You Are Not Behind. The Words Are Just Bad.
Read any article about AI right now and you will trip over five new words in the first paragraph. LLM. Model. Token. Inference. Fine-tuning. The writers act like everyone already knows what these mean. Most people do not, and that is fine.
Here is a short glossary you can come back to whenever a news story or sales pitch throws one of these at you. No code. No math. Just plain definitions in the language a business owner actually uses.
The Core Words
These are the four or five terms you will see most often. Get these and you will understand 80 percent of any AI conversation.
Start with these:
- AI (Artificial Intelligence): software that learns patterns from examples and uses those patterns to make decisions or write things. That is the whole umbrella.
- Model: the trained brain underneath an AI tool. ChatGPT is the app you use; the model is what does the actual thinking. Different models have different strengths and price tags.
- LLM (Large Language Model): a model that has been trained on huge amounts of writing so it can understand and produce human language. ChatGPT, Claude, and Gemini are all LLMs.
- Prompt: the question or instruction you type in. Writing a clear prompt is the same skill as briefing a new employee well.
- Token: a small chunk of text the AI counts in. About three quarters of a word, on average. You will see this on pricing pages: "1 cent per 1,000 tokens" just means a cent per ~750 words in or out.
Words You Will Hear in Sales Pitches
These come up when someone is trying to sell you something. Knowing them keeps you from nodding along when you should be asking questions.
Translate these on the fly:
- Generative AI: AI that creates new things (text, images, audio) instead of just sorting or predicting. ChatGPT writing an email is generative AI.
- Fine-tuning: paying to retrain a model on your own data so it sounds more like your business. Powerful, but rarely the right starting move for a small business.
- RAG (Retrieval-Augmented Generation): a fancy phrase for "let the AI look at your documents before answering." Useful when you want answers based on your own files instead of the open internet.
- Agent: an AI setup that does not just answer, it takes actions. Books an appointment, sends a follow-up, updates a spreadsheet. Worth a closer look but not always what the seller claims.
- Hallucination: when AI confidently makes something up. It looks right, but it is wrong. Always sanity-check anything that involves numbers, names, or dates.
Words That Sound Scarier Than They Are
A few terms get thrown around to make AI sound more complicated than it is. They are usually simpler than the word suggests.
Plain-English versions:
- Inference: the AI doing its job once. Every time you hit "send" on a question, that is one inference. Companies pay per inference; you usually pay a flat monthly fee.
- Parameters: the internal dials inside the model. People love to brag about '70 billion parameter models.' For your purposes, more is not always better. Bigger models cost more and run slower.
- Context window: how much the AI can hold in its head at one time. A bigger context window means you can paste in longer documents without it forgetting the start.
- Multimodal: a model that handles more than just text (images, audio, video). The newer versions of ChatGPT and Claude are multimodal.
- API: the plumbing that lets one piece of software talk to another. When a developer says they are connecting an AI to your CRM, they are using an API.
The Only One You Have to Memorise
If you only remember one word, make it model. The model is the brain. The app is just the steering wheel.
When someone says 'we use AI,' the right follow-up question is, 'which model, and why that one?' The answer tells you whether they actually know what they are doing.
If you ever want a straight, no-jargon walkthrough of how this stuff fits into your business, that is exactly what our free discovery call is for. Bring the words. We will translate.
What is the difference between AI, an LLM, and a model?
"AI" is the whole field. A large language model (LLM) is one kind of AI — the kind trained on text that predicts what comes next. A "model" is a specific trained system you can actually use, like Claude Sonnet or GPT-5. In practice, when a vendor says "our AI", they almost always mean "we send your text to somebody else's LLM".
Do I need to understand any of this to use AI at work?
You need exactly one: the token. It determines what you pay, how much the model can hold in mind at once, and where it starts forgetting. Everything else on this page you can look up when someone says it. Tokens are the concept that changes decisions.
What does "fine-tuning" mean, and do I need it?
Fine-tuning further trains a model on your own examples so it adopts a style or format. Most small businesses do not need it. It is expensive, it goes stale as your information changes, and for the usual goal — getting AI to answer using your own documents — retrieval does the job better and can be corrected by editing a file.
Is a "prompt" the same as a "system prompt"?
No. A prompt is what you type each time. A system prompt is standing instruction set behind the conversation that the operator writes — tone, rules, what to refuse. You pay for the system prompt on every single call, because it is re-sent each time, which is one reason long ones get expensive.