claim
active
claim:superposition-exploits-the-geometry-of-high-dimensional-spaces-which-allow-exponentially-many-almost-orthogonal-vectors-but-only-n-strictly-orthogonal-onesSuperposition exploits the geometry of high-dimensional spaces, which allow exponentially many almost-orthogonal vectors but only n strictly orthogonal ones.
Mechanistic explanation for why superposition is geometrically feasible
Source paper
extracted_from(2020) · Chris Olah · Nick Cammarata · Ludwig Schubert · Gabriel Goh +2
Neighborhood — ranked by edge-count
Claims (1)
claim
- Interpretation of the cars-in-superposition circuit finding as an intentional representational strategy
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.hypothesis0.815Explanation for why dictionary learning can recover many more features than dimensions.
- Supported by Table 1.
- Central thesis.
- Analogy to many ice phases.
- Superposition begins with less important features in antipodal pairs, then transitions to other geometriesfinding0.777Toy model sparsity sweep.
- Mathematical derivation from Saxe-style loss.
- Cost of removing superposition.
- Superposition maximalist stance.