Glossary

LLM (Large Language Model)

The technology behind ChatGPT and Claude. A program trained on enormous amounts of text that learns to predict what should come next — turns out that's enough to write, summarize, explain, and reason.

Prompt

The thing you type to the AI. Better prompts → better answers. Most of this site is about writing better prompts.

Hallucination

When AI invents something — a fact, a quote, a citation — and presents it confidently. The fix: verify anything that matters.

Context

The information you give the AI to work with — facts about you, the situation, the goal. The single biggest lever for better answers.

Token

A small chunk of text (roughly 4 characters or ¾ of a word). AI is paid and limited by tokens. You don't need to count them — just know that "too long" is a real constraint.

Context window

How much text the AI can "see" at once. Modern AIs can read whole books at a time, but very long chats can still lose track of what you said up top.

Model

A specific version of an AI. "GPT-4o", "Claude Sonnet 4.6" are models. Newer is usually better, but not always cheaper or faster.

API

A way for software to talk to AI directly (instead of through a chat website). Comes up when you want to build something that uses AI, like an app or automation.

RAG (Retrieval-Augmented Generation)

When an AI looks things up in a specific document or database before answering. It's how AI products give answers grounded in real, current data.

Fine-tuning

Customizing a model on specific examples so it gets better at a narrow task. Usually overkill for personal use — better prompting does most of the work.

System prompt

A "behind the scenes" instruction that shapes how the AI behaves in a chat — like "you are a helpful tutor for a 6th grader." Some apps let you set this; most chats just have a default.

Multimodal

An AI that can handle more than text — images, audio, sometimes video. Photo-of-a-receipt → typed total is a multimodal task.

Agent

An AI that can do multi-step things on its own — not just answer, but browse, click, file, send. Still early, but rapidly getting useful.

Open source / open weights

AI models you can download and run yourself. Mostly relevant if you're technical. Less relevant for most family use.

Cloud / on-device

Whether the AI runs on someone else's computer (cloud — ChatGPT, Claude) or on your own (on-device — like Apple Intelligence). On-device is more private; cloud is usually more capable.

Privacy mode / temporary chat

A setting that tells the AI service not to use your conversation to train future models. Use it when sharing anything you don't want stored.

Hear a word that isn't here? Ask us to add it →