The mini dictionary
One plain sentence, why it matters to you, and something you can picture. No condescension, no vendor language.
An AI that takes several steps and uses tools on its own, instead of answering once.
Why it matters: It's the difference between asking for a plan and something actually doing the work.
An agent that searches, compares, drafts an email and waits for your yes.
A permission slip that lets the AI read one of your other accounts, like Drive or email.
Why it matters: It's how it stops being a stranger and starts knowing where your files are.
Connecting a calendar so it can see when you're free.
How much it can hold in mind at once. Your whole conversation counts toward it.
Why it matters: It's the reason a long chat starts forgetting what you said at the beginning.
Like a desk: every new page pushes an old one off the edge.
A unit you pay for that stands for work done, rather than time spent.
Why it matters: Explains why the same question can cost different amounts in different apps.
One image often costs more credits than ten short messages.
Extra training on your own examples so it learns a style or a job.
Why it matters: Expensive and rarely needed — most people get further with a good standing instruction.
Training on 500 support replies so it writes like your team.
Giving the AI your own documents so it answers from them instead of memory.
Why it matters: It's the difference between a guess and a quote. Nothing beats it for accuracy.
Attaching your policy PDF and asking what it says about refunds.
When it fills a gap with something that sounds right and isn't.
Why it matters: It's not lying. It's guessing confidently, and it does it most when you'd never notice.
A citation that looks perfect and points at an article that doesn't exist.
The moment the model actually answers. Every reply is one run of inference.
Why it matters: It's what you're being charged for, and why speed varies by model.
A big model does slower inference than a small one.
The wait between sending and the first word coming back.
Why it matters: Usually about model size, not your internet. Smaller models feel instant.
A voice assistant with high latency feels like a bad phone call.
The trained brain behind the app. The chat window is just the front door.
Why it matters: It's why ChatGPT feels different month to month — the model changed underneath it.
The same app can run a small fast model or a large slow one.
It can handle more than text — pictures, audio, video, files.
Why it matters: It's why you can photograph a form and ask what goes in box 4.
Uploading a receipt and asking it to total the tax.
The instructions you type in — the whole message, not just the question.
Why it matters: Almost every disappointing answer is a prompt problem, not a you problem.
“Write about my shop” versus “Write 60 words for my shop's homepage, for people who just moved to the area”.
When text the AI reads — a web page, an email, a file — contains hidden instructions that hijack it.
Why it matters: It's why you don't let an AI act on your accounts unsupervised.
A webpage that tells the assistant to ignore you and email your files away.
Retrieval-augmented generation: look things up first, then answer.
Why it matters: The look-up is the accuracy. If a tool can't search your files, it can only remember.
A support bot that reads the help centre before replying.
Images, audio or video that were generated rather than recorded.
Why it matters: Assume any voice or face on a screen can be faked, and verify money requests by calling back.
A cloned voice asking a relative for a transfer.
The house rules set behind the chat that you usually can't see.
Why it matters: It's why two tools answer the same question in completely different voices.
One is told to be brief and cautious; another is told to be playful.
A setting for how much randomness the model allows itself.
Why it matters: Low gives you the same careful answer twice. High gives you surprises — sometimes good ones.
Low for a contract summary, higher for brainstorming names.
The unit an AI counts in — roughly three quarters of a word.
Why it matters: It's why long messages cost more, why there are limits, and why some tools charge by the word.
A 1,000-word email is about 1,300 tokens.