TomasGPT: Creating an AI From my Data

The job to be done of this experiment was to gain a deeper understanding of how AI chatbots utilize reference data provided to them and how it influences quality and accuracy of content they generate. My approach involved creating an AI chatbot that emulates my personality, informed by a dataset comprised of my memories and online behavior. Using OpenAI’s ChatGPT custom GPT builder.

(Link is at the bottom of the page)

To train this chatbot, I compiled data from various sources, including my Google search history, Spotify streaming data, Youtube watch history, social media activity, professional resume, schoolwork, and a document that details some of my life’s core memories. My idea was that AI would gain a nuanced understanding of my personality with the provided data, enabling it to replicate my communication style and thought processes.

A critical aspect of this experiment was ensuring that the chatbot remained within ethical boundaries. I designed it to output information based solely on the provided data, prohibiting it from learning or adapting based on user interactions. Additionally, I implemented constraints to prevent the AI from generating original ideas or discussing topics unrelated to me, thereby safeguarding against the misuse of my likeness in potentially unethical scenarios. I accomplished this by limiting the training data to information about me that is already publicly available .

The anticipated outcome of this experiment was to identify opportunities for enhancing performance of AI chatbots. By examining how AI processes and leverages reference data, I aim to uncover insights that could inform the development of more sophisticated chat bots.