concept
active
concept:adversarial-examplesAdversarial examples
Linked to superposition interference.
Neighborhood — ranked by edge-count
Papers (2)
paper
- Toy Models of Superpositionmentions
Methods (1)
method
- Used instead of gradient descent attacks due to gradient masking.
Concepts (1)
concept
- Feature Interferenceassociated_withWhen non-orthogonal features cause logistic regression to identify a suboptimal probe direction
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.
- Competitive multi-agent setting with conflicting incentives and direct opposition via bidding and bluffing.
- Superposition maximalist stance.
- Intervention used to control robustness.
- Eight instruction variants appended to prompts to attempt to break superficial role-play and test depth of character
- Risk that multiple truth directions enable attacks that shift outputs without triggering the primary truth direction
- Adversarial example transfer may be predicted by correlated arrangement of features in superpositionhypothesis0.755Speculative.
- Technique used in VPD to enforce mechanistic faithfulness of parameter decompositions.
- Optimization-based jailbreak method appending strings to prompts to elicit harmful outputs.