question
active
question:how-can-we-control-whether-superposition-and-polysemanticity-occurHow can we control whether superposition and polysemanticity occur?
Most pragmatically important question.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Toy Models of Superpositionmentions
Claims (1)
claim
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.
- Major prediction checked against InceptionV1.
- A head can implement multiple behaviors via high-rank output without superposition.
- Will a head perform 'the same algorithm' across prompts, or are individual heads polysemantic?question0.775Head polysemanticity.
- Low-rank matrices W^TW.
- Interpretation of the cars-in-superposition circuit finding as an intentional representational strategy
- Neurons that respond to multiple unrelated concepts, limiting interpretability.
- Superposition begins with less important features in antipodal pairs, then transitions to other geometriesfinding0.767Toy model sparsity sweep.
- Core mechanism statement.