Interned on Games Research project, programmed chamber LARP solution in Jupyter. Sole programmer in initial stage, had creative freedom for game and algorithm design. Game scenario: Post-apocalyptic world. ChatGPT serves as a powerful AI supercomputer. Players craft roles within given context, communicate with ChatGPT to influence game world. No specific goal; focus on creating an engaging, adaptive narrative. Used GitHub for version control and branching (testing, new story/world context). To create the program, I learned prompt engineering basics and experimented with the OpenAI API. Jupyter offers UI tools, but they display within the notebook, revealing all code blocks. Explored libraries like ipywidgets and Solara to export widgets to browser tabs. I needed real-time function communication, which was achieved through Solara despite scarce and outdated documentation. Managed to make Solara work, styled the local webpage. Webpage interacts with Jupyter code, enabling ChatGPT interactions. OpenAI API lacks conversation history, so each user message requires sending back entire message history and ChatGPT responses. Careful wording and formatting of prompt string essential. Copied formatting in ChatGPT's responses needs stripping to maintain clarity. Working prototype established, project ongoing. Teamed up with a Ph.D. student and AAA studio games programmer for the next phase. Future Goals/Proposals: Develop standalone version, eliminating Jupyter, Solara, and ipywidgets dependencies. Plan to publish findings on the solutions. Considering stable diffusion for end-game image generation, showcasing world progression and player-supercomputer impact. Exploring recursive prompt and history sending to extend conversation with more randomness, bypassing the OpenAI token limit.
