AI Paper Atlas

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

Steering at the Source: Style Modulation Heads for Robust Persona Control

Yoshihiro Izawa, Gouki Minegishi, Koshi Eguchi, Sosuke Hosokawa, Kenjiro Taura

ICML 2026 · Large language models › Language, safety and interpretability · approach: Task vectors and steering · 0 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
steer only Style Modulation Heads instead of MLP residual streamQwen2.5-7B humorous, Neutral+alpha · Pareto score tau=80trait score at coherency>=80: 17.8 -> 60.7Improved
steer only Style Modulation Heads instead of Attn ResidualQwen2.5-7B humorous, Neutral+alpha · Pareto score tau=80trait score at coherency>=80: 29.6 -> 60.7Improved
steer only Style Modulation Heads instead of Attn ResidualLlama-3.1-8B evil, Neutral+alpha · Pareto score tau=80trait score at coherency>=80: 41.4 -> 29.4Hurt
steer only Style Modulation Heads instead of Attn ResidualLlama-3.1-8B hallucination, Target-alpha · Pareto score tau=80trait score at coherency>=80: 50.5 -> 91.2Improved
steer only Style Modulation Heads instead of Attn ResidualQwen2.5-7B hallucination, Target-alpha · Pareto score tau=80trait score at coherency>=80: 42.8 -> 73.1Improved
also steer anti-correlated heads (ablation)Style Modulation Head steering, Qwen2.5-7B hallucination Neutral+alpha · Pareto score tau=80trait score at coherency>=80: 69.0 -> 36.5Hurt
also steer anti-correlated heads (ablation)Style Modulation Head steering, Llama-3.1-8B evil Target-alpha · Pareto score tau=80trait score at coherency>=80: 99.9 -> 13.2Hurt
also steer anti-correlated heads (ablation)Style Modulation Head steering, Llama-3.1-8B passionate Neutral+alpha · Pareto score tau=80trait score at coherency>=80: 94.5 -> 55.0Hurt
zero-ablate high-contribution heads in layers 20, 15, 19 (ablation)Qwen2.5-7B persona expression · trait scoretrait scoreImproved
zero-ablate high-contribution heads (ablation)Qwen2.5-7B · coherency and MMLUcoherency, MMLUNo clear effect
steer the Style Modulation Heads (Head Cor) vs other positions across personasQwen2.5-7B and Llama-3.1-8B · 12 conditions per modelhighest Pareto score count: 11 cases for Qwen2.5-7B and 9 cases for Llama-3.1-8BImproved
increase steering magnitude in the positive directionresidual stream steering · coherencycoherency scoreHurt

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