Large Language Model (LLM)

Definition

A Large Language Model (LLM) is a neural network—typically a transformer architecture—trained on massive text corpora using self-supervised learning objectives to generate and understand natural language. LLMs learn statistical patterns across billions of text examples, developing emergent capabilities including reasoning, summarization, code generation, and instruction following. In AI agent systems, the LLM serves as the reasoning core: the component that interprets inputs, forms plans, makes decisions, and generates outputs.

Engineering Context

In production AI agents, the LLM is a stateless component called via API. Key engineering parameters: model selection (capability vs. cost vs. latency tradeoffs), temperature (determinism vs. creativity), context window size (limits working memory), and structured output support (required for reliable tool calling). For enterprise deployments, consider on-premise options (Llama 3, Mistral) for data privacy and cost control at high volumes. Treat the LLM as you would any external service dependency: implement timeouts, retries with exponential backoff, fallback models, and circuit breakers. Monitor token usage, latency, and error rates per model in production.

Related Terms