claim
active
claim:canonical-tiling-across-dictionary-sizes-suggests-feature-manifold-pathologies-do-not-occur-in-this-settingCanonical tiling across dictionary sizes suggests feature manifold pathologies do not occur in this setting
Appendix claim.
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (1)
finding
- Canonical featurization.
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.
- Explanation of SAE instability.
- Core concern.
- Neuroscience-motivated prediction.
- Latents near digit/alphabet manifolds may have tiled a manifold with one killed to reduce L0hypothesis0.768Speculative explanation.
- Many latents tiling a low-dimensional manifold would show decoder cosine similarities near 1hypothesis0.756Motivates Appendix E.
- Appendix A.1.
- Future-work perspective.
- Features may not be strictly one-dimensional objects; higher-dimensional feature manifolds may exist in model representationshypothesis0.750Extension of superposition hypothesis to account for continuous families of features