Learning Task Decomposition to Assist Humans in Competitive Programming

TL;DR

This paper introduces a method for decomposing complex programming solutions into simpler parts to help humans repair and understand them more efficiently, significantly improving non-expert performance.

Contribution

It proposes a novel objective called assistive value (AssistV) for learning effective task decomposition to aid human problem-solving in programming tasks.

Findings

Non-experts solve 33.3% more problems

Speed increases by 3.3x

Empowers non-experts to match expert performance

Abstract

When using language models (LMs) to solve complex problems, humans might struggle to understand the LM-generated solutions and repair the flawed ones. To assist humans in repairing them, we propose to automatically decompose complex solutions into multiple simpler pieces that correspond to specific subtasks. We introduce a novel objective for learning task decomposition, termed assistive value (AssistV), which measures the feasibility and speed for humans to repair the decomposed solution. We collect a dataset of human repair experiences on different decomposed solutions. Utilizing the collected data as in-context examples, we then learn to critique, refine, and rank decomposed solutions to improve AssistV. We validate our method under competitive programming problems: under 177 hours of human study, our method enables non-experts to solve 33.3\% more problems, speeds them up by 3.3x, and empowers them to match unassisted experts.