claim
active
claim:sometimes-what-matters-at-a-position-is-the-features-active-there-not-the-tokenSometimes what matters at a position is the features active there, not the token
Explains hard-to-read dataset patterns.
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (2)
finding
- K-composition example.
- QK attribution from Ivory to 'a' has top pair Ivory Coast query feature and Zambian key featuresupportsExplains attention edge.
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.
- Design choice validated empirically.
- At 10,000 tokens positional information is nearly nonexistent for first half of groups except firstfinding0.770Padding result.
- If the relationship is present in activations, better tools (e.g. trained on these datasets) could surface ithypothesis0.765Alternative explanation.
- Describes the properties of the functional token.
- Known limitation.
- Justifies the choice of response tokens for persona vector extraction in the pipeline
- Claim linking late-binding to enhanced runtime adaptability.
- Fig. 7.