LLMs nowadays are headed for superintelligence, they say. They should make good tabletop game partners then, I guess?
In this repo you can find a simple implementation of a Guess Who game. If you’re not familiar with the rules, here they are: each player receives a character and has to guess the opponent’s character name by asking one yes/no question at a time. For example: “Is your character blonde?” or “Does your character have glasses?”. Whoever guesses the opponent’s character first wins.
A very simple game that should be within the understanding of any multimodal language model, right? Try for yourself to find out.
You can also find the game on GitHub. Contributions of all kinds are welcome 🙇