Advancing Reasoning Abilities in AI through Local Open Source Methods

Presenter

Peter David Thornton

Campus

Salem State University

Sponsor

Komalpreet Kaur, Department of Computer Science, Salem State University

Schedule

Session 2, 11:30 AM - 12:15 PM [Schedule by Time][Poster Grid for Time/Location]

Location

Poster Board A25, Campus Center Auditorium, Row 2 (A21-A40) [Poster Location Map]

Abstract

This research explores the capabilities of open-source Large Language Models (LLMs) and their performance relative to leading proprietary models like OpenAI's GPT 3.5 and GPT 4. 

 Through a comprehensive investigation, this project aims to enhance the reasoning abilities of LLMs by integrating traditional machine learning fine-tuning techniques alongside novel approaches including prompt engineering, retrieval augmented generation, and chain of thought prompting.  

The study also focuses on the application of these models in educational settings, specifically to develop a chatbot capable of assisting students from a K-12 level with mathematics, including problem-solving and explanation of solutions.  

By employing open-source models and tools, this project seeks to democratize access to advanced AI technologies, providing an alternative to the major tech corporations for individual users, researchers, and hobbyists.  

The methodology includes training open-source LLMs on a mixture of high-quality open-source mathematics datasets, synthetically generated data from state-of-the-art LLMs, followed by a benchmarking process comparing the trained models against each other, and against state-of-the-art commercial counterparts against an evaluation dataset. 

This research underscores the potential of LLM fine-tuning and prompt engineering techniques to significantly enhance the logical and reasoning capabilities of LLMs, enabling them to perform complex tasks more efficiently with fewer parameters, on local hardware.  

The project hopes to contribute to the broader discourse on open-source AI accessibility and capabilities, ethical use of technology, and the advancement of AI in educational and reasoning contexts. 

Keywords

Artificial Intelligence, large language model, machine learning, mathematics, AI assisted education

Research Area

Computer Science

SIMILAR ABSTRACTS (BY KEYWORD)