RewardSpy AI Debugger
6.5
A library to wrap existing reward functions in reinforcement learning (RL) environments, continuously monitoring and reporting on potential reward hacking allowing developers to identify and correct unintended exploits, inspired by Reddit’s rewardspy project.
80h
mvp estimate
6.5
viability grade
5
views
technology stack
Python
Difficult
inspired by
Debugger for RL reward functions detects reward hacking