Reference
AI Glossary
Plain-language definitions for the AI terms that actually matter, with practical context on why each one is relevant.
An AI system that can autonomously plan, reason, and take actions to accomplish goals, often using tools and external APIs.
The challenge of ensuring AI systems act in accordance with human intentions and values — making them do what we actually want, not just what we literally ask for.
Systematic unfairness in AI outputs caused by skewed training data, flawed labelling, or model design choices that reflect and amplify existing societal inequities.
The coordination and management of multiple AI models, tools, data flows, and human inputs within a unified workflow to accomplish complex tasks that no single component could handle alone.
AI systems that can autonomously plan, reason, use tools, and execute multi-step tasks with minimal human oversight — going beyond simple question-answering to take actions on behalf of users.
AI systems that find patterns, anomalies, and insights in large datasets, used for tasks like fraud detection, medical imaging analysis, and business intelligence.
A broad field of computer science focused on building systems that can perform tasks normally requiring human intelligence, such as understanding language, recognizing patterns, and making decisions.
A prompting technique that improves LLM accuracy on complex tasks by guiding the model to show its reasoning step-by-step before arriving at a final answer.
The process of splitting documents into smaller, semantically meaningful segments optimised for embedding and retrieval in AI systems like RAG pipelines.
AI architectures that combine multiple models, retrievers, tools, and control logic to tackle tasks that no single model could reliably handle on its own.
The maximum amount of text (measured in tokens) that a language model can consider at once — including both the input prompt and the generated output.
AI systems designed to engage in natural language dialogue with humans, ranging from simple chatbots with scripted responses to advanced assistants powered by large language models.
An AI assistant embedded directly into a workflow tool (IDE, browser, email) that suggests actions, generates content, or automates tasks inline.
The set of concerns and practices around protecting personal and sensitive information when using AI systems, covering training data, user inputs, model outputs, and data retention.
A subset of machine learning that uses neural networks with many layers to learn increasingly abstract representations of data, powering breakthroughs in language, vision, and generation.
A generative AI architecture that creates images, video, and other media by learning to gradually remove noise from random static until a coherent output emerges.
A prompting technique where you include a few examples of the desired input-output format in your prompt, helping the model understand exactly what you want without any fine-tuning.
The process of further training a pre-trained model on a smaller, task-specific dataset to improve its performance on that particular task or domain.
The mechanism that allows LLMs to interact with external tools and APIs by outputting structured data — typically JSON — specifying which function to invoke and with what parameters.
Artificial intelligence that creates new content — text, images, video, audio, or code — by learning patterns from existing data and producing original outputs in response to prompts.
Programmatic constraints placed around AI model inputs and outputs to prevent harmful, off-topic, or policy-violating behavior.
When an AI model generates information that sounds plausible but is factually incorrect, fabricated, or unsupported by its training data or provided context.
A retrieval approach that combines traditional keyword matching (BM25) with semantic vector search to capture both the precision of exact term matches and the contextual understanding of meaning-based search.
The date beyond which a language model has no training data, meaning it cannot know about events, discoveries, or changes that occurred after that point.
A structured representation of information as a network of entities and their relationships, enabling machines to reason about connections between concepts.
The set of practices and tools for deploying, monitoring, and maintaining machine learning models in production — essentially DevOps principles applied to the ML lifecycle.
A subset of AI where systems learn patterns from data rather than following explicitly programmed rules, improving their performance as they see more examples.
A neural network architecture that scales model capacity efficiently by routing each input through only a small subset of specialized sub-networks ("experts"), keeping compute costs manageable even as total model size grows.
An open standard for connecting AI assistants to external data sources and tools through a unified, composable interface.
AI systems that can process, understand, and generate across multiple types of data — text, images, audio, video, and code — within a single model.
The initial training phase where a language model learns general language patterns from a massive text corpus, before being fine-tuned for specific tasks or behaviors.
AI systems that forecast outcomes based on historical data patterns, used for tasks like demand forecasting, risk assessment, and recommendation engines.
The practice of designing and refining inputs to language models to elicit more accurate, useful, and consistent outputs.
A training technique where human preferences are used to fine-tune a language model through reinforcement learning, teaching it to produce responses that humans judge as helpful, accurate, and safe.
LLMs trained with reinforcement learning to "think before they answer" by generating internal chains of reasoning — producing more accurate results on complex tasks like maths, coding, and multi-step logic at the cost of higher latency and token usage.
Systematic adversarial testing of AI systems to identify vulnerabilities, failure modes, and unintended behaviours before deployment — adapted from cybersecurity to probe AI-specific weaknesses like prompt injection and jailbreaks.
A machine learning approach where an agent learns by taking actions in an environment and receiving rewards or penalties, gradually discovering which strategies produce the best outcomes.
A second-stage retrieval technique that re-scores and reorders an initial set of retrieved documents using a more computationally expensive cross-encoder model to surface the most relevant results.
A technique that grounds a language model's output in external data by retrieving relevant documents before generating a response.
A search technique that understands the meaning and intent behind queries rather than matching exact keywords, using vector embeddings to find conceptually relevant results even when different words are used.
A compact AI language model — typically under 10 billion parameters — designed to run efficiently on edge devices and single GPUs while delivering strong task-specific performance.
A technique for constraining a language model's output to follow a specific format like JSON, XML, or a defined schema, ensuring the response can be reliably parsed by downstream code.
A machine learning approach where the model learns from labeled examples — input-output pairs where the correct answer is provided during training.
The foundational instruction set given to an LLM that defines its role, behaviour, tone, and constraints for a particular application — set once at the application level and shaping all subsequent user interactions.
A parameter that controls how random or deterministic an LLM's output is — lower values produce more predictable, focused responses while higher values increase creativity and variation.
The basic unit of text that a language model processes — typically a word, subword, or punctuation mark, roughly equivalent to 3/4 of an English word.
The dataset used to teach a machine learning model, containing the examples and patterns the model learns to recognize and reproduce.
A neural network architecture that powers modern AI by processing entire input sequences simultaneously through an attention mechanism, rather than reading them word by word.