Alignment is the problem of making AI systems reliably pursue what we actually intend, including when they are more capable than the people overseeing them. This guide covers the field at research depth: the threat models, the empirical results (much of it from Anthropic), the methods and their limits, and the open problems as of late 2026.
Each idea arrives in the same order: intuition in one sentence → an analogy (and where it breaks) → an interactive you can poke → the evidence (setups, numbers, caveats) and the math where there is math → pitfalls → what the frontier says → the questions worth asking about it. Every chapter ends with flashcards and a verified reading list, and chapters 00–09 also end with a "Check yourself" set of practice questions.
Full path (≈ 3–4 weeks at 1–2 h/day): chapters in order.
Short path: 00 → 01 → 03 → 04 → 06 → 08 → 10, then skim the "In one breath" box and the "Check yourself" questions of every other chapter.
Companion: The Post-Training Field Guide covers the training mechanics these chapters build on: SFT, reward models, RLHF, DPO, RL with verifiable rewards, character training.
Works offline: every library (D3, GSAP, KaTeX) ships with the repository. Your progress (chapters read) is stored only in this browser's localStorage.