Learn
Glossary
Plain-language definitions of the terms the AI field moves on — from agents to zero-shot.
66 terms
A
AI Agents
Autonomous AI systems that can perceive their environment, make decisions, and take actions to achieve specific goals, often using LLMs as their reasoning engine.
AI Alignment
The challenge of ensuring AI systems behave in accordance with human intentions, values, and goals, particularly as they become more capable.
AI Benchmarks
Standardised tests used to measure and compare AI model capabilities, from knowledge and reasoning to coding and multilingual understanding.
AI Bias
Systematic errors or unfair outcomes in AI systems, often resulting from skewed training data, flawed model design, or societal inequalities reflected in the data.
AI Data Centers
Purpose-built computing facilities that house the GPUs, networking, cooling, and power infrastructure needed to train and serve large AI models at scale.
AI Governance
The rules, processes, and institutions — inside companies and across governments — that decide how AI systems are built, deployed, audited, and held accountable.
AI Guardrails
The safety layer around an AI system — filters, policies, validation, and monitoring — that constrains what a model can input, output, or do before actions reach users or downstream systems.
AI Orchestration
The coordination and management of multiple AI tools, models, agents, and workflows to accomplish complex tasks that no single model can handle alone.
AI Safety
The field focused on ensuring artificial intelligence systems are developed and deployed in ways that are beneficial, aligned with human values, and do not cause unintended harm.
AI Watermarking
Embedding hidden, machine-detectable signals in AI-generated content — text, images, audio, or video — so its origin can be verified or its provenance recorded.
Artificial General Intelligence (AGI)
A hypothetical form of AI that possesses human-level cognitive abilities across any intellectual task, capable of learning and adapting to any problem domain without specialized training.
Artificial Intelligence (AI)
The simulation of human intelligence in machines that are programmed to think, learn, and make decisions. AI encompasses a broad range of techniques from rule-based systems to machine learning and deep learning.
Attention Mechanism
A technique that lets neural networks focus on the most relevant parts of input data, forming the core innovation behind the Transformer architecture.
B
C
Chain-of-Thought (CoT)
A prompting technique where models are instructed or trained to break down complex reasoning into intermediate steps, dramatically improving accuracy on multi-step problems.
Computer Vision
A field of AI that trains computers to interpret and understand the visual world, enabling machines to identify objects, scenes, and activities in images and videos.
Context Engineering
The practice of designing and managing everything that goes into a model's context window — instructions, retrieved documents, tool outputs, and memory — so the model has exactly the information it needs, and nothing it doesn't.
Context Window
The maximum amount of text (measured in tokens) that an LLM can process at once, including both input and generated output.
D
Deep Learning
A subset of machine learning that uses multi-layered neural networks (deep neural networks) to model complex patterns in large amounts of data.
Diffusion Models
A class of generative models that create data by gradually removing noise, reversing a process of adding random noise to training data. The technology behind modern AI image generators.
E
Edge AI
The deployment of AI algorithms directly on local devices (smartphones, IoT devices, cameras) rather than in the cloud, enabling real-time processing with lower latency and improved privacy.
Embeddings
Numerical vector representations of data (text, images, audio) that capture semantic meaning, enabling machines to understand relationships and similarity between concepts.
Emergent Abilities
Capabilities that appear in large AI models but are absent in smaller ones — behaviour that seems to switch on abruptly as models scale, rather than improving smoothly.
EU AI Act
The European Union's comprehensive law regulating artificial intelligence, which classifies systems by risk level and imposes obligations on developers and deployers accordingly.
F
Federated Learning
A training approach where models learn from data spread across many devices or organisations without the raw data ever leaving its source — only model updates are shared.
Few-Shot and Zero-Shot Learning
Techniques where models perform tasks with few or no examples. Few-shot provides a handful of examples; zero-shot requires none, relying purely on instructions.
Fine-Tuning
The process of taking a pre-trained model and further training it on a specific, typically smaller, dataset to adapt it for a particular task or domain.
Foundation Model
A large AI model trained on broad, diverse data that can be adapted to a wide range of downstream tasks through fine-tuning or prompting. The backbone of modern generative AI.
G
Generative AI
A category of AI that creates new content — text, images, audio, video, or code — by learning patterns from existing data and generating novel outputs.
GPT (Generative Pre-trained Transformer)
A family of large language models developed by OpenAI based on the transformer architecture, known for their ability to generate coherent and contextually relevant text.
GPU (Graphics Processing Unit)
A processor with thousands of small cores designed for parallel arithmetic. Its ability to run many calculations at once made it the standard hardware for training and running neural networks.
