hypothesis
active
hypothesis:adversarial-example-transfer-may-be-predicted-by-correlated-arrangement-of-features-in-superpositionAdversarial example transfer may be predicted by correlated arrangement of features in superposition
Speculative.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Toy Models of Superpositionmentions
Hypotheses (1)
hypothesis
- Adversarially robust models would have worse performance and be more interpretable because robustness requires giving up superpositionassociated_withPredictions of superposition maximalism.
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.
- Superposition may be a major contributing factor, or even the primary cause, of adversarial examples.hypothesis0.846Hypothesis under test.
- Central thesis.
- Supported by Table 1.
- Adversarially robust models have different interpretability because adversarial training affects superpositionhypothesis0.805Could be tested by crosscoder comparisons.
- Interpretation of Figure 3.
- superposition is deeply linked to the ability to identify and enumerate over all features in a modelquote0.798Safety rationale.
- Consistent but alternative explanations like amplification exist.