Skip to main content

What Is an LLM?

Beginner

A Large Language Model (LLM) — the technology behind Claude — does one deceptively simple thing: it reads text and predicts what comes next, one chunk at a time. That's it. Everything else emerges from doing that astonishingly well.

What you'll learn
  • Grasp the one-sentence mental model: an LLM is a very sophisticated autocomplete
  • See how the model builds an answer one token at a time, in a loop
  • Understand why this mechanism explains both its strengths and its quirks
  • Know what an LLM is NOT — and how that changes the way you use it

The one-sentence mental model

An LLM is a very sophisticated autocomplete that has read an enormous amount of text and learned the patterns of how language — and the ideas inside it — tend to continue.

When you ask a question, the model isn't "looking up" an answer. It's generating the most plausible continuation of your text, token by token (see Tokens & Context). Plausible continuations of a good question are usually good answers — which is why this works at all.

:::tip Analogy: predictive keyboard on steroids Think of the autocomplete on your phone that suggests the next word. Now imagine it had read most of the books, articles, and code on the internet — and suggested not just the next word, but a whole essay, translation, or program that fits. That's the intuition behind an LLM. :::

One token at a time

The whole engine is a loop: read everything so far, predict the next chunk, append it, repeat.

Guided walkthrough1 of 4
  1. The model takes in your prompt plus everything generated so far as a single block of text.

Each step only ever predicts one token, then feeds the slightly longer text back in. The model has no plan for the whole answer up front — coherence emerges from doing this prediction extremely well, thousands of times. How the "pick one token" step behaves (greedy vs. a bit random) is what sampling controls like temperature adjust.

Why this explains its strengths

Because it learned patterns across writing, code, and reasoning, an LLM can fluidly write, summarize, translate, explain, and code — tasks that are all "continue this text sensibly." Give it a clear setup and it produces a strong continuation. That's why prompting matters so much: you're shaping the start of the text it continues.

Why this explains its quirks

The same mechanism explains the rough edges:

  • It can be confidently wrong. A fluent-sounding continuation isn't always a true one — that's hallucination.
  • It doesn't truly "know" today's facts unless you provide them or it has a tool to look them up.
  • It has no memory between conversations unless you give it some.

What an LLM is not

:::warning Adjust your expectations and you'll get better results

  • Not a database or search engine. It generates, it doesn't retrieve verified records.
  • Not a calculator. It can reason about math but isn't guaranteed exact — give it tools for that.
  • Not a person. No feelings, intentions, or continuous memory. It's a powerful text engine. :::

Treat it as a brilliant, fast, well-read assistant that occasionally misremembers — and verify what matters.

Key terms

Review the core concepts
Press Enter or Space to flip the card. Use the left and right arrow keys to move between cards.Term shown.
1 / 5
Key takeaways
  • An LLM is a very sophisticated autocomplete — it predicts the next token, not looks up an answer
  • Coherence emerges from running that prediction loop one token at a time, thousands of times
  • The same mechanism explains its strengths (write, summarize, translate, explain, code) and its quirks (confidently wrong, no live facts, no memory)
  • It is not a database, a calculator, or a person — verify what matters

Check yourself

Check yourself

0/3
  1. What does an LLM fundamentally do when you ask it a question?
  2. Why can an LLM be confidently wrong?
  3. Which statement about an LLM is correct?

Next