The glossary: every term, decoded in one breath

Abliteration

A technical term for changing a language model’s internal settings, called weights, to reduce its tendency to refuse requests. It is associated with models whose weights are available for others to modify. Fewer refusals do not mean better judgment, and the changes can weaken safety protections.

Say it like this: “A model that says yes to more things has not necessarily become a better adviser.”

Account recovery

The steps for getting back into an account when you’re locked out — usually a backup email, phone number, or recovery codes. Set these up on a calm day, not a locked-out one.

Say it like this: “I set up account recovery before I needed it. Future me sends her thanks.”

Agent skills

Reusable instructions, and sometimes supporting tools or files, that help an artificial intelligence agent carry out a particular task. Often shortened to skills. Having a skill does not guarantee good work or give the agent permission to access all your information.

Say it like this: “It has a skill for formatting documents. I still read the finished handout.”

Agentic artificial intelligence (agentic AI)

Artificial intelligence that can work through several steps toward a goal, often choosing tools and deciding what to do next. The amount of independence varies. A system that can take action needs clear limits and your supervision.

Say it like this: “I let it gather information and make a draft. Anything sent in my name still needs my approval.”

AI agent

An AI that doesn’t just answer but takes multi-step actions for you — filling in forms, comparing prices, booking. In 2026 they’re built into ChatGPT, Gemini, Claude, and some browsers. Powerful; supervise it like a new employee, and never let it pay, sign, or send without your yes.

Say it like this: “I let an AI agent compare the flights. I clicked ‘buy’ myself.”

AI browser

A web browser with an AI assistant built in that can read the page you’re on, summarize it, and sometimes click through websites for you. Examples include Perplexity’s Comet and Gemini in Google Chrome. Handy — and worth extra caution, because a browser that can click can also be fooled (see Prompt injection).

Say it like this: “I let the AI browser summarize the article. I didn’t let it near my bank.”

AI companion

An app designed to chat like a friend — remembering your day, asking how you are. It can ease a quiet evening, but it’s a program run by a company, not a person who cares about you. Never share money details or secrets with one.

Say it like this: “My AI companion is pleasant company. My actual friends still get the Thursday lunch.”

AI image generator

A tool that paints an original picture from your written description — DALL·E and Midjourney are the famous ones. You describe; it illustrates.

Say it like this: “I typed ‘a lighthouse in a thunderstorm, oil painting’ and had four of them before the kettle boiled.”

AI Mode / AI Overviews

Google’s AI-written summaries that now appear at the top of many search results (AI Overviews), plus a chat-style search you can switch to (AI Mode). Convenient — but it’s still AI, so check the links underneath for anything important.

Say it like this: “The AI Overview said one thing; the Medicare website said another. Medicare wins.”

AI slop

Low-quality, mass-produced AI content — fake videos, made-up ‘news,’ and odd pictures — posted to grab clicks. A good reason to ask ‘who made this, and how do they know?’

Say it like this: “Six-fingered grandma, eagles carrying cats — that feed is pure AI slop.”

AI video generator

A tool that creates realistic video from a written description or a photo — Google’s Veo is one well-known example. Wonderful for creative projects; also why seeing is no longer believing.

Say it like this: “That video of the mayor looked real. It was made by an AI video generator in about a minute.”

AI writing assistant

Using AI to produce a first draft — emails, letters, toasts, complaints — which you then edit until it sounds like you. The draft is its job; the judgment is yours.

Say it like this: “The AI wrote the first draft of my complaint letter. I added the steel.”

Alexa+

Amazon’s newer, AI-powered Alexa that holds a real back-and-forth conversation. In 2026 it’s included with Amazon Prime; without Prime it costs $19.99 a month (the original ‘classic’ Alexa stays free).

Say it like this: “Alexa+ finally understands me when I ask two things at once. My husband is jealous.”

Algorithm

A recipe a computer follows, step by step. Every app runs on them; AI’s recipes are learned rather than hand-written.

Say it like this: “The photo app’s algorithm decided those were all pictures of the dog. It was right.”

Anthropic

The company that makes Claude, founded by former OpenAI researchers with a particular focus on making AI safe and careful.

Say it like this: “Anthropic is the company; Claude is the chatbot. Like Kellogg’s and the corn flakes.”

App

A program on your phone or tablet. You get them from the App Store (Apple) or Play Store (Android).

Say it like this: “The bird-identifying app is the best dollar I never spent — it was free.”

Apple Intelligence

Apple’s bundle of AI features built into newer iPhones, iPads, and Macs — writing help, smarter photos, a more capable Siri.

Say it like this: “Apple Intelligence arrived with a phone update. I didn’t go get an AI; one moved in.”

Artificial general intelligence (AGI)

The idea of artificial intelligence that could handle a very wide range of mental tasks at a human level or beyond. Researchers disagree about exactly what would qualify and how to measure it. The label does not mean a system is conscious, trustworthy, or good at everything.

Say it like this: “When someone says AGI is coming, I ask what they mean by it. A big claim deserves a plain explanation.”

Artificial intelligence (AI)

Software that performs tasks that normally require human judgment — writing, answering, recognizing, predicting.

Say it like this: “I used AI to draft the letter, then I fixed the parts that didn’t sound like me.”

