Summary
The episode works through four questions that students find difficult when they write with AI assistants, beginning with what the experiments of Noy and Zhang (2023) and Doshi and Hauser (2024) show about speed, quality and convergence, and what their short, self-contained tasks leave out. It examines why an assistant's report that an edit preserved the meaning needs its own check, using a hypothetical rewrite that adds a causal claim, drops a hedge, recasts a comparison between groups of schools as a change over time and states a difference in percentage points as a percentage. A harder worked example combines the evidence that AI-text detectors misclassify writing by second-language writers (Liang and colleagues, 2023) with base rates to show what a flag means for different groups of writers, and the hosts relate the result to SFU's guidelines, publisher disclosure rules and the tri-agency guidance on grant proposals. The episode then shows how splitting sentences lowers a Flesch-Kincaid grade while every word stays as hard as before, and why plain-language and translated materials are finished only after real readers have reviewed them. It closes with a discussion between the hosts about when an assistant should write a first draft.