What is a Large Language Model (LLM)?

The Short Version

A Large Language Model is an autocomplete machine on steroids.

You've used autocomplete on your phone — you type "see you" and it suggests "tomorrow." An LLM does the same thing, except it's been trained on billions of pages of text, so its predictions are sophisticated enough to write emails, summarise legal contracts, debug code, and hold a conversation that feels human.

That's it. That's the core mechanism. Everything else — the hype, the fear, the billion-dollar valuations — builds on top of this one idea: predicting the next word, over and over, very well.

How LLMs Actually Work (In One Minute)

I explain this in every training session I run, and it takes about sixty seconds. Here's the version that sticks:

Pattern spotting. During training, the model reads billions of word sequences and learns which words are likely to follow which. Not rules. Patterns. The way a child learns language by hearing it, not by studying grammar textbooks.

Next-word prediction. When you type a prompt, the model starts filling in words — constantly checking what it's written so far — until it completes the answer. It doesn't "know" things the way you know your address. It predicts what a good answer looks like based on everything it's seen.

Scaling laws. Researchers found that throwing more data, more parameters, and more compute (GPUs/TPUs) at the network makes its language skills leap. That's how we went from basic autocomplete to models that can reason through multi-step problems and write working software.

Why "Autocomplete Machine" Matters

I call them Autocomplete Machines because the label changes how you use them.

If you think of ChatGPT as an all-knowing oracle, you'll trust its outputs blindly. You'll paste in a vague question and expect a perfect answer. When it hallucinates — and it will — you'll feel betrayed.

If you think of it as a very powerful autocomplete, you'll give it better inputs. You'll check its outputs. You'll treat it like a brilliant but overconfident junior colleague who needs good briefings and careful review.

That framing is the single biggest unlock I see in my training sessions. The people who get the most out of AI aren't the ones with the fanciest prompts. They're the ones who understand what the machine is actually doing.

Why This Matters to You

You don't need to understand transformer architecture to use AI well. But understanding the basics changes what you expect from these tools — and what you get out of them.

AI literacy is the next "digital literacy." LLMs are being woven into most work and consumer apps. Experimenting now means you're ready when they're everywhere.

Productivity on tap. Offload routine writing, research, and coding so you can tackle higher-value work. A well-prompted LLM can draft a first version of almost anything in seconds.

Competitive edge. Colleagues who learn to delegate tasks to AI get more done. Organisations that adopt it deliver faster and cheaper. This isn't theoretical — I see it every week in sessions.

Democratised expertise. Need legal phrasing, statistical advice, or marketing angles? An LLM gives you a "PhD in your pocket" whenever you ask. It's not perfect, but it's a remarkable starting point.

Career resilience. Understanding how to direct (and critique) AI outputs positions you for roles that won't vanish but evolve. The people who'll struggle aren't the ones replaced by AI — they're the ones replaced by people who use AI.

The Six Ways People Use LLMs

Not everyone uses these tools the same way. Depending on your goal — whether it's answering a question, automating a workflow, or doing deep research — there are different levels of engagement:

Key Terms (Glossary)

Artificial Intelligence (AI)

A branch of computer science focused on creating systems that perform tasks typically requiring human intelligence: learning, reasoning, problem-solving.

Large Language Model (LLM)

An AI system trained on vast amounts of text data to understand and generate human-like text. LLMs predict the next word in a sequence based on context. ChatGPT, Claude, Gemini, and Grok are all LLMs.

Generative AI (GenAI)

A subset of AI focused on creating new content — text, images, audio, or video — based on patterns learned from training data.

Prompt Engineering

The practice of crafting effective prompts to guide LLMs. It involves structuring your inputs to get better outputs. This is the single most useful skill you can develop right now.

Context Window

The amount of text (in tokens) an LLM can process at once. It includes both your input and the model's previous responses. Bigger context windows mean the model can work with longer documents.

Token

A unit of text used by LLMs. Roughly one word, but sometimes part of a word or a punctuation mark. Context windows are measured in tokens.

Hallucination

When an LLM generates plausible-sounding but factually incorrect information. This happens because the model predicts likely text, not verified facts. Always verify important outputs.

Fine-tuning

The process of further training a pre-trained model on specific data to improve its performance on particular tasks or domains.

Transformer Architecture

The neural network design that made modern LLMs possible. It lets the model process all words in a sequence simultaneously while understanding their relationships. Published in the famous "Attention Is All You Need" paper in 2017.

Inference

The process of an AI model generating responses. When you send a prompt and get an answer back, that's inference. The time and computational cost of inference varies significantly between models.