Glossary — Every AI Word in Plain English
Every term you'll hear, defined in one or two sentences. No "see also," no rabbit holes. Bookmark this and come back as needed.
Bookmark this page. When you hit a word in the wild and have no idea what it means, come back. Every term is one or two sentences. No fluff.
Agent. An AI that takes multiple steps on its own to finish a task — like booking a flight, not just describing how to. Still rough in 2026, getting better fast.
API. Short for “Application Programming Interface.” A way for one piece of software to talk to another. You don’t need this to use AI in a chat — you only need it if you’re building software.
Bias. When AI gives systematically slanted answers because of what it was trained on. The internet is biased; AI inherits some of that.
Chatbot. A program you talk to through text. ChatGPT is a chatbot. Customer service “Hi, how can I help?” pop-ups are chatbots. Most of what people call “AI” is technically a chatbot.
Claude. The chatbot made by Anthropic. Known for thoughtful writing and longer attention span.
Context window. How much you can paste in at once before the AI starts forgetting the beginning. Modern context windows are huge — entire books fit.
Custom GPT. A version of ChatGPT you’ve configured for one specific job (e.g., “my resume coach”). Available on the paid tier. We have a whole course on this.
Data privacy. What AI companies do with your conversations. Most let you turn off “use my chats for training.” Read the settings; default behavior varies.
Deep learning. A specific kind of machine learning that uses many layers of math to find patterns. The technique behind almost all modern AI. You don’t need to understand it to use AI.
Embedding. A way to turn text into a list of numbers so a computer can compare meanings. Used under the hood for search and recommendations. You can ignore this.
Fine-tuning. Taking a general AI model and training it more on a specific topic, so it gets better at that topic. Most people don’t need this — clever prompting goes 90% of the way.
GPT. Stands for “Generative Pre-trained Transformer.” The technical name for the family of models behind ChatGPT. In casual use, “GPT” and “ChatGPT” are used interchangeably.
Hallucination. When AI confidently states something that isn’t true. The number-one thing to watch out for. See When Not to Trust AI.
Inference. The act of an AI generating an answer. Inference takes computing power, which is why fast or huge models cost money to run.
Jailbreak. Tricking an AI into ignoring its safety rules. Generally a waste of time, often against terms of service, sometimes legally questionable. Skip.
Knowledge cutoff. The date past which an AI doesn’t know about the world. If a model has a cutoff of January 2025, it doesn’t know what happened in March 2025 — unless it’s hooked up to web search.
LLM. “Large Language Model.” The specific type of AI behind ChatGPT, Claude, and Gemini. When people say “AI” today, they usually mean an LLM.
Machine learning. The bigger umbrella that AI fits inside — software that learns patterns from data instead of being told rules. AI is one application of machine learning.
Model. The actual AI program doing the work. “GPT-4,” “Claude 3.5 Sonnet,” “Gemini Advanced” — those are model names. New models come out frequently and are usually smarter than older ones.
Multimodal. AI that can handle more than just text — also images, audio, or video. Modern ChatGPT, Claude, and Gemini are all multimodal.
Open source. Models whose underlying code and weights are public, so anyone can run them. Llama (Meta), Mistral, and DeepSeek are popular examples. ChatGPT and Claude are not open source.
Plugin / Tool. An add-on that lets AI use external services — search the web, run code, look up real-time data. Mostly handled automatically inside paid tiers now.
Prompt. What you type into the AI. Your message. That’s all this word means.
Prompt engineering. A fancy term for “writing good prompts.” Used to be a real specialty; in 2026 it’s mostly common sense + the 5-pattern framework.
Prompt injection. A security risk where someone hides hostile instructions in a document or website that tricks an AI into doing something malicious. Mostly relevant if you’re building AI products.
RAG. “Retrieval-Augmented Generation.” A technique where AI looks up info in a specific database (like your company’s docs) before answering. You don’t need to know this to use AI day-to-day.
Reasoning model. A newer kind of AI that thinks step-by-step before answering — better at math, logic, complex problems. ChatGPT’s “o3,” Claude’s “extended thinking,” Gemini’s “thinking” mode.
Reinforcement learning from human feedback (RLHF). How AI gets trained to be helpful and not awful — by humans rating its answers. Largely why modern AI feels conversational instead of robotic.
Sora. OpenAI’s video-generating model.
System prompt. A behind-the-scenes instruction that sets up how the AI should behave for an entire conversation. Custom GPTs are essentially saved system prompts.
Temperature. A technical setting that controls how “creative” or “random” AI output is. Higher = more varied answers. You can ignore this in chat tools — they pick reasonable defaults.
Token. Roughly, a chunk of a word. AI processes text in tokens. Useful only if you’re paying per-token via the API. Otherwise ignore.
Training. The process of feeding huge amounts of data into a model so it learns patterns. Models are trained once at huge cost, then run forever after.
Transformer. The specific math architecture behind modern AI (the “T” in GPT). Invented in 2017. You will never need to know how it works.
Vector database. A type of database optimized for finding “similar meaning” rather than exact matches. Powers most AI search and RAG. Power-user stuff.
Voice mode. Talking to AI out loud. Available in ChatGPT, Claude, and Gemini. Surprisingly natural.
Don’t see a word? Let us know in the community. We add to this list every week.
Where to go next
You’ve finished the Beginner Track. 🎉 Three good next steps:
- Browse the Guides for tool comparisons and decision frameworks.
- Try the 50-Prompt Starter Library.
- If you’re ready to do real work with AI — pick a Pro Course.
Get the next lesson