AI glossary and learning path: from beginner to expert

Artificial intelligence (AI)

German: Künstliche Intelligenz

Artificial intelligence (AI) is the field of computer science that builds systems able to perform tasks that normally need human judgment — recognizing patterns, understanding and producing language, making predictions or recommendations — by learning from data rather than following only hand-written rules.

Beginner · Technical writers · Technical marketers · Technical project managers

Machine learning (ML)

German: Maschinelles Lernen

Machine learning (ML) is the approach to artificial intelligence in which a system learns patterns from example data and uses them to make predictions or decisions on new data, instead of being programmed with explicit rules for every case.

Beginner · Technical writers · Technical marketers · Technical project managers

Neural network

A neural network is a machine learning model made of layers of connected numeric units whose connection strengths — the weights — are adjusted during training so that the network maps inputs, such as words or pixels, to useful outputs. Networks with many layers are called deep learning.

Beginner · Intermediate · Technical writers · Technical marketers · Technical project managers · Developers

Transformer

The transformer is the neural network architecture behind today’s large language models. It uses a mechanism called attention to weigh how strongly every token in a sequence relates to every other token, which lets it model long-range context and train efficiently on very large amounts of text.

Intermediate · Expert · Developers

Generative AI

Generative AI is artificial intelligence that creates new content — text, images, audio, video or code — in response to an instruction, by producing output that follows the patterns of the data the model was trained on.

Beginner · Technical writers · Technical marketers · Technical project managers

Large language model (LLM)

A large language model (LLM) is a generative AI model, usually a transformer, trained on very large amounts of text to predict the next token in a sequence. From this single skill emerge abilities such as answering questions, summarizing, translating, writing code and following instructions.

Beginner · Technical writers · Technical marketers · Technical project managers · Developers

Token (AI)

A token is the unit of text a language model reads and writes — a word, part of a word, a punctuation mark or a space. Models measure input length, output length, context limits and usage-based prices in tokens.

Beginner · Technical writers · Technical marketers · Technical project managers · Developers

Context window

The context window is the maximum amount of text, measured in tokens, that a language model can take into account at once — the prompt, any attached documents, the conversation so far and its own answer. Anything outside the window is invisible to the model.

Beginner · Technical writers · Technical project managers · Developers

Training vs. inference

Training is the phase in which a model learns by adjusting its weights on large datasets; inference is the phase in which the finished model is used to produce outputs for new inputs. Training happens rarely and costs a lot of compute; inference happens with every request.

Intermediate · Technical project managers · Developers

Parameters and weights

Parameters are the numeric values inside a neural network — mostly the weights of its connections — that are learned during training and store what the model knows. Model size is usually given as the number of parameters, from a few billion to more than a trillion.

Intermediate · Technical marketers · Developers

Multimodal model

A multimodal model is an AI model that can process or generate more than one type of data — for example text and images, or text, audio and video — within the same model.

Beginner · Technical writers · Technical marketers · Technical project managers

Hallucination (AI)

A hallucination is output from a generative AI model that sounds plausible but is false, unsupported or invented — for example a nonexistent citation, a wrong specification value or a procedure step that doesn’t exist.

Beginner · Technical writers · Technical marketers · Technical project managers

Foundation model

A foundation model is a large AI model trained on broad data at scale that can be adapted to a wide range of tasks — through prompting, retrieval or fine-tuning — instead of being built for a single purpose.

Intermediate · Technical project managers · Developers

Reasoning model

A reasoning model is a large language model trained to work through a problem in intermediate steps before giving its final answer, spending more computation at inference time to improve results on complex tasks such as math, code and multi-step analysis.

Intermediate · Technical writers · Technical project managers · Developers

Open-weight model

An open-weight model is an AI model whose trained weights are published for download, so organizations can run, inspect and adapt it on their own infrastructure — under a license that may or may not meet the definition of open source.

