Cosine Similarity Ranking for Instruction Discovery

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• Cosine Similarity Measurementmethod0.815
  Used to measure alignment between DIM direction and cone basis vectors to assess overlap
• Cosine Similarity Binary Classifiermethod0.807
  Classifier using cosine similarity between activation vectors and steering vectors to detect deception with 89% accuracy
• Pairwise Cosine Similarity Analysismethod0.803
  Used to quantify the semantic clustering of adjective-set embeddings across model families and conditions
• Feature neighborhood exploration via cosine similarity of decoder weightsmethod0.791
  Identifying related features by cosine distance in SAE decoder space.
• Cosine Similarity-Based Deception Detectionconcept0.791
  Detection mechanism computing cosine similarity between activation vectors and steering vectors to classify deception
• Cosine similarity between truth probesmethod0.780
  Geometric evaluation of truth direction alignment across layers and prompt templates.
• Steering vector-based instruction discovery outperforms input embedding similarity baseline for reflection-inducing instruction selectionfinding0.773
  Demonstrates that surface-level embedding similarity fails to capture reflective semantics.
• Probes trained under different explicit instruction prompts (ask-correct, ask-t/f, ask-able, ask-arith) are highly aligned with each other in cosine similarity.finding0.771
  Shows the passive vs. active divide is more important than the specific wording of instructions.