I am a board game designer and my design partner and I have been playing with GPT-3 to see how we might bring deeper interactivity to the universe we have created with the game and the short film which accompanies it.
The game sees 6 Tech Billionaires vying to āsave the world, no matter the costā and is a deeply satirical and fun ātake thatā style game where you try and achieve your end goal of creating a ākiller appā and satisfying your overall ambition. If youāre interested and want a bit more context you can see the game here.
What we are currently building is a game-like experience which people visiting our exhibition space at the Brighton Digital Festival will get to play (weāre based in London, UK.)
The prompt for each of these tech billionaires are in three evolving parts, one where the player is trying to extract information from a given CEO, followed by one where the CEO becomes suspicious of the player, and one where the CEO discovers the treachery of the player and a chase ensues.
We are pretty happy with where we are with development and are seeing wonderfully consistent outputs, and I am posting here in the hopes that you may know of other game-like projects/experiments which we might learn from.
Iāve been involved in a couple of chat/character driven games. I also wrote about the possibility of using cognitive architectures to create narratives (probably a little advanced for your use-case). In short, you need to craft the prompts such that they include a brief character synopsis as well as the current state of the game, and then just ask GPT-3 what the character would do.
Thatās effectively what we have done (character synopsis, motivations, antagonists, and current state) and itās working wonderfully for our limited use case.
You raise an interesting notion though - if we were to train an instance on our gameās rule set as a new semantic domain we might be able to get interesting results doing what you suggest: describing the state of the board and asking what the character would do next. Thanks!
Your Core Objective Functions, btw, are extremely sound, and align well with the humanist position adopted by many modern cross-disciplinary philosophers (Sam Harris comes to mind.)
Thanks! Youāre a very fast reader! LOL. Also thanks for the name, Iāll have to look into Sam Harris. Looks like heās got a few books and a podcast.
I have indeed, and follow along with their medium updates. Itās a wonderful exploration of open world creation, certainly far beyond what weāre building. I guess Im looking for examples of live game-like experiences which are time boxed and perhaps simple.
