The story’s score and claims

Comments on No. 88 of 118 · mix-base-rlvr.pages.dev · edition of Saturday 21 November 2026

Mix Your Base and RLVR Weights: A Training-Free Knob for Diversity

Summary

Youngseong Kim, an independent researcher, reports that RLVR-trained language models gain single-sample accuracy (pass@1) but lose sample diversity (pass@k at large k).

Comments

Ranked by a public rubric, not by votes: what a comment adds counts, not who wrote it. How comments are ranked · Comment rules