Intermediate · Technical marketers · Technical project managers · Developers

OpenAI (ChatGPT, GPT models)

OpenAI is a US AI company, founded in 2015, that develops the GPT family of large language models and reasoning models and the ChatGPT assistant, whose public launch in November 2022 brought generative AI to a mass audience. It also offers image, audio and video models, the Codex coding agent and a developer API.

Beginner · Technical writers · Technical marketers · Technical project managers · Developers · AI brand

Anthropic (Claude)

Anthropic is a US AI safety and research company, founded in 2021, that develops the Claude family of large language models — offered in tiers from fast and economical to most capable — together with the Claude apps, a developer platform and the Claude Code coding agent. Anthropic introduced the Model Context Protocol (MCP) as an open standard in 2024.

Beginner · Technical writers · Technical marketers · Technical project managers · Developers · AI brand

Google DeepMind (Gemini)

Google DeepMind is Google’s AI research lab and the developer of Gemini, a family of natively multimodal models used in the Gemini app, Google Search, Workspace and Android, and offered to developers through Google AI Studio and Vertex AI. Google also publishes the open-weight Gemma models; its researchers introduced the transformer architecture in 2017.

Beginner · Technical writers · Technical marketers · Technical project managers · Developers · AI brand

Meta (Llama)

Meta, the company behind Facebook, Instagram and WhatsApp, develops the Llama family of open-weight large language models, released under Meta’s own community license, and the Meta AI assistant built into its apps.

Beginner · Intermediate · Technical marketers · Technical project managers · Developers · AI brand

Microsoft (Copilot, Azure AI)

Microsoft brings generative AI into its products under the Copilot brand — in Windows, Microsoft 365 apps such as Word, Excel, Outlook and Teams, and, through its subsidiary GitHub, in GitHub Copilot for developers. Azure AI Foundry gives businesses access to models from OpenAI, Microsoft’s own Phi models and other providers.

Beginner · Technical writers · Technical marketers · Technical project managers · Developers · AI brand

Mistral AI

Mistral AI is a French AI company, founded in 2023, that develops both open-weight and commercial language models and the Le Chat assistant, and is often cited as Europe’s leading model developer.

Intermediate · Technical marketers · Technical project managers · Developers · AI brand

xAI (Grok)

xAI is a US AI company founded by Elon Musk in 2023 that develops the Grok models and assistant, integrated into the social network X and offered through an API.

Intermediate · Technical marketers · Technical project managers · AI brand

DeepSeek

DeepSeek is a Chinese AI company based in Hangzhou that publishes open-weight large language models. Its reasoning model DeepSeek-R1, released in January 2025 under the MIT license, showed that competitive reasoning performance could be reached at much lower reported training cost.

Intermediate · Technical marketers · Technical project managers · Developers · AI brand

Alibaba Cloud (Qwen)

Alibaba Cloud develops Qwen (Tongyi Qianwen), a family of language and multimodal models released in many sizes, a large share of them open-weight under permissive licenses, and offered as a service on Alibaba Cloud.

Intermediate · Technical project managers · Developers · AI brand

Amazon (Bedrock, Nova)

Amazon offers generative AI through Amazon Web Services: Amazon Bedrock, a managed service that gives access to models from several providers — including Anthropic, Meta, Mistral AI and Amazon’s own Nova models — and the Amazon Q assistants for business and developers.

Intermediate · Technical project managers · Developers · AI brand

NVIDIA

NVIDIA is the US company whose graphics processing units (GPUs) and CUDA software platform provide most of the computing power used to train and run AI models. It also publishes its own models, such as the Nemotron family, and inference software for deploying models.

Intermediate · Technical marketers · Technical project managers · Developers · AI brand

Cohere

Cohere is a Canadian AI company, headquartered in Toronto, that focuses on enterprise use: its Command language models, Embed models for semantic search and Rerank models for sorting search results by relevance, with private-deployment options.

Intermediate · Technical project managers · Developers · AI brand

Perplexity

Perplexity is a US company that operates an AI answer engine: it searches the web, reads the results and answers questions in natural language with numbered citations to its sources.

