Abstract:
Sharing 1 - Large Language Model: Background, Techniques and Our Development
Speaker(s):
Dr. Jing MA (Assistant Professor, Department of Computer Science, HKBU)
Large Language Models (LLMs) have revolutionized the way we interact with machine learning in natural language processing. We will start from a background on the evolution of LLMs, tracing their lineage from early computational linguistics to the deep learning era. We will explore the foundational techniques that have enabled LLMs to achieve remarkable feats in understanding and generating human language. Then we will move to our recent research and developments in the field of LLMs: (1) WizardCoder is a programming assistant designed to augment the coding capabilities of developers; and (2) Detecting harmful meme that visual and textual content is often blended and have potential social harms.
Sharing 2 - Experiments in Generative Al Competences in Grading Assessments
Speaker(s):
Professor Ahti PIETARINEN (Professor, Department of Religion and Philosophy, HKBU)
Dr. Clara Mingyu WAN (Lecturer, School of Continuing Education, HKBU)
We conducted a prototype experiment on a sample of student essays graded by GPT-4 with multiple feedback with course-specific rubric priming. The output is analyzed qualitatively and quantitatively. The pilot suggests significant matching with human instructor with small average letter grade deviation. The preliminary results indicate model converge, even superiority, to human grading and feedback competences. Further examination with larger datasets should address questions such as GPT-4’s analytic competences, objectivity in assessment, and quality of iterative feedback, as well as stability and responsible uses from the instructor’s’ points of views. What are the foreseeable competitive pedagogical and institutional advantages as well as potential disadvantages from the learning perspective?
Sharing 3 - Customizing ChatGPT for Automating Grading and Feedback Generation: Promises and Challenges for University Language Teachers
Speaker(s):
Dr. Simon WANG (Lecturer, Language Centre, HKBU)
Mr. Junxin HUANG (Research Assistant, Department of Computer Science/ Language Centre, HKBU)
This talk delves into the potential of ChatGPT as an automated reasoning engine for educators. It showcases the customization of ChatGPT for automated grading and comprehensive feedback for written assignments. It emphasizes the need for a coherent assessment platform and how a tailored system can significantly reduce redundancy, save time, and introduce structured responses for improved workflow. Preliminary results show impressive feedback generation with a tendency towards lenient grading. The talk also explores the implications and potential benefits for university educators in the context of building a cohesive, automated assessment platform.