Large Language Models
The technology powering ChatGPT, Claude, Gemini, and Copilot — explained without requiring a technical background.
When people say "AI" today, they usually mean a type of system called a Large Language Model, or LLM. This is what powers ChatGPT, Claude, Gemini, and the AI inside Microsoft Copilot.
How does an LLM work?
Imagine reading every book, article, forum post, and website ever written — billions of documents. Your brain absorbs all the patterns of how words relate to each other, how sentences are structured, how arguments are made. When someone asks you a question, you generate an answer based on all those absorbed patterns. That is roughly what an LLM does.
It is a pattern-completion engine operating at enormous scale.More precisely, an LLM is trained to predict: "given these words, what word is most likely to come next?" — millions of times across billions of examples. When this training is done at sufficient scale, the model learns to reason, argue, summarise, translate, write code, and create.
The three stages of training
Every major AI model goes through the following stages before you use it.
Stage 1 — Pre-training
The model reads enormous amounts of text from the internet, books, code, and encyclopaedias. It learns language patterns. A model like GPT-4 or Claude is trained on trillions of words.
Stage 2 — Fine-tuning
Human trainers rate responses as helpful or unhelpful. The model is adjusted to produce better, safer, more useful outputs. This process is called RLHF — Reinforcement Learning from Human Feedback.
Stage 3 — Deployment
The trained model is made available to users. You send it text. It sends text back. The intelligence is built in — the model does not automatically get smarter from your conversations unless the company specifically designs it to do so.
Two limitations that matter every day
The following limitations are important to keep in mind whenever you use AI tools at work.
LLMs can hallucinate
LLMs generate plausible-sounding text. They do not "know" facts — they predict words. This means they can confidently produce wrong information. Always verify important facts, especially names, numbers, dates, and citations.
Knowledge cutoff dates
LLMs have a training cutoff date. They do not know about events after they were trained, unless given tools to search the web. Always ask yourself: does this question require current information?
Knowledge check
If you ask an LLM about a news event from this week, what should you expect?