bibkey: “gao2024scaling” authors: “Leo Gao; Tom Dupre la Tour; Henk Tillman; Gabriel Goh; Rajan Troll; Alec Radford; Ilya Sutskever; Jan Leike; Jeffrey Wu” year: 2024 title: “Scaling and evaluating sparse autoencoders” doi: null claim: “k-sparse (TopK) autoencoders control sparsity directly and admit scaling laws in dictionary size.” strata_touched: [] license: “citation-only” triage: “anchor” url: “https://arxiv.org/abs/2406.04093”

Scaling and evaluating sparse autoencoders

Verified locator

arXiv:2406.04093. Used only for the definition of the TopK objective; the fiber-law volume’s TopK statement concerns an ideal single-atom encoder, not the implemented amortized encoder.

Declared identifiers: https://arxiv.org/abs/2406.04093.