claim
active
claim:privileged-basis-and-superposition-are-two-countervailing-forces-determining-whether-features-align-with-neuronsPrivileged basis and superposition are two countervailing forces determining whether features align with neurons
Explanation for neuron interpretability.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Toy Models of Superpositionmentions
Findings (2)
finding
- Explains preference for negative interference.
- Mathematical derivation from Saxe-style loss.
Questions (1)
question
- Motivating question.
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 hypothesis: neural networks represent more features than dimensions using almost-orthogonal directions.hypothesis0.825Explanation for why dictionary learning can recover many more features than dimensions.
- Interpretation of the cars-in-superposition circuit finding as an intentional representational strategy
- Superposition hypothesis: networks represent more features than neurons as almost-orthogonal directionshypothesis0.811Central hypothesis.
- Load-bearing theoretical claim providing the conceptual foundation for DAS.
- The central research question that drives the paper's analysis.
- Discussion implication.
- superposition is deeply linked to the ability to identify and enumerate over all features in a modelquote0.784Safety rationale.
- Correlated feature results.