research
The World Model Knows the Pieces but Not Their New Arrangement
AnyGoal holds familiar state and action factors inside training support, then recombines them into underrepresented planning tasks.
Summary
AnyGoal holds familiar state and action factors inside training support, then recombines them into underrepresented planning tasks.
A world model can reproduce expert trajectories and still fail when familiar ingredients appear in a new combination. AnyGoal builds goal-shifted variants where each state factor and action remains known, but their pairing was rare in expert data. Across several model families and environments, the authors found consistent prediction and planning failures under those recombinations. The benchmark is a diagnostic rather than a solution, but it raises the standard for generalization: knowing every move independently is not the same as knowing the world's rules.
Why it matters
AnyGoal holds familiar state and action factors inside training support, then recombines them into underrepresented planning tasks.
Limits and context
- The benchmark is a diagnostic rather than a solution, but it raises the standard for generalization: knowing every move independently is not the same as knowing the world's rules.
Key claims
AnyGoal holds familiar state and action factors inside training support, then recombines them into underrepresented planning tasks.
Qualification: The benchmark is a diagnostic rather than a solution, but it raises the standard for generalization: knowing every move independently is not the same as knowing the world's rules.
Evidence: source-2026-10-20-011
Sources
- arXiv preprint 2610.26844, original v1arXiv · primary research
Corrections
No corrections have been recorded for this story.