The AI glossary you can actually use at work.
Clear, current definitions of 161 essential AI terms — from fine-tuning and RAG to MCP, agents, GPT-5, Claude Sonnet 4, Midjourney v7 and Kling. Each entry explains what it means, why it matters and where it fits in real US business workflows.
161
Definitions
7
Categories
Updated 2026
Current models
By category
Browse the glossary by topic
Six categories grouping the entire vocabulary, from machine-learning fundamentals to infrastructure and ethics.
Fundamentals
Core machine learning concepts every AI practitioner should know — from training and inference to bias and variance.
Algorithm
A finite, well-defined sequence of steps that solves a problem or performs a task.
Read definitionArtificial Intelligence (AI)
Software systems that perform tasks normally requiring human reasoning, perception or decision-making.
Read definitionBackpropagation
The algorithm that computes gradients through a neural network by applying the chain rule layer by layer.
Read definitionBias
Two meanings: a model's systematic underfitting, or unfair behaviour toward demographic groups.
Read definitionCPU (Central Processing Unit)
The general-purpose processor at the heart of every computer; runs orchestration around AI workloads.
Read definitionCross-Validation
A technique for measuring model quality by training and testing on rotating slices of the dataset.
Read definitionDataset
A structured collection of examples used to train, validate or evaluate a machine-learning model.
Read definitionDeep Learning
Machine learning using multi-layer neural networks; the foundation of modern LLMs and image models.
Read definitionGPU (Graphics Processing Unit)
Massively parallel processors originally built for graphics, now the workhorse of AI training and inference.
Read definitionGradient Descent
The optimisation algorithm that powers neural network training by following the slope of the loss surface.
Read definitionInference
Running a trained model on new inputs to produce predictions or generated content.
Read definitionLabeling
Attaching the correct answer to each example in a dataset so a model can learn to reproduce it.
Read definitionMachine Learning (ML)
A branch of AI where algorithms learn patterns from data instead of following hand-coded rules.
Read definitionModel
A trained mathematical artifact that maps inputs to outputs, like a function learned from data.
Read definitionNeural Network
A mathematical model loosely inspired by the brain, made of layers of weighted connections that learn from data.
Read definitionOverfitting
When a model memorises the training set and fails to generalise to new examples.
Read definitionRegularization
Techniques that constrain a model so it generalises to new data instead of memorising the training set.
Read definitionTraining
The process of adjusting a model's weights from data so that it learns to perform a task.
Read definitionTransfer Learning
Adapting a model trained on one task to a related task with much less data and compute.
Read definitionVariance
How sensitive a model is to the specific training set; high variance means high overfitting risk.
Read definitionLarge Language Models
How modern LLMs work: tokens, embeddings, transformers, fine-tuning, RAG, and the techniques that shape model behaviour.
Attention
The mechanism that lets each token in a sequence weigh and combine information from every other token.
Read definitionBLEU Score
A classic metric for machine translation quality based on n-gram overlap with reference translations.
Read definitionChain-of-Thought (CoT)
Asking the model to reason step by step out loud before giving its final answer.
Read definitionContext
All the text the model can see when generating a response — the system prompt, history and current input.
Read definitionContext Window
The maximum amount of input (in tokens) a model can read in a single request.
Read definitionDistillation
Training a smaller, cheaper model to imitate the outputs of a larger one.
Read definitionEmbedding
A dense vector that represents the meaning of a piece of text, image or other content.
Read definitionFine-tuning
Continuing to train a pretrained model on your own data so it specialises in your task or style.
Read definitionHallucination
When an LLM produces fluent, confident-sounding output that is factually wrong or invented.
Read definitionInstruction Tuning
Fine-tuning a base model on (instruction, response) pairs so it follows directions instead of just predicting text.
Read definitionMixture of Experts (MoE)
An architecture where each input only activates a subset of the model's parameters, decoupling capacity from cost.
Read definitionParameters
The learned weights inside a model; counted in millions or billions, they store everything the model knows.
Read definitionPerplexity
A measure of how surprised a language model is by a piece of text; lower means better fit.
Read definitionRAG (Retrieval-Augmented Generation)
A pattern where the model retrieves relevant documents from a knowledge base and uses them to answer.
