Large language model — Glossary

Large language models (LLMs) learn statistical patterns of language (and often code) at scale. After pretraining, they can be prompted or fine-tuned for summarization, question answering, drafting, translation, and tool use. Their interface feels general because language itself encodes many tasks.

LLMs can produce fluent errors, invent citations, or reflect training-data biases. Context windows limit how much information they consider at once. Grounding techniques such as retrieval and tool calling improve reliability for knowledge-intensive work, but verification remains necessary for high-stakes outputs.