Portrait of Matt Suiche

Matt Suiche

AI Security

2026

The Teacher's Accent: Fingerprinting DeepSeek V4.1, Then Stress-Testing Anthropic's Model Diff Tool
Model Diffing Crosscoder DFC Distillation Refusal Direction DeepSeek Activation Steering Mechanistic Interpretability
The Stream, Its Writers, and the Assumption in Appendix E
Activation Steering Abliteration Control Vectors GLP Refusal Direction LoRA Weightless Mechanistic Interpretability
Autoresearch: Abliteration Without Redistributing the Model
Activation Steering Abliteration Control Vectors Refusal Direction Mixture of Experts LoRA Model Alignment GGUF Mechanistic Interpretability