hypothesis
active
hypothesis:inverse-scr-scaling-is-simply-due-to-a-fixed-number-of-ablated-latents-being-a-smaller-fraction-of-larger-saesInverse SCR scaling is simply due to a fixed number of ablated latents being a smaller fraction of larger SAEs
Alternative hypothesis tested and rejected.
Source paper
extracted_from(2025) · Adam Karvonen · Can Rager · Johnny Lin · Curt Tigges +11
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (1)
finding
- Inverse SCR scaling persists across intervention sizes for non-hierarchical architecturescontradictsRules out fraction-modified explanation.
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.
- Hypothesis cited in paper suggesting deceptive capabilities may scale with model size
- Larger models behave worse on a metric; applied to faithfulness.
- Claim about broader applicability of the scaling argument
- Novel metric extending SHIFT via zero ablation.
- Discrete case.
- Better bases may give sparsity.
- Author's emphasis on the interpretive nature of scaling.
- SAEs assume sparsity (number of concepts) is time-invariant, drawn i.i.d. from a fixed distribution.claim0.713Corollary 4.1.1.