Large Language Model (LLM)
AI Fundamentals
A large language model (LLM) is an AI model trained on huge amounts of text to predict the next piece of text, so it can write, summarize, and answer questions.
AI Fundamentals
What Is an AI Model?
An AI model is a trained computer program that has learned patterns from example data and uses them to turn an input into an output. That output might be an answer, a product description, a photo label, or a sales forecast. What it learned is stored as numbers called parameters. Apps like chatbots are built around models: the model generates the responses, while the app adds the interface, memory, and links to other tools and data.
Also called machine learning model, ML model
Example
An online store that sells handmade kitchenware needs descriptions for hundreds of new products. The owner runs a small test. She gives the same 20 products, with spec sheets and a brand style guide, to three options: a large proprietary model (one you can use but not download), a smaller, cheaper proprietary model, and an open-weight model (one whose trained weights are published) that her developer runs on a rented cloud server.
She compares the drafts for accuracy, brand voice, speed, and cost. She also reads the two proprietary services’ data terms, plus the open-weight model’s license and the cloud host’s terms. Two options write equally good drafts, so the data terms decide it. She notes the exact model version she chose, to retest when it changes. A team member still reviews every description before it goes live.
How an AI model works
Artificial intelligence is the broad field. A model is one concrete thing the field produces: a specific design (its architecture) plus a set of learned parameters, which together can be copied, versioned, and run.
A model is built during training, when developers feed it example data and adjust its parameters until its outputs get closer to the desired results. Training a large model takes specialized hardware, huge datasets, and significant money, so most businesses never do it. Every later use, such as each question you ask a chatbot, is inference: the model applies what it learned without changing it.
There are many kinds of models. A large language model (LLM) is built mainly to understand and generate text, while others generate images, transcribe speech, or predict numbers such as next month’s demand. Some handle several types of input at once.
The model vs. the app vs. the agent
People often use one name for three different layers, so it helps to separate them.
- The model is the trained program that takes an input and produces an output. On its own it has no interface, no memory of past chats, and no access to your systems.
- The app or product is what you actually use, such as ChatGPT, Claude, Gemini, or Microsoft Copilot. It wraps one or more models with an interface, account settings, file uploads, safety filters, and other features. Some brand names, like Claude and Gemini, refer to both the app and the models behind it, which adds to the confusion.
- An AI agent is a system built around a model that can plan steps and use tools, such as searching a database or sending an email, to complete a task.
Models also change often. Providers typically release families of models, such as a large, highly capable model alongside smaller, faster ones, then update them with new versions and eventually retire older ones. An app may switch the model it uses behind the scenes. If a business process depends on consistent output, record which model and version you tested, and recheck results when it changes.
How businesses get access to a model
Most businesses never run a model on their own hardware. They reach it in one of three ways:
- Through an app such as ChatGPT, Claude, or Gemini, where the provider runs the model and decides which models you can use.
- Through an API (a way for your own software to send requests to a model and get answers back), from the developer or a cloud platform. The provider runs the hardware, handles updates, and sets the data terms.
- On servers you control. This usually means an open-weight model, one whose trained parameters are published for download, though some vendors offer private deployments of proprietary models to large enterprises. You gain control over where data goes and take on hosting, security, and maintenance.
Proprietary (closed) models keep their parameters private, so you reach them through an app, an API, or a vendor’s private deployment.
How to choose an AI model for your business
There is rarely one “best” model. The right choice depends on the job. Compare candidates on these points:
- Task fit: Can it handle the input and output you need, such as long documents, images, spreadsheets, or code?
- Quality: Test it on your own real examples, not only on published benchmarks (standard tests used to compare models) or reviews.
- Speed: Larger models are often slower. For high-volume or real-time tasks, a smaller model may fit better.
- Cost: When you use a model through an API, many providers charge by the token (a small chunk of text a model reads or writes), so price depends on how much you send and receive and on the model’s cost per token.
- Data and privacy terms: Check whether your inputs may be used for training, how long they are kept, and whether business terms differ from consumer terms.
- Where it runs: The provider’s cloud, your own cloud account, or your own hardware. This matters most for sensitive or regulated data.
Start with one well-defined task, test two or three models side by side, and keep a person reviewing outputs until you trust the results.
Further reading
- AI Tool LibraryBrowse AI tools organized by the job they do, a practical starting point when comparing options for a specific task.
- AI Fluency AssessmentA free way to see where your business stands on AI adoption before you start choosing models and tools.
Beyond the definitions
Put AI to work for your business.
Go beyond the terminology. Explore practical AI training, workflows, and tools from SN3 Group.