World model

As explained in How Large Language Models Work

A computational representation of a world’s entities, states, and dynamics that supports reasoning about outcomes without directly acting in that world.

A world model represents entities, states, and dynamics so a system can reason about possible outcomes without directly acting in the world. The term remains unsettled, so define it before comparing systems. section 7.3

A genuine world model could compute about an answer before generating it. An LLM instead reflects patterns in training data and may need intermediate text to reach a useful response. section 7.3

Use the distinction to interpret apparent reasoning carefully:

• Chain-of-thought can improve results, but it may reflect extra transformer computation or alignment with verbose training examples rather than a world model. section 7.3
• Neither step-by-step output nor fluent language guarantees complete calculations or logically valid reasoning. section 7.3

Research example

A 2018 project by David Ha and Jürgen Schmidhuber used a technical world-model approach and reported major performance improvements over methods available then. section 7.3

Figure 7.6 The context and reason why someone is wearing or doing something unusual may be in the realm of something that an LLM properly recognizes and for which it produces an appropriate response. However, it might not be possible for an LLM to reach that appropriate response without producing some intermediate text. For a math problem, this intermediate text could be useful, but the intermediate text may not always be appropriate or desirable for a user to see.

Shows why an LLM may recognize an appropriate answer yet require visible intermediate text to reach it. section 7.3

How How Large Language Models Work covers it

How deep it goes

Defines the concept, contrasts it with LLM behavior, and stops short of specifying a settled technical definition.

Examples

Uses chain-of-thought, mathematics, unusual behavior, and a 2018 world-model research project as examples.

What you can do after

Evaluate claims about LLM reasoning without treating fluent or stepwise output as proof of factual or logical validity.

How it connects

Connects world models to LLMs, chain-of-thought, transformers, attention mechanisms, and machine learning research.

liveBookmanning.com

Read How Large Language Models Work on liveBook Get the book at manning.com

Also explained in

• 

  liveBookmanning.com
  Sutskever's ListA world model is an internal representation of how entities, events, and causal relationships in an environment behave.
• 

  liveBookmanning.com
  The AI Pocket BookAn internal representation an AI system builds about how the world works, including object properties and common situations, used to generalize beyond training examples.

Compare how 3 books explain world model →

Related concepts

• language modelcontrasts with · A language model is a computational model that represents patterns in linguistic tokens to estimate sequence probabilities and predict or generate language from context.
• internal representationhas part · An internal representation is an information-bearing encoding or state within a system that represents aspects of its input, environment, or task.
• stateuses · State is information or a configuration describing the condition of a system, entity, or computation at a particular time and potentially influencing subsequent behavior or decisions.
• foundational modelkind of · A base or foundation model is a pretrained model that provides learned parameters or capabilities as a starting point for reuse or adaptation to downstream tasks.
• compressionuses · Compression is the transformation of information into a more compact representation to reduce storage or transmission requirements while retaining information needed for recovery or use.

This explainer was generated by Manning's knowledge graph from the text of How Large Language Models Work. The section links point to the parts of the book it draws on.