method
active
method:supernodesSupernodes
Manual grouping of related features to simplify attribution graphs.
Neighborhood — ranked by edge-count
Papers (2)
paper
- On the Biology of a Large Language Modelintroducesmentions
Concepts (1)
concept
- Feature splittingassociated_withPhenomenon where a feature in a small SAE splits into multiple finer features in a larger SAE.
Methods (1)
method
- Gradient-based technique using SAE features to estimate causal effects on completions; used to corroborate NLA findings.
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.
- Phenomenon where models represent more features than dimensions via almost-orthogonal directions.
- A being with superhuman moral status, distinguished from super-beneficiary by how moral claims arise
- Agent aggregating reports from many investigators, like best-of-N.
- Algorithm canonicalizing network representations by distances.
- Theoretical model of how neural networks encode more features than dimensions, informing linear representation work.
- Superposition configuration associated with a plateau and a dip in vulnerability.
- A being that is superhumanly efficient at deriving well-being from resources; the paper's central analytical category