Comment by yencabulator
2 years ago
Well there's a thought. A zorklike where the game content is whatever generative ML hallucinates (instead of the built-in fixed maps & interactions) -- as long as a second ML system agrees that the answer follows some more general rules.
For example: Rules say "In the beginning, the Enemy has a diamond. User cannot get the diamond from the Enemy if the Enemy is still alive. The Enemy is a fierce opponent and hard to kill." but nothing about the details of the enemy, shape of the map, or the available tools. Re-generate each response until it succeeds the verification.
Let the adventure be randomized by the hallucinations, while keeping some basic challenges in place.
An acid-tripping D&D dungeon master coming up with plot twists, combined with a rulebook-reading lawyer. Bonus points for adding generated "cut scene" visuals every now and then.
With the new function calling feature you may not need the second system. Only present options to ChatGPT that are valid. Feed it updated state information as JSON. Have it describe and elaborate on what the game engine is doing, or use functions to invoke entity creation that can then be tracked by the engine.
So for example the engine can do combat rolls and the LLM can give each a unique description of the type of attack and defense. Each monster or treasure can get its own unique description generated by the LLM that matches the stats given by the LLM.
Yes, but then I fear you're back to having limited "things that can happen", with predefined entities and so on. I'd prefer the acid trip to break more paradigms, tell a story, while the lawyer makes sure there remain challenges.
For example: with strict entities "behind an API", the diamond is the singular diamond and is a diamond. With an ML-based lawyer, well, maybe you can duplicate the diamond? Maybe you can transmogrify it temporarily into a non-diamond, which the Enemy drops as undesirable? Maybe you can wander into an elaborate system of mines full of dwarves who actually know how to mine a diamond, as long as you help them with this pesky dragon... No human has to come up with all these possibilities.
Good point. You could also have the system create the entities on-the-fly if necessary by calling a function. But having them there in the prompt as a structure it's supposed to adhere to some degree makes it more consistent and would give it tools such as for dice rolls or a precise inventory and game state database etc.
ChatGPT already does really good adventure games.
"Let's play an adventure game, you be the DM. I want it set on a spaceship arriving at a planet after 10,000 year journey. It should have a sense of mystery and a slight sense of foreboding and dread. It must have at least 20 locations. The objective of the game is to find 10 colonists in the ship and get them safely to the surface of the planet. Make it play in the style of an Infocomm adventure. Don't tell me all the locations in advance, make discovery part of the adventure."
As a form of story telling, yes.
As a challenge, not really. You can just convince it to let you win. (Said differently: the meta-game is too easy.)
You need the second layer of output validation[1] to re-add the challenge of solving a puzzle.
[1] or some such mechanism; more rigorous system vs user input separation could also work
True, but remarkably fun stories.
As an aside, there should be an AI encabulator.
Nethack procedurally generates a unique dungeon with constraints every time you start a new game and has since 1987.
Randomized according to fixed rules. Now imagine not needing to write those rules / not being bound by them. Consider generative ML coming up with whole new categories of monsters. Consider a Nethack variant that was never told to include a candelabrum or Amulet of Yendor.
Sidenote but AI bot companion for D&D session going "you can't do that in rules" would be funny addition.
It would be interesting experiment to use it to work as NPC characters in one too.