Rankings

I've always been confused about university rankings. Especially the ones that assign a single number and call it a tier list. You can't take something as multi-faceted as a university, reduce it to one number, and proclaim that #1 is "the best". "Best" isn't intrinsic to the thing being judged—it's a matter of who's doing the judging. Prospective undergraduate, postdoc, or faculty? All care about different things.

I remind myself how useless I find these rankings every time a new AI model drops (Opus 4.8, recently) and the reporting immediately fixates on which model is now the best. "The best at what, for whom, in which context?", I want to shout. And even when the evaluation breaks down into software engineering benchmarks, it's not that interesting: How well does it perform on my codebase, in my tech stack? I insider-joked the other day that I'll believe Artificial General Intelligence has finally arrived once Claude stops being surprised that my linter removes unused imports on every save.

When a new model drops and you suspect there's more performance on the table, spend the time to see what it does for you. There's no #1 model. Only the one that ranks first on your work.