Reward Hacking

1 selectedDifficulty 6-61 unseenView topic

IntermediateNew

0 answered

1 intermediateAdapts to your performance

Question 1 of 1

120sintermediate (6/10)conceptual

Reward hacking is when an RL agent finds a high-reward strategy that violates the designer's intent. What is the structural cause, framed as a Goodhart-type failure?