Definition
Plain language
Deliberately picking answers that differ from each other instead of picking the ten that look best.
As stated in the literature
Selecting a submission set to maximize dissimilarity among candidates (e.g. greedy farthest-point selection over normalized source-code chunks) so that graded feedback probes distinct algorithmic strategies rather than duplicates.
Also called: diverse selection, diversity selector
Why it matters: It makes every graded attempt buy new information, since ten copies of the same idea all fail for the same reason and teach you nothing.
For example, instead of submitting ten near-identical programs that all use the same sorting trick, the system picks ten that each attack the problem a different way.
Heard on the show
“And there’s an important methodological detail here: the diversity selector is blind to grades.”Episode 258 — The Same Weights Scored 291, Then 468 — What Changed Was the Loop