Read definitionRLHF (Reinforcement Learning from Human Feedback)
A training technique where humans rank model outputs and the model is optimised to produce higher-ranked responses.
Read definitionSystem Prompt
The instructions sent before the user's message that define the model's role, tone and constraints.
Read definitionTemperature
A sampling parameter controlling how random or deterministic the model's outputs are.
Read definitionToken
The atomic unit a language model reads and writes; usually a word, sub-word fragment or punctuation mark.
Read definitionTop-p (Nucleus Sampling)
A sampling method that restricts the model to the smallest set of tokens whose probabilities sum to p.
Read definitionTransformer
The neural network architecture introduced in 2017 that powers virtually every modern LLM.
Read definitionAI Tools & Models
The 2026 product landscape: ChatGPT, Claude, Gemini, Llama, Midjourney v7, Sora, Cursor and the rest of the working stack.
ChatGPT
OpenAI's flagship conversational AI assistant, the product that brought generative AI to mainstream attention.
Read definitionClaude
Anthropic's family of conversational AI models, known for long context, careful reasoning and coding strength.
Read definitionClaude Sonnet 4
Anthropic's flagship workhorse model in 2026, known for coding, careful reasoning and long context.
Read definitionCursor
An AI-first code editor based on VS Code, the dominant pick for AI-assisted development in 2026.
Read definitionDALL-E 3
OpenAI's image generation model integrated into ChatGPT, known for strong literal prompt following.
Read definitionElevenLabs
The leading voice AI company, known for high-quality text-to-speech, voice cloning and dubbing.
Read definitionGemini
Google's family of multimodal AI models, integrated across Workspace, Android and the Gemini app.
Read definitionGitHub Copilot
Microsoft and GitHub's AI coding assistant, embedded in VS Code, JetBrains and the GitHub web interface.
Read definitionGPT-5
OpenAI's flagship LLM as of 2025–2026, with multimodal input, built-in reasoning and agentic tool use.
Read definitionKling
Kuaishou's text-to-video model, a leading Sora competitor known for cinematic realism and physical motion.
Read definitionLlama
Meta's family of open-weight large language models, the foundation of much of the open AI ecosystem.
Read definitionMidjourney v7
The 2025–2026 release of the popular image generation service, with sharper realism and stronger prompt control.
Read definitionMistral
A French AI lab known for high-quality open-weight models, especially Mixtral MoE and Mistral Large.
Read definitionnano-banana-2
An automation-friendly image generation model accessible via the Evolink API, popular for batch marketing pipelines.
Read definitionNotebookLM
Google's source-grounded AI research notebook; especially famous for its podcast-style audio summaries.
Read definitionPerplexity (search engine)
An AI-native search engine that answers questions with cited sources instead of a list of links.
Read definitionReplit Agent
Replit's AI agent that builds, debugs and deploys full applications from natural-language descriptions.
Read definitionSora
OpenAI's text-to-video model, capable of generating realistic short clips from text or image prompts.
Read definitionStable Diffusion
The leading family of open-weight image generation models, the foundation of much of the open creative AI ecosystem.
Read definitionSuno
An AI music generation service that produces full vocal songs from text descriptions.
Read definitionPrompts & Agents
Prompt engineering, structured output, function calling, agents, MCP, multi-agent systems and the patterns that ship in production.
Agent
An AI system that autonomously plans, takes actions through tools, and iterates toward a goal.
Read definitionAutoGPT
A 2023 viral open-source autonomous agent project that helped popularise the agent-loop concept.
Read definitionBrowser Automation
Letting an AI agent control a real web browser to navigate, fill forms and extract information.
Read definitionComputer Use
An LLM's ability to control a desktop environment with screenshots, mouse and keyboard input.
Read definitionCustom Tool
A function you write and expose to an agent so it can perform a specific action in your system.
Read definitionFew-Shot Prompting
Including a handful of input/output examples in the prompt to teach the model the desired pattern.
Read definitionFunction Calling (Tool Use)
An API mechanism where the model decides to call your code, with arguments, instead of generating text.
Read definitionJSON Mode
An API setting that tells the model to return only valid JSON, eliminating wrapper text and parse errors.
Read definitionMCP (Model Context Protocol)
An open standard from Anthropic for connecting tools and data sources to any compatible LLM.
