Resolve the conflicting reported SAE steering settings

Resolve the discrepancy between the reported layer-and-strength settings and the conflicting prose description for SAE difference-in-means steering in the CLaS-Bench language-steering evaluation.

Background

The survey compares residual-space and SAE-based difference-in-means interventions for multilingual output-language steering on CLaS-Bench. It records selected settings using layers 20 and 25 with intervention strength 5, but notes that the source prose instead reports an SAE peak at strength 15.

Because the inconsistency concerns the experimental configuration, it prevents an unambiguous interpretation of the reported comparison and leaves the relevant source-level setting unresolved.

References

The reported selected settings use layers 20/25, respectively, with strength 5; the source prose instead gives an SAE peak at strength 15. This discrepancy remains unresolved.

— Sparse Autoencoders for Mechanistic Interpretability in NLP: A Critical Survey of Recoverability, Interpretability, and Utility  (2610.20139 - Lamsiyah et al., 14 Oct 2026) in Appendix E.1, source-level evidence case E10; Appendix F.2