claim
active
claim:superposition-makes-cosine-similarity-potentially-misleadingSuperposition makes cosine similarity potentially misleading
Even basic methods become unreliable.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Toy Models of Superpositionmentions
Claims (1)
claim
- Strategic picture.
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.
- Limit on Jacobian matching.
- Geometric evaluation of truth direction alignment across layers and prompt templates.
- Cosine similarity between feature activations restricted to tokens where one of the features fires; used to identify feature splitting relationships
- Used to measure alignment between DIM direction and cone basis vectors to assess overlap
- Open question.
- Detection mechanism computing cosine similarity between activation vectors and steering vectors to classify deception
- Core mechanism statement.
- In the residual stream a country and its own cities have positive cosine similarity (mean +0.17)finding0.771Head flips sign via W_Q and W_K.