Published on February 15, 2026 5:49 PM GMTWe introduce an automated activation‑steering approach that plugs into standard labeled datasets—no handcrafted prompt pairs or feature annotation. On 18 tasks and 3 open‑weight models, the introspective variant (iPAS) yields the strongest behavior improvements, and layers on top of ICL/SFT.Full write‑up: https://open.substack.com/pub/sashacui/p/painless-activation-steering-pasPaper: arxiv.org/abs/2509.22739 Discuss Read More
Painless Activation Steering
Published on February 15, 2026 5:49 PM GMTWe introduce an automated activation‑steering approach that plugs into standard labeled datasets—no handcrafted prompt pairs or feature annotation. On 18 tasks and 3 open‑weight models, the introspective variant (iPAS) yields the strongest behavior improvements, and layers on top of ICL/SFT.Full write‑up: https://open.substack.com/pub/sashacui/p/painless-activation-steering-pasPaper: arxiv.org/abs/2509.22739 Discuss Read More
Related Posts
Going Phoneless
Published on October 7, 2025 4:40 PM GMTI’ve gone Phoneless. I still own an iPhone…
[Paper] Does Self-Evaluation Enable Wireheading in Language Models?
Published on December 8, 2025 4:03 PM GMTTL;DR: We formalized and empirically demonstrated wireheading in…
A Proposal for a Better ARENA: Shifting from Teaching to Research Sprints
Published on January 10, 2026 4:56 PM GMTTLDRI propose restructuring the current ARENA program, which…