Large language model (LLM)

A large language model, or LLM, is the kind of AI model behind most modern chatbots and voice agents. During training it processes huge volumes of text and learns statistical patterns in how words follow one another. When you give it a prompt, it generates a response one small piece of text, called a token, at a time, choosing likely continuations.

What it can and cannot do

LLMs are good at working with language: answering questions conversationally, summarizing a call, drafting an email, pulling fields out of a messy message, or sorting an inquiry by type.

They also have clear limits:

• No built-in fact checking. An LLM produces plausible text, not verified text. It can state wrong things confidently, which is called AI hallucination.
• Fixed knowledge. What it learned in training stops at a cutoff date and does not include your prices, policies or calendar.
• Limited working memory. It can only consider a certain amount of text at once, known as its context window.

How it works in a business

Businesses rarely use an LLM on its own. They give it instructions, supply business facts at the moment of the question through retrieval-augmented generation, and connect it to tools so it looks up live data instead of guessing. An LLM that can take actions toward a goal is usually called an AI agent.

How it relates to Worcia

LLMs handle the language work across Worcia's automation services: conversations in Concierge, its AI receptionist; call summaries and follow-up drafts in Cadenza, its lead follow-up system; and reading documents in Maestro, its workflow automation.