method
active
method:yes-no-token-logit-probability-measurementYes/No token logit probability measurement
How persona behavior p(y|x) is measured.
Neighborhood — ranked by edge-count
Papers (1)
paper
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.
- Full distribution over tokens 0-9 at first generation step; contains more information than any single sampled token
- Behavioral belief estimate as probability-weighted expected rating after each sentence.
- §5.5.
- Logit-minus-mean-logit loss was most effective for attribution, vs log-loss which was confusing to interpretfinding0.739Attribution loss selection.
- Measures emotion feature persistence as correlation between z-scored activation at token 0 and token 100 across all eligible target model tokens
- Automated logit weight prediction achieves 74% mean accuracy for features vs 58% for neurons vs 50% chancefinding0.733Automated interpretability of logit weights confirms feature downstream effects are more interpretable than neuron effects
- Dot product of summed decoder vector with unembedding vector, following Hanna et al.
- With raw data, wider token strides give more accurate approximations at same token budget.finding0.728Cost tradeoff.