method
active
method:adversarial-search-for-causally-unimportant-subcomponentsAdversarial search for causally unimportant subcomponents
Procedure in VPD that actively searches for combinations that break the prediction of which subcomponents are unimportant, stress-testing the decomposition.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Interpreting Language Model Parametersintroduces
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.
- Superposition maximalist stance.
- Central thesis.
- Adversarial example transfer may be predicted by correlated arrangement of features in superpositionhypothesis0.764Speculative.
- Superposition may be a major contributing factor, or even the primary cause, of adversarial examples.hypothesis0.761Hypothesis under test.
- Competitive multi-agent setting with conflicting incentives and direct opposition via bidding and bluffing.
- Supported by Table 1.
- Eight instruction variants appended to prompts to attempt to break superficial role-play and test depth of character
- The atomic pieces into which weight matrices are decomposed by VPD; each rank-one component is interpretable.