claim
active
claim:per-token-sae-objective-does-not-compel-high-level-characteristics-to-separate-from-token-level-detailPer-token SAE objective does not compel high-level characteristics to separate from token-level detail
Motivation.
Source paper
extracted_from(2026) · Kevin Der · Harish Kamath · Thompson, Ben
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (1)
finding
- Table 1.
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.
- Standard baseline SAE operating on individual token activations.
- Example A interpretation.
- Special tokens have very large magnitude, so their SAE features also have large magnitude and sit far from other pointshypothesis0.790First explanation for isolated features.
- Table 1.
- Claim that feature grounding enables interpretability metrics.
- Surprising finding that the two evaluation methods diverge in their relationship with persistence
- Extension of mechanistic interpretability findings to the metacognitive domain
- Scaling SAE size increases granularity and discovers new features.