method
active
method:cosine-similarity-between-truth-probesCosine similarity between truth probes
Geometric evaluation of truth direction alignment across layers and prompt templates.
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.
- Detection mechanism computing cosine similarity between activation vectors and steering vectors to classify deception
- Used to measure alignment between DIM direction and cone basis vectors to assess overlap
- Probe cosine similarity becomes more peaked with deeper layers 0-3, with higher secondary ringsfinding0.817Sharpening.
- Geometric measure used to track persona vector drift and stability across checkpoints
- Cosine similarity between feature activations restricted to tokens where one of the features fires; used to identify feature splitting relationships
- Even basic methods become unreliable.
- Gemma Scope decoder nearest-neighbor cosine similarities skewed above random but few extremely highfinding0.786Fig 7.
- Supported by the geometric transition visible in cosine similarity heatmaps for F0-F3.