Encodes structural arrangement — rows, columns, grids, hierarchies.
What It Does
Space.Layout neurons activate on structural arrangement descriptions: grids and tables ('arranged in rows and columns', 'a 3x3 matrix'), hierarchical structures ('the nested folders', 'the org chart'), floor plans and schematics ('the kitchen is adjacent to the living room'), and abstract organizational structure ('the first tier is...', 'the top level contains...'). They encode how elements are spatially organized relative to each other.
How It Behaves
Layout neurons are distributed across all layers and are particularly active in technical documentation, user interface descriptions, and data structure discussions. They show the broadest distribution of any Space sub-type across different domains. Layout neurons interact strongly with Relation.PartWhole neurons (parts within a structural whole) and Number neurons (dimensions and counts of layout elements).
Research Example
In GPT-2 Small, Space.Layout neurons activate on 'the periodic table organizes elements in rows and columns' and 'the company's management structure has three layers' — both describe spatial-structural organization despite operating in completely different domains (chemistry vs. corporate structure). The Layout neuron encodes the structural pattern, not the domain content.