Reading list
Ask what a world model actually outputs
A reading guide for distinguishing rendered observations, structured state and predictions before comparing systems under the same label.
THE SHORT ANSWER
The output contract determines whether you should inspect pixels, structure, dynamics or predictions.
Read one proposed taxonomy
Google DeepMind’s archived Genie 3 post describes interactive environments from a text prompt. World Labs’ separate taxonomy distinguishes renderers, simulators and predictors by function, using state and observation as different concepts. This is an organization’s explanatory framework, not a universally settled definition. Interactivity and visual plausibility alone do not demonstrate every kind of physical or structural fidelity.
Write the consumer of the output
Ask who or what uses the result. A viewer may need coherent frames. A geometry tool may need inspectable structure. A planning component may need predictions that can be checked against later observations. Write one acceptance condition for the actual consumer before comparing models. For example, returning to a previously seen location poses a different question from exporting a consistent scene representation.
Follow a focused reading path
Begin with the taxonomy’s state-versus-observation distinction, then inspect its renderer and simulator sections. Return to the original announcement and list which required outputs are demonstrated, described or absent. Keep unanswered items open rather than filling them with the broad world-model label. No interactive environment, video or model was tested here; this resource organizes the technical questions a future evaluation would need to answer.
Put it to work.
- Name the output consumer.
- Choose a matching acceptance condition.
- Separate description and demonstration.
- Leave missing outputs unverified.
Go to the source.
Original material, with its own scope and publication date. This note is an explanation, not a CooScope performance test.
Source published · Editorial review
Archived English post text; primary sources reviewed separately. Generated annotations, media and missing replies are outside the evidence scope.
- @GoogleDeepMind — archived X post (2025-08-05)
- World Labs — A functional taxonomy of world modelsLink / source review: 2026-10-15