Activation Similarity

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

• Activationsconcept0.809
  Internal representations of the model on which probes operate; the method uses activations to rank datapoints.
• Activation Correlationmethod0.807
  Pearson correlation of feature activations across 40M tokens used to measure feature similarity and universality across models
• Other-Referencing Activationsconcept0.798
  Latent model activations when processing inputs framed from another agent's perspective
• Activation Additionmethod0.789
  Intervention method that adds a learned direction vector to residual stream activations to steer model behavior
• Functional Similarityconcept0.787
  Similarity measured with respect to network behavior/function rather than statistical correlation of activations.
• Attribution Similaritymethod0.787
  Correlating attribution vectors (feature activation × logit weight of next token) across model pairs to measure functional universality
• Reflection Enhancement via Activation Additionmethod0.765
  Adding steering vector in forward direction to push model activations toward stronger reflective behavior.
• Activation decompositionconcept0.762
  The conventional approach (e.g., SAEs, transcoders) of decomposing activations into interpretable features.