Working with LLMs from Zero
by Sergey Okinchuk

Working with LLMs from Zero

Welcome

You have talked to an AI chat box. Maybe it impressed you, maybe it confidently lied to you — probably both. This course teaches you to work with large language models properly, from the very first prompt to reading the request underneath the chat box: what the model actually is, how to ask so you get what you need, how to catch it when it is wrong, what your documents and data are doing in there, what must never be pasted in — and then the part most courses skip, the machinery under the surface: the request, the tokens, the API key, the errors, and the bill. No programming required at any point. By the end, the chat box will never look the same.

≈10.5 hours to complete Knowledge check on every topic

About This Course

Everyone is suddenly expected to "use AI at work," and almost nobody has been told how it works. The advice on offer splits into hype ("this changes everything!") and tricks ("ten magic prompts!"), and both leave you exactly where you started: typing into a box you do not understand, trusting or distrusting it on vibes. This course takes the third path. It explains the machine honestly, at a beginner's pace, and turns the explanation into working habits — because every practical skill with these tools, from asking well to catching errors to understanding the bill, follows from a handful of plain facts about what is actually happening.

The course follows Tessa, marketing coordinator at Waymark, a regional travel agency, through one working season with these tools: from her first pasted prompt to templates her whole team reuses, through a hallucinated hotel that nearly reached a client, to the day a developer turns his laptop around and shows her the request every chat box has been hiding — and, by the end, to automation she commissioned herself, with her own checks on its output. Her arc is the course's promise: you do not need to be technical to end up genuinely in command of this tool.

Who This Is For

Anyone who works in text — emails, documents, reports, reviews, plans — and wants to use language models properly: marketers, coordinators, analysts, managers, support staff, owners of small teams. Students choosing what to learn. And anyone who has been told "just use AI" and privately wondered what, exactly, they are trusting. No technical background is assumed at any point: every term is defined the first time it appears, and the one script in the whole book is read aloud in plain words, not written.

It is not a course about how models are built — training, neural networks, the generative family — which is its own book in this catalogue: Machine Learning from Zero, a natural companion to this one in either order. And it is not a course about building software around models — agents, tools, evaluation — which assumes exactly what this course teaches and is the natural next step after it: Agentic AI from Scratch.

What You Should Already Know

  • How to use a computer, a browser, and everyday apps — nothing more technical than that
  • No programming and no command line — the one program in this book is read to you in plain words
  • No prior AI knowledge — the chat box is explained from zero, and every term is defined at first use
  • A healthy curiosity about what is actually behind the tools you are told to trust

How the Course Is Built

The eleven chapters run in five movements. What you are talking to (Chapter 1) is the honest introduction: the model behind the box, tokens, the context window, why answers vary, and what the model cannot know. The conversation (Chapters 2–3) is the craft: asking well, and catching the model when it is confidently wrong. The work (Chapters 4–6) puts your real materials through it: documents, usable output, and the hard line around what must never be pasted. Under the chat box (Chapters 7–8) is the reveal most courses skip: the request underneath, the key, the errors, the bill, and the model inside every product you use. Application (Chapters 9–11) faces outward: AI that writes code, when not to use a model at all, and where to go next.

Every topic has the same gentle shape: a hook from Tessa's world, the idea built step by step, an everyday comparison wherever one earns its place, the mix-ups people usually run into, why it matters, and a short knowledge check. Patient, but it keeps moving — you are here to learn, not to be slowed down.

The machine, not the magic
Every skill is taught from the mechanism it follows from. Once you know the model produces text that fits, prompting stops being tricks; once you know the window is finite, long-chat weirdness stops being mysterious.
One company, all the way through
Waymark's problems accumulate chapter by chapter — the reviews, the contract, the invented hotel, the first bill — so every new idea lands on a story you already know.
Honest, never hyped
What these tools genuinely do, how often they fail, and what they cost — stated plainly. The invented hotel stays in the book as a permanent reminder. Nothing here will need unlearning.
Mechanisms over products
No product names, no model versions, no prices that expire. Tokens, windows, requests, verification — the parts that will still explain the AI news five years from now.

Chapter Map

Chapter 1
What You Are Talking To
The chat box is a window; the model is elsewhere. Tokens, the context window, why the same question gets different answers, and the three limits hiding behind one confident voice.
Chapter 2
Asking Well — The Craft of Prompting
The four specifics that kill generic answers, context only you can supply, examples that outteach descriptions, formats and templates, and iteration — the craft, without the mysticism.
Chapter 3
When the Model Is Wrong
A hotel that does not exist, recommended in a warm confident voice. Hallucination, why fluency is not truth, the cheap habits of verification, the model that agrees with you, and the average answer.
Chapter 4
Giving It Your Documents
Paste or attach, the 120-page contract against the window, the three workhorse tasks — summarize, extract, classify — where numbers stumble, and the search-plus-context idea behind a whole product category.
Chapter 5
Getting Usable Output
From answers you read to artifacts you use: output that lands in the tracker, text a program can read, templates that make results stack, and checking at scale — because tidy tables hallucinate too.
Chapter 6
What Not to Paste — Privacy and Safety
Where your words go, the three never-paste categories, the redaction habit that unlocks most tasks anyway, why company AI rules exist, and what a refusal actually means.
Chapter 7
The Request Under the Chat Box
The reveal: every message is one request to a provider's server. HTTP in plain words, the payload you can read, the key that spends money, errors and retries, and the bill — explained line by line.
Chapter 8
The Model Inside a Product
Every AI feature is a request wearing clothes: the system prompt, the settings dials, Tessa's twenty-line script read aloud in plain words — and what to check when the product breaks.
Chapter 9
AI That Writes Code
Coding assistants honestly: why vibe coding genuinely flies at first, what the third week costs, the four disciplines that keep the speed — and what an agentic tool actually is.
Chapter 10
When Not to Use an LLM
The three weak patterns and the one mechanism behind them, the cost-and-latency ladder, the four-question decision card for the office wall — and keeping the judgement that makes checking possible.
Chapter 11
Where to Go Next
The three-rung ladder this book stands on, what AI actually means for your work — task by task, honestly — how to keep up without drowning, and your concrete path from here.

Disclaimer

This course is an independent educational project created and maintained by Sergey Okinchuk. It is provided for learning and reference purposes only.

No affiliation. This course is not affiliated with, sponsored by, endorsed by, or officially connected to any company, product, or organization. Waymark and all characters in this course are fictional; any resemblance to real companies or persons is coincidental.

Trademarks. Product and company names referenced are the property of their respective owners. Use of any name is for identification and educational purposes only and does not imply endorsement.

Educational simplifications. This material teaches durable concepts for understanding, not operational instructions. Explanations are deliberately simplified for learning; where a simplification could mislead, the text flags it. This course is not professional, legal, or financial advice.

Accuracy and currency. The field evolves quickly — capabilities, products, and practices drift over time. Facts reflect the author's understanding at the time of writing. The course deliberately teaches mechanism over product specifics; always consult authoritative sources for the current state of any tool or model.

No warranty. This material is provided "as is" without warranty of any kind. The author accepts no liability for any loss or damage arising from reliance on the content.