Web World Models

arXiv 25.12 2025 53.8 method, system

TLDR

Introduces Web World Models, a hybrid approach using web code for deterministic world state and LLMs for generative narratives, enabling controllable open-ended environments.

Reasoning

Strengths include a novel middle-ground architecture combining code-defined rules with LLM-driven generation, and practical design principles derived from multiple implemented worlds. Weaknesses are that the abstract lacks quantitative results or comparisons to baselines, and the scope is limited to web-based environments.

Read-first score

Read-first score 53.8, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 38.

Recency 8%

86.7

Uses a gentle age decay so recent papers surface without erasing older foundations. 2025

Topical relevance 42%

54.3

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Methodology quality 25%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=result

Reproducibility 25%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=code,github

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 333.

Keyword Scores

world model

interactive world model

world simulator

generative world model

model-based reinforcement learning world model

world dynamics prediction

video world model

Deep Analysis

Innovations

Middle ground between conventional web frameworks and fully generative world models, combining web code for logical consistency with LLMs for generative context, narratives, and decisions.
Suite of Web World Models on a realistic web stack covering diverse domains: infinite travel atlas, fictional galaxy explorers, web-scale encyclopedic and narrative worlds, simulation- and game-like environments.
Design principles: separating code-defined rules from model-driven imagination, representing latent state as typed web interfaces, and utilizing deterministic generation for unlimited but structured exploration.

Methodology

The authors built a suite of Web World Models (WWMs) on a realistic web stack, implementing world state and physics in ordinary web code to ensure logical consistency, while using large language models to generate context, narratives, and high-level decisions on top of this structured latent state. They identified practical design principles from these systems, including separation of code-defined rules from model-driven imagination, representation of latent state as typed web interfaces, and use of deterministic generation for structured exploration.

Key Results

The results suggest that web stacks themselves can serve as a scalable substrate for world models, enabling controllable yet open-ended environments.

Resources

Paper