Faulty reward functions in the wild
OpenAI analyzes failures of reward functions in reinforcement learning. The article explores how misspecifying the reward function can cause unexpected and counterintuitive behaviors in RL algorithms.
2 articles
OpenAI analyzes failures of reward functions in reinforcement learning. The article explores how misspecifying the reward function can cause unexpected and counterintuitive behaviors in RL algorithms.
OpenAI releases Universe, a software platform to measure and train AI general intelligence across games, websites, and applications.