DCPO decouples reasoning and calibration in RLVR, matching GRPO accuracy while reducing overconfidence via a claimed fundamental gradient conflict.
Sconu: Selective conformal uncertainty in large language models.arXiv preprint arXiv:2504.14154
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Decoupling Reasoning and Confidence: Resurrecting Calibration in Reinforcement Learning from Verifiable Rewards
DCPO decouples reasoning and calibration in RLVR, matching GRPO accuracy while reducing overconfidence via a claimed fundamental gradient conflict.