AI
How modern AI actually works — from the math under the hood to the engineering that makes it run. Each chapter is a self-contained area; each topic inside is a single focused lesson on the concept, the math, and the code.
Foundations
The terms, ideas, and learning paradigms the rest of the field assumes you already understand. Start here if any vocabulary further in feels unfamiliar.
The Transformer
The architecture that swallowed the field. Attention, tokenization, positional encoding, KV cache — and what makes the same model handle text, code, images, and audio.
Training Large Models
How a model that can do anything is actually built. Pre-training at scale, the laws that predict what bigger gets you, and the fine-tuning and RL steps that turn next-token prediction into a useful assistant.
Inference
Turning a trained model into something a user can actually use. Decoding, sampling, quantization, caching, and the systems engineering that makes responses fast and cheap.
Prompting & Context
Everything you do at inference time to get more out of a model. The structure of a good prompt, the surprises of in-context learning, and the failure modes that emerge when context gets long.
Retrieval-Augmented Generation
Letting a model answer with knowledge it wasn't trained on. Embeddings, vector search, the chunking choices that decide whether the right passage shows up, and the pipelines that wire it all together.
Agents & Tool Use
What changes when a model can call functions, browse the web, run code, and chain decisions across hundreds of turns. The patterns, the harnesses, and the failure modes.
Multimodal & Generative Models
Beyond text — the diffusion models, VAEs, and architectures that generate images, video, audio, and 3D. And how the same ideas come back as vision-language models and world models.
Evaluation
How do you grade a model that can do almost anything? The benchmarks, the methodology, the metrics — and why every claim about model capability deserves scrutiny.
Safety & Alignment
The hardest unsolved problem in the field. Interpretability, red-teaming, jailbreaks, scaling policies — and the open questions about systems we don't yet fully understand.