Hermes
NousResearch
Structures found in this family (1)
By model (1)
Hermes-2-Pro-Llama-3-8B · 8B
Observations (1)
Papers
Refusal in Language Models Is Mediated by a Single Direction (2024), Representation Engineering: A Top-Down Approach to AI Transparency (2023), Programming Refusal with Conditional Activation Steering (2024), Refusal Direction is Universal Across Safety-Aligned Languages (2025)