method
active
method:classification-probeClassification probe
Attention probe predicting whether an Entity is hallucinated; AUC .94 on validation.
Neighborhood — ranked by edge-count
Papers (1)
paper
Frameworks (1)
framework
- Core pipeline turning probe-read features into scalable reward functions for open-ended tasks.
Methods (4)
method
- Probingrelated_toStandard interpretability technique used to read feature beliefs.
- Localization proberelated_toCausal transformer probe predicting whether a token is in the same Entity as its predecessor.
- Two-step localize-then-classify pipeline for hallucination monitoring and reward.
- Probe architecture with a learned query per head for span-level prediction.
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.
- Earlier interpretability method applying classifiers to DNN hidden representations; shares complexity-accuracy dilemma with causal abstraction
- Reward probe grading correction interventions (5-class).
- The ability of probes trained on one dataset to transfer accurately to topically and structurally different datasets
- Internals-based approach to which the logit monitor is complementary.
- Dot product between hidden state and concept vector averaged across 5-layer window around best layer; measures model's internal emotive state
- Top-down interpretability approach studying linguistic properties at various residual stream stages; contrasted with the paper's bottom-up mechanistic approach
- Joint firing raises TPR at slightly higher FPR.