What is an LLM? A five-minute explainer

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

Key ideas

1. 1

   A machine that predicts what comes next

   An LLM, or large language model, is a program trained on an enormous amount of text to predict what word comes next.

   Just as you'd guess "nice" after "the weather today is really," it repeats that prediction over and over, very precisely, to produce sentences and answers.

   Picking the word after "the weather today is really" (illustrative numbers)

   Next word	Likelihood
   nice	62%
   hot	21%
   cold	11%
   sunny	6%

2. 2

   It reads in tokens, not letters

   LLMs break text into pieces called tokens. The amount they can read at once is the context window — the bigger it is, the longer the documents they can handle in one go.

   When you use AI through an API, you're usually charged by the token too; chat app subscriptions are typically a flat monthly fee.

3. 3

   Convincing, but sometimes wrong

   An LLM on its own doesn't look facts up; it predicts plausible next words. That's why it can state wrong information confidently — a hallucination.

   Some services also search the web, but either way, make a habit of checking the source for important facts and figures.

   Check it

   • Double-check numbers, dates and quotes in an answer against the original source.
   • Even when you ask for sources, open them yourself: models sometimes invent convincing-looking ones.
4. 4

   Getting good answers: give it context

   LLMs work best with the information you give them. A prompt that spells out the situation, the goal and the format you want gets better answers.

   I'm new on the marketing team and need to explain what an LLM is to colleagues in under a minute at next week's meeting. Write a plain-language explanation that includes one analogy.

   An LLM is a program that has read a huge amount of text and practiced guessing the next word in a sentence.

   Think of your phone keyboard's next-word suggestions, scaled up enormously — precise enough to write whole sentences and paragraphs, not just suggest one word.

   That's why it's good at writing and summarizing. But it continues text plausibly rather than looking facts up, so important details need checking.

   Check it

   • Check that it reads in under a minute and the analogy suits your audience. If it runs long, ask for a 30-second version.