claim
active
claim:lower-off-target-effects-of-sae-methods-are-explained-by-feature-splittingLower off-target effects of SAE methods are explained by feature splitting
Hypothesized explanation.
Source paper
extracted_from(2026) · Leon Bergen · Usha Bhalla · Sidharth Baskaran · Max Loeffler +13
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (1)
finding
Hypotheses (1)
hypothesis
- If bold is split into context-specific features, performance is capped unless all such features are aggregatedassociated_withFeature-splitting hypothesis.
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.
- Observed across SAE scales, e.g., 'San Francisco' split into 11 features.
- Explains lower off-target effects of SAE methods.
- Lower SAE loss at fixed sparsity likely indicates less interference and less superposition.claim0.787Justifies SAE proxy.
- Core critique of sparse autoencoders: they break the geometric structure of representations, making it harder to see the big picture.
- A promising property for interpretability analysis off-distribution.
- Out-of-distribution generalization of SAE features.
- Backprop of logit-minus-mean-logit loss to surface causal intervention candidates, summed across token positions.
- Robustness influences superposition, predicting lower SAE loss in more robust models.hypothesis0.760Popperian credit for surprising prediction.