Artificial intelligence (AI) alignment

Work to make artificial intelligence behave in ways that fit people’s intentions, values, and safety needs. An instruction can sound simple and still lead to an unwanted result. Alignment is an ongoing challenge, not a promise that a system will always do the right thing.

Say it like this: “I asked it to save me time. I still want it to ask before it sends a message in my name.”

Artificial intelligence (AI) literacy

Understanding enough about artificial intelligence to use it thoughtfully: what it can help with, what it can get wrong, and what to keep private. It includes knowing that you cannot always tell whether a picture, voice, or piece of writing was made with it.

Say it like this: “I don’t need to build the engine. I do want to know when to check the answer.”

Artificial intelligence (AI) medical scribe

Software that listens to a medical visit and drafts a clinical note. An ambient scribe works from the conversation rather than requiring the clinician to dictate each line. You should be told how it is being used and asked for consent. The clinician remains responsible for checking the note; the software can make mistakes.

Say it like this: “Before we begin, is a computer taking notes? What is recorded, who can see it, and can we proceed without it?”

Artificial intelligence (AI) watermarking

Adding a mark or detectable pattern to material made or changed by artificial intelligence. A watermark may be visible or hidden. It is one approach to identifying such content, distinct from a fuller record of its history. Not all generated material has one, and marks can be lost or altered.

Say it like this: “No watermark does not mean no artificial intelligence was involved.”

Automation

Setting up a task to happen by itself — bills paid, photos backed up, lights off at eleven. You decide once; the machine remembers forever.

Say it like this: “I automated the thermostat. It’s the one member of the household who never forgets.”

Avatar

A digital stand-in for a person on screen — sometimes a cartoon, sometimes a lifelike AI video of a real human.

Say it like this: “The host in my video is my avatar. My face, my voice, considerably better lighting.”

Backup

An extra copy of your photos and files kept somewhere else, so one lost phone doesn’t mean lost memories. The cloud will do it automatically if you let it.

Say it like this: “My photos live in three places. The people in them are irreplaceable; so are their pictures.”

Bias

When an AI’s answers lean unfairly because its training data did. Machines inherit our habits — including the bad ones.

Say it like this: “The AI learned from people, so I read it the way I read people: generously, but with one eyebrow up.”

Biometrics

Measurements of physical or behavioral features, such as a fingerprint, facial features, or voice, used to help identify someone. Face or fingerprint unlocking are familiar examples. These systems can make mistakes, and a familiar-sounding voice alone is not proof of who is calling.

Say it like this: “I unlock my phone with my face. I still keep its passcode private.”

Bot

Short for robot: a program that performs tasks automatically. Some bots use artificial intelligence; many follow simple rules. A chatbot talks with you, while other bots might send reminders or post messages. A bot is not necessarily a person or a trustworthy source.

Say it like this: “That reply came from a bot. If I need a person, I’ll ask how to reach one.”

Chatbot

A program you converse with in plain language. ChatGPT, Claude, Gemini, Grok, and Copilot are the famous ones.

Say it like this: “I asked the chatbot to explain the insurance letter as if to an intelligent friend. It obliged.”

ChatGPT

The chatbot from OpenAI that started the boom in late 2022 — the fastest-adopted consumer technology in history. It has a free version (with limits), voice conversations, and paid plans. The engine’s name, Generative Pre-trained Transformer (GPT), describes the kind of system behind ChatGPT (GPT-6 as of October 2026).

Say it like this: “ChatGPT reached 100 million users in two months. None of us are ‘behind.’”

Claude

The chatbot from Anthropic, known for careful writing and handling long documents. Free version available, including voice mode. (It helped build this website — we’re fond of it.)

Say it like this: “I gave Claude the whole 40-page policy and asked what it means for a person my age.”

Clinical decision support

Tools that help a clinician consider information, such as a possible medication interaction, a recommended test, or a treatment guideline. Some use artificial intelligence; others use set rules. An alert supports a medical decision and still needs clinical judgment.

Say it like this: “The system flagged a possible interaction. I asked my clinician what it meant for my medicines.”

Cloud computing (the cloud)

Using computers reached over the internet to store files, run software, or process information. Those computers are often called servers. The cloud is physical equipment somewhere else, not an invisible place in the sky. Some services keep copies on your own device, too.

Say it like this: “My photos are stored online, so I can open them on more than one device.”

Community / forum

An online gathering place where people trade questions, answers, and tips. Odds are excellent that someone has already had your exact problem — and posted the fix.

Say it like this: “I found the answer on a forum, posted by a stranger in 2023. I lit a small candle of gratitude.”

Computer use (computer-using agents)

An artificial intelligence system operating a computer interface by clicking, typing, opening apps, and moving between windows. It may be able to act on what is on the screen. A mistaken click can have consequences, so permissions and approval points matter.

Say it like this: “It can fill in the form. I want to review every entry before anything is submitted.”

Content Credentials (AI labels)

Information attached to a photo, video, or other file to record details about its source and editing history. It may disclose artificial intelligence use. This is one approach to content provenance. Credentials can be missing or removed, and even a valid history record does not prove that a scene or claim is truthful.

Say it like this: “I can check the picture’s history, then check whether the story being told about it is true.”

Content provenance

Information about where a photo, recording, video, or document came from and how it was changed. Think of it as a record of its history. Content Credentials are one way to provide such a record. A history label does not prove that the message or scene is truthful.

