finding
active
finding:20-of-27-features-in-multilingual-nodes-are-active-across-all-three-language-prompts-only-10-27-appear-in-all-pruned-graphs20 of 27 features in multilingual nodes are active across all three language prompts; only 10/27 appear in all pruned graphs.
Evidence for language-independent representation.
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Multilingual circuits.
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.
- Multilingual features grow with scale.
- Features respond to concepts across languages and in images, not just text.
- We hypothesize that intervention efficiency can be scaled with multi-node and multi-GPU training as language models grow largerhypothesis0.734Future work hypothesis about scaling pyvene's computational efficiency for very large models
- English is mechanistically the default output.
- Operation component is language-independent.
- Fig. 9c, Fig. 37.
- Cited as enabling precise behavioral control through SAE features, extending the same methodological line
- Line width head shows three behaviors: years split tokens, multi-token words, and newlinesfinding0.724Identified from attention patterns in Haiku 3.5.