A practical beginning, without the magic tricks.
Generative AI can draft, summarize, compare, explain, and organize. It predicts useful responses from patterns and the context you provide. That makes it powerful, but not automatically correct.
Fluency is not evidence.
An AI response can sound certain even when it is filling a gap with a plausible inference. It does not “know” your policies, definitions, priorities, or current guidance unless those are available in its approved context.
Treat the system like a brilliant new colleague. Explain the assignment, provide the material it should use, describe the audience and constraints, and review the result before relying on it.
Start small
Begin with drafting, brainstorming, reformatting, or summarizing work that a person will review.
Give context
State the task, audience, purpose, source material, constraints, and desired form.
Protect information
Use only approved environments and do not enter protected or restricted information without authorization.
Check the answer
Verify names, dates, numbers, quotations, citations, and current claims against authoritative sources.
Plain-language terms for practical use.
Use the ranges below to jump through the glossary. These definitions are written for education leaders and new AI users.
A–D
- Agent
- An AI-enabled system designed to pursue a goal through a sequence of steps, sometimes using tools or taking approved actions.
- Algorithm
- A defined set of rules or calculations used to process information and produce a result.
- Artificial intelligence
- A broad category of computer systems that perform tasks associated with human capabilities such as language, prediction, recognition, and decision support.
- Automation
- The use of technology to complete a repeatable task or workflow with limited manual effort.
- Context
- The instructions, source material, conversation, constraints, and background information available to an AI system for a specific request.
- Context window
- The amount of information a model can consider at one time when producing a response.
- Data governance
- The roles, rules, standards, and processes used to keep organizational data secure, reliable, understandable, and appropriately used.
E–G
- Embedding
- A numerical representation that helps an AI system compare meaning and find related information.
- Fine-tuning
- Additional model training using selected examples to shape performance for a particular task or domain.
- Foundation model
- A large model trained on broad data that can be adapted or instructed to perform many different tasks.
- Generative AI
- AI that creates new content such as text, images, audio, video, or code in response to instructions.
- Grounding
- Connecting an AI request to trusted source material so the response relies less on unsupported inference.
H–M
- Hallucination
- A fluent but inaccurate, unsupported, or invented AI output.
- Human in the loop
- A workflow in which a qualified person reviews, approves, corrects, or stops an AI-supported action.
- Inference
- The process of using a trained model to produce an output from new input.
- Large language model
- A model trained on large amounts of language data to predict and generate text and related forms of content.
- Machine learning
- A method in which computer systems learn patterns from data instead of relying only on explicitly programmed rules.
- Model
- A trained computational system that maps inputs to predictions, classifications, recommendations, or generated content.
- Multimodal
- Describes a model that can work with more than one type of information, such as text, images, audio, or video.
N–R
- Natural language processing
- Methods that help computers analyze, interpret, and generate human language.
- Prompt
- The instructions and information given to an AI system to guide its response.
- Prompt injection
- An attempt, sometimes hidden inside source material, to make an AI system ignore its intended instructions or reveal information.
- Reasoning
- The structured intermediate work used to analyze a problem, compare possibilities, or plan steps before producing an answer.
- Responsible AI
- The practice of designing, selecting, and using AI with appropriate attention to people, rights, safety, fairness, transparency, privacy, and accountability.
- Retrieval-augmented generation
- A method that retrieves relevant material from selected sources and supplies it to a generative model when forming a response.
S–Z
- System prompt
- Higher-priority instructions that establish an AI system’s role, behavior, boundaries, or operating context.
- Token
- A small unit of text processed by a language model, which may be a word, part of a word, punctuation, or another symbol.
- Training data
- The information used to teach a model patterns during its development.
- Transformer
- A model architecture that identifies relationships among parts of an input and underlies many modern language models.
- Verification
- The deliberate process of checking an AI output against authoritative sources, requirements, calculations, or expert judgment before use.
Move from user to operator.
As consequences grow, so must the quality of your direction, sources, review, and controls. The AI Operator’s Manual turns these foundations into a repeatable professional practice.