early layers (0–5)

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

• deeper layers (16–28)concept0.789
  Layers where anchoring weakens systematically due to representational drift.
• Math/code tasks S ≈ -1.65 at layers 8–12finding0.739
  Task-specific peak anchoring score for structured reasoning domains.
• Early layers of convolutional networks are more interchangeable than later layers across different architecturesfinding0.735
  Lenc & Vedaldi finding on layer-wise alignment
• Layer-wise trajectories show early enrichment, mid-layer alignment, and late re-clustering.claim0.721
  Qualitative geometry pattern.
• Initial layers of QwQ-32B demonstrate relatively poor LAT performance, consistent with early layers capturing low-level featuresfinding0.712
  Confirms prior research on layer specialization: early layers insufficient for semantic deception detection
• Layer-wise anchoring peaks in a 'Goldilocks zone' between early and late layers.claim0.712
  Qualitative characterization of optimal anchoring depth.
• Single-layer analyses can be misleading because early-layer truth directions may reflect surface features with limited cross-task generalization.claim0.709
  Methodological critique of prior work that fixed a single layer for truth probing.
• Layer-wise geometry shows early dip, mid-layer alignment, and late standardization across tasksclaim0.705
  Qualitative pattern from E3.