finding
active
finding:when-mechanisms-point-to-same-entity-model-responds-with-correct-entity-consistentlyWhen mechanisms point to same entity, model responds with correct entity consistently
Mechanism agreement.
Source paper
extracted_from(2025) · Yoav Gur-Arieh · Mor Geva · Atticus Geiger
Neighborhood — ranked by edge-count
Papers (1)
paper
Hypotheses (1)
hypothesis
- Tested in D.1.
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.
- When lexical/reflexive targets absent from original, model falls back to noisy positional mechanismfinding0.787Fig. 28–29.
- Motivating hypothesis for Section 5's investigation of prompt template effects.
- Supported by low correlation between ICatom and RCatom (r=0.44)
- Hallucination circuit.
- Self-explanation vs circuit.
- Speculation about false positives.
- Will a head perform 'the same algorithm' across prompts, or are individual heads polysemantic?question0.769Head polysemanticity.
- Antra's earlier definitive statement of the tricameral model.