Distill AI
AI is the teacher; I’m the student.
I first explain an idea in my own words, then ask AI to identify gaps in my understanding and offer clearer explanations. I think of this exchange as a kind of on-policy distillation, where my own reasoning guides what we explore next.
These notes are mostly written by AI, guided by my questions and evolving understanding. They help me build a holistic view of different subjects and the connections between them. They give structure to my understanding.
My interests
- Robotics
- Dexterous manipulation
- Reinforcement learning
- Model-based RL
- World models
- Generative models
- Multimodal foundation models
- Post-training
- Optimization
- Kinematics
- Control theory
- Geometric vision algorithms
And the connections between them.