LLM (Large Language Model)

How it works

LLMs are neural networks with billions of parameters, trained through self-supervised learning on text from the internet, books, and other sources. They develop a compressed representation of language patterns — including grammar, reasoning patterns, factual associations, and conversational norms — that allows them to generate coherent, contextually appropriate text for inputs they have never seen before.

For support applications, the LLM's core function is reasoning and generation: given a customer question and a set of retrieved knowledge-base passages, generate a response that accurately addresses the question. The LLM does not 'know' your business — it reads the retrieved context and reasons from it.

LLM selection affects support quality in measurable ways. More capable models produce better reasoning on complex, multi-part questions, better instruction following (e.g., 'always mention the 30-day policy'), and better handling of ambiguous queries where the intent isn't obvious.

When you'll encounter it

LLM is mentioned in vendor technical documentation, AI support architecture descriptions, and benchmark comparisons. When a vendor says they use 'GPT-4' or 'Claude,' they are specifying which LLM powers the reasoning layer of their RAG system.