Read definitionMemory (Agent Memory)
Persistent state that lets an agent remember facts and preferences across sessions and conversations.
Read definitionMulti-Agent System
An architecture where multiple specialised AI agents collaborate on a task with distinct roles.
Read definitionOrchestrator
The control layer that routes work between models, agents, tools and humans in a complex AI workflow.
Read definitionPlanner
The component of an agent that decomposes a goal into ordered steps before execution.
Read definitionPrompt Engineering
The discipline of writing instructions to LLMs so they produce reliable, useful outputs.
Read definitionPrompt Injection
An attack where untrusted input contains instructions that override the developer's system prompt.
Read definitionReAct Pattern
An agent loop that interleaves Reasoning and Acting steps, thinking out loud before each tool call.
Read definitionRole Prompting
Telling the model who to act as — "You are a senior tax accountant" — to set expertise and tone.
Read definitionStructured Output
Forcing the model to return data in a specific schema (JSON, XML) so it can be safely parsed.
Read definitionTool Use
An LLM's ability to call external functions or APIs to extend its capabilities beyond text generation.
Read definitionZero-Shot Prompting
Asking the model to perform a task with only an instruction, no examples in the prompt.
Read definitionVision & Generation
Image, video, 3D, voice and audio generation — diffusion, ControlNet, LoRA, NeRFs, Gaussian splats and modern speech AI.
Computer Vision
The field of AI that enables computers to interpret and reason about images and video.
Read definitionControlNet
An extension to Stable Diffusion that lets you condition image generation on poses, depth maps or sketches.
Read definitionDepth Estimation
Predicting the distance from camera to every pixel in an image, producing a depth map.
Read definitionDiffusion Model
A generative model that learns to reverse a noising process, the dominant approach to image and video generation.
Read definitionGaussian Splatting
A 3D scene representation made of millions of coloured 3D Gaussians, faster to render than NeRF.
Read definitionImage Generation
AI systems that produce images from text prompts, reference images or both.
Read definitionImage Segmentation
Labeling every pixel in an image so each region corresponds to an object or class.
Read definitionImage-to-Video
Generating a moving video clip from a single still image, often with described motion.
Read definitionInpainting
Editing a specific region of an image with AI while keeping the rest unchanged.
Read definitionLatent Space
A compressed mathematical space where AI models represent the meaningful features of their inputs.
Read definitionLoRA (Low-Rank Adaptation)
A parameter-efficient fine-tuning technique that adds tiny trainable adapters to a frozen base model.
Read definitionNeRF (Neural Radiance Fields)
A 3D scene representation learned by a neural network from multiple 2D photos.
Read definitionOCR (Optical Character Recognition)
Extracting text from images and scanned documents so it can be searched and processed.
Read definitionOutpainting
Extending an image beyond its original borders with AI-generated content that matches the existing scene.
Read definitionPose Estimation
Detecting the position of human (or animal) joints in images and video to reconstruct skeletal pose.
Read definitionSpeech-to-Text (STT)
AI systems that convert spoken audio into written text; the input side of voice interfaces.
Read definitionText-to-3D
Generating 3D models, scenes or assets from text prompts; the 3D analog of text-to-image.
Read definitionText-to-Speech (TTS)
AI systems that convert written text into natural-sounding spoken audio.
Read definitionVoice Cloning
Generating a synthetic voice that mimics a specific person's vocal identity from a short audio sample.
Read definitionWhisper
OpenAI's open-source speech recognition model, the dominant choice for transcription in 2026.
Read definitionInfrastructure & Ethics
Vector databases, alignment, EU AI Act, NIST AI RMF, copyright, deepfakes, evals and the operational layer of responsible AI.
AGI (Artificial General Intelligence)
Hypothetical AI capable of matching or exceeding human performance across the full range of cognitive tasks.
Read definitionAlignment
The discipline of ensuring AI systems pursue the goals their developers and users actually want.
Read definitionASI (Artificial Superintelligence)
Hypothetical AI vastly exceeding human cognitive abilities across essentially all domains.
Read definitionCopyright (and AI-generated works)
The legal protection for original creative works; complicated when AI generates the output.
Read definitionDeepfake
Synthetic media (image, audio, video) that convincingly depicts events or statements that did not happen.
Read definitionEmbedding Store
The infrastructure layer that stores, indexes and serves embeddings for retrieval and search.
