finding
active
finding:large-and-detailed-cluster-of-phonetic-and-character-related-featuresLarge and detailed cluster of phonetic and character-related features
Specific cluster observed.
Neighborhood — ranked by edge-count
Papers (1)
paper
Questions (1)
question
- Open question about feature cluster.
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.
- Decoder cosine similarity maps onto concept similarity.
- Mode in feature density histogram around 1e-5 corresponding to interpretable features, contrasted with ultralow density cluster
- Specific cluster observed.
- Complex, behavior-level semantic attributes such as personality traits that the paper aims to steer.
- Possible explanation for isolated repetitive chat features.
- Hypothesized extension of superposition where features may be higher-dimensional manifolds rather than 1D directions
- Fraction of training tokens on which a given feature has nonzero activation; used as proxy metric for autoencoder quality
- Synthesizing inputs maximizing component activation.