A single-token feature's causal necessity under zero-ablation depends on which SAE family found it: GemmaScope and BatchTopK features stay causally anchored while LlamaScope features are locally redundant.
Title resolution pending
1 Pith paper cite this work, alongside 310 external citations. Polarity classification is still indexing.
1
Pith paper citing it
310
external citations · OpenAlex
fields
cs.LG 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Are Single-Token Sparse Autoencoder Features Causally Necessary? Layer-Depth and SAE-Family Effects
A single-token feature's causal necessity under zero-ablation depends on which SAE family found it: GemmaScope and BatchTopK features stay causally anchored while LlamaScope features are locally redundant.