Skip to main content Skip to navigation

An Eye for an AI: Evaluating GPT-4o's Visual Perception Skills and Geometric Reasoning Skills Using Computer Graphics Questions

Project Overview

The document examines the role of generative AI, specifically the GPT-4o model, in enhancing education through its application in solving Computer Graphics (CG) assessment questions. It showcases the model's potential to support learning by improving insights into visual perception and geometric reasoning skills. Although GPT-4o demonstrates improved performance over its predecessor, GPT-4, it still encounters notable challenges, particularly in accurately addressing CG tasks that require visual inputs. The findings underscore the importance of effective prompting techniques and the critical evaluation of AI-generated solutions to maximize the benefits of generative AI in educational contexts. The study advocates for innovative strategies to integrate AI into CG education, aiming to leverage its capabilities while acknowledging its limitations. Overall, the document highlights the transformative potential of generative AI in education while calling for careful implementation and assessment to ensure effective learning outcomes.

Key Applications

GPT-4o (Large Multimodal Model)

Context: Educational context for Computer Graphics students, targeting undergraduate courses.

Implementation: The study involved constructing datasets of CG questions and evaluating GPT-4o's responses to these questions using both text and multimodal inputs.

Outcomes: GPT-4o achieved 50.1% accuracy across CG questions, showing improved performance over its predecessor, particularly in multiple-choice questions.

Challenges: GPT-4o struggled with visual processing and geometric reasoning tasks, showing inconsistency and reliability issues, especially with image-based questions.

Implementation Barriers

Technical Limitation

GPT-4o's performance is inconsistent, particularly in visual and geometric reasoning tasks, limiting its reliability for complex CG questions.

Proposed Solutions: Educators can generate human-written image descriptions to assist the model and encourage students to critically evaluate AI-generated solutions.

Educational Challenge

Students may misuse AI tools like GPT-4o due to its inaccuracies, leading to a false sense of understanding.

Proposed Solutions: Incorporate exercises that involve evaluating incorrect AI solutions to enhance critical thinking and understanding.

Project Team

Tony Haoran Feng

Researcher

Paul Denny

Researcher

Burkhard C. Wünsche

Researcher

Andrew Luxton-Reilly

Researcher

Jacqueline Whalley

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Tony Haoran Feng, Paul Denny, Burkhard C. Wünsche, Andrew Luxton-Reilly, Jacqueline Whalley

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