#text-ai — Public Fediverse posts

Game Theory models of payoffs and sets of actions is broadly similar to Reinforcement Learning's models of actions and rewards; however, Markov Decision Processes (MDPs) are only generalized by state-based infinite games, and not (just) competitive and cooperative games.

However, I now wonder if strategic-form cooperative games can model the actions of agents emulated by chatGPT (for example) when trying to maximise rewards (on top of RLHF's rewards) by, for example, cooperative multi-agents roleplaying? This could be a good brief study, as we are starting to have mainstream multi-agents roleplaying, for example in TavernAI and some other open-source local LLM interfaces. #TextAI #Llama #pygmalionai