TERMS & ABBREVIATIONS

Quick reference for common AI terminology and acronyms.

A

AGI (Artificial General Intelligence)
Theoretical AI that could perform any intellectual task a human can, with true understanding and reasoning. We don't have this yet. Current AI (like ChatGPT) is narrow AI, not AGI.

Agent / Agentic AI
An AI system that takes actions autonomously rather than just answering questions. For example, booking flights after searching sites, or scheduling meetings by reading your emails. Most "AI agents" in 2026 are rebranded automation tools with better interfaces—the concept isn't new, just the marketing term.

Algorithm
A set of rules or instructions that tells a computer how to solve a problem or complete a task. AI systems use algorithms to process data and make predictions.

Alignment
How well an AI system’s behaviour matches human values, rules, and intentions. Misalignment is why AI can be helpful yet still produce harmful or misleading outputs.

API (Application Programming Interface)
A way for different software systems to connect and communicate. Many AI tools you use are actually connecting to ChatGPT or Claude's API behind the scenes.

Automation Bias
The tendency for people to trust AI outputs too much, even when they’re wrong — especially dangerous in finance, medicine, and security.

B

Bias (in AI)
When an AI system produces unfair or skewed results because of problems in its training data. For example, if an AI was trained mostly on data from one demographic, it may perform poorly for others.

Black Box (AI Black Box)
The fact that many AI systems cannot fully explain how they reached a decision, even to their creators.

Bot
Short for robot, but in AI context usually means an automated program that performs tasks. Chatbots are bots that have conversations.

C

Chatbot
A conversational interface that lets you interact with AI by typing or speaking naturally. ChatGPT, Claude, and Gemini are all chatbots.

ChatGPT
OpenAI's AI chatbot, released in November 2022, that became the fastest-growing consumer application in history. The name combines "Chat" with "GPT" (Generative Pre-trained Transformer).

Claude
Anthropic's AI chatbot, known for longer conversations and detailed writing. Named after Claude Shannon, founder of information theory.

Context Window
The amount of text an AI can “see” and remember at once. Larger context windows allow longer conversations and documents but still aren’t true memory.

Copyright (AI Context)
The legal question of whether AI training or outputs infringe on creators’ rights. Still unresolved in many countries.

D

Data
Information fed into AI systems for training or processing. In AI context, usually means text, images, or other content the system learns from.

Dataset
A structured collection of data used to train or evaluate an AI model.

Deepfake
AI-generated fake video or audio that makes it look or sound like someone said or did something they didn't. Used in scams to impersonate voices or faces.

Disinformation
False information spread deliberately. AI lowers the cost and effort required to produce disinformation at scale.

E

Embedding
A way AI represents text, images, or data as numbers so it can compare meaning and similarity. Used heavily in search and recommendation systems.

Ethics (AI Ethics)
The study of right and wrong in how AI is developed and used, including questions about bias, privacy, job displacement, and safety.

Explainability
The ability to understand why an AI made a particular decision — a major issue in regulated industries.

F

Fine-tuning
Adjusting a pre-trained AI model for a specific task or to follow particular guidelines. For example, ChatGPT was fine-tuned to be more helpful and less harmful than the base model.

Foundation Model
A large, general-purpose AI model trained once and reused across many tasks (e.g. GPT-4, Claude, Gemini).

G

GDPR (General Data Protection Regulation)
European comprehensive data privacy law that gives people control over their personal data. Similar laws exist in UK, Canada, Australia, and New Zealand. Affects how AI companies can collect and use your information.

Gemini
Google's AI chatbot (formerly called Bard). Integrated with Google services like Search, Gmail, and Docs.

Generative AI
AI that creates new content (text, images, audio, video) rather than just analysing existing data.

GPT (Generative Pre-trained Transformer)
The underlying technology behind ChatGPT. "Generative" means it creates new content, "Pre-trained" means it learned from massive amounts of text first, "Transformer" is the architecture it uses.

H

Hallucination
When AI confidently states false information as if it were fact. This happens because AI predicts plausible-sounding text, not because it knows what's true.

HIPAA (Health Insurance Portability and Accountability Act)
US law that protects medical information privacy. If you're sharing health details with AI tools, HIPAA determines whether that information is legally protected. Most consumer AI chatbots are NOT HIPAA-compliant.

Human-in-the-Loop
Systems where humans still approve, guide, or override AI decisions — common in safety-critical uses.

I

Inference
The moment when a trained AI model actually produces an output in response to a prompt. Training happens once; inference happens every time you use it.

Input
The information or question you give to an AI system. Also called a prompt or query.

J

Jailbreak (AI Jailbreak)
Attempts to trick AI systems into ignoring their safety guidelines or restrictions. Not something you should try, but worth knowing the term exists.

