This repository has the information and the code that I have used in my experimentation and writing my dissertation for my MSc in Advanced Computer Science with AI. Overall, my project, titled "Explanation and Game-Play Capabilities of Modern AI and Large Language Models (LLMs)", delves into the capabilities of advanced and most recently introduced LLMs. It examines their performance across diverse prompts, use-cases, and notably, in strategic gameplay, highlighting their proficiency in providing detailed explanations.
This documentation presents the development and exploration of a chatbot powered by OpenAI's GPT-3.5 (text-davinci-003 engine), specifically designed for strategic gameplay analysis and automation tasks. Throughout our study, we experimented with zero-shot, one-shot, and few-shot prompting techniques to analyse the explainable capabilities of the OpenAI's GPT-3.5 LLM. By harnessing these different approaches, the chatbot not only offers insightful strategic gameplay explanations but also showcases its prowess in automating various tasks and interactions.
Our project, meticulously developed within a Google Colab environment, showcases the versatile capabilities of the GPT-3.5 LLM in diverse scenarios and tasks.
Project.ipynb: This is our main notebook with all our project experimentations leveraging the capabilities of GPT-3.5 LLM.
-
🖋 Prompt-Based Interactions:
The code encompasses distinct use-cases for zero-shot, one-shot, and few-shot prompting, exhibiting the vast scope of GPT-3.5 LLM interactions. It also includes interactive features for the user to continue the conversation with their follow-up prompts or end the conversation.
-
🎮 Interactive Gameplay Demonstrations:
Engage with the LLM in strategic and logical games like Tic-Tac-Toe, Connect-Four, and Chess, observing its gameplay strategies and decision-making.
-
🧮 Automation Tasks:
Explore our mathematical automation tasks, purposefully designed to showcase GPT-3.5's capabilities in responding to the zero-shot prompts. Witness its capability to generate examples mirroring the provided zero-shot prompts, autonomously solve them, and subsequently validate its own solutions.
-
🧩 Modular Code Design:
Our code is structured for clarity, adaptability, and ease of understanding, ensuring users can follow along and modify as needed.
-
💡 Detailed Comments:
Navigate through our code effortlessly with the aid of comprehensive comments, providing context and clarity for each block of code.
Google Colab Notebook - Project
Prerequisites:
The notebook was developed in the Colab Jupyter Environment.
- Python (version 3.x recommended).
- Necessary libraries and dependencies as listed in the Colab Notebook.
- OpenAI Python client.
Installation:
git clone [gh repo clone ZerksisMistry/Project_GPT-3.5_LLM]
!pip install openai # for the use of OpenAI API Key
!pip install python-chess # for chess board representation
OpenAI API Setup:
import openai
openai.api_key = 'YOUR_OPENAI_API_KEY'
Replace YOUR_OPENAI_API_KEY with your OpenAI API Key.
Usage:
To engage with the AI chatbot and delve deep into its strategic gameplay insights:
- Open the Google Colab notebook.
- Execute the cells in sequence.
- Use your OpenAI API Key in the respective area to leverage the capabilities of the LLM.
- Experience the chatbot system powered by GPT-3.5 LLM for prompts and automation tasks.
- Experience the AI's strategic gameplay prowess in the interactive sections.
Tokenization with TikToken:
Leverage OpenAI's TikToken Tokenizer for efficient token counting within text strings.
Embark on a journey through our comprehensive exploration of various Large Language Models (LLMs). We've delved deep into models such as GPT-4, Google Bard, Claude-2, and LLaMA-2, emphasizing their capabilities with zero-shot prompts. A special focus has been laid on GPT-3.5, assessing its performance across zero-shot, one-shot, and few-shot prompts, as well as its application in automation tasks and interactive gameplay.
Research Objectives: Our core mission - assessing LLMs for gameplay and explanatory prowess.
Literature Summary:
- Historical Overview:
From the Turing Test's inception to today's advances, chart the journey of machine comprehension.
- Research Gap:
Our study targets an unexplored niche: LLMs' strategy and self-explanation in games.
- LLM Impact:
Recent literature underscores LLMs' transformative potential across varied sectors.
- Explainability in AI:
AI's explainability is emerging, but methods to assess LLMs need to be more comprehensive for better understanding of the capabilities of the LLMs.
- Towards AGI:
Lastly, discussions about Artificial General Intelligence, or AGI, are abundant but often speculative. The literature revealed a need for empirical evidence to support claims and theories about AGI’s potential and challenges.
Methodology:
Methodology encompasses diverse LLM evaluations, strategic gameplay tests, zero-shot maths task automations, in-depth GPT-3.5 prompt assessments, specific evaluation metrics, and insights into OpenAI's LLM framework.
Analysis:
In-depth evaluation of gameplay strategies and explanatory capabilities of LLMs.
Proposed OpenAI Framework:
Unveil our tailored framework for understanding OpenAI API interactions.
Conclusion & Recommendations:
Enlisted key findings and our propositions for the broader AI ecosystem.
This repository contains various documents and files related to our comprehensive research on Large Language Models (LLMs) and their applications. While the primary content, including the Project.ipynb, Project_Presentation, and other associated files, are distributed under the Open Software License 3.0 (OSL 3.0), it's essential to be aware of the specific licensing terms.
The detailed licensing terms governing this repository can be found in the LICENSE.txt file. We urge all users, contributors, and researchers accessing our work to thoroughly review the LICENSE.txt to ensure a clear understanding of the rights and limitations it imposes.
Your respect for the licensing terms is crucial in acknowledging the efforts and contributions made in this project.
Contributions are the backbone of open-source projects! Fork our repository, apply your insights, and propose a pull request. Every bit of constructive feedback and enhancement is treasured.