It would be really interesting to use different language models and compare the results regarding player's perception of intelligence.
Another interesting issue to tackle would be the delay between answering the agent and the game and model processing the response.
Thank you for the feedback!