Assignment 3 – Chatbot OpenAI

Link to the chatbot

https://platform.openai.com/playground/p/nLg9mEfBf41xOlr5uhZh9cbH?model=ft:gpt-3.5-turbo-1106:personal::8MrG6V1H&mode=chat

Link to 30 questions

https://docs.google.com/document/d/1jKmYRuDghv8CrLqVXD7XIwb_b7IAtxXPjOHKk-VzPn0/edit

My key takeaway from this assignment

            Creating fine-tuning data, it’s hard for me since I have to think about how I usually answer questions from others. First, I have to customize the prompt by not using the exact prompt from the homework last week since I want it to be more specific and as precise as possible. After getting the prompt as a role for my chatbot, I decided to list out questions and answers for my chatbot. This might be the hardest part since first I just list questions that I believe are usually asked when it comes to flight recommendation. So when I used those types of questions, my answer was not what I expected since I typed the answer without adding my style to it. Therefore, I decided to list the first 15 questions that I could answer by using my style and it turned out well. After that, I continued to list out the next 15 questions and upload them as fine-tuning for my chatbot.

            By using the process above, I typed question & answer and made sure that they all sounded like me the most. So after uploading the data for fine-tuning, it turned out to be very similar to my style of answer. I was impressed by the answer of the chatbot after fine-tuning.

For the process of data and training, I tried to adjust my data around 2-3 times until I was satisfied with my question and answer. The training took a lot of time too since I needed to find the right temperature for my chatbot. I decided to choose 1.1 since it presented the most likable style of answer that match my style. It wasn’t a repeat answer and provided useful information.

            However, there were some aspects that I hope it could answer to be more like my style. But I understand that there is some restriction on deciding which airline is better than others since it has to be compared by using many factors. But overall, I am satisfied with the result since the chatbot sounds like me than before using fine-tuning.

            This assignment opened a new world for me since I never tried this kind of work before and far from what I have been done in my entire life. I learned to create my own chatbot and even customized it to sound like me. I hope I can learn more about this in more detail after this class. I enjoy learning a lot of new things that I never tried before in this class!