AI vocabulary arrives faster than anyone can learn it, and many words are used loosely in news and sales pages. You do not need a technical background to build with AI, but a handful of terms will help you read claims carefully and write better requests.
Below are the terms you will meet most often, explained without jargon, with a note on how each one shows up when you build a site or app.
AI is the broad idea of software doing tasks that usually need human judgement. An LLM is the kind that reads and writes text, a prompt is what you ask it, a token is the small piece of text it counts, and a hallucination is a confident answer that is wrong.
AI, machine learning and LLM
Artificial intelligence (AI) is the umbrella term. Machine learning (ML) is one way of building it: instead of writing rules by hand, people train a system on many examples so it finds patterns itself.
A large language model (LLM) is a machine learning system trained on large amounts of text. It predicts what text should come next, which is why it can answer questions, summarise, translate and write code.
- AI: the whole field, from spam filters to assistants.
- ML: learning patterns from examples.
- LLM: a model focused on language and code.
Prompt, token and context
A prompt is simply your instruction: the text you give the model. Clearer prompts usually give better results because the model has less to guess.
A token is a small chunk of text, often a word or part of a word, and models read and write in tokens. Context is how much text the model can keep in view at once, including your instructions and its earlier replies. In a very long conversation, early details can be dropped or blurred.
Hallucination: confident but wrong
A hallucination is an answer that sounds fluent and sure but is invented or incorrect, such as a made-up reference or a wrong figure. It happens because the model generates likely text, not because it checks a database of facts.
It can be reduced by giving the model your real information and asking it to say when it does not know, but nobody can promise it never happens. Anything with money, health or legal weight deserves a human check.
Where these words appear when you build
When you describe a site to Uttra, you are writing a prompt, and the length of the conversation is the context. If the result contains an invented detail such as a price or opening hours you never gave, that is the same effect in miniature, so check content against your real information before publishing.
- Give real names, prices and hours instead of letting the model fill gaps.
- Keep one request focused rather than packing ten changes into one message.
- Start a clearer, shorter brief when a long thread begins to drift.