Abstract

The paper introduces Alfred, a system for programmatic weak supervision (PWS) that creates training data for machine learning by prompting. It enables users to encode their subject matter expertise via natural language prompts for language and vision-language models.