Most correlated neuron A/neurons/470 has correlation of only 0.18 with base64 feature A/1/2357 and responds to code, HTML labels, URLs

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

• No neuron found with Hebrew Unicode block in top dataset examples; most correlated neuron A/neurons/489 has correlation of only 0.1finding0.870
  Hebrew feature is effectively invisible in the neuron basis
• Features in A/1 have median activation correlation of 0.72 with most similar feature in B/1; neurons have median 0.46finding0.848
  Systematic comparison showing features are substantially more universal than neurons across models
• 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 languagesfinding0.813
  Demonstrates that the Arabic feature is not aligned to any single neuron
• Base64 feature A/1/2357 and B/1/2165 have activation correlation of 0.85finding0.812
  Universality of base64 feature across two transformers
• Claude achieves significantly higher Spearman correlation predicting feature activations vs neuron activationsfinding0.802
  Automated interpretability analysis of activations confirms features are more interpretable than neurons
• 82% of features in 1M SAE had maximum Pearson correlation ≤0.3 with any MLP neuron, and manual inspection showed no semantic resemblance.finding0.785
  SAE features are not simply mirroring individual neurons.
• DNA feature A/1/2937 and B/1/3680 have activation correlation of 0.92finding0.782
  Universality of DNA feature across two transformer models with different random seeds
• Feature pair A/1/3949 and B/1/3321 have activation correlation 0.98 but negative logit weight correlation, firing on PLOSOne journal abbreviationsfinding0.768
  Demonstrates that activation similarity can diverge from logit weight similarity due to interference