I read a paper called Checklists Are Better Than Reward Models for Aligning Language Models. It got me thinking about ways to get students more involved in their own learning.
In this episode, I unpack the idea of “reinforcement learning with checklist feedback,” where AI builds an atomic yes/no checklist to complete a task, and wonder what it would look like if students did the same. Could making their own checklists for projects increase buy-in, help them actually understand the task, and cut down on busywork or cheating?
I don’t have it figured out - at all - but I share some half-baked thoughts about atomic checklists, student-generated assessments, faculty feedback, and the tradeoff between added work and added value.