Search papers, labs, and topics across Lattice.
This paper addresses the challenge of continual adaptation in world models by proposing stratified retention, recognizing that environments change and previously accurate knowledge can become false. It argues that world models require a mechanism to distinguish between knowledge that must be permanently retained (e.g., physics) and knowledge that should be revised (e.g., instance-level facts). The key result is the introduction of differential retention, a metric that jointly evaluates invariant regression testing and revision latency without aggregation, offering a more accurate assessment of continual world model performance.
World models need to forget selectively, discarding outdated instance facts while rigidly retaining fundamental invariants like physics, a nuance standard forgetting metrics miss.
Continual learning treats degradation on previously seen data as evidence of failure, a convention inherited from settings with a stationary prediction target, where a correct label remains correct indefinitely. World models do not satisfy this condition. Their prediction target is the environment, which changes, so knowledge that was accurate when acquired may later become false, and discarding it is required behavior rather than a defect. Non-stationary ground truth is well studied in the concept drift literature and in the temporal factuality of language models, but has not been formulated for world models, which are distinctive in that they also encode knowledge that must never be revised. We argue that continual world models require retention stratified by invariance timescale, separating invariants such as physics and object permanence, which must never be revised, from instance-level facts that should be revised as soon as the environment changes. Standard forgetting metrics cannot distinguish a world model that has correctly revised outdated knowledge from one that has suffered catastrophic forgetting, and consequently rank a frozen model highest, while existing physical-reasoning benchmarks evaluate only frozen checkpoints. We propose differential retention, which reports invariant regression testing across the adaptation stream jointly with revision latency, without aggregation.
