The AI Slang Glossary

An AI system that doesn't just answer a question — it takes actions in sequence to complete a task. A basic AI tool responds to each prompt individually. An agent can be given a goal, break it into steps, use tools (search engines, calendars, email), and work through the steps without you prompting each one. They are genuinely useful. They also do things autonomously, which raises questions worth understanding before you hand over access to your inbox.

Last updated: March 2026

Application Programming Interface. A way for one piece of software to talk to another. When someone says "we connected our app to the ChatGPT API," they mean they wrote code that sends requests to ChatGPT and gets answers back — without anyone opening a browser. APIs are how AI gets built into apps, websites, and tools you already use.

Last updated: March 2026

Using technology — including AI tools — to perform a task that would otherwise require human time and attention. In an AI context, automation often means setting up a tool to handle a repeated task: drafting responses to routine emails, organising data, generating regular reports. The goal is to remove yourself from the loop for low-value repetitive work.

Last updated: March 2026

Any AI tool you interact with through a conversation-style interface — you type something, it types back. ChatGPT is a chatbot. So is Claude, Gemini, and Grok. The word has been around much longer than modern AI — early chatbots followed simple rules and were mostly terrible. Today's chatbots are powered by large language models, which makes them vastly more capable.

Last updated: March 2026

The amount of text an AI tool can "hold in mind" at once during a conversation. Older models had small context windows — they'd forget what you said at the beginning of a long conversation. Newer models have much larger windows. When a tool starts giving answers that seem to ignore something you said earlier, it has usually run out of context window.

Last updated: March 2026

Taking an existing AI model and training it further on a specific set of data so it becomes better at a particular task. Think of a general-purpose AI as a broadly educated graduate — fine-tuning is like sending them to a specialist course. A company might fine-tune a model on their own support tickets so it gives answers that match their products and tone.

Last updated: March 2026

AI that creates new content — text, images, audio, video, code — rather than simply analysing or organising existing content. When people say "AI" in most everyday conversations, they mean generative AI. ChatGPT generates text. DALL-E generates images. These are all generative AI tools.

Last updated: March 2026

When an AI tool produces information that is confidently stated but factually wrong. Not a glitch, not a virus — the AI is doing exactly what it's designed to do (predict the most likely next word based on patterns), and sometimes that produces plausible-sounding nonsense. It's an inherent limitation, not a scandal. It means you should verify important facts, especially for anything consequential.

Last updated: March 2026

Using AI to create images from text descriptions. You type "a watercolour painting of a cat reading a newspaper" and the AI produces an image matching that description. Tools like DALL-E, Midjourney, and Stable Diffusion do this. The results can be stunning, bizarre, or both — and they're improving rapidly.

Last updated: March 2026

The date after which an AI model has no training data. If a model has a knowledge cutoff of April 2024, it genuinely does not know about anything that happened after that date — elections, product launches, news events. Some tools compensate by connecting to the internet for current information, but the base model itself stops at that line.

Last updated: March 2026

The type of AI behind tools like ChatGPT, Claude, and Gemini. "Large" refers to the enormous amount of text it was trained on. "Language model" means it works by predicting which words are most likely to come next in a given context — which, when done with enough data and computing power, produces responses that can feel surprisingly thoughtful. It is not thinking. It is very sophisticated pattern recognition.

Last updated: March 2026

The trained AI system that powers a tool. When someone says "GPT-4" or "Claude 3.5," they're referring to a specific model. A model is the result of training — the thing that has absorbed patterns from data and can now generate responses. One model can power many different tools and products. ChatGPT is a product; GPT-4 is the model behind it.

Last updated: March 2026

An AI model that can work with more than one type of input — not just text, but also images, audio, or video. GPT-4o is multimodal: you can show it a photo and ask what's in it, or upload a document and ask it to summarise the contents. A text-only model can't do that.

Last updated: March 2026

Whatever the AI produces in response to your prompt. Could be text, an image, code, a table, a list — anything the tool generates. The quality of the output depends almost entirely on the quality of the input. Better prompts, better output. That's the whole game.

Last updated: March 2026

The instruction or question you type into an AI tool. If you ask ChatGPT "write me an email declining a meeting," the whole thing — every word of it — is your prompt. The quality of what an AI produces depends heavily on the quality of what you ask. A vague prompt produces a vague result. A specific prompt produces something useful.

Last updated: March 2026

The practice of carefully crafting your prompts to get better results from AI tools. It sounds more technical than it is. In practice, it means being specific about what you want, giving the AI a role to play, and telling it what format you want the answer in. Role + Task + Format. That's the structure. That's what Prompt School teaches.

Last updated: March 2026

A technique where an AI tool looks up relevant information from a specific source — like a company's documents or a database — before generating its answer. Instead of relying solely on what it learned during training, it retrieves current, specific data first. This makes answers more accurate and grounded in real, up-to-date information.

Last updated: March 2026

A setting that controls how creative or predictable an AI's output will be. Low temperature (closer to 0) makes the AI stick to safe, expected answers. High temperature (closer to 1 or 2) makes it more creative and unpredictable. For factual tasks, low temperature is better. For brainstorming or creative writing, higher temperature can be useful.

Last updated: March 2026

The basic unit of text that an AI model processes. A token is roughly three-quarters of a word in English. "Hamburger" is one token. "I love hamburgers" is about four tokens. AI tools charge by the token and measure context windows in tokens. When you hit a limit, it's usually a token limit.

Last updated: March 2026

The text that an AI model learned from before it was made available to the public — books, articles, websites, code, and other written material, totalling billions of words. The model doesn't "remember" any of it the way you remember reading a book. It absorbed patterns from it. What an AI knows, and the limits of what it knows, come from what was in that training data.

Last updated: March 2026