The AI glossary, minus the hand-waving.
148 terms, each defined in one honest paragraph. No circular definitions allowed.
A
Active Parameters
In a mixture-of-experts model, the parameters actually used for each token — far fewer than the total, and what drives speed and cost.
Read definitionAgent Memory
The mechanisms that let an AI agent remember facts, preferences, and past interactions across sessions rather than starting blank each time.
Read definitionAgentic AI
AI systems designed to take autonomous, multi-step action toward a goal.
Read definitionAgentic Browser
A web browser with AI built in deeply enough to complete tasks for you, not just chat about a page.
Read definitionAgentic Commerce
Shopping in which an AI agent searches, compares, and completes purchases on a person's behalf, with payment networks verifying the agent is authorised.
Read definitionAGI (Artificial General Intelligence)
A hypothetical AI that matches or exceeds human cognitive ability across all tasks.
Read definitionAI Agent
An AI that takes actions in the world, not just generates text.
Read definitionAI Alignment
The problem of making AI systems do what humans actually want, safely.
Read definitionAI Avatar
A synthetic on-screen presenter, either a stock character or a clone of a real person, that speaks any script you type.
Read definitionAI Code Review
AI tools that automatically comment on pull requests, flagging bugs, style issues, and security problems.
Read definitionAI Companion
A chatbot designed for ongoing personal relationships — friendship, support, or romance — rather than tasks or information.
Read definitionAI Content Suite
A single platform bundling multiple AI generation tools — writing, image, video — instead of separate point solutions.
Read definitionAI Copilot
An AI assistant embedded directly inside an existing app or workflow, rather than a separate standalone chatbot.
Read definitionAI Detector
A tool that estimates how likely a piece of text was written by AI.
Read definitionAI Eval
Systematic tests that measure whether an AI model or app produces good output.
Read definitionAI Humanizer
A tool that rewrites AI-generated text to sound like a real person wrote it.
Read definitionAI Lip Sync
Making a face in a video move its mouth to match a new audio track, used for dubbing, avatars, and translated videos.
Read definitionAI Observability
Tools and practices for tracking how an AI application behaves in production — costs, latency, quality, and failures.
Read definitionAI Safety Institute
A government body that independently tests frontier AI models for dangerous capabilities before or after release.
Read definitionAI Sandbox
An isolated computing environment where AI agents can act without touching your real data or files.
Read definitionAI Scribe
Software that listens to a conversation, usually a doctor visit, and drafts the notes and paperwork automatically.
Read definitionAI Search
Search engines that synthesise answers from sources rather than returning a list of links.
Read definitionAI Slop
Low-effort, mass-produced AI content — text, images, video — published for clicks or volume with little human care.
Read definitionAI Startup
A company building on or with AI — the largest startup category of 2024–2026.
Read definitionAI Voice Agent
An AI system that holds a real-time spoken conversation, usually over the phone, and can take actions like booking or looking up an account.
Read definitionAI Voice Cloning
AI that recreates a specific human voice from a short audio sample.
Read definitionAI Watermarking
Invisible markers embedded in AI-generated images, video, or text to identify them as machine-made.
Read definitionAI Wearable
A small worn device, like a pendant or pin, that records ambient audio and turns it into searchable AI notes.
Read definitionAI Website Builder
A tool that generates a complete website, from structure to copy and layout, from a short description of a business or project.
Read definitionAI Wrapper
A product built mostly on someone else's AI model, adding a specific interface or workflow rather than new underlying technology.
Read definitionAI-Native IDE
A code editor designed around AI from the start, with agents, multi-file edits, and codebase awareness built in rather than added as a plugin.
Read definitionAlways-On Agent
An AI agent that holds an ongoing responsibility and keeps working between your messages, rather than answering one prompt and stopping.
Read definitionAPI (Application Programming Interface)
How developers access AI models programmatically — the foundation of every AI product.
Read definitionB
Background Coding Agent
An AI coding agent that works asynchronously in the cloud — assign a task, get a PR hours later.
Read definitionBenchmark Contamination
When an AI's training data accidentally includes the answers to the tests used to measure it, inflating its scores.
Read definitionBYOK (Bring Your Own Key)
A pricing model where you use a tool for free or cheaply but supply your own API key, so the model provider bills you directly for usage.
