hypothesis
active
hypothesis:many-similar-narrow-greater-than-features-exist-that-activate-for-different-promptsMany similar narrow greater-than features exist that activate for different prompts
Inferred from Feature 461858.
Source paper
extracted_fromNeighborhood — ranked by edge-count
Papers (1)
paper
Findings (1)
finding
- Narrow-range greater-than feature.
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.
- Interpretation.
- Third empirical pattern.
- Main conclusion.
- Shows interpretability correlates with activation strength, most model effect comes from high activations
- Mechanism from Theorem 4.
- Universalist claim predicting cross-cultural generality.
- Width buys features in order of utility, so only larger models learn infrequent or complex tasksclaim0.755Interpretation of Theorem 3.
- Refusal is not a single linear classifier.