What is a large language model (LLM)?

The engine inside ChatGPT, Claude, and Gemini. It predicts the next word from patterns it read across the internet. Brilliant at fluency, indifferent to truth, and that one fact is most of what a CEO needs to understand about it.

You have used a large language model. You may not have used the term. When you open ChatGPT and ask it to rewrite an email, the chat window is the car and the LLM is the engine under the hood. Knowing how the engine works does not make you a mechanic. It makes you a better driver, and it tells you exactly when to keep your hands on the wheel.

An LLM is a prediction machine for words, not a database of facts. Hold onto that sentence; it explains both why these tools feel like magic and why they confidently tell you things that are not true.

What it is (in plain English)

A large language model is a program that has read a staggering amount of text, most of the public internet plus a lot of books and documents, and learned the patterns in how words follow other words. When you give it a prompt, it does one thing very fast, over and over: it predicts the most likely next chunk of text, then the next, then the next, until it has written you an answer.

That is the whole trick. It is the world's best-read autocomplete. The same instinct that finishes "see you next..." with "week" on your phone is, scaled up across billions of examples, what lets an LLM draft a board memo or summarize a contract. It is not looking anything up. It is generating the most plausible continuation based on everything it has seen.

The products you already know are LLMs wrapped in an interface: ChatGPT runs on OpenAI's models, Claude is Anthropic's, Gemini is Google's. New versions leapfrog each other every few months. The interface is the app; the LLM is the engine inside it.

Why CEOs care

Because the engine has one quirk that decides how you should use it: it predicts plausible text, not verified truth. When it knows the answer, it tells you. When it does not, it does not stop and say so. It generates the most plausible-sounding answer anyway, in the same confident voice. The industry word for this is hallucination, and it is not a bug they are about to patch away. It is a direct consequence of how the thing works.

That single fact gives you your whole operating manual. Use an LLM where you can check the output, not where you have to trust it blind. A draft you will read before sending, a summary of a document you have, a first pass at an analysis you will sanity-check: perfect. An unsupervised answer to a customer about your refund policy, a number you will paste into the board deck without verifying: dangerous. You are an excellent judge of whether an output is right. That judgment is exactly the skill the model lacks, which is why you keep a human, you, in the loop.

The other reason to understand the category: the specific model is rented, and the vendors trade the lead constantly. A CEO who learns "the LLM is a verifiable-draft engine" can switch from one to another in an afternoon. A CEO who only learned one product's buttons starts over every time the lead changes.

What it's great at, and the one limit that matters

Great at: anything that transforms text you give it. Drafting, rewriting in your tone, summarizing a long thread, pulling action items out of a transcript, turning messy notes into a clean plan, translating jargon into plain English. Feed it your context and a clear ask, and it is genuinely a force multiplier.

The limit, again, is truth. It will state a wrong fact as confidently as a right one, and it cannot reliably tell you which is which. So the highest-value pattern is always the same shape: give it your real context, ask for a draft, and verify the output. The model supplies the speed. You supply the truth.

Where you'll see it

• In every chat tab: ChatGPT, Claude, Gemini. Same kind of engine, different vendor.
• Inside a coding agent like Claude Code, where the LLM is the brain and the tooling around it gives it hands to read files and run commands.
• Under your AI meeting-notes tool, your inbox assistant, and the "summarize this" button creeping into every app you use.
• In any sentence where someone says "the model." They mean the LLM.

What to do next

Pay for one frontier model (the free tiers are demos, not tools), and use it this week on one task you do every week and can check yourself. Then, if you want the bigger map of where the real leverage is, read where a CEO should start with AI and which use cases actually pay. The model is the engine. Knowing what it is for is how you stop being impressed by it and start getting leverage from it.