claim
active
claim:aos-are-expensive-at-inference-time-compared-to-probes-or-saesAOs are expensive at inference time compared to probes or SAEs
Multiple forward passes needed.
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.
- Limitation relative to SAEs.
- AOs can serve as probes on demand using only a natural-language property specificationhypothesis0.769Hope for generalization beyond trainable supervised probes.
- Argument for scalability.
- Overall positioning.
- AOs are not best on Secret Side Constraint, where black-box prompting reveals the secretfinding0.744The one task where AOs lose.
- A cascade of reward functions, escalating to expensive verifiers when probe confidence is low, is worthwhile.hypothesis0.737Future work.
- Headline auditing result.
- Scaling result over Qwen-3 8B, Gemma-2 9B, Llama-3.3 70B, Claude Haiku 3.5.