Features in A/1 have median activation correlation of 0.72 with most similar feature in B/1; neurons have median 0.46

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

• Most correlated neuron A/neurons/470 has correlation of only 0.18 with base64 feature A/1/2357 and responds to code, HTML labels, URLsfinding0.848
  Shows base64 feature is polysemantic at neuron level but monosemantic as learned feature
• DNA feature A/1/2937 and B/1/3680 have activation correlation of 0.92finding0.839
  Universality of DNA feature across two transformer models with different random seeds
• Arabic feature A/1/3450 has 27 neurons with coefficient magnitude ≥0.1 and three largest coefficients are negative; most correlated neuron responds to mixture of non-English languagesfinding0.818
  Demonstrates that the Arabic feature is not aligned to any single neuron
• 82% of features in 1M SAE had maximum Pearson correlation ≤0.3 with any MLP neuron, and manual inspection showed no semantic resemblance.finding0.807
  SAE features are not simply mirroring individual neurons.
• Base64 feature A/1/2357 and B/1/2165 have activation correlation of 0.85finding0.805
  Universality of base64 feature across two transformers
• Hebrew feature A/1/416 and B/1/1901 have activation correlation of 0.92finding0.803
  Universality of Hebrew script feature across two transformers
• Arabic feature A/1/3450 and B/1/1334 have activation correlation of 0.91 across 40M tokensfinding0.803
  Demonstrates universality of the Arabic script feature across two independently trained transformers
• Claude achieves significantly higher Spearman correlation predicting feature activations vs neuron activationsfinding0.803
  Automated interpretability analysis of activations confirms features are more interpretable than neurons