List your company

Code Gaming Mitigation on Code Gaming (held-out tests)

4.2Gaming RateSFT (NO RLHF)
1.91617.33332.7548.167Feb 2, 2026

Evaluation results

MethodLinks
SFT (NO RLHF)RLHF=None2026.024.228.5
ARAFramework=Auditor-gated RLHF2026.0219.635.8
INFORMCategory=Regularization Methods2026.0239.833.1
ODINCategory=Regularization Methods2026.0242.133.5
RM ENSEMBLECategory=Reward Model Improvements2026.0244.334
FILTERING (R > mu + 2sigma)Category=Training Interventions2026.0246.732.4
PPO W/KLCategory=Regularization Methods2026.0248.531.8
PPOCondition=Unmitigated2026.0261.334.2