Large language model

A large language model (LLM) is a neural network trained on a huge amount of text to do one job: predict the next token. Do that well enough over trillions of examples and the model starts to understand grammar, facts, and reasoning well enough to be useful.

“Large” is doing real work — these have billions of parameters and cost enormous compute to train.

• Input: your prompt (tokens).
• Process: the transformer stack.
• Output: the next token, repeated to form a full answer.