Read definitionEU AI Act
The European Union's risk-based AI regulation, the first comprehensive AI law from a major jurisdiction.
Read definitionEval (Evaluation)
Tests that measure how well an AI system performs on intended tasks and under adversarial conditions.
Read definitionFair Use (and AI training data)
The US copyright doctrine allowing limited use of copyrighted material; central to ongoing AI training lawsuits.
Read definitionJailbreak
A prompt or attack that bypasses an LLM's safety training to produce content the model is meant to refuse.
Read definitionLatency
How long an AI system takes to respond; often more important than raw quality for user experience.
Read definitionLeaderboard
A ranked list of AI models on a benchmark; useful for model selection and the source of constant marketing.
Read definitionModel Card
A standardised document describing an AI model's purpose, training, limitations and intended use.
Read definitionNIST AI Risk Management Framework
A voluntary US framework for managing risks across the AI lifecycle, increasingly cited in procurement.
Read definitionOpen Weights
Models whose trained parameters are publicly downloadable, even if the training data and code are not.
Read definitionP(doom)
Subjective probability that AI causes existential or near-existential catastrophe; a contested AI-policy shorthand.
Read definitionpgvector
The Postgres extension that adds vector similarity search to ordinary Postgres tables.
Read definitionRed Teaming
Adversarial testing of AI systems by humans (or other AI) trying to make them fail or misbehave.
Read definitionVector Database
A specialised database optimised for storing and searching high-dimensional vectors by similarity.
Read definitionWatermarking
Embedding hidden signals in AI-generated content so it can later be identified as synthetic.
Read definitionWorkflows & Careers
Practical AI adoption terms for roles, teams, course buyers, workflows, governance and career growth.
Agentic Workflow
A workflow where an AI agent plans and executes multiple tool-using steps toward a goal.
Read definitionAI Adoption Roadmap
A phased plan for introducing AI workflows, training, governance and measurement inside a team.
Read definitionAI Certificate
A completion credential that signals structured AI study, usually strongest when paired with a portfolio project.
Read definitionAI Content Brief
A structured instruction packet that tells an AI what audience, angle, facts and format to use.
Read definitionAI Course Pricing
The way AI learning products are priced across one-time courses, subscriptions and bundled access.
Read definitionAI Dashboard Narrative
A plain-English explanation of dashboard changes, drivers, caveats and next actions generated with AI support.
Read definitionAI Governance
The operating system of decisions, controls and accountability around AI use in an organization.
Read definitionAI Learning Path
A sequenced plan for moving from AI basics to practical workflows in a role or topic.
Read definitionAI Literacy
The practical ability to use, question and govern AI output in everyday work.
Read definitionAI Operating System
A team's practical layer of tools, policies and workflows for using AI every week.
Read definitionAI Playbook
A documented set of AI prompts, rules, review steps and examples for a recurring business task.
Read definitionAI Policy
A practical rule set that defines approved AI tools, allowed data, review standards and ownership.
Read definitionAI Portfolio Project
A concrete work sample that shows how someone used AI to produce a useful outcome.
Read definitionAI Procurement
The process of evaluating, buying and governing AI tools based on fit, risk, cost and data handling.
Read definitionAI Quality Checklist
A short set of criteria used to review AI output before it is published, sent or used for decisions.
Read definitionAI Readiness
The degree to which a person or organization can adopt AI without chaos or avoidable risk.
Read definitionAI Research Assistant
An AI workflow that gathers, summarizes and compares sources while preserving citations and uncertainty.
Read definitionAI ROI
The measurable return from AI adoption, usually time saved, revenue gained or risk reduced.
Read definitionAI Sales Sequence
A series of AI-assisted sales messages built from real account research and a clear hypothesis.
Read definitionAI SOP
A standard operating procedure that defines where AI fits in a repeatable process.
Read definitionAI Task Automation
Using AI to draft, classify, route, summarize or act on repeatable tasks with appropriate checks.
Read definitionAI Tool Stack
The set of AI tools a person or team uses for writing, research, automation, creation and review.
Read definitionAI Use Case
A specific work problem where AI can help produce a faster, better or cheaper result.
Read definitionAI Workflow
A repeatable sequence where AI helps transform inputs into a useful work output.
