The AI jargon translator

Every term that makes you nod along while secretly thinking "no idea." Click any word for the plain-English version, no judgment.

Prompt▶

The thing you type (or say) to AI. That's it. Your question, your request, your ramble. "Write me an email" is a prompt. People make careers out of this word, but it just means "what you asked."

You'll hear: "You just need better prompts." Translation: "Give it more detail about what you actually want."

LLM (large language model)▶

The technology under the hood of ChatGPT, Claude, and Gemini. It learned from enormous amounts of text how words tend to follow each other, which turns out to be enough to write, summarize, and explain scarily well. You never need to say this word out loud.

You'll hear: "LLMs don't really think." Translation: "It predicts words rather than understanding like a person, so double-check facts that matter."

Hallucination▶

When AI makes something up and says it confidently. It's not lying on purpose. When it doesn't know, it fills the gap with something that sounds right. This is why you double-check names, numbers, and anything you'd repeat in public.

You'll hear: "It hallucinated a whole court case." Translation: "It invented a very official-sounding thing that does not exist."

Model▶

A specific version of an AI brain. Like car model years: same company, different vintage and power. When an app lets you "pick a model," it's asking whether you want the faster one or the smarter one. When in doubt, use the default.

You'll hear: "The new model just dropped." Translation: "There's an upgrade and the internet has feelings about it."

Token▶

How AI counts text, in little chunks smaller than words. Why care? Because limits and prices are measured in them. For you, it just means: really long conversations and giant documents can hit a ceiling. That's the whole practical takeaway.

You'll hear: "I ran out of tokens." Translation: "I hit my usage limit and have to wait or pay."

Context window▶

How much a chat can "hold in mind" at once. In very long conversations, the earliest things can fall out of view, which is why AI sometimes forgets what you said an hour ago. Fix: start fresh chats for new topics, and re-paste anything important.

You'll hear: "It forgot my instructions." Translation: "The chat got long and the early parts fell off the table."

Agent▶

AI that does multi-step tasks instead of just answering. You say the goal ("find the double charges in these statements"), it works through the steps. The word gets thrown around a lot; if someone says "agent," think "AI that takes actions, not just gives answers."

You'll hear: "Agents are the future of work." Translation: "Soon you'll delegate whole tasks, not just questions."

Automation▶

Anything set up to happen without you doing it each time. In AI land: a scheduled task that runs every Monday, a follow-up that sends itself. Automations are just recipes: when THIS happens, do THAT.

You'll hear: "I automated my whole week." Translation: "I set up two or three recipes and I'm very proud, as I should be."

Custom instructions / system prompt▶

The standing orders you give AI once so you don't repeat yourself: who you are, how you like answers, what to avoid. Lives in settings. Ten minutes of setup that upgrades every chat afterward.

You'll hear: "Set your custom instructions." Translation: "Tell it once that you're a busy salon owner who hates jargon, and it remembers."

Projects / Gems▶

Folders with a memory. You set the instructions and files once, and every chat inside starts already knowing the job. ChatGPT and Claude call them Projects, Gemini calls them Gems. Same idea everywhere.

You'll hear: "I have a project for my newsletter." Translation: "One workspace already knows my voice and audience, so I never re-explain."

Training data▶

The mountain of text an AI learned from before you ever met it. Two practical things: its knowledge has a cutoff date (ask it to search the web for anything recent), and most apps have a setting to keep YOUR chats out of future training. Worth one minute to flip.

You'll hear: "Is it trained on my data?" Translation: "Check the privacy settings, there's an off switch."

Generative AI▶

The umbrella term for AI that creates things: text, images, audio, video. As opposed to older AI that mostly sorted and recommended things (your spam filter, your Netflix suggestions). If it makes something new, it's generative.

You'll hear: "GenAI is transforming everything." Translation: "Computers can write and draw now and business people are excited."

Voice mode▶

Talking to AI like a phone call: you speak, it speaks back. Wildly underrated for busy people. Brainstorm on your commute, think out loud while folding laundry, then say "summarize everything I just said into a plan."

You'll hear: "I talk to ChatGPT in the car." Translation: "This person has discovered the cheat code and is not crazy."

Machine learning▶

Teaching computers by showing examples instead of writing rules. Show it a million cat photos, it learns "cat." It's the broad field that modern AI grew out of. You will never need this term at dinner, but now you have it.

You'll hear: "It uses machine learning." Translation: "It learned from examples. Also, marketing wanted this in the brochure."

The secret: the people using these words confidently learned them about a month before you. Now you're caught up. If a new one appears, just ask your AI: "explain [term] like I'm smart but busy." That prompt never misses.