method
active
method:sae-reconstruction-loss-as-superposition-proxySAE reconstruction loss as superposition proxy
Proxy since superposition cannot be directly measured in real models.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Adversarial Examples Are Not Bugs, They Are Superpositionintroducesmentions
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.
- Robustness influences superposition, predicting lower SAE loss in more robust models.hypothesis0.787Popperian credit for surprising prediction.
- Authors' hedged interpretation.
- Lower SAE loss at fixed sparsity likely indicates less interference and less superposition.claim0.765Justifies SAE proxy.
- Source of mismatch.
- Quantifies SAE reconstruction damage.
- No features correspond to Oxford study.
- Open question in limitations.
- Justifies focusing on direction-based SAEs.