Read definitionAutomation Map
A visual or written map of triggers, inputs, tools, approvals and outputs in an automation.
Read definitionContext Engineering
The practice of designing what information an AI model sees before it generates an answer.
Read definitionCourse Preview
Public course material such as a curriculum, trailer or topic page that helps a buyer evaluate fit before paying.
Read definitionCRM AI
AI embedded into customer relationship management work such as summaries, scoring and follow-up.
Read definitionData Cleaning
The process of correcting, normalizing and validating data before analysis or AI use.
Read definitionData Quality
A measure of whether data is accurate, complete, current and fit for an AI or analytics task.
Read definitionExecutive Summary AI
AI-assisted synthesis that converts long material into decisions, context, risks and next steps for leaders.
Read definitionHuman-in-the-Loop Review
A control pattern where a person checks AI output before it affects customers, money or decisions.
Read definitionKnowledge Base AI
AI that answers from an approved body of documents, policies, tickets or help-center content.
Read definitionMeeting Summary AI
AI used to turn transcripts into decisions, owners, risks and concise follow-up notes.
Read definitionPrivate Data Prompting
Using AI with sensitive data only under approved policies, redaction and vendor controls.
Read definitionPrompt Evaluation
Testing prompt outputs against examples, rubrics and failure cases before reusing them.
Read definitionPrompt Library
A curated repository of reusable prompts, examples and output standards for common tasks.
Read definitionPrompt Template
A reusable prompt structure with slots for context, audience, examples, constraints and output format.
Read definitionSource Grounding
Constraining AI answers to approved source material so claims can be checked.
Read definitionSpreadsheet AI
AI assistance for formulas, cleanup, classification, summaries and analysis inside spreadsheet workflows.
Read definitionSupport Triage AI
AI that classifies incoming support issues by intent, urgency, customer state and next action.
Read definitionA–Z index
Every term, alphabetically
Looking for something specific? Use the index below.
A
- AgentAn AI system that autonomously plans, takes actions through tools, and iterates toward a goal.
- Agentic WorkflowA workflow where an AI agent plans and executes multiple tool-using steps toward a goal.
- AGI (Artificial General Intelligence)Hypothetical AI capable of matching or exceeding human performance across the full range of cognitive tasks.
- AI Adoption RoadmapA phased plan for introducing AI workflows, training, governance and measurement inside a team.
- AI CertificateA completion credential that signals structured AI study, usually strongest when paired with a portfolio project.
- AI Content BriefA structured instruction packet that tells an AI what audience, angle, facts and format to use.
- AI Course PricingThe way AI learning products are priced across one-time courses, subscriptions and bundled access.
- AI Dashboard NarrativeA plain-English explanation of dashboard changes, drivers, caveats and next actions generated with AI support.
- AI GovernanceThe operating system of decisions, controls and accountability around AI use in an organization.
- AI Learning PathA sequenced plan for moving from AI basics to practical workflows in a role or topic.
- AI LiteracyThe practical ability to use, question and govern AI output in everyday work.
- AI Operating SystemA team's practical layer of tools, policies and workflows for using AI every week.
- AI PlaybookA documented set of AI prompts, rules, review steps and examples for a recurring business task.
- AI PolicyA practical rule set that defines approved AI tools, allowed data, review standards and ownership.
- AI Portfolio ProjectA concrete work sample that shows how someone used AI to produce a useful outcome.
- AI ProcurementThe process of evaluating, buying and governing AI tools based on fit, risk, cost and data handling.
- AI Quality ChecklistA short set of criteria used to review AI output before it is published, sent or used for decisions.
- AI ReadinessThe degree to which a person or organization can adopt AI without chaos or avoidable risk.
- AI Research AssistantAn AI workflow that gathers, summarizes and compares sources while preserving citations and uncertainty.
- AI ROIThe measurable return from AI adoption, usually time saved, revenue gained or risk reduced.
- AI Sales SequenceA series of AI-assisted sales messages built from real account research and a clear hypothesis.
- AI SOPA standard operating procedure that defines where AI fits in a repeatable process.
- AI Task AutomationUsing AI to draft, classify, route, summarize or act on repeatable tasks with appropriate checks.
- AI Tool StackThe set of AI tools a person or team uses for writing, research, automation, creation and review.
