Skip to main content Skip to navigation

ChatGPT and general-purpose AI count fruits in pictures surprisingly well

Project Overview

The document explores the integration of generative AI, particularly ChatGPT and the T-Rex foundation model, in the field of education, showcasing their potential in enhancing learning experiences and facilitating various educational tasks. Key applications include automating content creation, providing personalized tutoring, and assisting in complex problem-solving, which can significantly alleviate the workload of educators and improve student engagement. The findings indicate that these AI models, due to their user-friendly nature and minimal coding requirements, can be widely adopted by educators, enhancing their teaching capabilities and making advanced technology accessible. Moreover, the document emphasizes that generative AI can outperform traditional educational tools, offering more efficient and effective solutions for diverse learning needs. However, it also acknowledges certain limitations, such as the need for careful oversight to ensure the accuracy and appropriateness of the generated content. Overall, the document presents a compelling case for the transformative potential of generative AI in education, advocating for its broader implementation to enrich learning environments and optimize educational outcomes.

Key Applications

ChatGPT and T-Rex for counting coffee cherries

Context: Used in agricultural settings by local farmers to count coffee cherries in images taken with smartphones

Implementation: ChatGPT utilized zero-shot learning by prompting with queries; T-Rex employed few-shot learning with bounding boxes drawn by users

Outcomes: T-Rex outperformed the YOLOv8 model in accuracy and required significantly less time for processing; ChatGPT showed potential, especially with user feedback

Challenges: ChatGPT had issues with reproducibility, and both models faced challenges in adapting to agricultural-specific conditions like lighting and object variability

Implementation Barriers

Technical barrier

The inability to set random seeds or control parameters in LLMs affects reproducibility in research.

Proposed Solutions: Ongoing research and development to address these methodological constraints.

Practical barrier

Traditional deep learning approaches require extensive data labeling, programming skills, and significant time investment.

Proposed Solutions: Foundation models reduce the need for extensive training and programming, allowing for easier implementation.

Project Team

Konlavach Mengsuwan

Researcher

Juan Camilo Rivera Palacio

Researcher

Masahiro Ryo

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Konlavach Mengsuwan, Juan Camilo Rivera Palacio, Masahiro Ryo

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