Learning Goals

  • Master multivariable calculus as a foundation for advanced math and science courses
  • Develop conceptual understanding beyond computational memorization
  • Cultivate independent learning skills and confidence in seeking resources

 

The Implementation

  • Instructor-generated course materials (primary use)—GenAI was used to draft and refine examples, topic-specific practice sets (e.g., partial derivatives, double integrals), and exam questions calibrated to an appropriate difficulty
  • Exam question refinement—As one example, GenAI helped adjust parameters for a surface-integral question with messy radicals, so the final answer was a clean perfect square; Park then verified by hand
  • Mathematical software support—Park originally planned to have students use GenAI as a software tutor; instead, she used it herself to generate Desmos, Mathematica, and SageMath visualization code shared in class
  • Concept explanation—Park originally planned a student group activity peer teaching Green's and Stokes' Theorems; instead, she used it to develop alternative explanations that she then brought into lecture

 

What Worked

  • Strategic AI prompting—specific wording (e.g., "conceptual," "geometry-aware," textbook references) produced significantly better content than generic requests
  • Content generation efficiency—saved preparation time while maintaining quality across problem sets and exam variants
  • Responsive teaching within a fixed structure—generating materials quickly let Park react to the students’ learning experience without falling behind the shared course schedule
  • Quality control checks—instructor verification of every AI output ensured mathematical accuracy

 

Challenges Encountered

  • Multisectional course constraints—shared syllabus, common exams, and no latitude for adding credited assignments made student-facing activities hard to require
  • Optional GenAI activities had no takers—students focused on the heavy weekly workload instead of optional, non-credited GenAI activities
  • Arithmetic errors and hallucinated references—GenAI made confident mistakes; every output and citation needed instructor verification
  • Default outputs were rudimentary—generic prompts produced repetitive, basic textbook-drill questions

 

Road Forward

  • Pilot GenAI use in smaller, project-based courses like Math 28 (Combinatorics) or Math 38 (Graph Theory), where adding credited student activities is possible
  • Establish honor-code boundaries up front in the syllabus—naming where GenAI is permitted, encouraged, or inappropriate
  • Pair instructor materials with light student touches—for example, give students an AI-generated problem with its AI-generated solution and ask them to grade it as if they were the instructor
  • Use GenAI for iterative refinement (adjusting numbers, varying problem difficulty) rather than initial generation of assessments tied to learning goals

 

Faculty Voice

“I would advise against using GenAI for coordinated multisectional classes. Without the liberty to assign credited work, it's difficult to motivate students to participate in optional activities. There are higher-level math classes—more theoretical, project-based—that allow both instructors and students to explore. I would love to try again in courses like Math 38: Graph Theory or Math 28: Introduction to Combinatorics.”