AI glossary

The terms you'll run into, explained without jargon. Short on purpose.

Agent: An AI system that can take several steps on its own, such as searching, reading, and drafting, to reach a goal you set. You stay responsible for checking the result.
AI assistant: A chat-style tool you can ask to draft, summarize, explain, or brainstorm. It responds to what you type, and you decide what to do with the answer.
API: A way for one piece of software to talk to another. Many automations use APIs behind the scenes to move information between your apps.
Automation: A set of steps that runs by itself when something triggers it, like saving every emailed receipt to a folder.
Bias: A pattern in AI output that favors some people or views over others, usually learned from the data it was trained on.
Chatbot: A program you talk to in everyday language. Some follow scripts; modern AI chatbots generate their replies.
Context: The information you give an AI in a conversation: your instructions, examples, and pasted material. Better context usually means better answers.
Context window: The amount of text an AI can consider at once. If a conversation grows past it, earlier parts may be forgotten.
Data privacy: Keeping personal or confidential information out of places it shouldn't go. Check a tool's policy before pasting sensitive material.
Delegation: Handing a task to someone, or something, else with clear instructions and a way to check the work. It is the core skill of working with AI.
Fine-tuning: Extra training that adapts an existing AI model to a specific style or task. Most everyday users never need it.
Generative AI: AI that creates new content such as text, images, audio, or code, based on patterns it learned from examples.
Guardrails: Rules or checks that keep an AI tool from doing things you don't want, such as sending an email without your approval.
Hallucination: When an AI states something false with confidence, like an invented quote or statistic. Verify anything that matters.
Human in the loop: A design where a person reviews or approves AI output before it has any effect. A good default for anything important.
Image generation: Creating pictures from a text description. Check the tool's rules on commercial use and on likenesses of real people.
Inference: The moment an AI model produces an answer from your input, as opposed to the earlier training phase.
Integration: A connection between two tools so they can share information, like linking your calendar to an assistant.
Large language model (LLM): An AI model trained on huge amounts of text to predict and generate language. It powers most chat assistants.
Machine learning: A way of building software that learns patterns from examples instead of following only hand-written rules.
Memory: A feature that lets an assistant remember details between conversations. Review and control what it stores.
Model: The trained system that produces AI output. Different models vary in speed, cost, and ability.
Multimodal: Able to work with more than one kind of input or output, such as text and images together.
No-code: Building automations by connecting blocks or filling in forms instead of writing programs.
Open-source model: A model whose weights or code are published so others can run or adapt them, under the terms of its license.
Output: What the AI gives back to you. Treat it as a draft until you have checked it.
Parameter: A number inside a model that was adjusted during training. Models are often described by how many they have.
Prompt: The instructions and information you give an AI. Clear prompts state the goal, the context, and the format you want.
Prompt engineering: The practice of wording and structuring prompts to get more useful results. Mostly it means being specific.
RAG (retrieval-augmented generation): A setup where the AI looks up relevant documents first and then answers using them, which helps keep answers grounded in your material.
Reasoning model: A model designed to work through problems step by step before answering, often slower but better at multi-step tasks.
Role prompt: A prompt that tells the AI what perspective to take, like a patient tutor or a careful editor.
System prompt: Background instructions that shape how an assistant behaves across a whole conversation.
Temperature: A setting that controls how predictable or varied the output is. Lower is steadier; higher is more creative.
Token: A small chunk of text, often a word or part of a word, that models read and write. Limits and prices are often counted in tokens.
Training data: The examples a model learned from. It shapes what the model knows, and what it gets wrong.
Trigger: The event that starts an automation, such as a new email arriving or a form being submitted.
Voice assistant: An assistant you talk to out loud. Some now use the same models as chat assistants.
Workflow: A repeatable sequence of steps to get something done, whether by hand, by AI, or both.
Zero-shot / few-shot: Asking for a result with no examples (zero-shot) or with a few examples included in the prompt (few-shot). Examples usually improve consistency.