Entropy Warrior #
Select Output #
"Elara, what spice evoked such a profound connection?"
The Spice was cardamom, Ignatius, its melody rich with quirky undertones"
Game explanation #
Entropy warrior is a game name which helps to illustrate that one of the core features of GPTs is that you can modify the temperature. When creating a game that allows for tweaking different variables and allowing the players to observe the outcome of the game scenario, it is interesting to consider that modifying entropy between rounds of dialogue could have large impacts on the content and outcome of the game.
The player can select a “player” persona, or player 1 and a “match” persona, or player 2, both of which are constructs that are then used to create ai generated speech. The player can influence the dialogue and the outcome of the game as the game proceeds.
Consider, that you can describe yourself, and describe a close friend or partner, and you can select a setting, you can create your own game, and you can watch the dialogue and situation unfold as the two AI charactes progress through the story.
At this phase the current functionality exists
- select player 1 character persona
- select player 2 character persona
- select game option
- be able to continue an existing game round from a previous save
- player and match have dialogue that outputs to the display simultanously
- entropy level function controls the level of randomness in the AI output and is displayed on screen
- entropy level can be modified by the game player
In the first stage of development, I chose to use a dating game as a structure for the player 1 and player 2, originally concieved as “player” and “match”. The idea is to have an example game structure which gives the two characters a dialogue sequence to go throuh.
Experimenting with the descriptions of the characters and the prompts structure and prompts pipeline led me to giving the characters attributes and being able to add additional game modes.
The game modes current are:
- dating
- survival
- escape
The core concept is to have a character prompt pipelines to GPT4 that will understand the context of the game mode and still be able to have a coherent series of dialogue interactions that can end in a somewhat predictable or controlable sequence.
Creating logic that will evaluate end the game in a conclusive way is still in development.
However I have experimented with this first dating simulation concept and have been able to successfully implement different character attributes, coherent, dialogue progression, the addition of a seperate location concept, and keeping state management to Postgres to handle data storage. The implementation of keeping the conversation data and the character and the game and location data is currently only used for state management. However in future implementation this could allow for more complex game interactions and designs as this project evolves.
Here is sample output from a conversation between Ignacio and Elera