Angehefteter Tweet
AI in games. Yuck.
Or at least that's what most gamers currently think. How can we change that ?
Players first.
We started with a simple idea: great LLMs should be able to automate and replace a lot of game logic. We built a prototype, and realized the technology wasn't there yet. There were still too many variables and logic the LLM couldn't handle.
We then took a different approach: build the skeleton of the game logic and let the LLM fill out the rest with variability. This ensures the game actually works as intended, but still keeps the magic of full customization for each character, storyline, and world building. This is the version we're launching for our early alpha testing!
This is very similar to somewhat standard game development: write some game logic, and assign that logic to more than one game asset (i.e. the long bow mechanic is assigned to Rare Long Bow, Elven Bow, etc)
But where's the innovation?
Our game logic for each item or ability is compartmentalized, annotated and marked with very specific instructions on how the logic works, and how it relates to the game engine. We can then go back to our first idea, armed with all that knowledge, and get much better results out of our LLM calls : we get both the what (consequence of an action), and the how ( code to execute that consequence).
This makes the players an integral part of the game creation and introduces fully emergent gameplay options. They now have the agency to make decisions they weren't able to make in a world without AI.
All while keeping us in control of balance and overall direction.
English
