This page displays the seed examples from the ARC (Abstraction and Reasoning Corpus) training dataset that were used to generate synthetic examples in the BARC framework. These seeds serve as inspiration for generating new, similar reasoning tasks.

Each seed example shows:

• Training pairs demonstrating the pattern or transformation
• Test input and expected output
• Python code implementing the solution
• Unique identifier for referencing in synthetic examples

Key Features:

• View the original ARC tasks used as seeds
• Examine the human-written solutions that explain each pattern
• Copy and run the solution code
• Shuffle examples to explore different patterns
• Filter seeds to see specific examples

Navigation:

• Click "Model Performance" to see how different models solve ARC tasks
• Visit "Synthetic Examples" to see how these seeds are used to generate new tasks
• Use the shuffle button to randomize the display order
• Click the code button on any example to view its implementation