claim
active
claim:reshaping-the-training-signal-can-compel-sae-features-toward-desired-characteristics-and-extends-to-other-activation-analysesReshaping the training signal can compel SAE features toward desired characteristics and extends to other activation analyses
Conclusion.
Source paper
extracted_from(2026) · Kevin Der · Harish Kamath · Thompson, Ben
Neighborhood — ranked by edge-count
Papers (1)
paper
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.
- Out-of-distribution generalization of SAE features.
- A promising property for interpretability analysis off-distribution.
- Standard interpretability approach that VPD critiques and proposes an alternative to.
- SAE reconstruction error direction and norm are largely linearly predictable from the activation vectorfinding0.776Errors are structured not random.
- Prediction following diversity-undertraining logic.
- Central claim.
- Explains why variance correction is needed to see the self-evaluation–persistence relationship
- Chat-trained SAEs find behaviourally relevant features because many causal features occur on chat-specific special tokenshypothesis0.762Explanation for LMSYS superiority.