Layer Normalization (Ba et al., 2016b)

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• At layer 12 (the layer analyzed by Burger et al. 2024), tP and tG explain similar fractions of truth-related variance (~0.33 each).finding0.748
  Shows that Burger et al.'s layer choice corresponds to a transitional phase, not a universal property.
• Math/code tasks S ≈ -1.65 at layers 8–12finding0.739
  Task-specific peak anchoring score for structured reasoning domains.
• maximum-norm gradient normalizationmethod0.727
  Training-free technique normalizing all task gradients to the maximum gradient norm magnitude
• Layer-wise geometry shows early dip, mid-layer alignment, and late standardization across tasksclaim0.726
  Qualitative pattern from E3.
• Cross-layer superpositionconcept0.725
  Representation of features spread across multiple layers, complicating dictionary learning.
• Layer-wise geometry summaries (Sbmax, AUSN) predict internal few-shot thresholds θ50claim0.721
  Claim that geometry-to-behavior correlates exist
• Layer-wise trajectories show early enrichment, mid-layer alignment, and late re-clustering.claim0.721
  Qualitative geometry pattern.
• Peak layer ℓ* median 10, IQR 0.384finding0.721
  Median layer where S(ℓ) peaks, across seeds.