Cosine similarity between truth probes

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

• Cosine Similarity-Based Deception Detectionconcept0.837
  Detection mechanism computing cosine similarity between activation vectors and steering vectors to classify deception
• Cosine Similarity Measurementmethod0.831
  Used to measure alignment between DIM direction and cone basis vectors to assess overlap
• Masked Cosine Similaritymethod0.797
  Cosine similarity between feature activations restricted to tokens where one of the features fires; used to identify feature splitting relationships
• The model converges to a more stable truth direction in middle-to-late layers, as evidenced by increasing cosine similarity between layer-wise probes.claim0.781
  Supported by the geometric transition visible in cosine similarity heatmaps for F0-F3.
• Pairwise Cosine Similarity Analysismethod0.781
  Used to quantify the semantic clustering of adjective-set embeddings across model families and conditions
• Cosine Similarity Ranking for Instruction Discoverymethod0.780
  Method to discover new reflection-inducing instructions by ranking candidate tokens by cosine similarity to steering vectors.
• In Qwen-2.5-9B, only v1 has meaningful cosine similarity to DIM direction; all additional basis vectors have cosine similarities ~1e-9finding0.775
  Appendix E replication of DIM alignment finding in Qwen model
• 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.finding0.774
  Shows the passive vs. active divide is more important than the specific wording of instructions.