In plain English
An LLM learns the patterns of language by processing an enormous amount of writing. Once trained, it predicts which words should come next in a way that produces useful text: an email reply, a summary of a report, an answer to a question. “Large” refers to the size of the model and the amount of text it learned from.
The LLM is the engine, and the products you use are built around it. ChatGPT runs on OpenAI’s models, Claude is Anthropic’s, and Microsoft Copilot uses several models behind the scenes, including OpenAI’s. The product adds the chat window, the connections to your apps and the safeguards.
On its own, an LLM knows only what it was trained on, up to a cut-off date, and nothing about your business. That’s why business tools connect it to your own documents (retrieval-augmented generation), and why its answers need checking (AI hallucination).
An example from a small business
Take a hypothetical 20-person manufacturer whose sales team spends hours each week answering similar technical questions from customers. An LLM on its own would give generic answers. Connected to the firm’s product sheets and past replies, it can draft an answer to each enquiry in the firm’s own words, which a salesperson checks, adjusts and sends. The LLM does the writing. The firm’s documents supply the facts.
What to watch out for
- Free and business versions aren’t the same. Some free consumer AI tools may use what you type to improve their models unless you switch that off. Business versions generally don’t by default. Check before anyone pastes in client data (see shadow AI).
- Fluent isn’t the same as right. LLMs write convincingly even when they’re wrong.
- They don’t know your business. Without your documents, an LLM can only guess at your prices, policies and clients.
- Know who handles your data. The company running the model should be covered by a data processing agreement.
Related terms
- Retrieval-augmented generation (RAG): giving an LLM your documents to answer from.
- AI hallucination: when an LLM makes things up.
- AI agent: an LLM given tools and a goal.
- Data processing agreement (DPA): the contract that covers your data.
Where AlphaOps can help
We build with Anthropic’s Claude through its business service, which doesn’t train on customer data and is covered by a data processing agreement. In-house AI on our own hardware in Cardiff is coming soon for work where data shouldn’t leave the building. See our AI consultancy page, or book a free automation audit.