LLMs explained simply
What a large language model actually does — no math, no hype — and why 'very good autocomplete' is both accurate and misleading.
Every AI tool you'll use at work — ChatGPT, Claude, Copilot, Gemini — is powered by the same kind of engine: a large language model (LLM). Before you can use one well, you need an honest mental model of what it is. Not the marketing version, and not the doomer version. The real one.
The one-sentence version
An LLM is a system that was shown a huge portion of human writing and learned one skill extremely well: given some text, predict what text should come next. That's it. Everything else — answering questions, drafting emails, summarizing reports, writing code — falls out of that one skill applied over and over, one word-fragment at a time.
When you type "Draft a polite reply declining this meeting", the model isn't looking up an answer in a database. It's performing billions of calculations to decide, over and over: given everything so far, what's the most plausible next piece of text? The reply it produces is a fresh composition — plausible, fluent, and usually right, but generated, not retrieved.
Predicting the next word well turns out to require absorbing an enormous amount about how the world works — grammar, facts, reasoning patterns, professional conventions, tone. The prediction engine is simple to describe; what it learned in order to predict well is not. That's why the same tool can draft a contract clause and explain photosynthesis.
Three properties that follow — and matter at work
- It's fluent even when it's wrong. The model optimizes for plausible, not true. A confident, polished paragraph is not evidence of accuracy. (Module 5 is entirely about this.)
- It has no memory of you between conversations unless the product adds one. Each chat starts fresh; anything it must know, you must tell it.
- It's a generalist you can specialize with words. The same model becomes a financial analyst, an editor, or a tutor depending on what you write in the prompt. Your words are the programming.
One more thing, named plainly because it's on most people's minds: will this thing replace you? The honest answer for most professional work: AI is replacing tasks, not judgment. Every workflow in this course keeps a human — you — deciding what's true, what matters, and what ships. The realistic risk isn't the tool taking your job; it's falling behind colleagues who learned to direct it. That's the skill you're here to build, and Modules 3 through 5 show exactly where the human part stays essential.
Meet Sam, an operations manager at Harbor Lane, a 200-person home-goods retailer. Sam will show up throughout this course. Sam doesn't code. In the next four weeks, Sam will go from pasting questions into a chatbot to running a set of AI-assisted workflows that save him about five hours a week — in this course you'll build the first one with him.
Set up your AI assistant (5 minutes)
- 1Pick one tool to learn with: Claude (claude.ai), ChatGPT (chatgpt.com), or Gemini (gemini.google.com). Any of them works for every lab in this course. No preference? Go with Claude.
- 2Sign up with a personal email — unless your company provides an AI workspace account, in which case use that. The free tiers are genuinely free: no credit card, no trial clock.
- 3Know who can see what: on a personal account your employer can't see your chats, but the AI provider stores them on its servers. Module 4 covers exactly what that means for work content.
- 4Bookmark it. Every lab in this course uses it.
One caveat before you type anything: if your company has an AI policy or an approved tool, use that tool and follow the policy — and if you're not sure, check with IT first.
Keep your chosen assistant open in another tab. Every lesson has prompts marked for copy-paste. Type them, read the output, and notice what the lesson told you to notice. The course only works if your hands are on the keyboard.