AI Paper Atlas

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

Steer Like the LLM: Activation Steering that Mimics Prompting

Geert Heyman, Frederik Vandeputte

ICML 2026 · Large language models › Language, safety and interpretability · approach: Mechanistic analysis of attention and representations

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
add token-specific steering coefficients (PSR) to single-layer MSE steering (ablation)S-Const MSE|QR · Persona Vectors, Llama-3.2-3BTA@Cp: 57.4 -> 66.8Improved
add token-specific steering coefficients (PSR) to single-layer LL steering (ablation)S-Const LL|QR · Persona Vectors, Llama-3.2-3BTA@Cp: 42.9 -> 52.6Improved
add token-specific coefficients (PSR) to all-layer MSE steering (ablation)A-Const MSE|QR · Persona Vectors, Llama-3.2-3BTA@Cp: 95.8 -> 92.5Hurt
add token-specific coefficients (PSR) to all-layer MSE steering (ablation)A-Const MSE|QR · Persona Vectors, Llama-3.1-8BTA@Cp: 91.3 -> 96.4Improved
use MSE instead of loglikelihood objective (ablation)S-PSR|QR · Persona Vectors, Llama-3.1-8BTA@Cp: 45.0 -> 74.7Improved
use loglikelihood instead of MSE objective (ablation)S-PSR · IFEval format, Gemma-2-9b-itIF accuracy: 47.5 -> 66.1Improved
add token-specific coefficients (PSR) to single-layer LL steering (ablation)S-Const LL · IFEval format, Gemma-2-9b-itIF accuracy: 13.4 -> 66.1Improved
steer all layers instead of a single layer (ablation)PSR LL · IFEval format, Gemma-2-9b-itIF accuracy: 66.1 -> 71.9Mixed
use PSR MSE instead of constant MSE on AxBench (ablation)A-Const MSE · AxBench Gemma-2-2B L20steering score: 0.783 -> 0.871Improved
use PSR MSE instead of constant MSE on AxBench (ablation)A-Const MSE · AxBench Gemma-2-9B L20steering score: 1.053 -> 1.120Improved
use PSR LL instead of constant LL on AxBench (ablation)A-Const LL · AxBench Gemma-2-2B L20steering score: 0.792 -> 0.690Hurt
use MSE instead of LL for single-layer PSR on AxBench (ablation)S-PSR · AxBench Gemma-2-9B L20steering score: 0.667 -> 0.900Improved

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