claim
active
claim:aos-are-a-complementary-approach-to-mechanistic-interpretabilityAOs are a complementary approach to mechanistic interpretability
Overall positioning.
Source paper
extracted_from(2025) · Adam Karvonen
Neighborhood — ranked by edge-count
Papers (1)
paper
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.
- Argument for scalability.
- AOs articulate responses with natural-language flexibility rather than a fixed concept setclaim0.815Contrast with SAE features.
- AOs can serve as probes on demand using only a natural-language property specificationhypothesis0.810Hope for generalization beyond trainable supervised probes.
- Limitation; confabulation risk.
- Limitation relative to SAEs.
- Injecting the activation difference between fine-tuned and original model into the AO.
- Proposed future direction.
- Training procedure for AOs on activation QA data.