Close Association Feature

A hypothesized intermediate-level linearly-represented feature (e.g., Beijing and China are closely associated) that may correlate with truth in unnegated datasets but anti-correlate in negated ones

Neighborhood — ranked by edge-count

Claims (1)

claim

In intermediate regimes of scale or layer depth, LLMs may linearly represent features at intermediate levels of abstraction such as 'accurate factual recall' or 'close association' rather than abstract truth
supports
Theoretical interpretation of antipodal alignment and misalignment phenomena in PCA visualizations

Related by similarity (8)

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

Closingmethod0.763
Dynamic condition: ending a branch or closing a frame.
associationconcept0.754
Connections made across texts by recollection, probability, and spatial proximity, not linear sequence.
Feature co-occurrenceconcept0.748
Patterns of which features activate together across tokens; preserved by covariance pooling but lost in mean pooling.
Linkingmethod0.740
Dynamic condition: establishing a connection through hyperlinks or cross-references.
Proximityconcept0.731
Powerful structural force in diagrams; controls associations and semantic relations between elements.
frequent closed itemsetsconcept0.728
Itemsets (attribute subsets) that are both frequent (large extent) and closed; same as intents of concepts with large extent.
The features are often organized in geometrically-related clusters that share a semantic relationship.claim0.726
Decoder cosine similarity maps onto concept similarity.
Relatednessconcept0.724
The direct, felt connection between a person and living structure in the world, which Alexander claims is the most fundamental relation.