Logit Weight Similarity

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

• Logit Weight Analysismethod0.826
  Computing each feature's linear effect on output token logits via path expansion through MLP output weights and unembedding matrix
• Automated logit weight prediction achieves 74% mean accuracy for features vs 58% for neurons vs 50% chancefinding0.742
  Automated interpretability of logit weights confirms feature downstream effects are more interpretable than neuron effects
• Attribution Similaritymethod0.739
  Correlating attribution vectors (feature activation × logit weight of next token) across model pairs to measure functional universality
• global logit shiftconcept0.719
  The methodological confound identified by this paper: injection biases model toward 'YES' for any binary question regardless of content
• Self-Similarityconcept0.719
  Structural and functional property exhibited by living systems but currently absent from most engineered machines.
• Logit-based self-reportmethod0.716
  Primary self-report measure: probability-weighted expected value over all ten digit-token logits, yielding a continuous rating that preserves full distributional signal
• Functional Similarityconcept0.706
  Similarity measured with respect to network behavior/function rather than statistical correlation of activations.
• Logit Lensmethod0.704
  Unsupervised interpretability technique that projects activations through unembedding matrix; provides comparison point for NLA approach.