Say it like this: “The label tells me something about the picture’s history. I still check the claim made alongside it.”

Context

The background you give an AI so its answer fits your actual life — your age, your state, your situation. More context, better answer.

Say it like this: “I told it I’m 78, in California, and on Medicare. Suddenly the advice fit like a good shoe.”

Context window

The amount of information an AI model can work with at once, measured in tokens. This can include your instructions, recent conversation, and documents, with room also needed for its reply.

Say it like this: “Think of the context window as the AI’s working desk: only so much fits on it at a time.”

Conversation history

The record of everything said in one chat. The AI remembers it within that conversation — handy for follow-ups, and worth remembering before you overshare.

Say it like this: “I scrolled the conversation history and found Tuesday’s recipe, right where we’d left it.”

Copilot

Microsoft’s AI assistant, built into Windows, Word, and Outlook.

Say it like this: “Copilot tidied my letter right inside Word — no new apps required.”

Copy

Duplicating text or a picture so you can place it somewhere else. Nothing moves — a twin is made.

Say it like this: “I copied the AI’s directions straight into an email to my sister. Retyping is a thing of the past.”

Copyright

Legal ownership of creative work. Who owns AI-made writing and art is genuinely unsettled law — a rare subject where the lawyers are as new to it as the rest of us.

Say it like this: “Who owns an AI painting? Decades in law, and my honest answer is: stay tuned.”

Credit freeze

A restriction on access to your credit report that makes it harder to open new credit accounts in your name. In the United States, placing or lifting one is free. Set it up separately with Equifax, Experian, and TransUnion, the three nationwide credit bureaus. It does not stop every kind of fraud or protect charges on existing accounts.

Say it like this: “I froze my reports with all three bureaus. When I apply for credit, I can temporarily lift the needed freeze.”

Custom instructions

A setting where you tell the AI, once, how you’d like it to talk to you — ‘use short sentences,’ ‘large print, please,’ ‘explain like I’m smart but new to this.’ It remembers for every future chat.

Say it like this: “My custom instructions say: plain English, no exclamation points, and assume I’m clever.”

Dashboard

One screen that gathers the important numbers and controls in a single view — like a car’s dashboard, but for an app or account.

Say it like this: “The health app’s dashboard shows my steps, sleep, and heart rate at a glance. My doctor is mildly jealous.”

Data

Information a computer can store: your photos, messages, purchases, steps. Guard it like mail — most is routine; some is precious.

Say it like this: “Before I use a free app, I ask the only question that matters: what data does it want?”

Data breach

When thieves break into a company’s computers and walk off with customer information. Not your fault — but your cue to change that password.

Say it like this: “The store had a data breach, so I changed that one password within the hour. Only that one — because I never reuse them.”

Data broker

A business that gathers information about people, often from several sources, and sells or shares it. That information can include contact details, purchases, interests, or public records. Changing one app’s privacy settings does not erase information already held elsewhere.

Say it like this: “That company may know things about me even though I never gave it my information directly.”

Data center

A building, or group of buildings, containing computers, networking equipment, power systems, and cooling. Data centers help run websites, online storage, and many artificial intelligence services.

Say it like this: “The cloud has an address after all: there are real computers in real buildings doing the work.”

Deep learning

Machine learning with many layers stacked deep, letting a system find patterns inside patterns. It’s the ‘deep’ behind most modern AI breakthroughs.

Say it like this: “Deep learning helps my phone recognize familiar faces in photos. It still makes mistakes.”

Deep research

A mode in several AI apps that spends several minutes searching many websites and writes you a report with sources. Excellent for big questions — still check the sources it lists before you act.

Say it like this: “I asked deep research to compare three hearing-aid options. Twelve minutes later, a tidy report with links.”

Deepfake

A picture, voice recording, or video made or altered with artificial intelligence to imitate a real person. Some are labeled jokes or parodies; scammers may present one as genuine to mislead you. It might arrive in a text or email, or appear in a social media post or advertisement.

Say it like this: “That advertisement looks like a celebrity, but I’ll check whether the endorsement is real before I buy anything.”

How to check a suspicious picture or video →

Dictation

Speaking instead of typing: tap the microphone on your keyboard or in an AI app, talk, and your words appear as text you can check before sending. The single kindest feature for tired eyes and stiff fingers.

Say it like this: “I don’t type my questions anymore. I dictate them, check them, and send.”

Digital estate

Your digital belongings and accounts: photographs, documents, email, online subscriptions, and other assets or obligations. A digital estate plan sets out what you would like to happen if you die or cannot manage them. Access can depend on account rules and applicable law, not just having a password.

Say it like this: “I want the people I trust to know where my photographs are and what I would like done with them.”

Digital will

An informal phrase that may mean instructions for digital accounts or a will prepared electronically. Those are different things. A list of wishes or passwords is not automatically a legally valid will. Get qualified help with the legal documents and account permissions for your situation.

Say it like this: “I can organize my account instructions now, then ask my lawyer how they fit with my estate plan.”

Download

Bringing a file from the internet onto your own device. Down from the cloud, into your hands.

Say it like this: “I downloaded the boarding pass, so it works even when the airport Wi-Fi doesn’t.”

Edge computing

Processing information near where it is collected—on a device, a local hub, or nearby equipment—rather than sending every task to a distant computer. An edge system may still send some information online.

