Language models: prediction, training and context
3Blue1Brown gives a compact introduction to how language models generate text and learn patterns.
Open the black box
This November 2024 introduction builds a chatbot explanation around repeated next-word prediction. Grant Sanderson introduces learned parameters, pretraining, human feedback and attention through accessible examples. It is a conceptual starting point for readers who want to understand what happens behind a conversation.
Add precision and limits
Google's language-model lesson clarifies that tokens may be words, word parts or characters. It also discusses hallucinations and bias. Keep these limitations in mind when reading fluent answers: understandable language does not guarantee a correct prediction.
The guide uses the creator's written adaptation of this dated foundations lesson.
