Robustness of SAE Features Under Steering

M. VogelJanuary 20, 2026
BlogpostSAEsSteeringNegative-results

TLDR

Negative result showing many SAE features drift under moderate fine-tuning and steering interventions.

Lorem ipsum dolor sit amet, consectetur adipiscing elit. Ut tincidunt, libero a mattis luctus, ligula mauris volutpat urna, sit amet cursus nibh sapien quis lectus.

Aliquam erat volutpat. Donec tristique odio non leo pretium, nec pharetra sem pharetra.

Author Note

Example artifact page (offline demo mode).

Comments

Comments are not available on example artifacts.

Related artifacts