Banca de DEFESA: PAULO CEZAR KROTH DOS SANTOS

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
DISCENTE : PAULO CEZAR KROTH DOS SANTOS
DATA : 22/12/2023
HORA: 08:00
LOCAL: https://meet.google.com/zqu-opzg-dyf
TÍTULO:

GENERATIVE AI IN MATHEMATICS EDUCATION: EXPLORING CHATGPT IN LEVEL I QUESTIONS OF THE UNEMAT MATHEMATICS OLYMPIAD

PALAVRAS-CHAVES:

Artificial Intelligence, Unemat Mathematics Olympiad, ChatGPT

PÁGINAS: 104
GRANDE ÁREA: Ciências Exatas e da Terra
ÁREA: Matemática
SUBÁREA: Matemática Aplicada
RESUMO:

This research explores the use of Generative Artificial Intelligence (GAI), specifically OpenAI's ChatGPT, as an aid tool for mathematics teachers in developing support materials for 4th and 5th year students, motivated by participation in the Mathematics Olympiad. Unemat.
The UNEMAT Mathematics Olympiad is structured into three distinct phases, each one presenting an increasing level of difficulty and complexity, covering students from the early years of elementary school to high school. For the purpose of this research, we focused specifically on Level I, which covers students in the 4th and 5th grades, with all questions from the years 2017, 2018, 2019, 2022 and 2023 being used for ChatGPT analysis and evaluation.
An assessment of the potential and limitations of the ChatGPT-4 and ChatGPT-3.5 versions was carried out, organized into three methodological stages. The first stage focused on the accuracy of ChatGPT-4, analyzing its successes and errors in mathematical questions. The second stage evaluated ChatGPT-3.5's ability to generate relevant questions for students' preparation for the Unemat Mathematics Olympiad. The third step compared the performance of the two versions of ChatGPT, examining their hit and error rates.
The results revealed that ChatGPT-4 has an accuracy rate of 82.20%, with the majority of errors occurring in questions with visual elements, indicating a limitation of the model in this area. The questions generated by ChatGPT-3.5 were mostly relevant, although some were too complex for the target audience. Comparison between versions showed that ChatGPT-4 is superior to ChatGPT-3.5 in terms of accuracy and detail in responses.
The research concludes that ChatGPT has significant potential as an auxiliary tool in mathematics teaching, but highlights the need for improvements in the interpretation of visual elements and in calibrating the difficulty of questions. The identified limitations offer valuable directions for future research and improving the integration of IAG into mathematics education. As a product, the research provides a valuable documentary record of the responses and questions generated by the ChatGPT versions, contributing to future comparative studies in the area.

MEMBROS DA BANCA:
Presidente - 105502004 - ERICO FERNANDO DE OLIVEIRA MARTINS
Interno - 82417001 - MIGUEL TADAYUKI KOGA
Interno - 80963002 - RODRIGO BRUNO ZANIN
Externo à Instituição - EDSON PEREIRA BARBOSA - UFMT