Say it like this: “The doorbell may handle part of the job nearby. I still want to know what gets sent elsewhere.”

Edit (a prompt)

Changing your original question and letting the AI answer again — no apology required. Rewording is a feature, not a failure.

Say it like this: “My first question missed, so I edited it. The second cast caught the fish.”

Electronic health record (EHR)

A digital medical chart that can contain diagnoses, medicines, allergies, test results, and visit notes. Authorized care teams may share information through connected systems, but records from different offices are not always complete or connected. A patient portal may show you parts of this record.

Say it like this: “I checked my electronic health record and asked the office to correct an old medication listing.”

Encryption

Scrambling information so only the intended reader can unscramble it. The little padlock in your browser means it’s working.

Say it like this: “Encryption is the sealed envelope. Without it, everything you send is a postcard.”

Ethics in AI

Thinking carefully about what’s fair, honest, and humane when building or using AI — the questions of should, not just can.

Say it like this: “Ethics in AI is the gap between what the machine can do and what we ought to let it.”

Fact-checking

Verifying a claim against a reliable source before you repeat it or act on it. In the age of AI, this is the new washing your hands.

Say it like this: “The chatbot sounded certain, so I checked anyway. Certainty is a tone, not a source.”

Foundation model

A big, general-purpose AI model that many different tools are built on top of — the base of the cake, before anyone adds frosting.

Say it like this: “Half these new AI apps are the same foundation model wearing different hats.”

Fraud

Deception aimed at taking your money or identity. Older than the telephone — AI has merely given it better manners.

Say it like this: “Fraud used to arrive with typos. Now it arrives polished, so I verify anything that asks for money.”

Frontier artificial intelligence (frontier AI)

A term for systems near the leading edge of artificial intelligence capabilities. You may also hear frontier models. The label changes as technology develops, and being advanced does not make a system accurate at every task.

Say it like this: “That is a frontier model. I still check its answer before I rely on it.”

Gemini

Google’s AI chatbot, woven through Google Search, Gmail, Chrome, and Android phones. Free with a Google account; Gemini Live is its voice conversation mode.

Say it like this: “Gemini lives inside my Gmail, which is convenient — and worth knowing.”

Generative AI

AI that creates new things — essays, pictures, music, plans — rather than just finding existing ones. It’s the kind that started all the current excitement.

Say it like this: “Generative AI doesn’t fetch a poem about my roses. It writes one.”

Generative Pre-trained Transformer (GPT)

The type of engine behind ChatGPT. Its long name describes how the system learns patterns and generates responses; you do not need to memorize it.

Say it like this: “GPT is the engine; ChatGPT is the car built around it.”

Google DeepMind

Google’s AI research lab — the team behind Gemini and some of the field’s biggest scientific feats, like predicting the shapes of proteins.

Say it like this: “DeepMind taught a computer to fold proteins. I’m still working on fitted sheets.”

Grok

The chatbot from Elon Musk’s AI company (xAI, now called SpaceXAI after SpaceX bought it in 2026). Built into X (formerly Twitter), with its own app and website, grok.com. Known for up-to-the-minute answers and a cheeky streak — and for more public controversy than the others about what it will produce. The free version has tight limits.

Say it like this: “Grok answers with a wink. Charming — and I fact-check the winks like everything else.”

Guardrails

Rules, checks, and limits intended to reduce harmful or unwanted behavior by an artificial intelligence system. These may restrict what it can say, which tools it can use, or which actions need approval. Guardrails can fail and do not guarantee a correct answer.

Say it like this: “The guardrails help. I still read what it wrote before I use it.”

Hallucination

When an AI states something false with complete confidence. Not lying — fluent guessing beyond its knowledge.

Say it like this: “It cited a book that doesn’t exist. Charming, fluent, wrong — a textbook hallucination.”

Harness (agent harness)

The software around an artificial intelligence model that manages tools, saved progress, permissions, and when to stop. Harness engineering means designing that surrounding system. A harness helps organize and control the work; it does not guarantee that every action will be safe.

Say it like this: “The harness gives the assistant its tools, keeps track of the job, and sets its limits.”

Human-in-the-loop

A way of working that includes a person at a review, approval, or correction point. The details matter: being shown a result after a message was sent is different from approving it before it goes out.

Say it like this: “It drafts the email. I check it and decide whether to send it. That is my place in the loop.”

Identity theft

When someone uses your personal details to pose as you — opening cards, filing claims. Freezing your credit is the strongest defense, and it’s free.

Say it like this: “I froze my credit years ago. Identity thieves are welcome to knock; the door is bricked over.”

Integration

When two apps work together as one — AI inside your email, the calendar talking to the phone. Less copying and pasting; things simply appear where you need them.

Say it like this: “The AI is integrated into my email now. It didn’t knock — it was just there one Tuesday.”

Knowledge cutoff

The date used to describe how recent an artificial intelligence model’s training information generally is. It does not mean the model knows everything before that date. Newer information may come from searches, connected documents, or what you tell it—and those sources still need checking.

Say it like this: “Before I trust an answer about something recent, I ask whether it checked a current source.”

Large Language Model (LLM)

The engine inside chatbots: trained on oceans of text to predict the next word astonishingly well. The best-known come from OpenAI (ChatGPT), Anthropic (Claude), Google (Gemini), SpaceXAI (Grok), and Meta (Meta AI).

