Large Language Model (LLM)
Last updated:
What a large language model is
A language model estimates which word, or more precisely which token, is likely to come next in a text. A large language model does this with a neural network that has billions of parameters and was trained on large collections of public text, books and code. That scale is what lets it summarize, translate, answer questions and follow instructions without being programmed separately for each task.
Well-known LLM families include OpenAI’s GPT models, Anthropic’s Claude, Google’s Gemini, Meta’s Llama and models from Mistral. They differ in size, cost, speed, language coverage and how reliably they follow instructions.
How an LLM works
Building and using an LLM happens in three broad stages:
An LLM has no live connection to the internet or your systems unless the application around it provides one. Its built-in knowledge ends at its training cutoff, and everything it reads during a conversation has to fit into its context window.
- Pre-training: the model reads huge amounts of text and learns to predict the next token. Along the way it picks up grammar, facts, styles and reasoning patterns.
- Instruction tuning and feedback: further training on examples of helpful answers and on human ratings teaches the model to follow instructions and decline harmful requests.
- Inference: when you send a prompt, the text is split into tokens and the model generates its answer one token at a time, each choice based on everything that came before.
Example
Ask a general-purpose LLM “What are the opening hours of Rossi Bakery on Main Street?” and it cannot know, because that information was never in its training data or has changed since. It may say so, or it may produce a plausible guess. Give the same model the bakery’s current opening-hours page as part of the prompt and it can answer correctly, in the visitor’s language.
Why LLMs matter for business chatbots
The LLM is the engine of an AI chatbot, but on its own it is not a reliable source of facts about your business. Business chatbots therefore combine an LLM with retrieval from your own content, clear instructions and rules for what to do when information is missing. The model affects answer quality, speed and cost, and providers update their models regularly, so whoever runs the chatbot has to test and maintain that choice.
How intoCHAT uses LLMs
intoCHAT runs every agent on one OpenAI model that intoCHAT selects and manages. You do not pick a model or tune its settings; you shape answers through your content, your instructions and Q&A pairs. For each message, intoCHAT retrieves relevant excerpts from your sources and passes them to the model together with rules that keep answers grounded in that material.