[Paper Review] ChatGPT in the Classroom: An Analysis of Its Strengths and Weaknesses for Solving Undergraduate Computer Science Questions

Ishika Joshi, Ritvik Budhiraja+6 more|arXiv (Cornell University)|Apr 28, 2023

Artificial Intelligence in Healthcare and Education7 citations

TL;DR

This study evaluates ChatGPT's reliability in answering diverse undergraduate computer science questions using a quantitative approach across multiple question types. It finds significant inaccuracies, especially in objective and coding questions, warning of self-sabotage risks from overreliance, while recommending strategic use as a supplementary tool for ideation and explanation rather than solution generation.

ABSTRACT

ChatGPT is an AI language model developed by OpenAI that can understand and generate human-like text. It can be used for a variety of use cases such as language generation, question answering, text summarization, chatbot development, language translation, sentiment analysis, content creation, personalization, text completion, and storytelling. While ChatGPT has garnered significant positive attention, it has also generated a sense of apprehension and uncertainty in academic circles. There is concern that students may leverage ChatGPT to complete take-home assignments and exams and obtain favorable grades without genuinely acquiring knowledge. This paper adopts a quantitative approach to demonstrate ChatGPT's high degree of unreliability in answering a diverse range of questions pertaining to topics in undergraduate computer science. Our analysis shows that students may risk self-sabotage by blindly depending on ChatGPT to complete assignments and exams. We build upon this analysis to provide constructive recommendations to both students and instructors.

Motivation & Objective

• To assess the reliability of ChatGPT in answering a broad spectrum of undergraduate computer science questions across different formats.
• To identify the risks associated with overreliance on ChatGPT for assignments and exams, including compromised learning and academic integrity.
• To provide evidence-based recommendations for students and instructors on how to use ChatGPT constructively to enhance learning without undermining understanding.
• To evaluate the impact of prompt design and question type on ChatGPT's response accuracy and consistency.

Proposed method

• A quantitative evaluation framework was developed to test ChatGPT’s responses across seven question types: true/false, multiple-choice, multi-select, short answer, long answer, design-based, and coding-related questions.
• Questions were drawn from core undergraduate computer science courses, coding interviews, and competitive examinations to ensure real-world relevance.
• Responses were systematically analyzed for factual accuracy, logical consistency, and completeness, with a focus on identifying hallucinations and reasoning errors.
• The study compared ChatGPT’s answers against gold-standard solutions to measure error rates and reliability across different question formats.
• The analysis included an evaluation of how prompt phrasing and contextual cues influenced response quality and consistency.

Experimental results

Research questions

• RQ1 What are the strengths and weaknesses of ChatGPT when answering various types of computer science questions?
• RQ2 How does ChatGPT’s performance vary across different question formats such as multiple-choice, coding, and design-based questions?
• RQ3 What are the risks of students relying on ChatGPT for assignments and exams, particularly in terms of learning outcomes and academic integrity?
• RQ4 How can students and instructors use ChatGPT constructively to enhance teaching and learning in computer science education?

Key findings

• ChatGPT demonstrated high unreliability in answering objective and coding-related questions, with significant error rates despite providing detailed explanations.
• In multiple-choice and true/false questions, ChatGPT frequently selected incorrect answers, even when explanations were plausible and well-structured.
• For short and long answer questions, responses were often verbose and contextually plausible but factually inaccurate, indicating a risk of misleading students.
• The model showed particular weakness in reasoning-based and design-oriented questions, where it failed to apply correct principles or constraints.
• In coding tasks, ChatGPT generated syntactically correct but logically flawed or incorrect solutions, especially for complex algorithmic problems.
• The study found that prompt phrasing significantly influenced response quality, with subtle changes leading to inconsistent or incorrect answers.