The language
of AI.

Computer systems that perform tasks such as recognizing patterns, generating language, or making predictions.

Picture it: a photo sorter and a writing assistant are different kinds of AI.

A way to build systems by learning patterns from data instead of specifying every rule by hand.

Picture it: showing a sorter many labeled examples instead of listing every possible object.

AI that produces content, such as text, images, audio, or code, using learned patterns.

Try it thoughtfully: request three possible story settings, then invent your own fourth.

A trained mathematical system that transforms inputs into predictions or generated outputs.

Useful distinction: the model is one part of an app. Search, file storage, and other features may be separate.

A model trained on large amounts of language data to process and generate language.

Picture it: a system that can help compare explanations, draft dialogue, or discuss a coding problem.

Adjusting a model’s parameters using data and a learning objective.

Keep separate: giving an assistant feedback in a conversation is not the same as directly updating its model.

Using a trained model to produce a prediction or response.

Picture it: training is preparation; inference is the model working on an input now.

A unit a model processes. In text, it can be a word, word fragment, or punctuation.

Useful reminder: token counts and word counts are different. How text is split depends on the model’s tokenizer.

Context is the available information. The context window limits how many tokens the model can work with at once.

Try it: restate the goal and important rules when a long conversation starts drifting.

The input that asks or guides a model to do something.

Example: “Ask me one question about fractions and wait for my answer.”

Application-level directions that guide an assistant’s behavior alongside other inputs.

Useful reminder: your message is not its only instruction. A tutor app may have rules about giving hints first.

Including a few examples to show the pattern, style, or format you want.

Example: give two question-and-hint pairs before asking for another practice question.

Boundaries for the task and the shape you want the response to take.

Example: “Use everyday words. Give me a table with claim, evidence, and uncertainty.” Check that the result follows it.

Trying, reviewing, and refining in repeated rounds.

Example: “The second idea fits my goal. Make the first step smaller, and keep the ending for me to write.”

A generation setting that changes how output choices are sampled. Higher values often allow more variation.

Useful reminder: it is not a truthfulness control. Not every app exposes this setting.

An application feature for retaining and reusing information across interactions; behavior varies by product.

Try it: inspect memory settings and saved information instead of assuming the assistant remembers everything—or nothing.

Finding relevant information in a source or collection for use in a task.

Example: locating the right passage in your permitted class notes before discussing a question.

A system that retrieves material and supplies it to a model when generating a response.

Useful reminder: retrieving a passage does not ensure it is the right one or that the answer uses it correctly.

Connecting a response to specific information or sources rather than relying only on learned patterns.

Try it: ask the assistant to show which provided passage supports each claim. Then check the match yourself.

Generated content that is false or unsupported, even when it sounds convincing.

Example: an impressive-looking citation to a paper that does not exist. Open the source before using it.

A systematic skew in data or outputs that can create misleading or unfair results.

Try it: ask whose experiences are missing from an example, then seek evidence beyond the assistant’s answer.

A system that can work with more than one kind of information, such as text and images.

Example: discussing a diagram alongside its written description. Check small labels and visual details yourself.

An assistant calling an external capability, such as a calculator, search service, or code runner.

Useful question: did it actually run the test, or only describe how you could test?

Software that uses a model and available tools to plan and take steps toward a goal.

Try it safely: define the task and require review before sharing, buying, deleting, or publishing anything.

A defined way for software to request information or actions from another system.

Picture it: your learning app asks a service for a practice question. Keep API keys private.

A sequence of steps for completing a task, which may include people, AI, and other tools.

Example: read a source → draft a claim → request feedback → verify → revise in your own words.

A structured check of whether a system’s outputs meet the criteria you care about.

Try it: test your study helper with an easy question, an ambiguous question, and one it should say it cannot answer.

Using ordinary-language descriptions to guide software creation or behavior.

Useful reminder: describing a program is a starting point. Read the resulting code, test it, and learn how it works.