I prepared these notes for myself to understand core AI concepts easily. I find it easier when maths is accompanied by visual examples. So I drafted these with Claude and then went through each one to fix the parts that were wrong or unclear.
Policy gradients, value-based methods, and the full post-training pipeline. PPO, DPO, GRPO, Q-learning, and …
Transformer internals and the ideas that make large language models work. Attention, positional encoding, …
The mechanics underneath every training run. Optimizers, loss landscapes, gradient flow, and the tricks that …
How to build models that generate. Diffusion from first principles — DDPM, score matching, SDEs, DDIM — and …