Say it like this: “The chatbot is the shop window; the large language model is the machinery in back.”

Legacy project

Using AI to help capture what only you can tell: memoirs, family recipes, letters to grandchildren, the story of how you two met.

Say it like this: “My legacy project is sixty years of recipes with the stories attached. The AI types; I remember.”

Link (hyperlink)

Clickable text or a button that jumps you to another page — usually blue, often underlined. If it arrived in an email, check where it really leads before clicking; links can lie.

Say it like this: “I hover over a link before I click it, the way I check the peephole before I open the door.”

Local computing

Processing or storing information on your own device, or on nearby equipment you control. It can work without a distant online service for that task. An app running locally may still sync files, send usage information, or use other online features.

Say it like this: “The editor works on my computer. I check its sharing and sync settings as well.”

Long-running agents

Artificial intelligence systems designed to work on a task over an extended period, sometimes across several sessions. They need a way to save progress and resume. Whether work continues when you close your laptop depends on the service and setup; it is not automatic.

Say it like this: “Before I leave it working, I ask where the job runs, how to stop it, and how I’ll know it has finished.”

Machine learning

Teaching computers by example instead of by rules — show millions of cat photos rather than defining ‘cat.’

Say it like this: “Machine learning is how the bank knew that charge in another state wasn’t me.”

Malware

Malicious software — the umbrella term for viruses, ransomware, and other digital pests built to damage or steal. Updates and healthy suspicion are the vaccines.

Say it like this: “I don’t open attachments from strangers. That one habit retires most malware before lunch.”

Meme

Said “meem,” rhyming with “team.” An image, short video, phrase, or joke people share online, often copying it and changing the caption. Think of a pet photo with a funny line underneath. Memes are not necessarily made with artificial intelligence, and being widely shared does not make their claims true. A meme can contain a deepfake, but the two words do not mean the same thing.

Say it like this: “That cat picture with the joke about Mondays is a meme. It made me laugh; it wasn’t a news report.”

Memory (in AI apps)

When an AI app remembers things about you between conversations — your name, your interests, how you like answers. Useful, and you can usually see, edit, or turn it off in Settings. Don’t let it remember anything private.

Say it like this: “I told it to forget my address. Then I checked the memory settings to make sure it had.”

Meta AI

The chatbot from Meta, tucked into Facebook, Instagram, and WhatsApp — the little blue ring in the search bar.

Say it like this: “Meta AI turned up inside WhatsApp uninvited. I said hello anyway.”

Model

The trained AI itself — the bundle of learned patterns. New models ship yearly, like car models, usually smarter.

Say it like this: “The new model is better at math, they say. It’s still not better than checking.”

Model names and versions

AI models get new numbers every few months. As of October 2026 the big ones include OpenAI’s GPT-6 (inside ChatGPT), Anthropic’s Claude Opus and Sonnet 5.5, Google’s Gemini 3.1, and SpaceXAI’s Grok 4.7. Don’t chase the numbers — the app name is what matters.

Say it like this: “Whatever version it’s on this month, I still check the important answers.”

Multimodal

An AI that can work with more than words — it can look at a photo, listen to your voice, or read a picture of a document. Point your phone at a plant, a pill bottle label, or a confusing form and ask.

Say it like this: “I photographed the thermostat and asked what the little snowflake meant. Multimodal, and very patient.”

Neural network

A computing design loosely inspired by the brain’s web of connections — layers of simple parts that together learn complex things.

Say it like this: “A neural network is millions of dimmer switches that tuned themselves until the answers came out right.”

New chat

Starting a fresh conversation so the AI arrives with no memory of the last one. A clean slate, on demand.

Say it like this: “When the answers get muddled, I open a new chat. It’s the machine’s version of a good night’s sleep.”

No-code / low-code

Building an app, website, or tool without writing programming code — you describe, drag, or click, and the software handles the technical part.

Say it like this: “I built the garden club’s sign-up page with a no-code tool. Nobody needs to know it took one afternoon.”

Notification

An alert from an app: a message arrived, a package shipped, someone liked the photo. You’re allowed to turn most of them off — and you should.

Say it like this: “I silenced every notification except family and the bank. The phone works for me, not the other way around.”

On-device artificial intelligence (local AI)

Artificial intelligence processing that runs on your own phone, tablet, or computer. This can reduce the information sent elsewhere, but an app may also use online services or share other data. On-device does not automatically mean everything stays private.

Say it like this: “This feature runs on my phone. I still check whether the app sends anything else online.”

Open source

Software whose recipe is public — anyone may inspect or improve it. Some AI models are open source; most famous chatbots aren’t.

Say it like this: “Open source means the recipe’s on the box, not just the cake.”

OpenAI

The company that makes ChatGPT and the DALL·E image generator — the lab whose 2022 launch set off the whole AI boom.

Say it like this: “OpenAI is the company; ChatGPT is the product. The names could not be more confusable, and yet.”

Opt out

Telling a company no — no to data collection, no to marketing, no to training AI on your words. The button exists; it’s just rarely on the first page.

Say it like this: “I found the opt-out in the settings, three menus deep. They hide; I seek; I win.”

Overreliance

Depending so heavily on artificial intelligence that you stop checking its work or using your own judgment. A confident answer, a friendly voice, or several good results can make this easy to slip into.

Say it like this: “It has helped me before. That does not make this answer automatically right.”

