claim
active
claim:superposition-is-a-key-challenge-to-using-interpretability-for-ai-safety-because-it-blocks-enumeration-of-featuresSuperposition is a key challenge to using interpretability for AI safety because it blocks enumeration of features
Strategic picture.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Toy Models of Superpositionmentions
Concepts (1)
concept
- Any method enabling enumeration of features.
Claims (1)
claim
- Even basic methods become unreliable.
Quotes (1)
quote
- superposition is deeply linked to the ability to identify and enumerate over all features in a modelsupportsSafety rationale.
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.
- Central thesis.
- Supported by Table 1.
- Discussion implication.
- Core mechanism statement.
- Adversarially robust models have different interpretability because adversarial training affects superpositionhypothesis0.782Could be tested by crosscoder comparisons.
- Superposition maximalist stance.
- Features smeared across layers cannot be fully disentangled by SAE on a single residual stream.
- Open question.