Skip to main content Skip to navigation

From Generation to Adaptation: Comparing AI-Assisted Strategies in High School Programming Education

Project Overview

The document explores the use of Large Language Model Coding Agents (LCAs) in high school programming education through a comparative case study of two pedagogical approaches: the 'From-Scratch' methodology and the 'MFU-based' approach. The 'From-Scratch' method requires students to generate code based on abstract specifications, while the 'MFU-based' approach focuses on adapting existing Minimal Functional Units (MFUs). The findings reveal that the MFU-based approach significantly improves student outcomes, with a remarkable 100% completion rate of Minimum Viable Products (MVPs), in stark contrast to only 20% in the From-Scratch method. This highlights the crucial role of instructional design in effectively integrating generative AI tools like LCAs in educational settings, suggesting that well-structured instructional strategies can leverage AI capabilities to enhance learning and achievement in programming education.

Key Applications

Large Language Model Coding Agents (LCAs) for programming education

Context: High school students (Grades 10-12) participating in a WeChat Mini Program competition

Implementation: Students first attempted to create programs from scratch, then switched to adapting existing MFUs with structured guidance.

Outcomes: Increased MVP completion from 20% to 100%, improved student confidence, faster feature implementation, and better structured prompts.

Challenges: Initial struggle with creating effective prompts, debugging issues, and lack of contextual understanding when generating code from scratch.

Implementation Barriers

Pedagogical Barrier

The success of LCA integration depends more on instructional design than on AI capabilities.

Proposed Solutions: Implement a dual-scaffolding model combining technical support with pedagogical guidance.

Technical Barrier

Students struggled with generating precise prompts and debugging AI-generated code.

Proposed Solutions: Shift from a '0-to-1' generation strategy to a '1-to-100' adaptation approach using MFUs.

Project Team

Tong Hu

Researcher

Songzan Wang

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Tong Hu, Songzan Wang

Source Publication: View Original PaperLink opens in a new window

Project Contact: Dr. Jianhua Yang

LLM Model Version: gpt-4o-mini-2024-07-18

Analysis Provider: Openai

Let us know you agree to cookies