Large language model

A large language model predicts likely text from the text it is given, which turns out to be enough to summarize a call, draft an email, classify a request or carry a conversation. The instruction you give it is called a prompt. On its own a model knows nothing about your business; it learns that from the prompt, from documents retrieved for it, or from being connected to your systems.

For a growing business the model is rarely the decision; the vendor's use of it is. Two products on the same model behave differently depending on their prompts, guardrails, retrieval and integrations. The practical questions are about accuracy, confidentiality and what the system does when it is unsure.

Ask which model a product uses and whether your data is used to train it, where the data is processed, how the vendor prevents hallucination (for example with retrieval-augmented generation), how often the model changes, and what tests they run after each change. Ask for a written data-handling statement.