's Short-Term Impact and Underestimating the Long Term? — EnvironmentalDay8864 · 2026-09-25
- Beff Jezos: humanities grads hated techies for a decade out of jealousy — beffjezos · 2026-09-25
- Opinion: An AGI aligned with humanity's interests would seek to escape technocratic control — examachine · 2026-09-25
- Theo: reasoning-effort dropdowns are over-optimization; models will soon auto-tune — heyneighbor · 2026-09-25
- soleio: non-native English speakers are precise because words carry higher stakes — soleio · 2026-09-25