Problem
Research questions and friction points this paper is trying to address.
ARC-AGI-3
executable world models
coding agents
generalization
agent evaluation
Executable World Models for ARC-AGI-3 in the Era of Coding Agents
🤖 AI Summary
This work addresses the challenge of achieving generalization in ARC-AGI-3 tasks without relying on task-specific code. It proposes a novel approach based on an executable Python world model that integrates the principle of minimum description length with automated model refactoring. By combining model validation, abstraction-based simplification, and model-based planning, the method balances generality and parsimony without handcrafted rules. The system architecture comprises a world model, a validator, an abstraction-refinement module, a planner, and an executor. Evaluated on 25 public tasks, the approach fully solves 7 tasks and achieves a human-relative action efficiency (RHAE) exceeding 75% on 6 tasks, yielding an average score of 32.58% RHAE—establishing a new baseline for ARC solvers that eschew task-specific customization.
📝 Abstract
We evaluate an initial coding-agent system for ARC-AGI-3 in which the agent maintains an executable Python world model, verifies it against previous observations, refactors it toward simpler abstractions as a practical proxy for an MDL-like simplicity bias, and plans through the model before acting. The system is intentionally direct: it uses a scripted controller, predefined world-model interfaces, verifier programs, and a plan executor, but no hand-coded game-specific logic. We report results on the 25 public ARC-AGI-3 games. Each recorded playthrough uses a fresh agent instance with no access to previous playthrough-specific files or conversation state. Most games have a single recorded playthrough; for a few games, we report multiple independent fresh-agent playthroughs to expose run-to-run variability. The agent fully solved 7 games, achieved a Relative Human Action Efficiency greater than 75%, on 6 games, and obtained a mean per-game RHAE of 32.58%. Because the system uses no game-specific code, it can serve as a game-general baseline for ARC-AGI-3. Performance on the private validation set remains to be tested. Overall, the results provide preliminary evidence that verifier-driven executable world models are a promising approach for ARC-AGI-3 agents.
Innovation
Methods, ideas, or system contributions that make the work stand out.
executable world model
coding agent
model verification
simplicity bias
planning
🔎 Similar Papers
No similar papers found.
