Effective concept steering needs curved trajectories, not one linear direction

measured in 1 paper

Jin et al. note prior activation steering assumes a single fixed-direction additive shift, and replace it with FLAS, a concept-conditioned velocity field integrated over frozen model activations [jin-etal-2026-flow-based-activation-steering-inference-time-intervention] The learned steering trajectories are curved and token-varying: every concept leaves the origin in a shared direction, bends sharply, then enters a concept-specific region [jin-etal-2026-flow-based-activation-steering-inference-time-intervention] This two-phase nonlinear shape directly challenges the single-linear-direction assumption underlying standard steering [jin-etal-2026-flow-based-activation-steering-inference-time-intervention] FLAS outperforms prompting and prior steering (held-out HMean 1.015 vs 0.762/0.608 on Gemma-2-2B-IT) at under 1/26th the parameters and transfers zero-shot to Qwen3-4B-Instruct [jin-etal-2026-flow-based-activation-steering-inference-time-intervention]

Structure

Linear Direction

Context

steering intervention trajectories that are curved rather than straight lines through representation space, a shared-direction-then-bend two-phase trajectory structure common across concepts, a learned nonlinear velocity field replacing a fixed additive steering direction

Confirmed in models

Gemma-2-2B-it · Gemma, Gemma-2-9B-it · Gemma, Qwen3-4B-Instruct · Qwen

Method

Flow-based Activation Steering (FLAS)

Papers

Beyond Steering Vector: Flow-based Activation Steering for Inference-Time Intervention — Jin, Zehao, Deng, Ruixuan, Wang, Junran, Shen, Xinjie, Zhang, Chao2026 · arXiv:2605.05892