BARC is a framework for generating synthetic examples to enhance AI reasoning capabilities. This page showcases examples generated using different models and approaches.

Available Models:

• Heavy with Suggestions: Generated using GPT4o descriptions and GPT4o-mini with suggestion functions for code generation
• Heavy: Uses GPT4o descriptions and GPT4o-mini code generation
• GPT-4: Combines GPT-4 descriptions with GPT4o-mini code generation
• GPT4o-mini: Uses GPT4o-mini for both descriptions and code generation

Each example card shows:

• Training input-output pairs that demonstrate the pattern
• Test input and output examples
• Associated seed examples used for generation
• Generated Python code (viewable by clicking the code button)

Features:

• Switch between different model outputs using the dropdown
• Shuffle examples to view them in random order
• View and copy the generated Python code
• Examine seed examples used in generation
• Navigate to related examples through seed connections