Yesterday I made an experiment with connecting several (chatGPT-driven) agents: 2 players and a game-runner and put them in somewhat difficult situations and have the agents act autonomously. The interesting thing was that it got really boring real quick. The players began to collaborate and being very nice and courteous with each other, which killed any tension. 
Wondering if this would be different with an uncensored LLM. https://gw.ipfs-lens.dev/ipfs/Qmab2ifgaTXLeKZNL6zJvFMgm9gJuzaUfMzbGP8UQUMcMe