AI comes with a lot of jargon. Terms like "neural network", "tokens", "parameters", and "hallucination" get thrown around constantly — often without explanation. This module gives you plain-English definitions for the most important terms you'll encounter.

How to use this page

Think of this as your AI dictionary. You don't need to memorise everything — come back here whenever you encounter an unfamiliar term in the other modules.

Core AI concepts

Algorithm
A set of step-by-step instructions a computer follows to solve a problem. All AI systems are built on algorithms — from simple ones that sort a list, to complex ones that generate human-like text.
Model
The output of training an AI system on data. A model is a mathematical structure that has learned patterns and can make predictions or generate outputs. When people say "GPT-4" or "Claude", they mean the model.
Training
The process of feeding data to an AI system so it can learn patterns. Training is expensive and slow — it can take weeks on thousands of specialised processors. It only needs to happen once to produce a model.
Inference
Using a trained model to make predictions or generate outputs. When you type a message to ChatGPT and it replies, that's inference. Much faster and cheaper than training.
Parameters
The internal numbers inside a model that get adjusted during training. More parameters generally means a more capable model. GPT-4 has an estimated 1.7 trillion parameters. When you hear "a 7 billion parameter model", this is what it means.
Dataset
The collection of data used to train a model. Modern LLMs are trained on trillions of words scraped from the internet, books, and other sources. The quality and breadth of the dataset strongly shapes what the model learns.

Language model terms

LLM
Large Language Model. An AI model trained on massive amounts of text, capable of generating, summarising, translating, and reasoning about language. Examples: GPT-4, Claude, Gemini, Llama.
Token
The basic unit an LLM processes. Not quite a word — tokens are chunks of text, roughly 3–4 characters on average. "Artificial intelligence" is about 4 tokens. Models have a "context window" — a limit on how many tokens they can process at once.
Context window
The maximum amount of text (measured in tokens) an LLM can "see" at once — both your input and its output. Older models had small windows (~4k tokens). Modern models can handle 100k–1 million tokens, enabling very long documents and conversations.
Prompt
The input you give to an AI model — your question, instruction, or context. Prompting is a skill: how you phrase your input significantly affects the quality of the output. "Prompt engineering" is the practice of crafting effective prompts.
Temperature
A setting that controls how random or creative an AI's outputs are. Low temperature (e.g. 0.1) = predictable, focused responses. High temperature (e.g. 1.0) = more varied, creative, and sometimes surprising outputs. Used mainly by developers.
Hallucination
When an AI confidently states something that is factually wrong or entirely made up. A major limitation of LLMs. The model doesn't "know" it's wrong — it generates plausible-sounding text based on patterns, not verified facts. Always verify important claims.
Fine-tuning
Taking a general-purpose model and training it further on a specific dataset to specialise it for a task. A medical company might fine-tune a general LLM on clinical notes to create a model that understands medical language better.

Machine learning terms

Neural network
A type of AI model loosely inspired by the structure of the human brain. Built from layers of interconnected "neurons" (mathematical functions). Deep neural networks — with many layers — are behind most modern AI breakthroughs.
Deep learning
Machine learning using deep (many-layered) neural networks. Responsible for the major AI advances since 2012 — image recognition, speech, language, and more. The "deep" refers to the depth of layers, not profundity.
Overfitting
When a model learns the training data too well — including its noise and quirks — and performs poorly on new, unseen data. Like a student who memorises past exam papers but can't handle new questions. A common challenge in ML.
Supervised learning
Training a model on labelled examples — where the correct answer is provided. Example: showing a model thousands of photos labelled "cat" or "dog" so it learns to classify new photos. The most common form of ML.
Unsupervised learning
Training a model on data without labels, letting it find patterns on its own. Used for clustering similar items, detecting anomalies, or compressing data. LLMs use a form of this during pre-training.

Practical terms

AI Agent
An AI system that can take actions in the world — browsing the web, writing and running code, sending emails, or controlling software — not just answer questions. Agents combine an LLM with tools and memory. Covered in depth in the Agentic AI module.
RAG
Retrieval-Augmented Generation. A technique where an AI retrieves relevant documents from a database before generating a response. Helps models answer questions about specific, up-to-date, or private information they weren't trained on.
Multimodal
An AI model that can handle multiple types of data — text, images, audio, and video — not just one. GPT-4o and Claude 3 are multimodal: you can show them a photo and ask questions about it.
Foundation model
A large model trained on broad data that can be adapted for many tasks. GPT-4, Claude, and Gemini are foundation models. Most AI products you use are built on top of foundation models rather than trained from scratch.
GPU
Graphics Processing Unit. Originally designed for gaming graphics, GPUs can perform many calculations in parallel — making them ideal for training AI models. NVIDIA's GPUs are now the backbone of the AI industry.

Key takeaways

  • A model is what you get after training — the AI system itself
  • Tokens are how LLMs process text — roughly 3–4 characters each
  • Temperature controls creativity vs consistency in outputs
  • Hallucination is when AI states incorrect things confidently — always verify
  • Fine-tuning specialises a general model for a specific task or domain
  • Multimodal models can handle text, images, audio — not just words