claim
active
claim:if-the-sae-sparsity-budget-k-is-at-most-the-manifold-dimension-support-switching-must-occur-under-low-errorIf the SAE sparsity budget K is at most the manifold dimension, support switching must occur under low error.
Proposition 4.2.
Source paper
extracted_from(2025) · Lubana, Ekdeep Singh · Rager, Can · Hindupur, Sai Sumedh R. · Costa, Valerie +12
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (1)
finding
- Predictive code shows high within-event similarity (0.56) versus across (-0.04), while SAEs show low within (0.12–0.29).associated_withFig. 7c over 50 GPT-5 stories.
Concepts (2)
concept
- Feature splittingaboutPhenomenon where a feature in a small SAE splits into multiple finer features in a larger SAE.
- Support switchingaboutDifferent latent sets activating for nearby points.
Claims (1)
claim
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.
- Lower SAE loss at fixed sparsity likely indicates less interference and less superposition.claim0.786Justifies SAE proxy.
- Core concern.
- Better bases may give sparsity.
- Motivates multi-metric evaluation.
- Probing limited.
- Main proposal.
- Unsure.
- Authors' experience-based claim.