AI

Systems that use learned patterns to predict, classify or generate outputs.

Generative AI

AI that produces text, images, audio or other new outputs from learned patterns.

Training

Adjusting model parameters using examples.

Inference

Using a trained model to produce an output.

Token

A chunk of text processed by a model; it can be part of a word.

Context window

The limited amount of input and output a model can work with at once.

Hallucination

A generated claim or reference that is unsupported or false.

Grounding

Connecting an answer to supplied or retrieved evidence. It still needs checking.

Prompt injection

Untrusted content that attempts to redirect a system or its tools.

Redaction

Removing or replacing sensitive information, including identifying context.

Agent

A system that can select steps or tools towards a goal within set boundaries.

Idempotency

Designing retries so the same request does not create duplicate effects.

Human oversight

A responsible person checks outputs and approves consequential actions.

Bias

A systematic skew that can produce unfair or misleading outputs.

Structured output

An output following a specified format or schema; accuracy still needs validation.

Useful original sources

External resources are optional. Their content, availability and accessibility may change. Our original written lessons cover the learning objectives without third-party videos.

Video review status

The original publisher 3Blue1Brown is a candidate for supplementary technical explanations. Playback, captions, duration and embedding could not be verified from this environment. No unverified provider IDs are embedded. Editorial review is required before adding a video verified for playback in lessons. Candidate source links and review notes are available beside the relevant written lessons.

Back to my learningExplore the courses →