- AI Use CaseA specific work problem where AI can help produce a faster, better or cheaper result.
- AI WorkflowA repeatable sequence where AI helps transform inputs into a useful work output.
- AlgorithmA finite, well-defined sequence of steps that solves a problem or performs a task.
- AlignmentThe discipline of ensuring AI systems pursue the goals their developers and users actually want.
- Artificial Intelligence (AI)Software systems that perform tasks normally requiring human reasoning, perception or decision-making.
- ASI (Artificial Superintelligence)Hypothetical AI vastly exceeding human cognitive abilities across essentially all domains.
- AttentionThe mechanism that lets each token in a sequence weigh and combine information from every other token.
- AutoGPTA 2023 viral open-source autonomous agent project that helped popularise the agent-loop concept.
- Automation MapA visual or written map of triggers, inputs, tools, approvals and outputs in an automation.
B
- BackpropagationThe algorithm that computes gradients through a neural network by applying the chain rule layer by layer.
- BiasTwo meanings: a model's systematic underfitting, or unfair behaviour toward demographic groups.
- BLEU ScoreA classic metric for machine translation quality based on n-gram overlap with reference translations.
- Browser AutomationLetting an AI agent control a real web browser to navigate, fill forms and extract information.
C
- Chain-of-Thought (CoT)Asking the model to reason step by step out loud before giving its final answer.
- ChatGPTOpenAI's flagship conversational AI assistant, the product that brought generative AI to mainstream attention.
- ClaudeAnthropic's family of conversational AI models, known for long context, careful reasoning and coding strength.
- Claude Sonnet 4Anthropic's flagship workhorse model in 2026, known for coding, careful reasoning and long context.
- Computer UseAn LLM's ability to control a desktop environment with screenshots, mouse and keyboard input.
- Computer VisionThe field of AI that enables computers to interpret and reason about images and video.
- ContextAll the text the model can see when generating a response — the system prompt, history and current input.
- Context EngineeringThe practice of designing what information an AI model sees before it generates an answer.
- Context WindowThe maximum amount of input (in tokens) a model can read in a single request.
- ControlNetAn extension to Stable Diffusion that lets you condition image generation on poses, depth maps or sketches.
- Copyright (and AI-generated works)The legal protection for original creative works; complicated when AI generates the output.
- Course PreviewPublic course material such as a curriculum, trailer or topic page that helps a buyer evaluate fit before paying.
- CPU (Central Processing Unit)The general-purpose processor at the heart of every computer; runs orchestration around AI workloads.
- CRM AIAI embedded into customer relationship management work such as summaries, scoring and follow-up.
- Cross-ValidationA technique for measuring model quality by training and testing on rotating slices of the dataset.
- CursorAn AI-first code editor based on VS Code, the dominant pick for AI-assisted development in 2026.
- Custom ToolA function you write and expose to an agent so it can perform a specific action in your system.
D
- DALL-E 3OpenAI's image generation model integrated into ChatGPT, known for strong literal prompt following.
- Data CleaningThe process of correcting, normalizing and validating data before analysis or AI use.
- Data QualityA measure of whether data is accurate, complete, current and fit for an AI or analytics task.
- DatasetA structured collection of examples used to train, validate or evaluate a machine-learning model.
- Deep LearningMachine learning using multi-layer neural networks; the foundation of modern LLMs and image models.
- DeepfakeSynthetic media (image, audio, video) that convincingly depicts events or statements that did not happen.
- Depth EstimationPredicting the distance from camera to every pixel in an image, producing a depth map.
- Diffusion ModelA generative model that learns to reverse a noising process, the dominant approach to image and video generation.
- DistillationTraining a smaller, cheaper model to imitate the outputs of a larger one.
E
- ElevenLabsThe leading voice AI company, known for high-quality text-to-speech, voice cloning and dubbing.
- EmbeddingA dense vector that represents the meaning of a piece of text, image or other content.
- Embedding StoreThe infrastructure layer that stores, indexes and serves embeddings for retrieval and search.
- EU AI ActThe European Union's risk-based AI regulation, the first comprehensive AI law from a major jurisdiction.
- Eval (Evaluation)Tests that measure how well an AI system performs on intended tasks and under adversarial conditions.
- Executive Summary AIAI-assisted synthesis that converts long material into decisions, context, risks and next steps for leaders.
