Why directional ablation fails to detect refusal vector rotation after fine-tuning, and how calibration decouples estimate drift from safety.
Research. Experiment. Build. Repeat
I'm an AI Engineer and Researcher passionate about understanding, building, and advancing intelligent systems. My work spans Large Language Models (LLMs), reinforcement learning, deep learning, computer vision, and AI engineering. I enjoy exploring how models learn, designing experiments, investigating research questions, and translating ideas from papers into practical implementations. Here, I share research insights, technical deep dives, experiments, implementation guides, and lessons learned while building AI systems. * My goal: Bridge the gap between AI research and real-world engineering through rigorous experimentation, reproducible work, and open knowledge sharing.
Markdown supported · ⌘↵ to post
No responses yet.