concept
active
concept:lookup-table-features

Lookup table features

Features translating between input digit properties and output properties in addition.

Neighborhood — ranked by edge-count

Methods (1)

method
  • Visualizations of feature activity on '=' token across 10,000 a+b prompts.

Related by similarity (8)

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

  • Featuresconcept0.767
    Interpretable concept representations hypothesized as basic units of computation.
  • Method of optimizing input to cause a neuron to fire maximally, used to characterize what a neuron detects; establishes causal link
  • The single primitive operation of looking up a value from a set of n keys.
  • Planning featureconcept0.725
    Features that activate before the action sequence they bring about, fulfilling functions of intention
  • Datapoint featureconcept0.719
    Transcoder feature that activates on p and produces p, memorizing something the model did not.
  • Attentional featuresframework0.712
    Proposed attention analog of sparse monosemantic features for decomposing attention layers.
  • Feature Sparsityconcept0.709
    Property that features activate on only a small fraction of inputs; enables compressed sensing and is what allows superposition to work
  • Feature utilityconcept0.708
    Frequency-weighted spectral value πkλkj determining which features are retained.