Where Steering Signals Come From: Activation Source Selection in Activation Steering
cs.CL, cs.AI, cs.LG
Submitted: 2026-07-28
Updated: 2026-08-28
Terminology
Sources
- Improving Activation Steering in Language Models with Mean-Centring
- Contrastive Principal Component Analysis
- Understanding intermediate layers using linear classifier probes
- Improving Steering Vectors by Targeting Sparse Autoencoder Features
- Persona Vectors: Monitoring and Controlling Character Traits in Language Models
- Llama Scope: Extracting Millions of Features from Llama-3.1-8B with Sparse Autoencoders
- SAIF: A Sparse Autoencoder Framework for Interpreting and Steering Instruction Following of Language Models
- Steering Large Language Models using Conceptors: Improving Addition-Based Activation Engineering
- Steering Language Models With Activation Engineering
- Refusal Direction is Universal Across Safety-Aligned Languages
- Qwen2.5 Technical Report
- Spherical Steering: Geometry-Aware Activation Rotation for Language Models
- Gemma 3 Technical Report
- Representation Engineering: A Top-Down Approach to AI Transparency
- The Llama 3 Herd of Models
- Gemma 2: Improving Open Language Models at a Practical Size
Related papers
- Exploring Solution Divergence and Its Effect on Large Language Model Problem Solving
- Ishigaki-IDS-Bench: A Benchmark for Generating Information Delivery Specification from BIM Information Requirements
- Subliminal Steering: Stronger Encoding of Hidden Signals
- MedStruct-S: A Benchmark for Key Discovery, Key-Conditioned QA and Semi-Structured Extraction from OCR Clinical Reports
- The End of Transformers? On Challenging Attention and the Rise of Sub-Quadratic Architectures
- Untangling the Mechanisms of Misleading Context in Medical Question Answering