The AI Dictionary

LLM (Large Language Model)

A model trained on huge amounts of text to predict the next word — the engine behind ChatGPT, Claude, Gemini and Grok.

Token

The chunk of text a model actually reads and writes — roughly ¾ of a word on average.

Context window

How much text a model can 'see' at once — your conversation, files and instructions, all measured in tokens.

Parameters

The billions of internal numbers a model learns during training — its 'knowledge', stored as weights.

Transformer

The neural-network design (from Google, 2017) that nearly every modern AI model is built on.

Attention

The mechanism that lets a model weigh which earlier words matter most for the next one.

Hallucination

When a model states something false — a fake quote, citation, or fact — with total confidence.

Multimodal

A model that handles more than text — images, audio, video or files — as input, output, or both.

Inference

Running a trained model to get an answer — as opposed to training it.

Reasoning model

A model that 'thinks' step by step before answering — slower, but much better at math, logic and code.

Open weights

A model whose trained parameters are published so anyone can download and run it themselves.

Benchmark

A standardized test used to compare models — coding, math, reasoning, knowledge.

Pre-training

The first, most expensive phase: teaching a model language by having it predict the next token across trillions of tokens.

Fine-tuning

Extra training on a smaller, focused dataset to specialize a model's style, format or skill.

RLHF (Reinforcement Learning from Human Feedback)

Training a model on human ratings of its answers so it becomes more helpful, honest and safe.

Distillation

Training a small 'student' model to imitate a big 'teacher' model — most of the smarts at a fraction of the cost.

Synthetic data

Training data generated by AI models rather than written by humans.

LoRA (Low-Rank Adaptation)

A cheap fine-tuning method that trains a tiny add-on instead of the whole model.

Quantization

Shrinking a model by storing its numbers with less precision, so it runs faster on smaller hardware.

Mixture of Experts (MoE)

A model built from many specialist sub-networks where only a few 'experts' activate for each token.

Alignment

Making AI systems reliably do what people actually intend — helpful, honest, and not harmful.

Prompt

The instructions and context you give a model — the single biggest lever on the quality you get back.

System prompt

Behind-the-scenes instructions that set a model's role, rules and style for the whole conversation.

Temperature

A setting that controls randomness — low for precise, repeatable answers; high for creative variety.

Few-shot prompting

Showing the model a few examples of what you want before asking it to do the real task.

Chain of thought

Having the model work through a problem step by step before giving its answer.

Structured output

Getting answers in a strict, machine-readable format like JSON or a fixed table.

Embeddings

Lists of numbers that represent meaning — texts about similar things end up close together.

Vector database

A database built to store embeddings and quickly find the most similar ones.

RAG (Retrieval-Augmented Generation)

Looking up relevant documents first, then having the model answer using them — grounded, current, citable.

Grounding

Tying a model's answer to real, provided sources — search results, documents, data — instead of memory.

Context engineering

Deliberately choosing what goes into the model's context — instructions, files, examples, tools — to get reliable results.

API

The programmatic way to use a model from your own code or app, usually paid per token.

Latency

How long you wait for a response — time to first token plus generation speed.

Agentic AI (AI agents)

AI that doesn't just answer — it plans, takes actions with tools, checks results and keeps going until the job is done.

Tool use (function calling)

A model's ability to call external tools — search, calculators, code, APIs — and use the results.

MCP (Model Context Protocol)

An open standard for plugging AI apps into tools and data — a 'USB-C port' for AI.

Orchestration & subagents

Coordinating multiple AI agents — a lead agent splits work, hands pieces to subagents, and merges the results.

Evals

Repeatable tests that measure whether an AI system does your task well — the unit tests of AI.

Guardrails

Rules and checks around an AI system that block unsafe, off-topic or wrong outputs and actions.

Prompt injection

Hidden instructions in a web page, email or file that try to hijack an AI agent.

Human in the loop

Designing AI workflows so a person reviews or approves key steps.

Vibe coding

Building software by describing what you want to an AI and iterating on the result, often without reading every line.

Computer use

AI that operates a computer like a person — seeing the screen, moving the mouse, typing and clicking.