The short version
A language model learns patterns from data and predicts sequences of tokens. A large language model uses that ability to generate, transform and analyse text.
Learning patterns
Training adjusts many numerical parameters so the model becomes better at predicting examples. Later training can shape how it follows instructions and responds to people.
What they can do
Models can help with drafting, summarising, explaining and writing code. Their performance varies with the task, the context and the checks around the output.
Where they fall short
A fluent answer can contain mistakes or unsupported statements. Models do not automatically know whether their answer is correct, current or appropriate.
Key takeaways
Treat a model's output as something to inspect. Give it useful context and verify important claims with an original source.