Reference

AI glossary

One page, not one thin page per term. Each definition explains what the term means and why it matters when you are using these tools.

Bias (in AI output): Systematic skew in what a model produces, traceable to its training data, its objectives or how a question is framed.
Context window: The amount of text a model can consider at once, including your prompt and its own reply. Older parts of a long conversation can fall outside it.
Deep learning: Machine learning using many-layered neural networks. It underpins current language and image models.
Fine-tuning: Further training of an existing model on a narrower set of examples so it behaves more predictably for a specific task.
Generative AI: Software that produces new text, images, audio or code based on patterns learned from large amounts of example data, rather than retrieving a stored answer.
Hallucination: A confident answer that is not supported by fact or by the sources provided. It is a normal failure mode of text prediction, not a rare bug.
Large language model (LLM): A model trained to predict likely continuations of text. It powers chat assistants and has no separate store of verified facts.
Machine learning: Building systems that improve at a task by finding patterns in data instead of following rules written by hand.
Personal data: Information that identifies a person. Worth keeping out of chatbot conversations unless you know how the provider stores and uses it.
Prompt: The instruction and context you give an AI system. It sets the task, the audience, the format and the limits of the answer.
Prompt injection: Text hidden in a document or web page that tries to issue instructions to an AI system reading it.
Retrieval-augmented generation (RAG): Supplying a model with documents fetched at question time so its answer can be grounded in, and checked against, those sources.
System prompt: Background instructions set by the product rather than the user, shaping tone, scope and refusals.
Temperature: A setting that controls how varied a model's wording is. Lower values give more repeatable answers; higher values give more variety.
Token: The chunk of text a model reads and writes — often a word fragment. Context limits and pricing are usually measured in tokens.
Training data: The examples a model learned from. Its coverage, age and gaps shape what the model does well and badly.