Definition
A large language model (LLM) is a deep neural network trained on massive amounts of text; “large” typically refers to parameter counts and training data far beyond those of earlier, smaller language models. It takes text in and produces text out, covering many kinds of language tasks.
How it works
The model works in tokens: pretraining teaches it to predict the next token, and generation proceeds the same way, producing text autoregressively one token at a time. Everything the model can use to shape its answer is carried in the context.
Usage
LLMs power chatbots, coding assistants, translation, summarization, and agents. They also serve as a base that is further adapted through fine-tuning or reinforcement learning from human feedback to align with specific uses.
Related and confusable terms
An LLM is one concrete class of artificial intelligence, not AI as a whole; its underlying architecture is the transformer. Three concepts unlock how LLMs work: tokens, next-token prediction, and context length.