concept
active
concept:feature-discovery-count-d-n-and-efficiency-d-n-nFeature discovery count D(N) and efficiency D(N)/N
Metric of features discovered.
Neighborhood — ranked by edge-count
Papers (1)
paper
Concepts (1)
concept
- Pathological scaling regime (β<α)associated_withSAEs learn far fewer features than latents.
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.
- Warm-up derivation.
- Measures features learned.
- Optimal number of features scales faster than optimal number of training steps with compute budget.finding0.748Allocation result from scaling laws.
- Fraction of training tokens on which a given feature has nonzero activation; used as proxy metric for autoencoder quality
- The central axiom of Nicholson's theory, establishing the causal relationship between environmental complexity and human cognitive/creative capacity.
- Claude Opus 4.1 and 4 show greatest reduction in apology rate in the prefill detection taskfinding0.730Injecting a concept matching the prefilled word reduces the rate at which the model apologizes, maximally for Opus models.
- Claim that building a scalable associative memory engine is a significant research challenge.