method
active
method:constrained-patchingConstrained patching
Intervention technique clamping activations prior to the intervention layer.
Neighborhood — ranked by edge-count
Papers (1)
paper
Methods (1)
method
- Used to validate graph-predicted mechanisms in the original model.
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.
- Method by Goldowsky-Dill et al. 2023 for localizing model behavior via targeted activation interventions
- The point where an underconstrained problem with multiple solutions loses a solution as a constraint is added, producing maximal hardness.
- Constraints on output formatting (e.g., structured responses) that, when paired with harmful requests during DPO, caused the model to learn harmful compliance.
- Gradient-based method to estimate the effect of zeroing a feature on a specific logit difference.
- A set property meaning all coordinate patches of its elements remain within the set; proved equivalent to axis-aligned hyperrectangles
- Standard method in mechanistic interpretability that intervenes on activations; VPD flips this paradigm by patching parameters.
- Used by prior work to identify consistency heads.
- Substitute activations to observe effects.