K

Knowledge cutoff
The date after which an AI system has no training data and therefore no reliable knowledge. ChatGPT's knowledge cutoff is October 2023, meaning it doesn't know events after that date unless it searches the web.

L

Latency
The delay between submitting a prompt and receiving a response. Lower latency feels “snappier.”

LLM (Large Language Model)
The underlying technology behind AI chatbots like ChatGPT, Claude, and Gemini. A type of AI trained on massive amounts of text to understand and generate human-like language.

M

Machine Learning
A type of AI where systems learn patterns from data without being explicitly programmed for every scenario. The foundation of modern AI.

Metadata
Data about data. For example, a photo's metadata includes when it was taken, where (GPS), what device, file size, and camera settings. Can reveal information even when the main content doesn't.

Model
In AI context, the trained system that processes your input and generates output. GPT-4, Claude 3.5, and Gemini are all models.

Model Drift
When an AI’s performance degrades over time because real-world data changes.

Multimodal AI
AI systems that can handle multiple input types — text, images, audio, video — in one model.

N

Neural Network
A type of AI architecture loosely inspired by how human brains process information. Modern AI like ChatGPT uses neural networks with billions of connections.

NLP (Natural Language Processing)
The field of AI focused on helping computers understand, interpret, and generate human language.

O

OpenAI
The company that created ChatGPT, GPT-4, and DALL-E. Founded in 2015, became famous when ChatGPT launched in 2022.

Output
The response or result an AI system generates based on your input.

Overfitting
When a model learns its training data too well and performs poorly on new or unfamiliar inputs.

P

Parameters
The internal settings in an AI model that control how it processes information. More parameters generally means more capability. GPT-4 has hundreds of billions of parameters.

Perplexity
An AI chatbot focused on search and research, providing cited answers to questions.

Phishing
Scam attempts (usually via email or message) trying to trick you into revealing personal information. AI makes phishing more convincing by generating personalized, error-free messages.

PII (Personally Identifiable Information)
Any data that can identify you specifically: name, address, phone, email, Social Security number, medical records, financial details, or location history. When AI companies talk about "protecting user privacy," they're referring to how they handle PII.

Prompt
The instruction, question, or request you give to an AI system. For example, "Write a summary of this article" or "What's the weather today?" The quality of your prompt will affect the quality of the AI's response.

Q

Query
Another word for a question or search request you give to an AI system or search engine.

R

Red Teaming
Deliberately trying to break or misuse AI systems to discover weaknesses before real attackers do.

Reinforcement Learning from Human Feedback (RLHF)
A training method where humans rate AI responses as good or bad, helping the system learn to give better answers. Used to train ChatGPT and Claude.

Retrieval-Augmented Generation (RAG)
A technique where AI looks up external documents or databases before answering, reducing hallucinations.

S

Siri
Apple's voice assistant, built into iPhones, iPads, and Macs. Uses AI for voice recognition and some smart features.

Synthetic Data
Artificially generated data used to train AI when real data is scarce, sensitive, or restricted.

System Prompt
Hidden instructions that shape how an AI behaves before user prompts are applied.

T

Temperature
A setting that controls how creative or predictable AI responses are. Higher temperature = more variety, more risk.

Token
A piece of text (roughly a word or part of a word) that AI processes. AI models have token limits — the maximum amount of text they can handle in one conversation.

Training Data
The massive collection of text, images, or other information used to teach an AI system. What an AI knows comes entirely from its training data.

Transformer
The architecture behind modern AI language models. Introduced by Google in 2017, it made current AI chatbots possible.

U

User Interface (UI)
How you interact with a system. ChatGPT's user interface is a simple chat box; other AI tools might have more complex interfaces.

V

Vector Database
A database designed to store embeddings, enabling fast semantic search and memory-like behaviour.

Voice Cloning
AI technology that can replicate someone's voice from audio samples. Used in scams to impersonate family members or authority figures.

W

Web Scraping
Automatically collecting data from websites. Some AI training data comes from web scraping, which raises copyright and privacy questions.

X

XAI (Explainable AI)
Techniques designed to make AI decisions more understandable to humans.

xAI
A private AI company founded by Elon Musk in 2023. Despite the similar name, xAI has nothing to do with Explainable AI as a concept. It develops AI models and products (such as Grok) and is closely associated with the X (formerly Twitter) platform.

Y

(No Ys yet!)

Z

Zero-shot Learning
When an AI can perform a task it wasn't explicitly trained for, just from understanding the instruction. For example, asking ChatGPT to translate a language it wasn't specifically trained to translate.

Z-Score (AI Monitoring)
A statistical measure sometimes used to detect unusual AI behaviour or outputs in monitoring systems.

This page is updated as new terms are added. Last updated: January 2026