A neural model trained to predict the next token over broad text, then adapted for chat, tools, and tasks—your default “general text AI” in most engineering tools.
An LLM learns statistical patterns from large corpora. At inference time it generates text (or structured outputs) by repeatedly predicting likely continuations. It is not a database of facts: strengths are language, code, and reasoning when context and verification are right.
How it connects to your stack
LLMs pair with function calling, RAG, and context windows. Expectations should align with hallucination risk and update behavior described under training vs inference.