Randomized trial: Oxbridge-style tutorials lift exam grades

The paper's claim is simple. When students know a language model can hand them a homework answer, the homework stops doing its job. An arxiv preprint by John C. Duchi, Will Hartog, and Dennis L. Sun proposes a fix borrowed from Oxford and Cambridge: weekly small-group tutorials where students present problems in person, from memory, to an instructor or TA.

They tested it. "We conducted a randomized experiment in an introductory probability class to evaluate the effectiveness of tutorials, finding a significant positive effect of tutorial enrollment on exam grades," the abstract reports. Two of the researchers we follow circulated the preprint shortly after it posted.

The abstract stops there. It publishes no effect size, no class size, and no description of how students were assigned to tutorial sections. The authors frame the design as a response to the "ubiquity of Large Language Models" devaluing "the intended productive struggle" of homework, and promise a later discussion of "implementation, challenges, and opportunities."