Continuous Function

Legacy Concept Lab

Activation Steering: Feature-Guided Interventions for Inference-Time Control

Inference-time control without fine-tuning—enforce format/length/constraints, style shifts, reduce/induce refusals at deployment

Concept 29 of 100RepresentationsPhase 5
Open DemoAll Foundations
#29Activation SteeringRepresentations
key equation
hℓ,tsteer=hℓ,t+αvℓ,th_{\ell,t}^{\text{steer}} = h_{\ell,t} + \alpha v_{\ell,t}
Phase 5: Representation & interpretabilityConcept 29 of 100

Why It Matters for Modern Models

  • Inference-time control without fine-tuning—enforce format/length/constraints, style shifts, reduce/induce refusals at deployment
  • Steering is causal test—if adding feature k induces behavior, you've localized mechanism and can validate circuits (#28)
  • Minimal interventions trend: activation scaling learns sparse scalars to strengthen/weaken existing directions, more interpretable than dense edits
  • After SAEs (#27) and circuit discovery (#28), steering makes interpretability actionable—debugging & control, not just analysis
  • Bridge to safety: controllable refusal/style/format constraints as post-training knobs, without expensive retraining loops

What Tutorials Skip

What is still poorly explained in textbooks and papers:

  • Steering is geometry on model's manifold—small α nudges within-distribution, large α throws you off-manifold → incoherence and capability loss
  • Interpretable steering isn't just what direction—it's which basis: SAE features give human handles, but decoder can entangle effects
  • Compositionality isn't guaranteed—adding two "instruction vectors" can cancel or amplify depending on where they write in residual space
  • Strength-capability tradeoff is fundamental—too weak has no effect, too strong breaks generation quality
  • Steering reveals what model "knows"—if you can steer to behavior, the capability exists in the weights, just not naturally expressed

Interactive Visualization

Core Math (Optional Deep Dive)

If you want intuition first, start with the key equation and the visualization. Come back here for the full walkthrough.

Key Equation
hℓ,tsteer=hℓ,t+αvℓ,th_{\ell,t}^{\text{steer}} = h_{\ell,t} + \alpha v_{\ell,t}

Activation steering edits hidden states (or SAE latents) to change behavior without retraining, turning interpretability into control knobs for inference-time behavioral modification.

Classic steering vector addition (one layer/position):

hℓ,tsteer=hℓ,t+αvℓ,th_{\ell,t}^{\text{steer}} = h_{\ell,t} + \alpha v_{\ell,t}

where α\alpha controls strength and vℓ,tv_{\ell,t} is the steering direction.

Contrastive steering vector (difference of means):

vℓ,t=E[hℓ,t∣desired]−E[hℓ,t∣undesired]v_{\ell,t} = \mathbb{E}[h_{\ell,t}\mid \text{desired}] - \mathbb{E}[h_{\ell,t}\mid \text{undesired}]

Core idea behind "instruction vectors" and CAA-style methods—find direction that separates desired/undesired behaviors.

SAE-latent steering (feature toggle → decode back):

z=fenc(hℓ,t),hℓ,tsteer=fdec ⁣(z+δek)z = f_{\text{enc}}(h_{\ell,t}), \quad h_{\ell,t}^{\text{steer}} = f_{\text{dec}}\!\left(z + \delta e_k\right)

with eke_k a basis vector selecting feature kk—interpretable steering via learned feature directions.

Canonical Papers

Activation Scaling for Steering and Interpreting Language Models

Stoehr et al.2024EMNLP Findings
Read paper →

Improving Instruction-Following in Language Models through Activation Steering

Stolfo et al.2025ICLR (arXiv)
Read paper →

Feature Guided Activation Additions

Soo et al.2025OpenReview
Read paper →

Connections

Prerequisites

⊗Attention◎Embeddings🔍Sparse Autoencoders🔬Circuit Discovery

Next Moves

Explore this concept from different angles — like a mathematician would.

Semantic Connections

⚠️ Breaks When
🔍Feature entanglement→Sparse Autoencoders
🔄 Same Technique
◎Linear directions→Embeddings
🔧 Invented to Fix
🔍Targeted intervention→Sparse Autoencoders🔬Inference interventions→Circuit Discovery
≈ Analogy
🎲Inference-time control→Decoding