What Is a Large Language Model, in Plain Terms?

The Simple Version

A large language model is software trained to be very good at one core task: predicting what text should come next, given some text as a starting point. That sounds narrow, but when a model gets good enough at this prediction task across an enormous and varied body of text, it ends up able to answer questions, summarize documents, write code, translate languages, and carry on conversations — all as byproducts of learning to model language well.

Instead of following hand-written rules (“if the user says X, respond with Y”), an LLM learns patterns directly from data. During training, it’s shown vast quantities of text — books, articles, websites, code, and more — and adjusts millions or billions of internal numerical values, called parameters, so that its predictions get closer and closer to what actually appears in real text.

Why It Works the Way It Does

The “large” in large language model refers to two things at once: the size of the training data and the size of the model itself, measured in parameters. Researchers found that as both of these scale up, along with the computing power used to train the model, the resulting system tends to get noticeably better at a wide range of language tasks, not just the narrow prediction task it was trained on. This is part of why LLMs can handle tasks they were never explicitly programmed for, like writing a poem in a specific style or debugging a piece of code.

Under the hood, most modern LLMs use a neural network architecture called a transformer, which is particularly good at figuring out which earlier words in a piece of text are most relevant to predicting the next one — even across long stretches of text. This lets the model track context, follow multi-step instructions, and maintain coherence across a long conversation or document.

After the initial training phase, most LLMs go through additional steps to make them more useful and safer as assistants, such as being fine-tuned on examples of helpful responses and adjusted based on human feedback about which outputs are preferred. This is what turns a raw text-predicting model into something that behaves more like a helpful assistant.

Where You Actually Encounter LLMs

You interact with LLMs any time you use a chatbot like ChatGPT, Claude, or Gemini, but they also sit quietly behind many other features: the autocomplete suggestions in some writing tools, AI summaries in search engines, coding assistants that suggest the next line of code, and customer service bots on company websites. In each case, the same underlying capability — predicting and generating coherent text — is being applied to a specific task or wrapped in a specific interface.

It helps to think of the LLM as the engine and the chatbot or app as the car built around it. The engine’s core skill is the same across products, but the surrounding design — what data it can access, what instructions guide it, what guardrails are in place — shapes how it actually behaves for the end user.

Bottom Line

A large language model is, at its core, a pattern-matching engine trained on massive amounts of text to predict and generate language — a capability that, once developed at scale, turns out to be useful for an enormous range of tasks well beyond simple text prediction.

Look Up AI Terms

Search plain-English definitions of AI and machine learning terms in our free AI Glossary.

Go deeper

• What Is ‘Hallucination’ in AI and Why Does It Happen?
• How Should You Prompt AI to Draft a Clear Refund or Return Policy Explanation?
• How Does the Language an AI Model Is Trained on Affect Its Cultural Understanding?