The short answer
A large language model (LLM) is an AI system trained on very large collections of text to predict and generate language. Models such as Claude, GPT, Gemini and Llama power chat assistants, writing tools, coding assistants and business automation.
How an LLM works, briefly
An LLM learns statistical patterns in language during training. When you give it a prompt, it generates the most likely useful continuation, one token at a time. Modern models can also follow instructions, reason through steps, call external tools and work with images and documents.
Common business uses
- Drafting and summarising emails, reports and proposals.
- Answering customer and employee questions from company documents (with RAG).
- Extracting structured data from invoices, contracts and forms.
- Writing and reviewing code.
- Powering AI agents that complete multi-step tasks.
Limitations to plan for
LLMs can produce confident but incorrect answers, their built-in knowledge has a cut-off date, and they do not automatically know your company's private information. Production systems address this with retrieval of trusted sources, evaluation sets, guardrails and human review for high-stakes decisions.