A position paper proposing feature consistency, measured by PW-MCC, as a core SAE evaluation criterion, with evidence that TopK SAEs achieve high consistency on LLM activations.
Turner, Cem Anil, Carson Denison, Amanda Askell, et al
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Position: Mechanistic Interpretability Should Prioritize Feature Consistency in SAEs
A position paper proposing feature consistency, measured by PW-MCC, as a core SAE evaluation criterion, with evidence that TopK SAEs achieve high consistency on LLM activations.