method
active
method:linear-probe-weight-steeringLinear Probe Weight Steering
Uses normalized probe weights as steering vector.
Neighborhood — ranked by edge-count
Papers (1)
paper
Methods (2)
method
- Linear ProbeusesSimple linear classifiers trained on model activations used as the probing technique within the introduced method.
- Activation Addition SteeringextendsAdd scaled steering vector to layer activations.
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.
- Typical approach that adds a scaled steering vector to representations; the paper argues this is mismatched with actual representation geometry.
- Constructing steering vectors from the difference of mean activations on positive and negative examples, for comparison.
- Method for fitting a linear classifier on collected activations to predict task-relevant features
- Causal evidence.
- Self-repair effect.
- Used to evaluate representation quality across VTAB tasks
- Used to characterize counting representations.
- The paper's critique of the standard linear steering baseline, supported by the days-of-week demo.