method
active
method:concept-based-probesConcept-based probes
Classifiers trained to predict concepts from activations.
Neighborhood — ranked by edge-count
Papers (1)
paper
Frameworks (1)
framework
- Second approach: propose concepts and find components corresponding to them.
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.
- Interpretability tools that decode information from internal model activations; here, linear probes are used for data attribution.
- Attention probe predicting whether an Entity is hallucinated; AUC .94 on validation.
- Internals-based approach to which the logit monitor is complementary.
- Causal transformer probe predicting whether a token is in the same Entity as its predecessor.
- Linear classifier approach applied to model activations to identify which training datapoints caused undesired behaviors in post-training.
- Central entity of Jackson's framework: a structure invented to give coherent account of immediate consequences of actions; the building block of software design
- Technique of injecting activation patterns associated with specific concepts into a model's internal states to test whether self-reports reflect ground truth.
- Standard interpretability technique used to read feature beliefs.