Passkey

A newer, safer replacement for passwords: your phone or computer confirms it’s you with your face, fingerprint, or screen personal identification number (PIN). Nothing to remember, and nothing a scammer can trick you into reading out.

Say it like this: “My bank offered a passkey. Now I sign in with my face, and there’s no password to phish.”

Password manager

An app that invents and remembers strong, unique passwords so you don’t reuse one everywhere. One of the best safety upgrades there is.

Say it like this: “My password manager knows 80 passwords. I know one — the one that opens it.”

Paste

Placing whatever you copied into its new home — an email, a document, a chat box. Copy’s other half.

Say it like this: “Copy, then paste: the two-step that saves me a thousand keystrokes a week.”

Patient portal

A website or app provided by a health-care practice or hospital for viewing parts of your record and communicating with the office. Depending on the service, you may see test results, request refills, or arrange appointments. Some add artificial intelligence summaries. A portal is not an emergency service.

Say it like this: “I use the portal for a routine question. For an emergency, I call 911 rather than wait for a message reply.”

Perplexity

A research-flavored AI that answers by searching the live web and showing clickable sources.

Say it like this: “I like Perplexity because it shows receipts — little footnotes I can actually check.”

Personal data

Any information about you — name, address, photos, habits, health details. Companies collect it; you’re entitled to ask why, and to say no.

Say it like this: “A flashlight app wanted my contact list. My personal data and I declined together.”

Phishing

Pretending to be a trusted person or organization to trick you into sharing private information or opening something harmful. It often arrives by email or text. The phone version is called voice phishing, or vishing. Check the request using contact details you find independently.

Say it like this: “The caller claimed to be my bank. I hung up and called the number on my card.”

What to do about a suspicious message or call →

Physical artificial intelligence (embodied AI)

Artificial intelligence connected to machines that sense or act in the physical world. A robot using cameras to recognize objects and guide its movements is one example. Not every robot uses artificial intelligence, and a demonstration does not prove a machine can handle every home.

Say it like this: “That robot has to deal with real chairs, real stairs, and the shoes I left in the hallway.”

Privacy

Your right to decide who sees your information and what they may do with it. Not secrecy — sovereignty.

Say it like this: “Privacy isn’t about having something to hide. I close the curtains at night, too.”

Prompt

Whatever you say or type to an AI. Better prompts, better answers. It’s the one skill that transfers everywhere.

Say it like this: “My favorite prompt begins: ‘Assume I’m intelligent but new to this.’ Works on people too.”

Prompt engineering

The craft of writing prompts that get good answers: be specific, give context, say what you want it to sound like. Grand name; learnable skill.

Say it like this: “Prompt engineering sounds like a degree. It’s mostly asking the way I’d brief a bright new assistant.”

Prompt injection

A trick where hidden instructions on a web page or in an email try to take over an AI assistant — telling it to reveal information or do something you didn’t ask. It’s why AI agents and AI browsers should never be left alone with your accounts.

Say it like this: “The page whispered instructions to my AI. That’s prompt injection — and why my AI doesn’t hold my passwords.”

Quantum computing

A kind of computing that uses the rules of quantum physics and units called qubits. It is being developed for particular kinds of problems. It is not simply a faster version of every computer, and ordinary chatbot use does not require a quantum computer.

Say it like this: “Quantum computing is a different technology. I do not need it to ask a chatbot for help with a letter.”

Quishing

A scam that uses a quick response (QR) code—the square pattern you scan with a phone—to lead you to a deceptive website or request. It is a form of phishing: tricking you into sharing information or granting access. A code on a sign can be replaced with a scammer’s sticker.

Say it like this: “Before I pay through a scanned code, I check where it leads. If I’m unsure, I use a way I already trust.”

Ransomware

Malware that locks up your files and demands payment for the key. A current backup turns this catastrophe into an inconvenience.

Say it like this: “Ransomware could lock my files, but I keep backups — so I’d shrug, wipe, and restore.”

Reasoning model

An AI that ‘thinks’ step by step before answering — slower, but better at math, planning, and tricky questions. Many apps show a ‘Thinking’ label or let you choose it.

Say it like this: “For the tax question, I picked the thinking mode. It took a minute. Worth it.”

Recursive self-improvement (RSI)

The idea that an artificial intelligence system could help improve its own design or capabilities, then use those improvements to improve itself again. Having a chatbot revise an answer is not the same thing. The term describes a possibility being studied, not a feature you should assume every tool has.

Say it like this: “Rewriting my letter is one thing. Improving the system that wrote it is another.”

Regenerate

Asking the AI to take another swing at its last answer. Same question, fresh attempt — sometimes markedly better.

Say it like this: “The first draft was stiff, so I hit regenerate. The second one had charm.”

Remote patient monitoring

A care arrangement in which devices, such as a connected blood pressure cuff, send health readings from home to a care team. Ask how often readings are reviewed, who responds, and what you should do if something seems wrong. Transmission does not mean someone is watching every moment.

Say it like this: “My readings go to the clinic. I asked when someone checks them and whom to call between visits.”

Responsible AI use

Using AI with your judgment firmly in charge: check the facts, guard private details, and never hand it a decision that deserves a human.

Say it like this: “I use AI the way I used a paralegal: excellent help, but I sign everything.”

