The Library of Worlds

What will media be like in a world with generative language models, speech models, and image models?

I realized that world already exists. Language models can generate a script, speech models can read it aloud, and image models can illustrate (increasingly consistent) accompanying imagery. These ingredients afford a new kind of dynamic storytelling, a sort of physics simulator for narrative.

If the story is playing out in a simulator, you can intervene on the simulation. This led me to try "intervention cues": sometimes the player is asked to speak on behalf of the traveler character. The story generator then responds to the intervention, simulating the world accordingly.