claim
active
claim:aos-are-not-exhaustive-and-will-not-answer-questions-we-don-t-know-to-askAOs are not exhaustive and will not answer questions we don't know to ask
Limitation relative to SAEs.
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.
- Overall positioning.
- Limitation; confabulation risk.
- Headline auditing result.
- AOs articulate responses with natural-language flexibility rather than a fixed concept setclaim0.783Contrast with SAE features.
- AOs can serve as probes on demand using only a natural-language property specificationhypothesis0.782Hope for generalization beyond trainable supervised probes.
- Multiple forward passes needed.
- AOs are not best on Secret Side Constraint, where black-box prompting reveals the secretfinding0.767The one task where AOs lose.