question
active
question:should-we-expect-superposition-to-go-away-if-we-just-scale-enoughShould we expect superposition to go away if we just scale enough?
Open question.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Toy Models of Superpositionmentions
Claims (1)
claim
- Cost of removing superposition.
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.
- Core mechanism statement.
- If we don't give a model enough neurons to implement this solution, how does it do computation in superposition?question0.776Left to future work.
- Explains robust models being more robust than expected.
- Superposition begins with less important features in antipodal pairs, then transitions to other geometriesfinding0.770Toy model sparsity sweep.
- Limit on Jacobian matching.
- Because superposition is a phase, there may exist a regime for every model with no superpositionclaim0.763Cause for hope.
- Central thesis.
- Most pragmatically important question.