How AI actually works.
Nine interactive explainers for legal professionals. No jargon, no hype — the machine, its habits, and how to use it safely.
The series reads best in order — each page builds on the one before. But every explainer stands alone, so start wherever the question is burning.
How Transformers Work
The machine underneath every modern AI: tokens, attention, and a probability over the next word. Start here.
The Context Window
What the model can actually see, why long chats drift, and why a bigger window is a bundle, not a library.
Why AI Makes Things Up
Fabricated cases, invented citations, confident nonsense — why it happens, and the workflow that protects you.
Grounding & RAG
The most effective reliability upgrade there is: stop asking what the model remembers and show it the documents.
Prompting Opus 4.8 for Legal Work
Brief it like a junior: role, context, task, format — and a live prompt builder for common legal tasks.
Where Your Data Goes
Consumer chat, enterprise endpoint or local model: what happens to a privileged document when you press send.
What Is an AI Agent?
Chatbots answer; agents act. The think–act–observe loop, why errors compound, and how to delegate safely.
Choosing the Right Model
Families, tiers and effort levels — and a matcher for picking the right tool for each legal task.
How Models Are Trained
Pretraining, fine-tuning and human feedback: the upbringing that explains every habit and quirk.
Prefer it in person?
Everything in this series — and everything it leaves out — delivered as bespoke training for your chambers or firm.
Lextrapolate AI Training