Security of LLM Activation Functions and Architecture
Research on how architectural components like activation functions affect safety and security properties of LLMs
Security of LLM Activation Functions and Architecture
Research on Large Language Models in Security of LLM Activation Functions and Architecture
Hidden Dangers in LLM Optimization
How activation approximations compromise safety in aligned models
Multi-Dimensional Safety in LLM Alignment
Revealing the hidden complexity of safety mechanisms in language models
The Geometry of LLM Refusals
Uncovering Multiple Refusal Concepts in Language Models
Unlocking Precise Control of AI Behavior
Sparse Activation Steering: A New Approach to LLM Alignment
