claim
active
claim:superposition-is-at-least-one-causal-factor-in-the-existence-of-adversarial-examples-and-possibly-the-primary-oneSuperposition is at least one causal factor in the existence of adversarial examples, and possibly the primary one.
Central thesis.
Source paper
extracted_from(2025) · Gorton, Liv · Lewis, Owen
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (4)
finding
- Evidence in real models.
- Robustness controls superposition.
- Superposition controls robustness.
- Appendix replication.
Hypotheses (1)
hypothesis
- Hypothesis under test.
Claims (5)
claim
- Superposition maximalist stance.
- Interpretation of Figure 3.
- Discussion implication.
- Stated limitation.
- Supported by Table 1.
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 is deeply linked to the ability to identify and enumerate over all features in a modelquote0.840Safety rationale.
- Adversarial example transfer may be predicted by correlated arrangement of features in superpositionhypothesis0.834Speculative.
- Novelty claim.
- Interpretation of the cars-in-superposition circuit finding as an intentional representational strategy
- Core theoretical framework: neural networks represent more features than neurons by encoding features as directions in superposition
- Main contribution for interpretability.
- Mechanistic explanation for why superposition is geometrically feasible
- Superposition hypothesis: neural networks represent more features than dimensions using almost-orthogonal directions.hypothesis0.794Explanation for why dictionary learning can recover many more features than dimensions.
Restated by (1)
cosine ≥ 0.90Other entities that say roughly the same thing. May be merge candidates or independent restatements across papers.