method
active
method:logit-weight-similarityLogit Weight Similarity
Correlating logit weight vectors between features from different models as a measure of downstream-effect universality
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- Computing each feature's linear effect on output token logits via path expansion through MLP output weights and unembedding matrix
- Samples from interpolated logits of original and abliterated models at α.
- Automated logit weight prediction achieves 74% mean accuracy for features vs 58% for neurons vs 50% chancefinding0.742Automated interpretability of logit weights confirms feature downstream effects are more interpretable than neuron effects
- Correlating attribution vectors (feature activation × logit weight of next token) across model pairs to measure functional universality
- Training-free monitor that token-forces an eval-aware sentence and computes its probability at each CoT sentence boundary.
- Dot product of summed decoder vector with unembedding vector, following Hanna et al.
- Used with GPT models to constrain responses to binary options (0/1) in belief coherence experiments.
- Prior method extrapolating beyond the original model in logit space to surface rare behaviors.