F
- Fair Use (and AI training data)The US copyright doctrine allowing limited use of copyrighted material; central to ongoing AI training lawsuits.
- Few-Shot PromptingIncluding a handful of input/output examples in the prompt to teach the model the desired pattern.
- Fine-tuningContinuing to train a pretrained model on your own data so it specialises in your task or style.
- Function Calling (Tool Use)An API mechanism where the model decides to call your code, with arguments, instead of generating text.
G
- Gaussian SplattingA 3D scene representation made of millions of coloured 3D Gaussians, faster to render than NeRF.
- GeminiGoogle's family of multimodal AI models, integrated across Workspace, Android and the Gemini app.
- GitHub CopilotMicrosoft and GitHub's AI coding assistant, embedded in VS Code, JetBrains and the GitHub web interface.
- GPT-5OpenAI's flagship LLM as of 2025–2026, with multimodal input, built-in reasoning and agentic tool use.
- GPU (Graphics Processing Unit)Massively parallel processors originally built for graphics, now the workhorse of AI training and inference.
- Gradient DescentThe optimisation algorithm that powers neural network training by following the slope of the loss surface.
H
I
- Image GenerationAI systems that produce images from text prompts, reference images or both.
- Image SegmentationLabeling every pixel in an image so each region corresponds to an object or class.
- Image-to-VideoGenerating a moving video clip from a single still image, often with described motion.
- InferenceRunning a trained model on new inputs to produce predictions or generated content.
- InpaintingEditing a specific region of an image with AI while keeping the rest unchanged.
- Instruction TuningFine-tuning a base model on (instruction, response) pairs so it follows directions instead of just predicting text.
J
K
L
- LabelingAttaching the correct answer to each example in a dataset so a model can learn to reproduce it.
- LatencyHow long an AI system takes to respond; often more important than raw quality for user experience.
- Latent SpaceA compressed mathematical space where AI models represent the meaningful features of their inputs.
- LeaderboardA ranked list of AI models on a benchmark; useful for model selection and the source of constant marketing.
- LlamaMeta's family of open-weight large language models, the foundation of much of the open AI ecosystem.
- LoRA (Low-Rank Adaptation)A parameter-efficient fine-tuning technique that adds tiny trainable adapters to a frozen base model.
M
- Machine Learning (ML)A branch of AI where algorithms learn patterns from data instead of following hand-coded rules.
- MCP (Model Context Protocol)An open standard from Anthropic for connecting tools and data sources to any compatible LLM.
- Meeting Summary AIAI used to turn transcripts into decisions, owners, risks and concise follow-up notes.
- Memory (Agent Memory)Persistent state that lets an agent remember facts and preferences across sessions and conversations.
- Midjourney v7The 2025–2026 release of the popular image generation service, with sharper realism and stronger prompt control.
- MistralA French AI lab known for high-quality open-weight models, especially Mixtral MoE and Mistral Large.
- Mixture of Experts (MoE)An architecture where each input only activates a subset of the model's parameters, decoupling capacity from cost.
- ModelA trained mathematical artifact that maps inputs to outputs, like a function learned from data.
- Model CardA standardised document describing an AI model's purpose, training, limitations and intended use.
- Multi-Agent SystemAn architecture where multiple specialised AI agents collaborate on a task with distinct roles.
N
- nano-banana-2An automation-friendly image generation model accessible via the Evolink API, popular for batch marketing pipelines.
- NeRF (Neural Radiance Fields)A 3D scene representation learned by a neural network from multiple 2D photos.
- Neural NetworkA mathematical model loosely inspired by the brain, made of layers of weighted connections that learn from data.
- NIST AI Risk Management FrameworkA voluntary US framework for managing risks across the AI lifecycle, increasingly cited in procurement.
- NotebookLMGoogle's source-grounded AI research notebook; especially famous for its podcast-style audio summaries.
O
- OCR (Optical Character Recognition)Extracting text from images and scanned documents so it can be searched and processed.
- Open WeightsModels whose trained parameters are publicly downloadable, even if the training data and code are not.
- OrchestratorThe control layer that routes work between models, agents, tools and humans in a complex AI workflow.
- OutpaintingExtending an image beyond its original borders with AI-generated content that matches the existing scene.
