What ultimately gets displayed to the GM will be one or multiple of many suggestion categories. Develop a prompt that best fits the current session context into a pre-defined suggestion category (rule, encounter, shop, scenery detail).
R&D a method to optimize the llm to accurately detect when an encounter has occurred in a transcript stream.
One idea, take 'right before encounter' sections from 100 or n recorded live play sessions and across all n sessions set it up such that they are replied to simultaneously for quick sessions and massage the pre-prompt until the you get the highest count of high encounter. Then take the same amount of different samples of live that are different categories and ensure it does not return a high encounter rate.
keep in mind that encounter tables should be produced in sections where the players a navigating some dangerous part of the world, it's about encountering some random creature in that biome.
What ultimately gets displayed to the GM will be one or multiple of many suggestion categories. Develop a prompt that best fits the current session context into a pre-defined suggestion category (rule, encounter, shop, scenery detail).
R&D a method to optimize the llm to accurately detect when an encounter has occurred in a transcript stream.
One idea, take 'right before encounter' sections from 100 or n recorded live play sessions and across all n sessions set it up such that they are replied to simultaneously for quick sessions and massage the pre-prompt until the you get the highest count of high encounter. Then take the same amount of different samples of live that are different categories and ensure it does not return a high encounter rate.
keep in mind that encounter tables should be produced in sections where the players a navigating some dangerous part of the world, it's about encountering some random creature in that biome.