Beginner · Technical writers · Technical marketers · AI brand

Hugging Face

Hugging Face is a company and platform that hosts the largest public collection of open AI models and datasets, maintains the widely used Transformers library, and lets people publish demos as “Spaces.”

Intermediate · Developers · AI brand

Apple Intelligence

Apple Intelligence is Apple’s set of generative AI features built into iPhone, iPad and Mac — such as writing tools, summaries and image generation — that run on the device where possible and otherwise on Apple’s Private Cloud Compute, with optional handoff to ChatGPT.

Beginner · Technical writers · Technical marketers · AI brand

IBM (watsonx, Granite)

IBM offers enterprise AI through its watsonx platform — for building AI applications, managing data and governing models — and publishes the Granite family of open-weight models under the Apache 2.0 license.

Intermediate · Technical project managers · Developers · AI brand

Baidu (ERNIE)

Baidu, China’s largest search company, develops the ERNIE family of large language models and the ERNIE assistant, and integrates generative AI into Baidu Search and its cloud services.

Intermediate · Technical marketers · Technical project managers · AI brand

Midjourney

Midjourney is an independent US research lab whose image-generation model creates images from text prompts, used through its website and Discord and known for its distinctive, highly stylized output.

Beginner · Technical marketers · AI brand

Adobe Firefly

Adobe Firefly is Adobe’s family of generative models for images, vector graphics, video and audio, built into Photoshop, Illustrator, Express and other Adobe apps. Adobe states that Firefly is trained on licensed content such as Adobe Stock and on public-domain material, and attaches Content Credentials to generated assets.

Beginner · Technical writers · Technical marketers · AI brand

Model selection

Model selection is the structured choice of an AI model for a specific use case, weighing task quality, cost per token, speed, context size, openness, hosting and data terms, and regulatory fit — ideally based on tests with your own content rather than on public leaderboards.

Intermediate · Expert · Technical project managers · Developers

Prompt

A prompt is the input given to a generative AI model — an instruction, a question, examples, reference material or a combination — that the model uses as the starting point for its output.

Beginner · Technical writers · Technical marketers · Technical project managers

Prompt engineering

Prompt engineering is the practice of designing, testing and refining prompts so that a generative AI model produces the desired output reliably — through clear instructions, relevant context, examples, output formats and systematic comparison of variants.

Beginner · Intermediate · Technical writers · Technical marketers · Technical project managers

System prompt

A system prompt is the standing instruction that an application gives a language model before any user input — defining its role, tone, rules, knowledge boundaries and output format for the whole conversation. Users usually don’t see it.

Intermediate · Technical writers · Technical project managers · Developers

Few-shot prompting

Few-shot prompting is the technique of including a small number of input–output examples in a prompt so that the model infers the desired pattern, style or format from them; with no examples it is called zero-shot prompting.

Intermediate · Technical writers · Technical marketers · Developers

Chain-of-thought prompting

Chain-of-thought prompting asks a language model to work through a problem in explicit intermediate steps before giving its answer, which improves results on tasks that need reasoning, such as calculations, comparisons or multi-step checks.

Intermediate · Technical writers · Technical project managers · Developers

Temperature (sampling)

Temperature is a setting that controls how much randomness a language model uses when choosing each next token: low values make output more focused and repeatable, high values make it more varied and creative.

Intermediate · Technical writers · Developers

Structured output

Structured output is model output that follows a predefined, machine-readable format — typically JSON matching a schema — so that other software can process it reliably. Many AI APIs can enforce a schema during generation.

Intermediate · Expert · Technical writers · Developers

Context engineering

Context engineering is the discipline of deciding what information enters a language model’s context window for each step of a task — instructions, retrieved documents, examples, conversation history, tool results and memory — and in what form, so the model has exactly what it needs and nothing that misleads it.

Expert · Technical writers · Technical project managers · Developers

Prompt library

A prompt library is a shared, maintained collection of tested prompts for recurring tasks, with notes on purpose, inputs, model and known limits, so that a team gets consistent AI results and doesn’t reinvent prompts individually.

