question
active
question:to-what-extent-can-neural-networks-do-useful-computation-on-features-in-superpositionTo what extent can neural networks do useful computation on features in superposition?
Open question.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Toy Models of Superpositionmentions
Claims (1)
claim
- Supported by abs model.
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.
- Theoretical model of how neural networks encode more features than dimensions, informing linear representation work.
- Superposition hypothesis: neural networks represent more features than dimensions using almost-orthogonal directions.hypothesis0.854Explanation for why dictionary learning can recover many more features than dimensions.
- If we don't give a model enough neurons to implement this solution, how does it do computation in superposition?question0.827Left to future work.
- superposition is deeply linked to the ability to identify and enumerate over all features in a modelquote0.809Safety rationale.
- Superposition hypothesis: networks represent more features than neurons as almost-orthogonal directionshypothesis0.804Central hypothesis.
- Interpretation of the cars-in-superposition circuit finding as an intentional representational strategy
- Vision statement in the conclusion.
- Speculative extension of universality to neuroscience, with high-low frequency detectors as a candidate prediction