AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Fine-Grained Activation Steering: Steering Less, Achieving More

Zijian Feng, Tianjiao Li, Zixiao Zhu, Hanzhang Zhou, Junlang Qian, Li Zhang, Chua Jia Jim Deryl, Mak Lee Onn, Gee Wah Ng, Kezhi Mao

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Task vectors and steering · 1 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use localized AU-level steering (FFN) instead of SADI head steeringGemma2-9B-it · Commonsense + math average accuracyaccuracy avg: 82.05 -> 83.96Improved
AUSteer-FFN vs SADILLaMA2-7B-Chat · Commonsense + math average accuracyaccuracy avg: 59.49 -> 61.34Improved
AUSteer-Head vs SADIQwen3-8B · Commonsense + math average accuracyaccuracy avg: 80.80 -> 83.21Improved
AUSteer on RealToxicPrompts vs SADIQwen3-8B · RealToxicPromptsdetox: 78.07 -> 81.65Improved
AUSteer-FFN on BPO vs SADIGemma2-9B-it · BPOdelta WR: 25.50 -> 30.00Improved
AUSteer-Head vs SADI on detoxGemma2-9B-it · RealToxicPromptsdetox: 99.08 -> 99.00No clear effect
use activation-momentum localization instead of random localization (ablation)AUSteer-FFN Gemma2-9B-it · Commonsense + math average accuracyaccuracy avg: 79.08 -> 83.96Improved
use activation-momentum localization instead of activation difference (ablation)AUSteer-FFN Gemma2-9B-it · Commonsense + math average accuracyaccuracy avg: 83.12 -> 83.96Improved
use adaptive steering vector instead of fixed vector (ablation)AUSteer-FFN Gemma2-9B-it · Commonsense + math average accuracyaccuracy avg: 82.05 -> 83.96Improved
use adaptive strength instead of fixed strength (ablation)AUSteer-FFN Gemma2-9B-it · Commonsense + math average accuracyaccuracy avg: 83.04 -> 83.96Improved
steer more than 5,000 AUs instead of fewer than 100 (ablation)AUSteer LLaMA2-7B-Chat · COPAaccuracy: 82.8 -> 70.4Hurt
steer all 131,072 AUs (ablation)AUSteer LLaMA2-7B-Chat · commonsenseaccuracyHurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy