ARC-AGI Without Pretraining

2025-12-05

AI Summary

This work addresses IQ-style puzzles in the ARC-AGI-1 visual reasoning benchmark. We propose a fully pretraining-free, end-to-end solver that operates under a single-test-sample setting. Our method leverages inference-time optimization of the Minimum Description Length (MDL) objective to induce a compact neural network (76K parameters) to generate generalizable programs. Crucially, it abandons conventional reliance on large-scale training data and achieves, for the first time, deep learning–based solving purely at inference time—without any training samples. Experiments demonstrate successful resolution of 20% of puzzles in the ARC-AGI evaluation set, exhibiting strong generalization and diverse reasoning capabilities across unseen tasks. This provides critical empirical validation for MDL as a foundational principle for modeling general intelligence.

Technology Category

• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning
• Machine Learning: Deep Neural Architectures and Foundation Models
• Search and Optimization: Learning to Search

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Web Mining and Content Analysis: Large pretrained models with web data
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Conventional wisdom in the age of LLMs dictates that solving IQ-test-like visual puzzles from the ARC-AGI-1 benchmark requires capabilities derived from massive pretraining. To counter this, we introduce CompressARC, a 76K parameter model without any pretraining that solves 20% of evaluation puzzles by minimizing the description length (MDL) of the target puzzle purely during inference time. The MDL endows CompressARC with extreme generalization abilities typically unheard of in deep learning. To our knowledge, CompressARC is the only deep learning method for ARC-AGI where training happens only on a single sample: the target inference puzzle itself, with the final solution information removed. Moreover, CompressARC does not train on the pre-provided ARC-AGI "training set". Under these extremely data-limited conditions, we do not ordinarily expect any puzzles to be solvable at all. Yet CompressARC still solves a diverse distribution of creative ARC-AGI puzzles, suggesting MDL to be an alternative feasible way to produce intelligence, besides conventional pretraining.

Problem

• Solves IQ-test-like visual puzzles without pretraining.
• Minimizes description length for extreme generalization.
• Trains only on target puzzle, not external data.

Innovation

• Minimizes description length during inference
• No pretraining, only 76K parameters
• Trains solely on target puzzle sample