Component and Dimension Sparsity in Transformer Refusal Mechanisms
cs.CL, cs.AI, cs.LG
Submitted: 2026-09-30
Updated: 2026-09-30
Code: https://github.com/wangresearch-lab/Refusal
Terminology
Sources
- Phi-4 Technical Report
- Jailbreaking Leading Safety-Aligned LLMs with Simple Adaptive Attacks
- Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computation
- Jailbreaking Black Box Large Language Models in Twenty Queries
- Towards Automated Circuit Discovery for Mechanistic Interpretability
- Toy Models of Superposition
- Localizing Model Behavior with Path Patching
- The Llama 3 Herd of Models
- Finding Neurons in a Haystack: Case Studies with Sparse Probing
- How to use and interpret activation patching
- Spectral Superposition: A Theory of Feature Geometry
- Programming Refusal with Conditional Activation Steering
- LoRA Fine-tuning Efficiently Undoes Safety Training in Llama 2-Chat 70B
- Does Circuit Analysis Interpretability Scale? Evidence from Multiple Choice Capabilities in Chinchilla
- Refusal in LLMs is an Affine Function
- Locating and Editing Factual Associations in GPT
- Steering Llama 2 via Contrastive Activation Addition
- RepIt: Steering Language Models with Concept-Specific Refusal Vectors
- Gemma 2: Improving Open Language Models at a Practical Size
- Steering Language Models With Activation Engineering
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