Retrieval-augmented generation (RAG)

A method that finds relevant information in documents or another source and gives it to an artificial intelligence model to help answer a question. The information may be private, public, old, or current. Looking something up can help, but the model can still misunderstand it or miss a key passage.

Say it like this: “It searched the documents I supplied. I checked the passage it used before relying on the answer.”

Role prompt

Telling the AI who to be before you ask: ‘Act as a patient Medicare counselor.’ The costume changes the performance.

Say it like this: “I told it to act as a gentle physical therapist. Suddenly the exercises came with encouragement.”

Scam

Any trick designed to separate you from money, information, or trust. AI has made them smoother; the defense hasn’t changed — slow down and verify.

Say it like this: “Urgency is the scammer’s cologne. The moment I smell ‘act now,’ I act never.”

Screenshot

A photograph of whatever’s on your screen, taken by the device itself. The single best way to show someone exactly what you’re seeing.

Say it like this: “I send my friend a screenshot instead of a description. It ended our twenty-questions era forever.”

Search vs. chat

Search (Google) finds pages for you to read. Chat (ChatGPT) composes an answer for you. Search shows sources by default; chat must be asked.

Say it like this: “I search when I need the source. I chat when I need it explained.”

Second act

The projects taken up after retirement — a business, a book, a channel, a cause — with AI as tireless staff. The résumé isn’t finished; it changed departments.

Say it like this: “My second act has an assistant that never sleeps. My first act would have killed for one.”

Sensitive information

The handful of details that unlock your life: Social Security number, bank logins, full passwords, medical records. Never type these into a public AI chat.

Say it like this: “The chatbot gets my questions, never my Social Security number. Friendly is not the same as trustworthy.”

Siri / voice assistant

The talking helpers — Siri, Alexa, Google Assistant. Older, simpler cousins of chatbots, best at timers, calls, reminders, and lights.

Say it like this: “Siri sets the timer. ChatGPT explains why the roast needs one. Different tools.”

Smart speaker

A speaker with a voice assistant inside (Echo, Nest). Handy for radio, timers, and calling family hands-free.

Say it like this: “The smart speaker plays Ella Fitzgerald on request. The twentieth century had nothing like it.”

Smishing

A scam sent by text message, also called text-message phishing. It may ask you to tap a link, share information, pay a fee, or call a number. A familiar company name or an existing message thread does not guarantee that the request is genuine.

Say it like this: “The text says my package is delayed. I check the delivery company another way instead of using the message’s link.”

Social engineering

Manipulating a person into sharing information, sending money, or granting access. Scammers may use trust, fear, kindness, authority, or urgency. It can happen by phone, online, or in person; it does not require a clever computer trick.

Say it like this: “The caller sounded calm and official. I still ended the call and checked independently.”

Source citation

A note saying exactly where a fact came from, so anyone can check it. Ask your AI for sources — the good answers survive the request.

Say it like this: “I asked for its sources. Two were real, one was fiction — and that is precisely why I ask.”

Spoofing

Faking identifying information so a call, message, or website appears to come from someone else. Caller identification (caller ID) numbers and email sender details can be spoofed. What appears on the screen is not enough to establish who is contacting you.

Say it like this: “The screen showed my bank’s number. I called the number on my card myself.”

Subscription

Paying monthly for an app’s premium tier. Rule one: start free. Rule two: note what day it renews, in ink, somewhere you look.

Say it like this: “Every chatbot has a $20-a-month tier. Not one of them is necessary for learning.”

Summarization

AI condensing something long — a report, a meeting, a 40-page policy — into its key points. The details remain; you decide when to visit them.

Say it like this: “I asked for the homeowners association meeting minutes in five bullet points. Nothing was lost but the tedium.”

Symptom checker

A tool that asks about symptoms and suggests possible explanations or next steps. Some use artificial intelligence. It cannot confirm a diagnosis and may miss something serious or cause unnecessary worry. Do not wait for a chatbot answer when you think you need emergency help.

Say it like this: “It helped me organize what to tell the clinician. It did not tell me for certain what was wrong.”

Sync

Keeping the same information current across all your devices automatically — add a contact on the phone, find it waiting on the tablet.

Say it like this: “My contacts sync everywhere. I corrected Ruth’s number once, and every device got the memo.”

System prompt

Instructions supplied by the people who build or run an artificial intelligence service to guide how it behaves. They may set its role, rules, or limits before your message arrives. They are often behind the scenes, though some services explain or publish parts of them.

Say it like this: “My request is one set of instructions. The service also has instructions of its own.”

Telehealth (telemedicine)

Health care provided at a distance using technology. People often use telemedicine to mean a medical appointment by video or phone; telehealth also includes services such as remote monitoring. Captions or automated note-taking may be offered, but artificial intelligence is not required.

Say it like this: “My follow-up is by phone. I wrote down my questions before the appointment.”

Temporary chat

A conversation that isn’t saved to your history (ChatGPT calls it Temporary Chat; other apps have similar options). Good for a quick question you’d rather not keep.

Say it like this: “I used a temporary chat to ask about the lump in the mattress. Some questions don’t need an archive.”

Terms of service

The contract you accept when you tap ‘Agree.’ Nobody reads every word — do skim what they collect, what they share, and what canceling costs.

Say it like this: “I paste the terms of service into a chatbot and ask what I’m really agreeing to. Someone should read them — even if it’s a machine.”

