finding
active
finding:minimizing-projection-onto-the-refusal-direction-yields-efficient-red-teaming-inputsMinimizing projection onto the refusal direction yields efficient red-teaming inputs
Interpretability insight accelerates red-teaming.
Source paper
extracted_from(2025) · Lee Sharkey · Bilal I. Chughtai · Joshua Batson · Jack W. Lindsey +25
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Prioritization argument.
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.
- Stated generalization, untested here.
- Section 6.1 suggests future work on scaling automated red teaming.
- Practical application proposed based on mechanistic findings
- Consistent with refusal cone work.
- Causal validation.
- Layer-by-layer analysis of refusal direction properties
- Refusal is not a single linear classifier.
- Mechanistic basis and under-elicitation of introspective awareness in LLMs.