single mention · data curation
Distilling LLM-judge annotation axes into binary classifiers
Each scalar judgment axis is converted into a binary top-versus-bottom-quartile classification task, excluding the middle observations.
The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See data filtering for the methods filed beside it.
Filed under data curation › data filtering