method
active
method:difference-of-means-dom-probeDifference of Means (DoM) probe
Core method: class-mean difference direction in residual stream scored per token to detect reward hacking.
Neighborhood — ranked by edge-count
Papers (1)
paper
Datasets (1)
dataset
- 5,285 conversations in five hack families used to build DoM vectors.
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.
- Probe construction method: concept vector at each layer is L2-normalized difference between mean positive and mean negative representations from contrastive system prompts
- Method for extracting linear directions by subtracting mean activations of contrastive groups; used to define the Assistant Axis
- Key methodological claim: MM probes are both competitive in accuracy and superior in causal influence
- Attention probe predicting whether an Entity is hallucinated; AUC .94 on validation.
- 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
- Method for computing persona vector as difference between mean activations under aligned and opposite prompts
- Standard interpretability technique used to read feature beliefs.