What does a language model actually do?
Understand how text generation differs from looking up a fact.
In plain language
A large language model learns patterns from training examples. Given a sequence of tokens, it produces a continuation. A token can be a word, part of a word, or punctuation; it is not a fixed number of characters.
An AI application can combine a model with search, documents and tools. That application may look up information before responding. The model and the surrounding application are therefore useful to distinguish: generating text is one capability; retrieving evidence is another.
Ask an assistant to draft three invitation titles. Then ask it for a current train departure. The second task needs a reliable, current source in addition to fluent wording.
Try it yourself
Compare a creative writing request with a question about a current event. Which needs a current source?
Go deeper · for developers and architects
Tokenization and input/output limits vary by model. Check the actual model limits, reserve output capacity, and test your real workload. Avoid a universal word-to-token ratio or a fixed context-budget percentage.
Check your understanding
1 related question, with an explanation for each answer.
Explore the original sources
Google ML Crash Course ↗Report an issue
Tell the editors what needs correcting. Reports are reviewed and do not automatically remove content. Do not include personal or confidential information.
Finished this lesson?
Completion records your choice, not a measured skill level.