The Step That Must Be Yours

By Oliver Maynard, Statistics 

Teaching Effectiveness Award Essay, 2026

Concepts of Probability, Stat 134, equips students with a probabilistic toolkit. The core skill of the course is not computation but, rather, selection: given a problem, which tool applies? Over three semesters as a GSI, I have watched performance decline in ways that the professors I work with describe as unlike anything they have seen before.

The culprit is a subtle but damaging shift in how students study. Large language models have made it trivially easy to obtain the first step of any problem. Students ask in good faith, receive a nudge, and work through the remaining mechanics, leaving them with the impression that they understand the material. But in Stat 134, the first step is everything. Identifying the correct probabilistic framework and asking whether the setup actually makes sense in context is the skill the course is intending to build. By outsourcing it habitually, students arrive at exams having practiced everything except the one thing the exam will truly test. The evidence is stark: the pass threshold, normally set at 40%, had to be lowered to 18% last semester. Scores were clustered at the extremes, near full marks or zero, with almost no partial credit in between. Students either aced the first step or could not begin at all.

For much of my early teaching I made an error that felt like kindness. When a student was stuck I would offer the first hint freely. They would complete the computation and leave satisfied. When I asked for questions, hands stayed down. It looked like understanding. It was not. Thinking carefully about pedagogy led me to a diagnosis. The gratification students felt after receiving that first nudge was masking a critical gap. I began to deliberately withhold it. In practice, this meant spending the majority of section time not checking whether students had the right answer, but diagnosing whether they had chosen the right tool. Rather than working through problems on the board, I would sit with each group and press them on their setup: what is this random variable actually representing? Would it make sense for it to behave this way? Only once they could answer those questions would we touch the math. I was also explicit with students about AI. I told them plainly: using it for computation costs you little, calculus is not what this course is testing. Using it for the first step costs you everything.

The effects were measurable. Quiz averages shifted from 38% to 60% comparing equivalent assessments across semesters. Even more telling was the change in distribution. Previously, scores clustered at the extremes, the statistical signature of the first step problem. This semester they spread out, with partial credit appearing throughout the range. Students who once sat frozen in front of a blank page began putting pen to paper. The attempts were not always correct, but they reflected genuine probabilistic reasoning. Struggling with the first step is not an obstacle to learning. It is the learning. In an age where that struggle can be skipped with a single prompt, being deliberate about preserving it may be among the most consequential things a teacher can do.