Representation-Engineering

Activation Steering & Steering Vectors

Title	Link	Brief	Pipeline
Steering Llama 2 via Contrastive Activation Addition	paper code	CAA
Steering Language Models With Activation Engineering	paper code	ActAdd
Representation Engineering: A Top-Down Approach to AI Transparency	paper code	First Work for RepE
In-context Vectors: Making In Context Learning More Effective and Controllable Through Latent Space Steering	paper code	ICV
Steering Language Model Refusal with Sparse Autoencoders	paper	SAE
Personalized Steering of Large Language Models: Versatile Steering Vectors Through Bi-directional Preference Optimization	paper code	BiPO
AxBench: Steering LLMs? Even Simple Baselines Outperform Sparse Autoencoders	paper code	Benchmark
Improving Instruction-Following in Language Models through Activation Steering	paper code
Model Surgery: Modulating LLM's Behavior Via Simple Parameter Editing	paper code
Towards Inference-time Category-wise Safety Steering for Large Language Models / SafeSteer: Interpretable Safety Steering with Refusal-Evasion in LLMs	paper paper
Steering Away from Harm: An Adaptive Approach to Defending Vision Language Model Against Jailbreaks	paper code	VLM
SAIF: A Sparse Autoencoder Framework for Interpreting and Steering Instruction Following of Language Models	paper	SAE
Surgical, Cheap, and Flexible: Mitigating False Refusal in Language Models via Single Vector Ablation	paper code	Single-vector steering
MASteer: Multi-Agent Adaptive Steer Strategy for End-to-End LLM Trustworthiness Repair	paper	Agent
Steering Large Language Model Activations in Sparse Spaces	paper	SAE
Learning to Steer: Input-dependent Steering for Multimodal LLMs / Analyzing Fine-tuning Representation Shift for Multimodal LLMs Steering	paper paper code	MLLM
Multi-Attribute Steering of Language Models via Targeted Intervention	paper code	通过 attribute 减少“过度纠正”
Steering Large Language Models using Conceptors: Improving Addition-Based Activation Engineering	paper code	Conceptor Steering
Steering MoE LLMs Via Expert (De)Activation	paper code	MoE
Mitigating Overthinking in Large Reasoning Models via Manifold Steering	paper	缓解推理模型的“overthinking“现象
SAE-SSV: Supervised Steering in Sparse Representation Spaces for Reliable Control of Language Models	paper	SAE
REAL: Reading Out Transformer Activations for Precise Localization in Language Model Steering	paper	Attention
Probing the Robustness of Large Language Models Safety to Latent Perturbations	paper code	Probe
SARSteer: Safeguarding Large Audio Language Models via Safe-Ablated Refusal Steering	paper	AudioLLM
HyperSteer: Activation Steering at Scale with Hypernetworks	paper code	Hyper
Denoising Concept Vectors with Sparse Autoencoders for Improved Language Model Steering	paper	SDCV
KV Cache Steering for Controlling Frozen LLMs	paper	KV-Cache