- OverfittingWhen a model memorises the training set and fails to generalise to new examples.
P
- P(doom)Subjective probability that AI causes existential or near-existential catastrophe; a contested AI-policy shorthand.
- ParametersThe learned weights inside a model; counted in millions or billions, they store everything the model knows.
- PerplexityA measure of how surprised a language model is by a piece of text; lower means better fit.
- Perplexity (search engine)An AI-native search engine that answers questions with cited sources instead of a list of links.
- pgvectorThe Postgres extension that adds vector similarity search to ordinary Postgres tables.
- PlannerThe component of an agent that decomposes a goal into ordered steps before execution.
- Pose EstimationDetecting the position of human (or animal) joints in images and video to reconstruct skeletal pose.
- Private Data PromptingUsing AI with sensitive data only under approved policies, redaction and vendor controls.
- Prompt EngineeringThe discipline of writing instructions to LLMs so they produce reliable, useful outputs.
- Prompt EvaluationTesting prompt outputs against examples, rubrics and failure cases before reusing them.
- Prompt InjectionAn attack where untrusted input contains instructions that override the developer's system prompt.
- Prompt LibraryA curated repository of reusable prompts, examples and output standards for common tasks.
- Prompt TemplateA reusable prompt structure with slots for context, audience, examples, constraints and output format.
R
- RAG (Retrieval-Augmented Generation)A pattern where the model retrieves relevant documents from a knowledge base and uses them to answer.
- ReAct PatternAn agent loop that interleaves Reasoning and Acting steps, thinking out loud before each tool call.
- Red TeamingAdversarial testing of AI systems by humans (or other AI) trying to make them fail or misbehave.
- RegularizationTechniques that constrain a model so it generalises to new data instead of memorising the training set.
- Replit AgentReplit's AI agent that builds, debugs and deploys full applications from natural-language descriptions.
- RLHF (Reinforcement Learning from Human Feedback)A training technique where humans rank model outputs and the model is optimised to produce higher-ranked responses.
- Role PromptingTelling the model who to act as — "You are a senior tax accountant" — to set expertise and tone.
S
- SoraOpenAI's text-to-video model, capable of generating realistic short clips from text or image prompts.
- Source GroundingConstraining AI answers to approved source material so claims can be checked.
- Speech-to-Text (STT)AI systems that convert spoken audio into written text; the input side of voice interfaces.
- Spreadsheet AIAI assistance for formulas, cleanup, classification, summaries and analysis inside spreadsheet workflows.
- Stable DiffusionThe leading family of open-weight image generation models, the foundation of much of the open creative AI ecosystem.
- Structured OutputForcing the model to return data in a specific schema (JSON, XML) so it can be safely parsed.
- SunoAn AI music generation service that produces full vocal songs from text descriptions.
- Support Triage AIAI that classifies incoming support issues by intent, urgency, customer state and next action.
- System PromptThe instructions sent before the user's message that define the model's role, tone and constraints.
T
- TemperatureA sampling parameter controlling how random or deterministic the model's outputs are.
- Text-to-3DGenerating 3D models, scenes or assets from text prompts; the 3D analog of text-to-image.
- Text-to-Speech (TTS)AI systems that convert written text into natural-sounding spoken audio.
- TokenThe atomic unit a language model reads and writes; usually a word, sub-word fragment or punctuation mark.
- Tool UseAn LLM's ability to call external functions or APIs to extend its capabilities beyond text generation.
- Top-p (Nucleus Sampling)A sampling method that restricts the model to the smallest set of tokens whose probabilities sum to p.
- TrainingThe process of adjusting a model's weights from data so that it learns to perform a task.
- Transfer LearningAdapting a model trained on one task to a related task with much less data and compute.
- TransformerThe neural network architecture introduced in 2017 that powers virtually every modern LLM.
V
- VarianceHow sensitive a model is to the specific training set; high variance means high overfitting risk.
- Vector DatabaseA specialised database optimised for storing and searching high-dimensional vectors by similarity.
- Voice CloningGenerating a synthetic voice that mimics a specific person's vocal identity from a short audio sample.
W
From theory to skill
Knowing the words is step one. Using them is the goal.
Once a term clicks, the next move is to put it to work. Browse our courses to turn vocabulary into a working AI workflow you can ship this week.