← Explainers
2 min read

What is an LLM?

The technology behind ChatGPT, Claude and Gemini, explained without the jargon.

BuzzwordsAI basics
Watch it in about a minute

ChatGPT, Claude, Gemini and Llama all have one thing in common: they are powered by an LLM. It’s the engine under the hood of almost every AI chatbot you’ve used.

Term
LLM
Large Language Model

A computer program trained on huge amounts of text, so it can predict what comes next and use that to read and write human language.

Think of it like this

Think of the autocomplete on your phone keyboard. Now imagine it learned from a huge library of books, websites and code, and instead of suggesting one word, it can keep going for whole pages.

Breaking down the name

  1. 1Large — trained on enormous amounts of text, with billions of adjustable numbers inside
  2. 2Language — it works with human language, and with code too
  3. 3Model — a mathematical system that learned patterns from that data

How an LLM is made

From raw text to chat assistant
  1. 01Huge text dataset
  2. 02Pre-training
  3. 03Base model
  4. 04Extra training
  5. 05Chat assistant

First, the model reads a massive amount of text and practises one task over and over: guess the next piece of text. Every wrong guess nudges its internal numbers a little. After enough practice, it has absorbed grammar, facts, writing styles and reasoning patterns.

That “base model” is good at continuing text but not at being helpful. So it gets extra training on examples of good conversations, plus feedback from people rating its answers. That’s what turns it into an assistant you can chat with.

What it’s good at, and where to be careful

Strengths and weak spots
  • Writing, summarising, translatingStrong
  • Explaining ideas and codeStrong
  • Exact maths and countingUnreliable
  • Recent news or private factsNeeds tools

A common misunderstanding

“It looks things up.” Not by default. An LLM writes answers from patterns it learned during training. It doesn’t check a database before replying. Some apps add a search step on top, which is what RAG is about.

A typical case

A bank or airline chatbot can draft a reply in seconds because the model has read millions of messages like it. It has not read that company’s rules. Left alone, it can write a confident answer that breaks them, which is why companies connect it to their own documents with RAG and add guardrails.

Token · Parameters · Prompt · Hallucination