Token

The word-pieces AI actually reads and writes — ‘grandmother’ might be two or three tokens. Explains why limits are phrased so oddly.

Say it like this: “The machine doesn’t count words, it counts tokens. Like a butcher weighing by the ounce.”

Token budget

A limit on the amount of material an artificial intelligence system can process or generate for a task, measured in tokens—small pieces of text or other information. It can help control cost and length. It is not a reliable clock or a guarantee that a job will finish.

Say it like this: “A token budget limits how much it can process. I set a separate time or spending limit if the service offers one.”

Token economics

How token usage affects the cost of running an artificial intelligence system. Reading your request and producing an answer may be priced differently, and tools can add charges. Token economics concerns costs and trade-offs; a token budget is a specific limit.

Say it like this: “Before I run a big job, I want to understand what I’m paying for.”

Tool use (function calling)

An artificial intelligence system using a connected program or service to do something beyond writing an answer, such as searching, checking a calendar, or creating a file. Function calling is one way developers connect those tools. Access depends on what has been enabled and permitted.

Say it like this: “It can check my calendar because I connected that tool. Sending invitations is a separate decision.”

Training

The process of adjusting an AI model using examples so it learns patterns. Training data is the practice material; training is how the model learns from it.

Say it like this: “The examples are the practice material. Training is how the model learns what to do with them.”

Training data

Examples a model learned from, such as text, pictures, or recordings. They shape its abilities and limitations. A model does not know every fact in that material, and its answers may also draw on searches, connected sources, or information you supply.

Say it like this: “It hadn’t heard the news from last week — that happened after its training data was gathered.”

Transcription

AI turning spoken words into written text — voice memos become notes; rambling meetings become searchable documents.

Say it like this: “I recorded Aunt Ida’s stories at Thanksgiving. Transcription turned two hours of talk into forty tidy pages.”

Translation

AI converting one language into another, in text or out loud, nearly instantly — good enough now for travel, letters, and long-lost cousins.

Say it like this: “I write to my cousin in Palermo in English; she reads it in Italian. The machine handles the middle.”

Two-factor authentication (2FA)

A second lock on an account: your password plus a code sent to your phone. Mild nuisance, major protection. Turn it on for email and banking.

Say it like this: “Two-factor saved the day: some stranger had my password, but they didn’t have my phone.”

Update

A newer version of an app or system, delivered over the internet — mostly bug fixes and security patches. Installing them promptly is genuine self-defense.

Say it like this: “I install updates the day they arrive. A lock only protects you if you use the newest one.”

Upload

Sending a file from your device up to a website or app — attaching the photo, submitting the document. Download’s opposite.

Say it like this: “I uploaded the insurance letter and asked the AI about page three. It read the whole thing without sighing.”

Vibe coding

Building software by describing what you want in plain English and letting AI write the actual code. You supply the vision and the taste; it supplies the semicolons.

Say it like this: “I vibe-coded a little quiz for my friends. I described, the machine typed, and we both took credit.”

Virtual private network (VPN)

A service that wraps your internet connection in a private tunnel — most useful on public Wi-Fi, where the café’s network is anybody’s guess.

Say it like this: “On hotel Wi-Fi I switch on the VPN. It’s a curtain around my little booth of internet.”

Virus

Malware that spreads from device to device, like its biological namesake. Modern phones and computers resist them well — updates keep it that way.

Say it like this: “Updates are the flu shot. The virus prefers an unvaccinated computer.”

Vishing

Voice phishing: using a phone or voice call to trick you into sharing information, sending money, or giving access. The voice may be a real person, a recording, or a computer-generated imitation.

Say it like this: “The caller claimed to be from the Internal Revenue Service and demanded immediate payment. I ended the call and checked independently.”

Voice clone

An AI copy of a real person’s voice, made from a short sample. Beautiful for preserving stories; dangerous in scams. Consent is everything.

Say it like this: “Three seconds of audio can clone a voice. That’s why our family has a code word.”

Voice mode

Having a spoken, back-and-forth conversation with an AI — you talk, it answers out loud, like a phone call. ChatGPT, Claude, Gemini (Gemini Live), and Grok all offer it; how much is free varies. You can ask it to slow down or speak more simply.

Say it like this: “I asked voice mode to slow down. It did, and it didn’t sigh once.”

Wearable

An electronic device you wear, such as a smartwatch, fitness tracker, or certain health monitors. It may track movement, heart rate, sleep, or other measurements. Features and medical uses vary. A consumer gadget’s reading or alert is not, by itself, a diagnosis.

Say it like this: “My watch noticed a change. I asked my care team what that reading could—and could not—tell me.”

Wi-Fi vs. cellular data

Two roads to the internet: Wi-Fi (your home network, free to use at home) and cellular data (the phone company’s network, metered on many plans).

Say it like this: “On Wi-Fi at home, stream away. On data at the beach, maybe not the whole opera.”

Workflow

The sequence of steps that gets a job done, start to finish. Name yours out loud and an AI can often shorten it.

Say it like this: “My newsletter workflow had nine steps. The AI absorbed four; I kept the fun five.”

xAI

Elon Musk’s artificial intelligence company, maker of Grok. In February 2026 it became part of SpaceX and now goes by SpaceXAI.

Say it like this: “xAI, SpaceXAI — whatever it’s called this year, Grok is the chatbot.”