I feel like an analysis of question complexity would be interesting, albeit hard to quantify. ChatGPT is great for dealing with "trivial" problems (i.e. which function from which package can do X), but complicated problems might still require a human to solve.