Part 16 — Where to go next
You've completed the core path. Parts 1–15 took you from IF-THEN rules to a working, guarded agent — the full arc this guide set out to teach. This part is a checkpoint, not a chore: an honest look at what you know, what's still missing, and where to go next. It's worth pausing here before reading on.
Five more parts follow this one (17–21), and they are optional deep-dives, not a continuation of the core arc: model compression, agent frameworks and governance, the internals behind a few things you used without deriving, agents in production (harnesses, long-running loops, evals, deployment), and the newest model designs beyond next-token prediction (decision models and world models). Read whichever one matches something you actually need, in any order, or skip them entirely — nothing past this point is required to say you've finished the guide.
Chapter 63 — What you know now, and the honest gaps
You can now
- Explain the whole arc from rule-based AI to LLMs, and why each step happened.
- Explain attention, tokens, pretraining, fine-tuning, inference, and context.
- Train a language model from scratch and fine-tune an open one.
- Build and evaluate a RAG application.
- Reason about cost, latency, and quality trade-offs.
That is genuinely more than most people working near AI understand.
Honest gaps this guide left
| Gap | Where to fill it |
|---|---|
| The maths in depth (linear algebra, probability, optimisation) | Mathematics for Machine Learning; 3Blue1Brown; Stanford CS229 |
| Computer vision, CNNs, diffusion models | fast.ai course; Hands-On Machine Learning part 2 |
| Multimodal models (vision + text + audio) | recent papers; the model cards of current multimodal models |
| Reinforcement learning properly | Sutton & Barto (free online); Spinning Up in Deep RL (OpenAI) |
| Distributed training at scale | Megatron-LM / torchtitan docs; the "Ultra-Scale Playbook" (HF) |
| Serving infrastructure at scale | vLLM docs and blog; TensorRT-LLM |
| Interpretability (what's inside the model) | Anthropic's transformer-circuits.pub; distill.pub |
| AI safety and alignment research | Anthropic/DeepMind/OpenAI alignment blogs; AI Safety Fundamentals course |
| Classical ML depth (trees, boosting, tabular) | Hands-On Machine Learning; Kaggle courses |
| Agent frameworks (Parts 12–15 teach the raw loop, not LangGraph/CrewAI/SDKs) | Each framework's own concepts docs — read them after building the loop yourself |
| Formal AI red-teaming and offensive testing | OWASP LLM Top 10; PortSwigger LLM labs; Gandalf/Lakera exercises |
| Compliance and governance (EU AI Act, model cards, audits) | Your regulator's guidance; NIST AI Risk Management Framework |
| Agents in production: harnesses, hours-long loops, reliability evals, deployment | Part 20 of this guide, then the Anthropic Engineering blog and your agent platform's docs |
| Decision models, calibration, and world models (JEPA) | Part 21 of this guide, then the V-JEPA 2 and LeJEPA papers |