Definition
Plain language
A small, carefully chosen subset of a big dataset that stands in for the whole thing.
As stated in the literature
A compact representative subset selected to preserve some property of a larger dataset; in RHO, a set of past trajectories chosen to balance difficulty and diversity for harness optimization.
Also called: coreset selection
Why it matters: It saves time and compute by letting you work with a small sample that behaves like the whole dataset.
For example, instead of testing a model on a million records, you might pick a few thousand carefully chosen ones that still capture the full range of cases.
Heard on the show
“Let me take the coreset one first, the study-group selection.”Episode 120 — How an AI Agent Rewrites Its Own Tools, Without an Answer Key