Beginner · Technical writers · Technical marketers · Technical project managers

Embedding

An embedding is a list of numbers — a vector — that represents the meaning of a piece of text, an image or other data, produced by an embedding model so that items with similar meaning end up close to each other in vector space.

Intermediate · Technical writers · Developers

Vector database

A vector database stores embeddings together with their source content and metadata and finds the entries most similar to a query vector quickly, which makes it the usual retrieval layer of semantic search and RAG systems.

Intermediate · Technical project managers · Developers

Retrieval-augmented generation (RAG)

Retrieval-augmented generation (RAG) is an architecture in which an AI system first retrieves relevant passages from a trusted content source and then has a language model generate its answer from those passages, often with citations — so answers reflect current, approved content rather than only the model’s training data.

Intermediate · Technical writers · Technical marketers · Technical project managers · Developers

Chunking

Chunking is the splitting of documents into smaller passages — chunks — before they are embedded and indexed for retrieval, so that a RAG system can retrieve just the relevant part of a document.

Intermediate · Technical writers · Developers

Grounding

Grounding is the practice of tying a generative AI model’s output to specific, verifiable sources — retrieved documents, databases, search results or tool outputs — so that answers can be traced and checked instead of relying on the model’s internal knowledge.

Intermediate · Technical writers · Technical marketers · Technical project managers · Developers

Knowledge graph

German: Wissensgraph

A knowledge graph is a network of entities — such as products, components, functions and documents — and the typed relationships between them, stored in a machine-readable form such as RDF, so that people and AI systems can query and reason over connected knowledge.

Intermediate · Expert · Technical writers · Technical project managers · Developers

AI-ready content

AI-ready content is content that AI systems can find, interpret and reuse correctly: modular, consistently structured, described with metadata, written with controlled terminology, kept current and clearly owned, and published in formats machines can parse.

Beginner · Intermediate · Technical writers · Technical marketers · Technical project managers

llms.txt

llms.txt is a proposed convention for a plain-markdown file at the root of a website (/llms.txt) that gives AI assistants a concise overview of the site and links to its most useful pages, so they can find and use the content without parsing complex HTML.

Intermediate · Technical writers · Technical marketers · Developers

Generative engine optimization (GEO)

Generative engine optimization (GEO) is the practice of making content more likely to be found, correctly understood and cited by AI answer engines and AI search features — through clear structure, authoritative facts, sources, structured data and unambiguous terminology.

Intermediate · Technical writers · Technical marketers

AI API

An AI API is a programming interface through which software sends inputs — prompts, documents, images — to an AI model hosted by a provider and receives its output, usually billed per token and governed by the provider’s usage and data terms.

Intermediate · Technical project managers · Developers

Fine-tuning

Fine-tuning is the further training of a pretrained model on a smaller, task-specific dataset so that it adopts a particular behavior, format, style or domain vocabulary more reliably than prompting alone achieves.

Expert · Technical project managers · Developers

Tool use (function calling)

Tool use, also called function calling, is the ability of a language model to request that the application run a defined function — such as a search, a database query or an API call — by returning a structured call with arguments, and then to use the result in its answer.

Expert · Developers

Model Context Protocol (MCP)

The Model Context Protocol (MCP) is an open standard, introduced by Anthropic in 2024 and since adopted widely across the industry, that defines how AI applications connect to external tools and data sources through MCP servers, so one integration works with many AI clients.

Expert · Technical writers · Technical project managers · Developers

AI agent

An AI agent is a system in which a language model pursues a goal over several steps on its own — planning, calling tools, observing the results and deciding what to do next — until the task is done or it needs human input.

Intermediate · Technical writers · Technical marketers · Technical project managers · Developers

Agentic workflow

An agentic workflow is a business process in which one or more AI agents carry out defined steps — alone or in sequence with other agents and humans — with explicit handoffs, checkpoints and approvals built into the process design.

Expert · Technical writers · Technical project managers · Developers

AI coding assistant

An AI coding assistant is a tool that uses language models to suggest, write, explain, test and refactor code inside an editor or terminal — from inline completion to agents that change several files and run tests.

