claim
active
claim:harmful-weights-reflect-the-tradeoff-between-implementing-more-or-more-complex-circuits-and-the-interference-between-themHarmful weights reflect the tradeoff between implementing more or more complex circuits and the interference between them
Discussion.
Source paper
extracted_fromNicholas L. Turner · Jeffrey Wu · Joshua Batson
Neighborhood — ranked by edge-count
Papers (1)
paper
Concepts (1)
concept
- Interference WeightsaboutLogit weight contributions from a feature that arise due to superposition with other features, not from the feature's own causal role
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.
- Interpretation of motivating example.
- Explains why Head Cor does not fully dominate for safety-related persona amplification in MMLU/IFEval metrics
- Concluding statement.
- Identifying interference weights is necessary but not sufficient for reading global circuitsclaim0.768Discussion.
- Motivating example.
- Helpful and harmful weights are scattered across the entire range of virtual weight magnitudesfinding0.757IN Tokens→Logits shows weak VW-helpfulness relationship.
- Three regimes by effectiveness.
- Effectiveness doesn't fully clean.