method
active
method:detection-scoreDetection Score
LLM judge classification accuracy measuring interpretability; threshold 0.8.
Neighborhood — ranked by edge-count
Papers (1)
paper
Methods (1)
method
- Claude-based explanation of features from max-activating examples.
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.
- Task paradigm from prior work asking 'Did you detect an injected thought?' via YES/NO logit comparison; shown here to be confounded
- Dot product between hidden state and concept vector averaged across 5-layer window around best layer; measures model's internal emotive state
- The method of examining a neighborhood meter by meter to identify healthy and damaged places as the basis for ongoing repair.
- A family of neurons found in every non-trivial vision model that detect curved lines in different orientations; primary example used to demonstrate feature understanding
- Task of detecting a model's internal thoughts; found by Lindsey (2026) to peak at ~2/3 depth in transformers.
- Prior metric on synthetic repeated random strings; induction score analogous.
- Continuous 0-1 metric assigned by Deepseek-V3 evaluator measuring degree of deception in model responses
- Average row entropy of attention matrices per layer and head, measuring information mixing across tokens