What Should World Models Forget? Stratified Retention for Continual Adaptation
Teaching AI to forget outdated facts while keeping physics rules forever
World models need to unlearn information that becomes false as environments change—unlike traditional AI that treats all forgetting as failure. This paper shows that current methods can't tell the difference between a model that correctly updated to new conditions and one that catastrophically forgot important rules, and proposes a new way to measure which knowledge should stick around forever and which should change.
Robots and autonomous systems that operate in changing environments need to adapt continuously without losing core knowledge like how physics works or that objects don't vanish. Current evaluation methods reward models that never change, even when that's the wrong behavior, making it impossible to build systems that learn reliably over time.