Skip to main content Skip to navigation

Navigating Compiler Errors with AI Assistance -- A Study of GPT Hints in an Introductory Programming Course

Project Overview

The document explores the application of generative AI, particularly the GPT-4 model, in an introductory programming course to provide personalized hints for resolving compiler errors. The study revealed that students appreciated the usefulness of the hints generated by GPT, resulting in increased focus and decreased confusion and frustration among those in the experimental group. However, despite these positive subjective experiences, the impact of the AI-generated hints on actual academic performance was inconsistent, particularly with more complex programming errors, where students did not always perform better when using the hints. Overall, while generative AI shows promise in enhancing student engagement and support in learning programming, its effectiveness in improving performance remains variable and suggests the need for further investigation into its optimal use in educational settings.

Key Applications

GPT-4 model for generating personalized hints for compiler errors

Context: Introductory programming course for computer science students at a university in Poland

Implementation: Hints generated via OpenAI API when a compiler error was detected during automated assessment of programming assignments.

Outcomes: Students rated the hints as useful, reported increased focus and decreased confusion; experimental group performed better on complex tasks after hints were disabled.

Challenges: Mixed results in performance; some students found hints unhelpful; challenges in automating hint generation due to the variety of potential compiler errors.

Implementation Barriers

Technical

The complexity and variety of compiler error messages make it difficult to provide relevant hints.

Proposed Solutions: Refined prompt engineering and potential fine-tuning of the AI model to improve hint relevance.

User Experience

Approximately 20% of students found the hints not beneficial.

Proposed Solutions: Gathering user feedback to tailor hints more closely to student needs.

Project Team

Maciej Pankiewicz

Researcher

Ryan S. Baker

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Maciej Pankiewicz, Ryan S. Baker

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