How I changed teaching after AI managed to do all my homework assignments
11 hours ago
- AI advancements have forced the instructor to redesign assessments, shifting focus from written homework to in-person interactions, oral check-ins, and video demos.
- Despite deviating from evidence-based best practices (like frequent low-stakes homework), the instructor prioritizes authentic engagement over traditional methods to counter AI abuse.
- Written reflections have been replaced with 15-minute TA conversations, graded pass/fail, with retry options to reduce fairness concerns and maintain learning outcomes.
- Reading quizzes were abandoned due to LLM delegation; readings are now integrated into in-class discussions without points, acknowledging most students rely on AI.
- Coding tasks now involve larger, production-scale assignments (e.g., Zulip) that require interactive AI agent use, with in-person knowledge checks and video demos for verification.
- Grading shifted toward classroom activities, with exams now worth 25% (up from 15%), and AI-autograding (LLM-as-a-judge) reduces TA workload while maintaining human review for deductions.
- The instructor intentionally designs some tasks where AI is confidently wrong (e.g., security problems) to teach students about automation bias and calibrate trust.
- Students generally accept these changes, though learning outcomes are hard to measure; grade distributions have slightly declined, but this started before AI changes.
- Continuous adaptation is necessary as AI improves, requiring semester-by-semester reassessment of assignments and teaching strategies.