Large Language Models (LLMs)

A large language model (LLM) is an AI model trained on very large volumes of text that predicts and generates language one token at a time. It can reason over text, summarize, classify, translate, and hold a conversation, which makes it the reasoning engine behind AI assistants and agents. It does not hold an organization's own facts; those are supplied through knowledge sources and tools.

An LLM is a transformer neural network: given some text, it repeatedly predicts the most likely next piece. Everything it produces is generated this way. When building an agent, the model choice is a real decision, since different models behave differently on the same instructions and vary in quality, speed, and cost. The common misconception is that the model "knows" your company. Out of the box it knows general patterns from training data with a cut-off date and has no access to your systems.