claim
active
claim:aos-articulate-responses-with-natural-language-flexibility-rather-than-a-fixed-concept-setAOs articulate responses with natural-language flexibility rather than a fixed concept set
Contrast with SAE features.
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 can serve as probes on demand using only a natural-language property specificationhypothesis0.820Hope for generalization beyond trainable supervised probes.
- Overall positioning.
- Limitation; confabulation risk.
- Limitation relative to SAEs.
- Scaling result over Qwen-3 8B, Gemma-2 9B, Llama-3.3 70B, Claude Haiku 3.5.
- Articulates why a one-layer transformer with MLP is the appropriate starting target for mechanistic interpretability
- Antra's earlier definitive statement of the tricameral model.