Acknowledgments
This project leverages a Large Language Model (LLM) known as "im-a-good-gpt2-chatbot." While I did not write any of the code myself, I was responsible for training the model and converting the Python code to Jupyter Notebook format. It has been an exciting journey to witness the advancements in LLMs as of May 2024.
Motivation
As a first-year university student with no prior experience in machine learning, I decided to participate in this challenge to explore the field further. I had not selected machine learning as part of my coursework, making this an invaluable opportunity to see if it aligns with my interests.
I joined the challenge after receiving an email from "Informatics Peer Assisted Learning," which mentioned the chance to earn a Limited Edition Signed Ian Mackie Certificate. The thought of obtaining such a unique recognition was incredibly motivating!
Discovering "im-a-good-gpt2-chatbot"
In May 2024, I learned about "im-a-good-gpt2-chatbot," a powerful yet little-known LLM that is available for free at chat.lmsys.org. Speculation suggests it may be linked to OpenAI, but no confirmation exists as of now (09/05/2024).
Developers have indicated that this LLM surpasses the capabilities of OpenAI's GPT-4-Turbo-0409 and other predecessors. Given the opportunity to experiment with such a tool, I decided to explore its potential. My analysis of AI's impact on machine learning jobs suggested that it may play a significant role in the future.
Project Summary
I dedicated approximately 5 hours to this project, and I am curious if my solution achieved one of the highest accuracy rates. Although I occasionally feel like I might be "cheating" by using an LLM, there were no restrictions on the GitHub page regarding its use. Since this project is not part of my coursework, I feel confident moving forward.
This experience has been both enjoyable and educational, enhancing my understanding of machine learning despite my lack of formal study in the area. Ultimately, I am thrilled about the opportunity to earn the Limited Edition Signed Ian Mackie Certificate!
Results
I regret to mention that I utilized an LLM (which some may consider "cheating") to attain a 99.89% accuracy rate. I plan to upload a screenshot demonstrating how I prompted the LLM to achieve this outcome.
Reflection
According to the guidelines on the GitHub README page, I believe "Creative solutions" could include the use of LLMs, although I'm not entirely certain. Nevertheless, it was an enjoyable experience, especially since "any computer science approach is valid." Perhaps I can even say I used machine learning to facilitate my machine learning efforts!
I find it amusing and exciting that, despite having never formally studied machine learning, I was able to achieve a high accuracy rate in this challenge. It remains a significant hurdle for many machine learning students, yet I managed to succeed (albeit with some assistance).