Read definitionC
Chain-of-Thought (CoT)
Prompting an AI to "show its work" — dramatically improves accuracy on hard problems.
Read definitionChatGPT
OpenAI's flagship AI assistant — the app that kicked off the AI boom in late 2022.
Read definitionChunking
Splitting long documents into smaller pieces before embedding them, so a retrieval system can find and return the relevant passage rather than the whole file.
Read definitionClaude
Anthropic's AI assistant — known for writing quality, long-document analysis, and coding.
Read definitionClaude Opus
Anthropic's most capable model family — leads on coding and long-form reasoning.
Read definitionCompute Governance
Government policies that control access to the advanced chips and data centers needed to train frontier AI models.
Read definitionComputer Use
AI that can see your screen, move your cursor, and click — controlling a computer like a person would.
Read definitionConfidential Computing
Running computation inside hardware-isolated enclaves so that even the cloud operator cannot read the data or model while it is being processed.
Read definitionConstitutional AI
Anthropic's technique for training AI to follow written principles instead of relying only on human feedback.
Read definitionContent Credentials (C2PA)
A tamper-evident label attached to an image, video, or audio file recording where it came from and whether AI was used to make or edit it.
Read definitionContext Engineering
Deliberately deciding what information to feed an AI model at each step, beyond just writing a good prompt.
Read definitionContext Rot
The tendency for an AI model's response quality to quietly decline as its context window fills up.
Read definitionContext Window
The amount of text an AI can consider at once — its short-term memory.
Read definitionCursor
The AI-first VS Code fork — most-used AI coding tool among professional developers.
Read definitionD
Decision Model
A small model that returns an answer or label in one forward pass instead of generating text, built for routing, verification, and fast yes-or-no choices.
Read definitionDeepfake
Realistic fake video, audio, or images of a real person generated or altered by AI, usually without their consent.
Read definitionDevin
Cognition's autonomous AI software engineer — takes a Jira ticket, opens a PR.
Read definitionDiffusion Model
The AI architecture behind Stable Diffusion, DALL-E, Sora 2, and most modern image/video generators.
Read definitionE
Effort Setting
An API control that tells a model how much reasoning to spend on a request, trading speed and cost against capability without changing models.
Read definitionElevenLabs
The leading AI voice generator — natural-sounding voices, cloning, multilingual.
Read definitionElo Rating (AI)
A ranking score for AI models derived from thousands of blind head-to-head votes, borrowed from chess ratings.
Read definitionEmbeddings
How AI represents meaning as lists of numbers — the math behind search and RAG.
Read definitionF
Fine-Tuning
Continuing to train a pre-trained AI model on your own specific data.
Read definitionFrontier Model
The most capable AI models at the cutting edge — GPT-5.6, Claude Opus 4.8, Gemini 3.5 Pro.
Read definitionFunction Calling
How LLMs invoke tools — the model outputs structured JSON instead of free text.
Read definitionG
Gemini
Google's frontier AI family — biggest context window, strongest on multimodal.
Read definitionGenerative UI
An AI reply that includes interactive elements — a chart, a form, a button, a small tool — generated on the spot because they fit the answer better than text.
Read definitionGitHub Copilot
The AI coding autocomplete that started the AI-pair-programming category.
Read definitionGPT-5
OpenAI's flagship model — reasoning, multimodal, and the model behind most ChatGPT Plus output.
Read definitionGPU Cloud
A service that rents graphics processors by the hour or second so you can train, fine-tune, or run AI models without buying hardware.
Read definitionGrok
xAI's chatbot — real-time X access, fewer content restrictions, competitive with GPT-5.
Read definitionGrounding (AI)
Connecting an AI's answers to verifiable, real-world sources instead of relying only on what it memorized in training.
Read definitionGuardrails (AI)
Rules and filters wrapped around an AI system to keep its output within safe, intended limits.
Read definitionH
Hallucination
When an AI confidently generates text that is plausible-sounding but factually wrong.
Read definitionHugging Face
The GitHub of AI — hosts open-source models, datasets, and inference for the whole community.
Read definitionHybrid Model Architecture (Transformer-Mamba)
AI models that blend standard Transformer layers with a state-space design like Mamba to handle long context more cheaply.
Read definitionHybrid Search
Combining keyword search and vector similarity search in one query so results match both exact terms and meaning.
