AI Jargon, Translated: 12 Terms Every Office Worker Should Know

Every AI conversation at work eventually hits a word that makes half the room nod and the other half quietly google under the table. This page is for the second half, and there is no shame in it: the jargon is newer than most people's job titles.

Twelve terms, each in plain English, each with why it matters to you. No math, no diagrams, two minutes each.

The words you hear every day

Prompt

The instruction you type into an AI. That is the whole definition. Writing better prompts (more context, clearer goal, explicit tone) is the single highest-leverage AI skill for a non-technical person, and it is learnable in an afternoon.

AI model (or LLM)

The engine behind the chat window. LLM stands for large language model: a system trained on enormous amounts of text until it became very good at predicting what words come next. ChatGPT, Claude and Gemini are products; underneath each is a model doing the work.

Generative AI

The family name for AI that creates things (text, images, audio, video) instead of just sorting or ranking things. The AI that writes your email and the AI that recommends your playlist are different species; generative is the new one.

Chatbot vs assistant

Mostly marketing. Chatbot is the older, narrower word (think of rigid customer-service widgets). Assistant implies the modern kind that can handle open-ended tasks. People use both loosely; do not read too much into either.

The words that explain weird AI behavior

Hallucination

When an AI states something false with total confidence: a made-up statistic, a nonexistent article, a wrong date. It is not lying, it is autocomplete overreaching. This is why you verify numbers, names and dates before repeating them, a habit our summarizing guide builds in.

Token

The chunks AI reads text in, roughly three-quarters of a word each. You mostly meet tokens in limits: a model that handles 200,000 tokens can hold about 150,000 words in mind at once. Which brings us to...

Context window

The AI's working memory: how much of the conversation and any pasted documents it can consider at once. When a long chat gets forgetful or a huge document gets mangled in the middle, you have hit the context window. The fix is starting fresh chats or feeding documents in parts.

Training data

The mountain of text the model learned from, with a cutoff date. Anything after that date, the model simply does not know unless it can search the web. If an AI seems oddly out of date, that is why.

The words on the settings page

Custom instructions

A place to tell the AI about yourself once (your job, your audience, your preferred tone) so every future chat starts pre-briefed. Ten minutes of setup, permanent payoff. Lesson six of our beginner course walks through it.

Temporary chat

A conversation that is not saved to history and not used for training. The right mode for anything borderline sensitive. Details in our guide to using ChatGPT safely at work.

Multimodal

An AI that handles more than text: it can look at images, listen to audio, read PDFs. Practically, it means you can photograph a whiteboard and ask for the action items, or paste a chart and ask what it says.

AI agent

An AI that does not just answer but takes multi-step actions: browsing, filling forms, running tasks while you do something else. The frontier of the field right now, arriving in office tools gradually. When a vendor says agent, ask exactly what it can do without you watching.

That is the whole vocabulary

Twelve words cover essentially every AI conversation you will have at work this year. Bookmark this page for the next meeting, and when someone uses a term that is not here, there is a decent chance they are improvising too.

Ready to go from vocabulary to skill? The free course ChatGPT from Zero: Your First Week with AI turns these concepts into daily habits, six short lessons, no jargon, we promise. The deeper technical rabbit hole, if you ever want it, starts at Wikipedia's large language model page.