claim
active
claim:features-are-better-viewed-as-low-dimensional-manifolds-evolving-over-time-rather-than-isolated-directionsFeatures are better viewed as low-dimensional manifolds evolving over time rather than isolated directions.
Future-work perspective.
Source paper
extracted_from(2025) · Lubana, Ekdeep Singh · Rager, Can · Hindupur, Sai Sumedh R. · Costa, Valerie +12
Neighborhood — ranked by edge-count
Papers (1)
paper
Frameworks (1)
framework
- Linear Representation HypothesiscontradictsThe hypothesis that models internalize concepts as approximately linear directions in representation space; used to interpret MDS injection behavior
Claims (2)
claim
- Proposition 4.1 central claim.
- Motivates structured protocols.
Quotes (1)
quote
- Core future-work framing.
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- Implication for interpretability.
- Features may not be strictly one-dimensional objects; higher-dimensional feature manifolds may exist in model representationshypothesis0.826Extension of superposition hypothesis to account for continuous families of features
- Proposes that nonlinear geometric structure is superior to linear feature spaces for capturing semantic content.
- Mere correlation of feature directions, as used in prior work, need not suffice to find manifoldsclaim0.800Contrast with Engels et al.
- General principle supported tangentially by covariance pooling work; relates to feature co-occurrence structure.
- Load-bearing interpretive claim about the layer-specificity of Burger et al.'s finding.
- Discrete features and feature manifolds are dual descriptions of the same underlying objectclaim0.793Duality.
- Core concern.