AI glossary: 24 terms explained in plain English
No circular definitions, no jargon explaining jargon — just what each term actually means, grouped by where you'll actually run into it.
Core AI concepts
AI model
The trained system behind a tool's output — the thing that actually generates the response, not the app wrapped around it.
Machine learning
The broader technique behind most modern AI: a system improves at a task by learning from examples, rather than following hand-written rules.
LLM (large language model)
A model trained on huge amounts of text, built to predict and generate language — what powers most AI chat tools.
Training data
The examples a model learned from before it was released — its knowledge is shaped by what was, and wasn't, in that data.
Neural network
The layered mathematical structure most AI models are built from, loosely modeled on how neurons connect in a brain.
Multimodal
A model that can handle more than one type of input or output — text plus images, audio, or video, not just words.
Prompting and interaction
Prompt
What you type to tell an AI tool what you want — the input, whether it's a question, instruction, or example.
Zero-shot
Asking a model to do something with no examples given, relying entirely on its existing training.
Few-shot
Giving a model one or more examples of the output you want before asking it to produce something similar.
System prompt
Background instructions set before a conversation starts, shaping how the model behaves throughout, invisibly to the end user.
Context window
The maximum amount of text a model can consider at once — think working memory, not permanent storage.
Temperature
A setting controlling how predictable versus varied a model's output is — low for consistency, high for creative variation.
Technical and model terms
Token
A chunk of text a model processes at a time — roughly three-quarters of a word on average, not a whole word or letter.
Fine-tuning
Further training an existing model on specific, narrower data to specialize it for a particular use case.
Hallucination
A confident-sounding output that's actually incorrect or made up — one of the main reasons to fact-check AI-generated claims.
Inference
The act of a trained model actually generating a response — as opposed to training, which happens beforehand and separately.
API
A way for one piece of software to talk directly to an AI model, without a human typing into a chat window.
Parameter
One of the internal values a model adjusts during training — roughly, a proxy for how large and complex a model is.
Business and pricing terms
Worth knowing before reading any AI tool's pricing page — these patterns repeat constantly across the tools covered on this site.
Freemium
A genuinely usable free tier with capped features, plus a paid plan to raise the limits.
Credit-based pricing
You get or buy a set number of credits, spent per action — a generation, a lookup, a query.
Per-seat pricing
Price scales directly with how many people on a team have an account, regardless of how much each uses it.
Usage-based pricing
Cost scales with volume — contacts on a list, messages sent, or data processed — rather than a flat monthly fee.
Free trial
Temporary full or partial access to paid features, usually requiring a card and converting to a paid plan automatically.
Rate limit
A cap on how many requests can be made in a given time window, common on both free tiers and API access.
Where to go next
This covers the terms that come up constantly, not every term that exists — new vocabulary shows up as AI tools evolve, and this list will grow with it. For a deeper look at putting some of these ideas into practice, the Prompt Engineering guide covers prompting terms in more depth, and Advanced Guides goes further into agents, protocols, and context windows.