Keegan Kang (Bucknell University)
Abstract
The widespread use of LLMs have made assigning out-of-class work challenging, yet having students do work in class might mean covering less content. In this breakout session, we discuss means of assessing student learning via out-of-class homework that cannot be easily completed by an LLM. We want to distinguish two broad groups of homework assignments: one where an LLM gives a plausible but wrong answer (in this case, a student might not recognize that the LLM is ineffective, hence no learning is done); and a second where an LLM gives an easily recognizable wrong answer (in this case, students recognize an LLM is ineffective, and does their own learning).
The intended audience is statistics instructors who want to design homework assignments where students recognize LLMs are not useful on their own, and do their own learning. The goal for this breakout session is to share ideas on how to construct effective homework assignments for statistics, and perhaps come up with a crowd-source compilation of ideas. Participants would first have a sample of "homework questions", and are encouraged to "plug the text" into an LLM to see the output, and record the types of answers seen. Discussion will then follow on what features lead to useful homework assessments for learning.