method
active
method:decomposed-probing-protocolDecomposed probing protocol
Two-step localize-then-classify pipeline for hallucination monitoring and reward.
Neighborhood — ranked by edge-count
Papers (1)
paper
Methods (2)
method
- Attention probe predicting whether an Entity is hallucinated; AUC .94 on validation.
- Causal transformer probe predicting whether a token is in the same Entity as its predecessor.
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.
- Standard interpretability technique used to read feature beliefs.
- Top-down interpretability approach studying linguistic properties at various residual stream stages; contrasted with the paper's bottom-up mechanistic approach
- Earlier interpretability method applying classifiers to DNN hidden representations; shares complexity-accuracy dilemma with causal abstraction
- Probing approach avoiding supervision to sidestep complexity-accuracy tradeoff
- Technique of reading out model beliefs from internal activations before the final answer token is generated
- Method from Gurnee et al. 2023 for finding feature directions including individual neuron analysis
- Interpretability tools that decode information from internal model activations; here, linear probes are used for data attribution.
- Method of using base models (no post-training) to observe spontaneous self-referential behaviors without confound of memorized introspection language.