Multiverse
Multi-agent world model approach concatenating agent states into split-screen representation
Multiverse is an early approach to multi-agent world models that concatenates agent states into a single split-screen representation, effectively treating multiple players as one unified world state.
In this approach, each agent's observed state is concatenated alongside others in a unified representation. The world model processes all concatenated states simultaneously and generates output for each agent, effectively running multiple views through a single model but with separation baked into the input/output structure.
This approach has a fundamental limitation: it does not scale efficiently with the number of agents, and both Multiverse and similar approaches struggle to robustly maintain consistency when players lose sight of one another. The split-screen representation does not preserve true shared state — each view is generated from the concatenated states rather than from a single authoritative source.
Multiverse represents an important stepping stone but has been superseded by architectures that maintain explicit shared world state.
Research this in Signals
Signals turns a topic into a sourced research record you can inspect and rerun. Your first scan is free, and this one starts with Multiverse already loaded, so edit it or scan as is.