The tasks in this competition are based on the Abstraction and Reasoning Corpus (ARC-AGI), a groundbreaking dataset created by François Chollet, a prominent AI researcher at Google. The original dataset is designed as a benchmark for AGI, focusing on abstract reasoning rather than pattern recognition. Each task is a unique puzzle, intended to be easily solvable by humans but challenging for current AI models.

For the purpose of this hackathon, the dataset has been specially curated and modified by Hackade. While the core challenge of abstract reasoning remains the same, this version has been adapted to create the specific train/test splits and structure for this competition on the Zindi platform.

Each task in the dataset consists of:

Training Pairs: A small number of example input-output grids that demonstrate a specific abstract transformation.

Test Input: One or more input grids for which your model must predict the corresponding output grid.

Resources:

ARC Official website: https://arcprize.org/

ARC Prize 2024 Technical Report: https://arxiv.org/html/2412.04604v1

The dataset will be divided into:

• Training Pairs: A small number of example input-output grids that demonstrate a specific abstract transformation.
• Test Input: One or more input grids for which your model must predict the corresponding output grid.

The leaderboard will be based on a 70% evaluation on the public and 100% evaluation on the private test set to ensure a simple evaluation strategy.