LLM stands for Large Language Model. Take each word in turn:
- Language — it works with text: reading it and writing it. Many modern LLMs also handle images, audio, or video, but text is their core.
- Model — it's a trained model, as in the previous lesson: lots of numbers tuned from examples.
- Large — it's huge. LLMs have billions of parameters and were trained on enormous amounts of text from books, websites, code, and more.
That size is what makes them general: the same model can summarise an email, explain a concept, translate a sentence, and write code — without being trained separately for each task.
Some LLMs you've heard of
| Model family | Made by |
|---|---|
| Gemini, Gemma | |
| GPT | OpenAI |
| Claude | Anthropic |
| Llama | Meta |
| Mistral | Mistral AI |
New versions come out often, so the exact model names change. The ideas in this course stay the same.
The app is not the model
This trips people up. ChatGPT is an app; GPT models power it. The Gemini app is an app; Gemini models power it. The app adds a chat window, conversation history, file uploads, and safety features around the model.
As a developer, you usually skip the app and talk to the model directly through an API — which you'll meet in the next lesson. That lets you build the model into your own product.
What LLMs are good and bad at
Good at: writing and rewriting text, summarising, answering general questions, explaining, translating, extracting information from documents, writing and explaining code.
Weak at: knowing recent events, knowing anything private to your company, exact arithmetic on large numbers, and admitting when they don't know. You'll learn how to work around these later in the course.
Quick check: what does each letter in LLM stand for, and what does "large" refer to? (Large Language Model — "large" refers to the huge number of parameters and the vast amount of training text.)