Large Language Models
A large language model, or LLM, guesses the next word again and again. At a huge scale, this simple game gives you a model that can chat, write code and summarise. In this lesson you see how an LLM is made, what "large" means and how to choose one.
New words
- LLM
- Large language model: a very big model that guesses the next token of text.
- Pre-training
- The first stage, where the model reads a huge amount of text.
- Parameters
- The numbers inside a model that it learns. They are also called weights.
- Alignment
- Training a model to be helpful, honest and safe, using human feedback.
- Open-weight model
- A model whose weights you can download and run yourself.
- Knowledge cutoff
- The date after which the model has seen no new text.
- Benchmark
- A fixed set of test questions used to compare models.
What you will learn
- One student who read every book
- Three stages: how an LLM is made
- What does "large" mean?
- Context window and knowledge cutoff
- Try it: a model card comparer
- Open-weight or closed? How people choose
- Why does an LLM sound sure when it is wrong?
- Benchmarks, and why your own tests matter more
- In a real system, and interview tip
- Key takeaways
10 quiz questions at the end.
The full lesson, with interactive demos and the quiz, opens after you sign in with Google. It is free.