Introducing PROWL!

We’ve built RL agents that explore game environments, tasked with discovering failures in world models across physics, visuals, and actions.

Those failures then become training data in an automated loop that advances world model performance.