Build your AI fluency

A broad field where machines perform tasks that normally need human intelligence.

AI systems learn patterns from data instead of being explicitly programmed for every rule.

Machine learning using neural networks with many layers, often used for language, images, and speech.

AI that creates new content such as text, images, audio, video, code, summaries, and designs.

A large general-purpose model that can be adapted for many tasks.

A model trained on huge amounts of text to predict and generate language.

AI that can work with more than one input or output type, such as text, image, voice, video, or files.

The trained AI system that receives input and produces output.

Learned numerical values inside a model that help it make predictions.

A small unit of text. It can be a word, part of a word, punctuation mark, or symbol.

The process of breaking text into tokens before the model processes it.

A numerical representation of meaning. Similar meanings are represented by vectors that are closer together.

A list of numbers that represents information mathematically.

Converting text, image, audio, or other data into vectors.

A mathematical structure that learns patterns through connected layers of calculations.

The architecture behind most modern LLMs, especially strong at handling sequences of text.

The mechanism that helps a model decide which tokens matter most for the current output.

A process where tokens compare themselves with other tokens in the same input to understand relationships.

The basic generation mechanism: predict the next token repeatedly.

Using a trained model to produce an answer.

A setting that controls randomness. Lower is more predictable, higher is more varied.

The amount of information the model can consider at one time.

Tokens sent to the model, including prompt, files, chat history, instructions, and retrieved context.

Tokens generated by the model in response.

The maximum number of tokens the model can handle in one interaction.

Previous messages included in the current context.

Information currently available inside the context window.

Information saved outside the context window and brought back later when relevant.

Designing what information should be included in the model's context to improve output.

The instruction or question given to the AI.

Writing prompts in a way that improves the answer.

Higher-priority instruction that defines how the AI should behave.

Product or app-level instruction that guides the model's behavior.

The user's actual request to the AI.

Asking the model to perform a task without examples.

Giving examples so the model understands the desired pattern.

Asking for step-by-step reasoning, though many systems hide private reasoning and show a summary.

Asking the model to act as a specific role, such as coach, teacher, analyst, or interviewer.

The priority order of instructions, usually system, developer, user, and context.

Data used to train the model.

The first major training stage where the model learns general patterns from large datasets.

Additional training to make a model better at a specific task, domain, or style.

Reinforcement Learning from Human Feedback. Humans rank or evaluate outputs to improve model behavior.

AI-generated data used for training, testing, or simulation.

Training a smaller model to imitate a larger model.

The process of making model behavior better match human instructions, preferences, and safety expectations.

Retrieval-Augmented Generation. The model retrieves external knowledge before generating an answer.

A collection of documents, data, or content the AI can search.

Breaking large documents into smaller parts for retrieval.

How large each chunk of text is.

A database that stores embeddings and helps find similar content.

Preparing and organizing documents so they can be searched quickly.

Finding relevant information from a knowledge base.

Search based on meaning rather than exact keywords.

Search based on exact words or phrases.

Making the model's answer rely on provided sources or retrieved data.

Showing where the answer came from.

The model uses an external tool such as search, calculator, code, calendar, email, CRM, or database.

A structured way for the model to call a predefined function.

A way for software systems to talk to each other.

An AI system that can plan steps, use tools, check results, and continue working toward a goal.

Using AI plus tools to complete a repeatable process.

A person checks, approves, or guides the AI before important actions.

Model Context Protocol. An open standard for connecting AI applications to external systems, tools, and data.

The system that exposes tools, data, or resources to an AI application.

The AI application that connects to MCP servers.

An action the model can perform through a protocol or connector.

A bridge between AI and external software such as Drive, Slack, GitHub, databases, CRMs, or calendars.

A standard set of rules that systems follow to communicate.

A model optimized for harder tasks involving planning, logic, math, coding, or multi-step thinking.

Revising an answer after detecting an issue.

The model's expression of certainty, which may not always be reliable.

Converting text into spoken audio.

A model that can understand images or visual inputs.

A common model type used for image generation.

Optical Character Recognition. Extracting text from images or scanned documents.

When AI gives an answer that sounds correct but is false or unsupported.

Patterns in output that reflect skewed training data or system behavior.

The date after which a model may not know events unless connected to live sources.

The same input can produce different outputs.

A prompt designed to bypass safety or instruction rules.

Malicious or hidden instructions that try to manipulate the AI, often through external content.

Rules or systems that keep AI behavior safe, accurate, and appropriate.

Protecting user and organizational data.

How long data is stored.

Testing AI output for quality, accuracy, safety, and usefulness.

A standard test used to compare model performance.

How long the model takes to respond.

How many requests the system can handle.

The maximum number of requests allowed in a time period.

Pricing based on input and output tokens.

Showing output as it is generated.

Reusing previous results to save time or cost.

A system that chooses which model to use for a task.

Monitoring how the AI system performs in real usage.

Answers questions conversationally.

Helps a user complete work such as writing, coding, designing, or analyzing.

Takes multi-step action using tools.

Automates tasks across systems.