Tag: reward hacking
Browse our exclusive articles!
Preventing Reward Hacking in RLHF with Sign-Certified PO
Discover how Sign-Certified Policy Optimization improves RLHF by mitigating reward hacking through advantage sign robustness for better AI alignment.
PROGRS: Enhancing LLM Reasoning with Process Rewards
Discover PROGRS, a framework improving LLM mathematical reasoning by combining process rewards and outcome correctness for accurate, efficient AI solutions...
Extending MONA for Reward-Hacking Mitigation in RL
Explore MONA extension in Camera Dropbox for reward-hacking mitigation, with learned approval and PPO training enhancing AI safety in reinforcement learnin...
Understanding Reward Hacking in AI under Finite Evaluation
Explore how reward hacking forms a structural equilibrium in AI systems under finite evaluation, impacting alignment and optimization strategies.
Avoiding Faulty Reward Functions in Reinforcement Learning
Learn how to design effective reward functions in reinforcement learning to prevent failures and ensure AI agents behave as intended.