Gradient Descent
An optimisation algorithm that improves a model by repeatedly nudging its parameters in the direction that most reduces error, following the slope of the loss function downhill.
H
Hallucination
A phenomenon where an AI model generates plausible-sounding but factually incorrect or nonsensical information, often with high confidence.
Human-in-the-Loop (HITL)
A system design where a person reviews, approves, or corrects an AI system's decisions at defined points, combining model speed with human judgement for high-stakes actions.
I
K
L
Large Language Model (LLM)
A deep learning model trained on massive text datasets to understand, generate, and manipulate human language. LLMs power modern chatbots, code generators, and text-based AI tools.
LoRA (Low-Rank Adaptation)
A parameter-efficient fine-tuning technique that adds small, trainable matrices to existing model weights, enabling fine-tuning of large models with minimal computational cost.
M
Machine Learning (ML)
A subset of AI that enables systems to learn and improve from experience without being explicitly programmed, using statistical techniques to find patterns in data.
Mixture of Experts (MoE)
A neural network architecture that uses multiple specialized sub-networks (experts) with a gating mechanism that activates only relevant experts for each input, improving efficiency and capability.
MLOps
The set of practices, tools, and automation for taking machine learning models from experimentation to reliable production — covering data pipelines, training, deployment, monitoring, and retraining.
Model Context Protocol (MCP)
An open standard that lets AI applications connect to external tools, data sources, and services through a single common interface, instead of every tool and every model needing custom integrations.
Model Parameters
The numerical weights and biases that define a neural network's learned knowledge. Larger parameter counts generally indicate greater model capacity and capability.
Multimodal AI
AI systems capable of processing and generating multiple types of data simultaneously, such as text, images, audio, and video, often within a single model.
N
Natural Language Processing (NLP)
A subfield of AI focused on enabling computers to understand, interpret, and generate human language in a meaningful way.
Neural Network
A computing system inspired by biological neural networks in the human brain, consisting of interconnected nodes (neurons) organized in layers that process information.
O
Open-Source AI
AI models, tools, and frameworks that are publicly released with permissive licenses, allowing anyone to use, modify, study, and distribute them.
Overfitting
When a model memorises its training data — including noise and quirks — instead of learning general patterns, so it performs well on seen examples but poorly on new ones.
P
Prompt Engineering
The practice of designing and refining input prompts to guide LLMs toward producing desired outputs more reliably and accurately.
Prompt Injection
An attack where untrusted text — a web page, email, document, or tool output — contains instructions that hijack an AI model, causing it to ignore its original instructions or leak data.
Q
R
Reasoning Models
AI models trained to spend extra computation on a problem before answering — producing internal chains of thought, checking their work, and trading speed for higher accuracy on maths, coding, and logic.
Reinforcement Learning (RL)
A machine learning paradigm where agents learn optimal behavior through trial and error, receiving rewards for good actions and penalties for bad ones.
Reinforcement Learning from Human Feedback (RLHF)
A training technique that uses human preferences to fine-tune language models, aligning them with human values and making them more helpful, harmless, and honest.
Retrieval-Augmented Generation (RAG)
A technique that combines information retrieval with text generation, allowing LLMs to access external knowledge sources to produce more accurate and up-to-date responses.
S
Scaling Laws
Empirical relationships showing that model performance improves predictably as you increase parameters, training data, and compute — and that the three must be scaled together to avoid waste.
Stable Diffusion
An open-source deep learning model for generating images from text descriptions, released by Stability AI in 2022.
Synthetic Data
Artificially generated data that mimics real-world data, used to train AI models when real data is scarce, expensive, or privacy-sensitive.
T
Temperature
A sampling setting that controls how random a language model's output is — low values make responses focused and repeatable, high values make them more varied and creative.
Test-Time Compute
Spending more computation while a model is answering — generating longer reasoning, sampling multiple solutions, or searching over options — instead of only scaling the training run.
Token
The basic unit of text that an LLM processes. Tokens can be words, parts of words, or individual characters. LLMs have a maximum context window measured in tokens.
Transfer Learning
Reusing a model trained on one task as the starting point for a different task, so new problems can be solved with far less data and compute.
Transformer
A neural network architecture introduced by Google in 2017 that uses self-attention mechanisms to process sequential data. It is the foundation of virtually all modern LLMs.
Turing Test
A test proposed by Alan Turing in 1950 in which a machine passes if a human judge, chatting via text, cannot reliably tell whether responses come from the machine or another person.
V
W
About this glossary
This glossary is curated and updated regularly to reflect the evolving language of artificial intelligence. Terms range from foundational concepts to cutting-edge techniques and tools. Definitions are written to be accessible to beginners while remaining accurate for professionals.