AUG 12, 2026 · PREPRINT
GCPO: Diagnosing and Constraining Subspace Geometry in Rollout RL for LLMs
arXiv
This is an unreviewed preprint proposing a novel method (GCPO) with empirical results on benchmark tasks, but lacks the peer review, independent validation, and clinical/deployed-setting evidence needed for stronger evidence grades.
Reported
Improvement over base modelsup to 27.69 points
Improvement over strongest baseli…up to 2.37 points