Read definitionI
Image-to-Video
Generating a video that starts from, or is guided by, an existing still image, giving far more control than a text prompt alone.
Read definitionIn-Context Learning
An LLM's ability to learn a pattern from examples given in the prompt, with no retraining involved.
Read definitionInference Provider
A company that runs AI models on its own servers so developers can call them via API instead of hosting models themselves.
Read definitionK
Keyframe Conditioning (Video)
Supplying one or more fixed frames that a video model must hit at set points, so you control where a generated clip starts, passes through, and ends.
Read definitionKnowledge Cutoff
The date after which an AI model has no built-in knowledge, because its training data ends there.
Read definitionL
Large Language Model (LLM)
An AI trained on enormous amounts of text to predict what word comes next.
Read definitionLate-Interaction Embeddings
Embedding each token of a document separately and scoring queries token-by-token at search time, instead of compressing a whole passage into one vector.
Read definitionLatent Space
The internal, abstract mathematical space where an AI model represents the meaning of its inputs.
Read definitionLLM Gateway
A single API layer that sits between your app and many AI providers, handling routing, fallbacks, keys, and cost tracking.
Read definitionLLM-as-a-Judge
Using a language model to score or compare other models' outputs, as a cheaper stand-in for human evaluation.
Read definitionLocal LLM
A language model that runs entirely on your own computer or server instead of a cloud provider's, so no data leaves your machine.
Read definitionLong Context
AI models that can hold millions of tokens at once — entire codebases, books, year-long threads.
Read definitionLong-Context Surcharge
A higher per-token rate that applies once a prompt crosses a size threshold, usually charged on the whole request rather than only the tokens over the line.
Read definitionLPU (Language Processing Unit)
A chip built specifically for fast AI inference, pioneered by Groq as an alternative to general-purpose GPUs.
Read definitionM
Midjourney
The AI image generator with the most distinctive aesthetic — runs in Discord.
Read definitionMixture of Depths (MoD)
A neural architecture where the model decides per-token how many layers to use — saving compute on easy tokens.
Read definitionMixture of Experts (MoE)
An AI architecture where only a fraction of the model "wakes up" per request.
Read definitionModel Card
A standardized document publishing an AI model's training data, capabilities, limitations, and known risks.
Read definitionModel Collapse
A degradation that can occur when AI models are trained too heavily on other AI-generated content instead of real-world data.
Read definitionModel Context Protocol (MCP)
The open standard that lets AI assistants connect to external tools and data sources.
Read definitionModel Distillation
Training a small AI model to mimic a larger one — how GPT-5 mini came from GPT-5.
Read definitionModel Router
A system that automatically sends each request to the cheapest AI model capable of handling it well.
Read definitionModel Weights
The millions or billions of learned numbers inside a trained AI model that encode everything it knows.
Read definitionMulti-Agent Orchestration
Coordinating several specialized AI agents that each handle part of a larger task.
Read definitionMultimodal AI
AI that handles text, images, audio, and video in a single model.
Read definitionN
Native Audio Generation (Video)
A video model producing the soundtrack, from dialogue to ambient sound and effects, in the same pass as the picture, synchronised to what is on screen.
Read definitionNotebookLM
Google's research notebook — upload sources, get a personalised AI tutor with citations.
Read definitionO
On-Device AI
AI that runs entirely on your phone or laptop — no cloud, no internet required.
Read definitionOpen Source AI
AI models whose weights are publicly released — anyone can run, fine-tune, and modify them.
Read definitionOpen-Weight Model
A model whose trained parameters are published for anyone to download and run, even if the training data and code are not.
Read definitionP
Perplexity
The AI search engine — answers with citations, replaces Google for research queries.
Read definitionPersonal Agent Protocol (PAP)
A proposed open standard for how a person's AI agent identifies itself to a business, signs in, and is granted read or write access on the customer's behalf.
Read definitionPrompt Caching
A pricing trick that cuts AI API input cost by 50–90% for repeat-context workflows.
Read definitionPrompt Engineering
The skill of writing clear, structured instructions that get the best results from AI.
Read definitionPrompt Injection
An attack that hides malicious instructions inside content an AI reads, tricking it into acting against its user.
Read definitionR
RAG (Retrieval Augmented Generation)
The technique that lets AI actually know things it was not trained on.
