method
active
method:global-weights-tweraGlobal weights (TWERA)
Prompt-independent feature interaction estimates for traversing harm features.
Neighborhood — ranked by edge-count
Papers (1)
paper
Datasets (1)
dataset
- Sorry Benchuses200 diverse harmful prompts used to corroborate refusal features.
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.
- Target-weighted variant of ERA introduced in prior work.
- Computing each feature's linear effect on output token logits via path expansion through MLP output weights and unembedding matrix
- Baseline MTL approach minimizing sum of task losses with equal weights; suffers from task balancing
- Implicit weights directly connecting any pair of layers computed by multiplying output weights of one layer with input weights of another through the residual stream
- Used to project features through QK weights and find aligned features.
- Virtual weights between features forced into superposition by low-dimensional projections.
- Correlating logit weight vectors between features from different models as a measure of downstream-effect universality
- Helpful and harmful weights are scattered across the entire range of virtual weight magnitudesfinding0.683IN Tokens→Logits shows weak VW-helpfulness relationship.