Intermediate · Technical writers · Technical project managers · Developers

Evaluation (evals)

Evaluation — “evals” for short — is the systematic measurement of how well an AI system performs a task, using a fixed set of test inputs with expected outputs or scoring criteria, scored by rules, by people or by another model, and repeated whenever the model, prompt or content changes.

Expert · Technical writers · Technical project managers · Developers

Guardrails (AI)

Guardrails are the technical and procedural controls that keep an AI system’s inputs and outputs within defined limits — such as topic restrictions, content filters, checks for personal data, fact or format validation and escalation to humans.

Intermediate · Technical writers · Technical marketers · Technical project managers · Developers

Prompt injection

Prompt injection is an attack in which text supplied to a language model — typed by a user or hidden in a web page, email or document the model reads — contains instructions that override or subvert the application’s intended instructions.

Expert · Technical project managers · Developers

Bias (AI)

Bias in AI is a systematic skew in a model’s outputs — favoring or disadvantaging certain groups, viewpoints, languages or cases — that usually stems from imbalances in its training data, its design or the way it is used.

Beginner · Technical writers · Technical marketers · Technical project managers

Human in the loop

German: Mensch in der Schleife

Human in the loop is a design principle in which a qualified person reviews, corrects or approves an AI system’s output or action at defined points before it takes effect, and remains accountable for the result.

Beginner · Technical writers · Technical marketers · Technical project managers · Developers

AI content disclosure

AI content disclosure is the practice — and in some cases the legal duty — of telling audiences that text, images, audio or video were generated or substantially altered by AI, through labels, notices, metadata or provenance credentials.

Beginner · Technical writers · Technical marketers

Responsible AI

Responsible AI is the set of principles and practices for developing and using AI in ways that are safe, fair, transparent, privacy-preserving, secure and accountable — translated into concrete policies, controls and documentation throughout an AI system’s life cycle.

Beginner · Technical writers · Technical marketers · Technical project managers

AI usage policy

An AI usage policy is an organization’s internal rule set for using AI tools — which tools are approved, which data may be entered, which tasks require review or disclosure, and who is accountable — so that employees can use AI productively without creating legal, security or quality risks.

Beginner · Technical writers · Technical marketers · Technical project managers

Model card

A model card is a short, structured document published with an AI model that describes its intended uses, limitations, training data, evaluation results, risks and ethical considerations, so that users can judge whether the model fits their purpose.

Intermediate · Technical writers · Technical project managers · Developers

Accountability (AI)

German: Rechenschaftspflicht

In AI governance, accountability is the assignment of a named person, role or organization that answers for a decision, output or system — including the duty to explain it, correct it and bear its consequences. It presupposes intent, authority and the ability to act, so it rests with people and organizations rather than with a model or an agent, however autonomous the system appears. <cite index="7-1,7-4">The OECD AI Principles put it as a duty of organisations and individuals that develop, deploy or operate AI systems to be held accountable for their proper functioning, based on their roles, the context and their ability to act.</cite> In practice, accountability becomes visible only when it is documented: a role, a review step and a record that shows who decided what.

Intermediate · Technical writers · Technical marketers · Technical project managers · Developers

Accountability gap (AI)

German: Verantwortungslücke

In AI governance, an accountability gap is a situation in which an AI system contributes to an outcome while no identified person or organization is clearly answerable for it. It typically arises when AI output is accepted without a named reviewer, when the tool chain records no attribution, or when roles were defined for a workflow that no longer matches how the system is actually used. The term is an analytic label from the ethics and policy debate rather than a definition fixed by a standard or a law; frameworks such as the OECD AI Principles and the NIST AI Risk Management Framework instead state the accountability expectation the gap violates.

Intermediate · Technical writers · Technical project managers · Developers

Data privacy in AI tools

Data privacy in AI tools concerns what happens to the information users enter into or connect to AI services — whether it is stored, used for model training, shared, or processed outside the region — and the legal and contractual controls, such as the GDPR and enterprise terms, that govern it.

Beginner · Technical writers · Technical marketers · Technical project managers · Developers