This was really fun! What a cool way to integrate ChatGPT for the interrogation dialogue. I tried to reason with him and offer him a battle 1 vs 1 and if I win he sets me free... he wouldn't bite! Im curious how you determined which phrases or responses would make him take me in? Especially if I was using custom ones.
Viewing post in Smack to the Future: Clone Wars Edition jam comments
Thanks for the comment kokutouchichi and thanks for trying out our game. So one of the places LLM based AI agents really shine is in decision making. So there are no hardcoded rules about what works and what doesn't. Instead the prompt is taking in a threatscore from the game state based on prior dialogue with the police and the amount of chaos in the brawler section. After each piece of dialogue the Agent can decide to adjust the threatscore up or down. Reaching a certain threshold gets you arrested.
This is a very basic implementation of https://react-lm.github.io/ . You can build way more complex decision making loops using this pattern and it is one of the things that @drusepth and I are most excited about incorporating into our games!
More good reading : https://arxiv.org/abs/2304.03442