claim

active

claim:the-features-are-often-organized-in-geometrically-related-clusters-that-share-a-semantic-relationship

The features are often organized in geometrically-related clusters that share a semantic relationship.

Decoder cosine similarity maps onto concept similarity.

Source paper

extracted_from

Scaling monosemanticity: Ex-tracting interpretable features from claude 3 sonnet

Neighborhood — ranked by edge-count

Findings (1)

finding

Golden Gate Bridge feature neighborhood includes Alcatraz, Presidio, Lake Tahoe, Yosemite; decoder cosine similarity maps onto semantic relatedness.
supports
Example of geometric clustering of features.

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.

Properties in each cluster appear to be similar or interrelatedclaim0.831
Interpretive claim that the statistically derived clusters reflect conceptual similarity or interdependence among the properties.
Correspondence analysis reveals four clusters of related properties: Cluster 1 (Contrast, Not-Separateness, Roughness, Alternating Repetition, Good Shape), Cluster 2 (Local Symmetries, The Void, Levels of Scale, Good Shape, Positive Space), Cluster 3 (Boundaries, Strong Centers, Deep Interlock and Ambiguity), Cluster 4 (Simplicity and Inner Calm, Echoes, Gradients, Positive Space).finding0.820
Statistical grouping of properties based on dependency patterns, enabling deeper understanding of their coherence and interaction.
There is a bidirectional relationship between the geometry of representation and behavior across tasks and modalities.claim0.811
Author’s interpretive claim that the shared geometry is general and robust.
Correspondence analysis reveals four major clusters of fifteen properties with overlapping membership.finding0.809
What is the relationship between different dimensions or clusters of dimensions in LLM representations? Do they and/or how do they interact with each other?question0.809
Motivates the RN hypothesis by pointing to the unknown relational structure within high-dimensional representation vectors.
Geometry of features matters for representation quality.claim0.804
General principle supported tangentially by covariance pooling work; relates to feature co-occurrence structure.
Features are connected by weights forming circuits, and these circuits can be rigorously studied and understood as meaningful algorithms.claim0.802
Second of three speculative claims asserting that subgraphs of neural networks are tractable and meaningful objects of study
There appears to be a systematic relationship between the frequency of concepts and the dictionary size needed to resolve features for them.claim0.801
Feature presence depends on concept frequency in training data, with a threshold scaling inversely with alive features.