finding
active
finding:attribution-graph-for-the-1711-prompt-shows-dense-feature-concentration-from-token-11-to-final-position-concentrated-in-later-layersAttribution graph for the 1711 prompt shows dense feature concentration from token 11 to final position, concentrated in later layers
Structure of the computational pathway.
Source paper
extracted_fromNeighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Main conclusion.
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.
- Contrast in §6.3.
- Quantifies method limitations.
- Design choice validated empirically.
- Mechanistic discrimination of CoT faithfulness.
- Narrow-range greater-than feature.
- Feature attribution (gradient-based) correlates 0.8 with ablation effects on the 'John' and 'Kobe' examples.finding0.773Validation of attribution as a fast proxy for causal importance.
- Table 2; e.g., Identity +1.8 for turn-averaged.
- Blocking attention to query token in layers 11–16 drops accuracy from 98% to 37%; with next token 8%finding0.767Attention knockout.