AI Glossary

AI terms, no jargon

Every term you'll actually meet in AI news, papers and product docs — explained in one line.

Agent

An AI that can plan, use tools and take multi-step actions on your behalf — not just answer questions.

Context window

How much text (in tokens) a model can consider at once. Bigger = can read longer documents.

Embedding

A numeric fingerprint of text/image so a computer can measure meaning and similarity.

Fine-tuning

Continuing to train a base model on your own data so it learns your style or domain.

Foundation model

A very large model trained on broad data, used as the base for many downstream tasks.

Hallucination

When an AI confidently makes something up. Fix with grounding (RAG) and citations.

Inference

The act of the model actually generating an answer for a request.

LLM

Large Language Model. An AI trained on huge amounts of text to predict the next token.

LoRA

Low-Rank Adaptation. A cheap way to fine-tune only a small slice of a big model.

MoE

Mixture of Experts. Only a subset of the model's weights fires per token — faster and cheaper.

Multimodal

A model that understands more than one type of input (text + image + audio + video).

Prompt

The instructions you give an AI. Better prompt = better output.

Prompt injection

An attack where hidden instructions in content trick an AI into ignoring its real orders.

Quantization

Shrinking a model by using lower-precision numbers so it runs on cheaper hardware.

RAG

Retrieval-Augmented Generation. Fetch relevant docs first, then let the LLM answer using them.

Reasoning model

An LLM trained to 'think step by step' internally before answering (e.g. o-series, R1).

RLHF

Reinforcement Learning from Human Feedback — how models are aligned to be helpful and safe.

Temperature

How random the AI's output is. 0 = deterministic, 1+ = creative.

Token

A chunk of text (~¾ of a word). Models read and generate one token at a time.

Vector database

A database that stores embeddings so you can search by meaning, not keywords.

Zero-shot

Asking a model to do a task with no examples — relying only on its general knowledge.