method
active
method:activation-probesActivation probes
Internals-based approach to which the logit monitor is complementary.
Neighborhood — ranked by edge-count
Papers (1)
paper
Methods (1)
method
- Logit monitorassociated_withTraining-free monitor that token-forces an eval-aware sentence and computes its probability at each CoT sentence boundary.
Claims (1)
claim
- Positioning.
Hypotheses (1)
hypothesis
- Future work.
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.
- Technique of reading out model beliefs from internal activations before the final answer token is generated
- Interpretability tools that decode information from internal model activations; here, linear probes are used for data attribution.
- Internal representations of the model on which probes operate; the method uses activations to rank datapoints.
- Attention probe predicting whether an Entity is hallucinated; AUC .94 on validation.
- Discovery claim.
- Causal transformer probe predicting whether a token is in the same Entity as its predecessor.
- Standard interpretability technique used to read feature beliefs.
- Classifiers trained to predict concepts from activations.