concept
active
concept:feature-splitting-and-absorptionFeature splitting and absorption
Problems suggesting sparsity breaks down as interpretability proxy.
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- SDL assumption may break down.
Concepts (2)
concept
- Feature splittingrelated_toPhenomenon where a feature in a small SAE splits into multiple finer features in a larger SAE.
- Feature Absorptionrelated_toPhenomenon where sparsity yields gerrymandered latents for hierarchical concepts.
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.
- Authors argue the absence of a fixed feature count is a property of the superposition geometry, not a failure of the method
- Observed across SAE scales, e.g., 'San Francisco' split into 11 features.
- Adapted absorption metric detecting partial and shared absorption across layers.
- Open question about feature cluster.
- Synthesizing inputs maximizing component activation.
- Functionally significant directions in the residual stream that organize activations in mechanistic interpretability
- Features with similar decoder magnitude across both models.
- Explains lower off-target effects of SAE methods.