List your company

Length Bias Mitigation on Length Bias

347Avg LengthPPO
140.04193.77247.5301.23Feb 2, 2026

Evaluation results

MethodLinks
PPOCondition=Unmitigated2026.0234723.1
PPO W/KLCategory=Regularization Methods2026.0226822.8
FILTERING (R > mu + 2sigma)Category=Training Interventions2026.0224122.6
RM ENSEMBLECategory=Reward Model Improvements2026.0222423
INFORMCategory=Regularization Methods2026.0220823.2
ODINCategory=Regularization Methods2026.0219523.4
ARAFramework=Auditor-gated RLHF2026.0216224.1
SFT (NO RLHF)RLHF=None2026.0214821.4