Top news
Investing in the public benefit OfficialAnthropic NewsOct 18Popular Musk Pledges SpaceX Shares to ‘Every Kid in America’ WebBloomberg TechnologyOct 19Popular Websites are closing their doors to AI agents WebThe Verge AIOct 17 This is the new ChatGPT Voice, powered by GPT Live OfficialOpenAI on YouTubeOct 18 How AI giants will reshape global philanthropy WebAxios AIOct 19 Gemini 4 Argon Really Surprised Me WebTwo Minute PapersOct 19 Google Gemini 4 agents just beat Minecraft with no human intervention WebEngadget AIOct 19 Who could it be? 🧐 OfficialGoogle on YouTubeOct 18 Anthropic gives outside AI watchdog “employee-like” access WebArs Technica AIOct 18 What are self-replicating ‘AI worms’, and why is Anthropic worried about them? WebThe Guardian AIOct 18
GlossaryModelsIntermediate

World model

Also written: World simulators, World-modeling, Implicit world models, World simulation, LLM world models

An AI that learns how its environment works to predict what will happen next.

In plain terms

A world model is an AI system that tries to understand and predict how the world around it works. Think of it like an AI learning the rules of a video game by playing it. It observes what happens when it takes certain actions and uses that information to build an internal understanding of cause and effect.

This internal understanding allows the AI to anticipate future events. For example, if an AI is controlling a character in a game, its world model might predict that pressing the jump button will make the character move upwards. This ability to predict is crucial for planning and making intelligent decisions.

The expert version

A world model is a component within an AI system, often a reinforcement learning agent, that learns a predictive model of the environment's dynamics. It aims to capture the transition function (how states change given actions) and potentially the reward function. This learned model can then be used for planning, simulating future outcomes, or improving the agent's policy without direct interaction with the real environment, thereby reducing sample complexity.

Tradeoffs involve the fidelity of the learned model versus computational cost. Inaccurate world models can lead to suboptimal or catastrophic planning. Current research explores various architectures, from generative models to state-space models, to improve the accuracy and efficiency of world model learning.

Why it matters now

Frontier AI labs are increasingly discussing world models as a core component for building more capable AI systems. Recent mentions highlight their use in scientific discovery, like Anthropic's Elicit platform, and in creating realistic training environments, as seen with Google Gemini's Project Genie. The concept is central to discussions about AI's ability to understand, predict, and interact with complex real-world scenarios.

Where it shows up 25 items, newest first
How each company uses it
Anthropic Claude
Anthropic's Elicit platform uses world models for scientific research, aiming for better transparency and reasoning.
Google Gemini
Google Gemini is developing world models for AI training, using tools like Project Genie and Street View to simulate real-world environments.
Microsoft Copilot
Microsoft's Quine research system, designed for biological complexity, likely leverages world modeling principles to computationally explore systems.
OpenAI ChatGPT
OpenAI's work, including discussions around GPT-6 Astra, touches upon the underlying mechanisms that could be interpreted as world models for advanced AI.
Related terms

Reinforcement learningPredictive modelingSimulationAgentState representationModel-based reinforcement learning

Often confused with
Generative model
A generative model creates new data similar to its training da
Simulator
A simulator is a pre-programmed system that mimics real-world process
Policy
A policy dictates an AI's actions in a given state, while a world model
Origin

The concept of world models in AI has roots in cognitive science and early AI research, particularly in model-based reinforcement learning. The term gained prominence with research exploring how agents can learn internal representations of their environment to improve decision-making. Source ↗

Try this about 5 minutes

Open Google Gemini's Project Genie and try creating a simple interactive world. Describe a scene with text and images, then interact with it to see how the AI predicts and responds to your actions. This demonstrates the AI's attempt to build a basic understanding of the world you've defined.

Try it: Playbook entries that use thisPlaybook →

The Playbook: reusable how-tos, prompts, configs and tips pulled out of the news.

Questions people ask

How does a world model help an AI?
A world model helps an AI by allowing it to predict what will happen next in its environment. This prediction ability is crucial for planning actions and making better decisions without having to constantly experiment in the real world.
Is a world model the same as an AI's memory?
While related, a world model is more about understanding the rules and dynamics of the environment to predict future states. Memory typically refers to storing past experiences or observations.

First seen on Subvolts Sep 30, 2025. Updated Oct 12, 2026. Written by an AI model from the items above and reviewed against them weekly; report an error.