I've been working on ChatVTT, where I built a React App that chatGPT uses via an API action in a custom GPT to build maps. It can also use the map as a reference to not "forget" the range context, which I found to be a recurring issue. Goblins would make a melee attack from 20 feet away with a dagger, turns out LLM are terrible at theater of the mind.
Now the issue of story progression/tracking. I have been making lists, reference tables, etc., but this seems to take a lot of context for the LLM which can cause the AI to loose track of what it was trying to do. I have seen the new wiki methodology, is this a good way to track player context, like relationships with NPC/Enemies, or say there history with the local town guard (weather or not they are wanted in the area)?
Your project, ChatVTT, sounds like a fascinating application of AI in the realm of tabletop gaming, especially for managing complex scenarios and character interactions. The idea of using a map as a persistent reference to maintain spatial context is clever and addresses a common pain point when relying solely on text-based interactions. Regarding story progression and tracking, while the wiki methodology can indeed be a powerful tool for maintaining a detailed and easily accessible record of player actions, NPC relationships, and world history, it might still require significant manual input and maintenance. Have you considered integrating more automated ways to update this information, perhaps through natural language processing (NLP) techniques that could parse key details from the ongoing game sessions? This could potentially reduce the burden on the players or the game master and keep the narrative flowing smoothly. #AI