claim
active
claim:linear-parameter-decomposition-methods-do-not-suffer-from-feature-splittingLinear parameter decomposition methods do not suffer from feature splitting
Claimed by Braun et al.; demonstrated empirically here.
Source paper
extracted_from(2025) · Lucius Bushnaq · Dan Braun · Lee Sharkey
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (1)
finding
- Ground truth mechanisms identified.
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.
- Framework decomposing network parameters into sum of sparsely used vectors in parameter space.
- Calls for conceptual clarity.
- Authors argue the absence of a fixed feature count is a property of the superposition geometry, not a failure of the method
- The central object of study — the idea that a concept like truth is encoded as a direction in the LLM's latent space
- Phenomenon where a feature in a small SAE splits into multiple finer features in a larger SAE.
- Interpretation of Fig. 9.
- Author's open question about limitations of linear persona vector approach