Binarized DNA proxy has Pearson correlation of 0.80 with A/1/2937 feature activations

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

• DNA feature A/1/2937 and B/1/3680 have activation correlation of 0.92finding0.857
  Universality of DNA feature across two transformer models with different random seeds
• Pearson correlation of 0.74 between A/1/3450 activation and Arabic script proxy over 40M tokensfinding0.840
  Joint measure of sensitivity and specificity for the Arabic script feature
• Base64 feature A/1/2357 and B/1/2165 have activation correlation of 0.85finding0.787
  Universality of base64 feature across two transformers
• Feature pair A/1/3949 and B/1/3321 have activation correlation 0.98 but negative logit weight correlation, firing on PLOSOne journal abbreviationsfinding0.778
  Demonstrates that activation similarity can diverge from logit weight similarity due to interference
• Hebrew feature A/1/416 and B/1/1901 have activation correlation of 0.92finding0.777
  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 tokensfinding0.776
  Demonstrates universality of the Arabic script feature across two independently trained transformers
• Features in A/1 have median activation correlation of 0.72 with most similar feature in B/1; neurons have median 0.46finding0.761
  Systematic comparison showing features are substantially more universal than neurons across models
• 82% of features in 1M SAE had maximum Pearson correlation ≤0.3 with any MLP neuron, and manual inspection showed no semantic resemblance.finding0.743
  SAE features are not simply mirroring individual neurons.