A plain-language AI glossary

LLM (Large Language Model)

The kind of AI behind most chat assistants. Trained on large amounts of text, it predicts likely next words to generate answers, summaries, and code.Related: Token, Context window, Inference

Token

The unit an AI model reads and writes in — roughly a word-piece. Limits and pricing are often measured in tokens rather than words.Related: LLM, Context window

Context window

How much text a model can consider at once, measured in tokens. A larger window lets it work with longer documents or conversations without losing track.Related: Token, LLM

Prompt

The instruction or question you give an AI. Clearer, more specific prompts usually get better results.Related: System prompt, LLM

System prompt

Hidden instructions that set an assistant's role and rules before your own messages — for example, telling it to answer concisely or stay on a topic.Related: Prompt

Fine-tuning

Further training of an existing model on extra examples so it does better on a specific task or style, without training a whole model from scratch.Related: LLM, Open weights

Hallucination

When an AI states something false but plausible-sounding. It is why answers on important topics should be checked against a real source.Related: RAG, LLM

RAG (Retrieval-Augmented Generation)

A technique where the model first retrieves relevant documents and then answers using them, so responses are grounded in real sources rather than memory alone.Related: Hallucination, Embedding

Multimodal

A model that handles more than text — for example images, audio, or files — as input or output.Related: LLM

API

A way for your own software to send requests to an AI service and get results back, so you can build the model into an app instead of using a chat window.Related: Inference, Open weights

Inference

Running a trained model to get an answer. Each response you get from an assistant is an inference; on paid platforms it is what you are billed for.Related: LLM, API

Open weights

Models whose trained parameters are published so anyone can download and run them — often free to use and possible to run locally.Related: Fine-tuning, Inference

Embedding

A numeric representation of text (or other data) that captures meaning, letting software measure how similar two things are. The backbone of search and RAG.Related: RAG

Temperature

A setting that controls how varied a model's output is. Lower values give more focused, predictable answers; higher values give more creative, varied ones.Related: Prompt

Free tier vs freemium

A free tier is a no-cost level of a paid product; freemium means the core is free with paid upgrades. Both let you start without paying — the limits are what differ.

Ready to try some of these ideas in practice? Browse the free AI tools directory.