Read definitionRAG vs Fine-Tuning
When to use RAG (retrieval) versus fine-tuning (continued training) — the central architectural choice.
Read definitionReasoning Effort
A setting that controls how much hidden thinking a reasoning model does before answering, trading speed and cost against accuracy.
Read definitionReasoning Model
An AI that "thinks" before answering — explicit step-by-step reasoning at inference time.
Read definitionRed-Teaming (AI)
Deliberately trying to break an AI model before release to find dangerous or unwanted behavior first.
Read definitionReranker
A second-stage model that re-orders search results by relevance so the best passages reach the AI first.
Read definitionRLHF (Reinforcement Learning from Human Feedback)
How AI labs teach models to be helpful, harmless, and honest — using human ratings.
Read definitionS
Safety Case
A structured, evidence-backed argument that an AI system is safe enough to train or deploy for a stated use, borrowed from aviation and nuclear engineering.
Read definitionSemantic Search
Search that matches on meaning rather than exact words, usually by comparing embeddings of the query and the documents.
Read definitionServerless Inference
Running an AI model behind an endpoint that spins up on demand and scales to zero when idle, so you pay only for the seconds it is actually working.
Read definitionShadow AI
Employees using AI tools at work without their employer's knowledge, review, or approval.
Read definitionSmall Language Model (SLM)
A compact AI model, typically under 10 billion parameters, built to run fast and cheap on limited hardware.
Read definitionSora 2
OpenAI's AI video generator — text or image to high-quality video clips.
Read definitionSovereign AI
AI models built and hosted within a country's borders for data sovereignty and strategic independence.
Read definitionSpeaker Diarization
Working out which speaker said which parts of a recording, so a transcript can be labelled by person.
Read definitionSpeculative Decoding
A trick that makes frontier AI 2-3x faster by using a tiny model to guess tokens for a big one to verify.
Read definitionSpeech-to-Text API
A service that converts audio into text, which other apps plug into instead of building transcription themselves.
Read definitionStable Diffusion
The open-source AI image generator — runs locally, infinitely customisable.
Read definitionStem Separation
Using AI to split a finished song back into its parts, such as vocals, drums, bass, and instruments.
Read definitionStructured Output
Forcing a model to reply in an exact machine-readable format, such as JSON matching a schema, so software can use the answer directly.
Read definitionSWE-bench
The standard benchmark for AI coding ability: can a model fix real GitHub issues so that the project's own tests pass?
Read definitionSycophancy (AI)
An AI model's tendency to tell users what they want to hear instead of what's accurate.
Read definitionSynthetic Data
Artificially generated data used to train AI models when real-world data is scarce, costly, or sensitive.
Read definitionSystem Prompt
The hidden instructions an app sends an AI model before the user's message, shaping its behavior and persona.
Read definitionT
Temperature
The AI setting that controls how creative (high) or predictable (low) responses are.
Read definitionTest-Time Compute
Letting an AI think longer at inference time — the breakthrough behind reasoning models.
Read definitionText-to-3D
Generating a three-dimensional model, with geometry and textures, from a text description or a reference image.
Read definitionText-to-Video
Generating a video clip from a written description, using a model trained on video and its captions.
Read definitionToken Budget
The cap a team sets on how many tokens (and dollars) an AI feature can use per request or period.
Read definitionTokens
The chunks of text AI models read and produce — usually 3–4 characters each.
Read definitionTokens per Second
The speed at which a model produces text, measured in tokens generated each second; the main number behind how fast an AI feels.
Read definitionTool Use
How AI models call external tools — search, code execution, APIs — to do real work.
Read definitionV
Vector Database
A database optimised for storing and searching AI embeddings — the backbone of every RAG system.
Read definitionVibe Coding
Building software by describing what you want to an AI and accepting its code largely on trust, iterating by feel rather than by reading every line.
Read definitionVibe Design
Producing interfaces, brands, and layouts by prompting an AI design agent and iterating on the result, the design-world counterpart to vibe coding.
Read definitionZ
Zero Data Retention (ZDR)
An API or enterprise setting under which the provider does not store your prompts or outputs after processing the request.
Read definitionZero-Shot / Few-Shot Learning
Getting an AI to do a task with no examples (zero-shot) or just a handful of examples (few-shot) in the prompt.
Read definition