Program Synthesis can automate a wide variety of tasks for spreadsheet users (e.g., string transformations, table extraction, advanced data analysis), developers (e.g., debugging, repeated or associated edits), and students (e.g., grading, feedback or hint generation). In this talk, I will demonstrate that user intent can be expressed not only through natural language but also via input-output examples, static and temporal context, and broken artifacts for repair. The most natural way for a user to express intent depends on the task at hand. Additionally, I will advocate for leveraging neuro-symbolic techniques, which combine the power of large language models (LLMs) with logical-reasoning-based symbolic techniques, to build more effective solutions for specific verticals.
Back to Workshop III: Naturalistic Approaches to Artificial Intelligence
Copyright. All Rights Reserved.
