Reinforcement Learning · advanced · concept 127 of 187

🎮Reward Design & Reward Hacking

Designing reward functions is RL's hardest problem. Reward hacking: agents find unintended shortcuts to maximize reward without actually solving the task. A core challenge in AI alignment.