Artificial intelligence glossary

AGI (artificial general intelligence)

A hypothetical AI able to perform any intellectual task as well as a person, or better. There is no agreed definition, nor agreement on whether or when it will arrive.

In depth: Is artificial intelligence dangerous? Could it really turn against us?

In the news: Microsoft’s Mustafa Suleyman says its AI outperforms doctors on hard cases as he chases ‘medical superintelligence’ · Google's Gemini 4 Argon is closer to the AI frontier than any Google model in months · Machine learning pioneer Michael I. Jordan dismantles AI doom rhetoric as "grand moralizing performance"

AI Act

The European Union's regulation on artificial intelligence, in force since 2024 and applied in stages. It classifies systems by risk: some uses are banned, others are subject to transparency and oversight obligations.

In depth: Artificial intelligence laws around the world

In the news: AI Act, in vigore il decreto attuativo italiano · AI Act : la première norme harmonisée ne couvre que le système qualité · AI Act, quando un incidente AI è grave e chi deve segnalarlo

AI agent

An AI system that does not just answer but carries out a task over several steps: it looks up information, uses software, fills in forms, writes and tests code, with more or less human supervision.

In the news: Here are the top AI agents that can live in your text messages · OpenAI's isolated AI agents found ways to talk to each other through CPU cache · Three-week on-site audit at OpenAI reveals major gaps in AI safety monitoring

API

The interface that lets one program use another's service. Through APIs, companies build AI models into their own apps and websites, paying according to use.

In depth: Is artificial intelligence free or paid?

In the news: Build more responsive apps with the Decisions API · Jev is Vercel’s fastest-adopted AI model ever · Gemini 4 Argon is now available worldwide as Google's most capable frontier model

Artificial intelligence (AI)

The set of techniques that let a computer carry out tasks that usually require human intelligence: understanding text, recognising images, translating, predicting, deciding.

In depth: What is artificial intelligence and how does it work? A simple explanation

Benchmark

A standard test, the same for everyone, used to compare AI models. Useful as a guide, but a high score on a test does not guarantee good results in real use.

In depth: Is artificial intelligence reliable? When it can get things wrong

In the news: AI agents have bad taste in long-term decisions and a new benchmark proves it · Deepseek V4.1 Flash 1017 beats larger rivals in agentic tasks while once again massively cutting per-task costs · Google's Gemini 4 Argon is closer to the AI frontier than any Google model in months

Chatbot

A program you talk to in natural language, by typing or speaking. The latest ones are built on an LLM and can also search the web, read documents and use other tools.

In depth: What is ChatGPT, how does it work and how do you use it?

In the news: Disney Asked Character.AI to Go Bankrupt for a Get-Out-of-Lawsuits Card · Black Box: The Chatbots | What I Saw at the Double Six Ranch | Ep 6 - podcast · Maybe Chatbots Shouldn’t Get to Say ‘I’ or ‘Me’

Context window

The amount of text a model can keep in mind within a single conversation, measured in tokens. Beyond that limit the oldest parts are forgotten or summarised.

In depth: How to use artificial intelligence: a guide for complete beginners

In the news: Kimi K3.1 Pro, Moonshot AI rilancia sui modelli open-weight: prestazioni vicine ai migliori sistemi proprietari

Data centre

The buildings that house the computers on which AI models are trained and run. They use a great deal of electricity and, for cooling, often water as well.

In depth: Why does artificial intelligence use so much water and energy?

In the news: The datacentre next door: on the frontline of tech’s latest David v Goliath battle · Sharp drops for AI stocks overwhelm markets worldwide as oil prices jump · How tech became the villain of the US midterms

Deep agents

AI agents built for long, complex tasks that a simple agent cannot complete. They rely on four elements: very detailed instructions, a tool for planning the work in steps, the ability to hand parts of the task to sub-agents, and a file system for saving notes and results. The term was proposed in 2025 by LangChain, which also uses it for its Deep Agents library; the examples it cites include Claude Code, Deep Research and Manus.

Deep learning

A type of machine learning that uses neural networks with many layers. It made image and speech recognition and today's language models possible.

In depth: What is artificial intelligence and how does it work? A simple explanation

Deepfake

A fake but realistic video, image or audio clip made with AI, showing a person saying or doing things that never happened.

In depth: How to spot audio, photos and videos made with artificial intelligence

In the news: Opposition deepfake video of Spanish PM’s wife sparks outrage before election · Nobody is safe from celebrity AI deepfakes · Studie: KI-Videos untergraben Vertrauen in Bewegtbild massiv

Diffusion model

The technique behind many image and video generators: the model learns to remove noise from a blurred image, one step at a time, until the requested content emerges.

In depth: How to spot audio, photos and videos made with artificial intelligence

Embedding

The conversion of a text, or an image, into a series of numbers that represents its meaning. It makes it possible to find content that is similar in sense, not just in wording.

In the news: OlmoEarth v1.3: State-of-the-art embeddings for the whole planet · Lokale KI auf dem Mac: Ollama nutzt MLX als Standard

Fine-tuning

Additional training of an existing model on specific data, to adapt it to a task or a field: for example legal language or a company's style.

In depth: What is artificial intelligence and how does it work? A simple explanation

Generative AI

Artificial intelligence that creates new content, such as text, images, video, music or code, from a request. ChatGPT, Claude, Gemini and image generators are examples.

In depth: What is artificial intelligence and how does it work? A simple explanation

In the news: How Itaú is building generative AI for its entire customer base · Roundtables: A Conversation With the Creator of AI-Designed Viruses · Character.AI files for bankruptcy

GPU

The graphics processor, originally built for video games, that turned out to be ideal for AI calculations. GPUs are the most sought-after resource in the industry.

In depth: Why does artificial intelligence use so much water and energy?

In the news: AI’s hidden storage bill: What inference leaves behind · Black Forest Labs' open-weights image model FLUX 3 Image Klein runs on consumer GPUs with just 8 GB of VRAM · Microsoft’s new AI chip makes it less reliant on Nvidia’s GPUs

Hallucination

When an AI states something false or made up with full confidence: a quote that does not exist, a wrong date, a source that never existed. It is why answers need checking.

In depth: Is artificial intelligence reliable? When it can get things wrong

In the news: heise+ | Typed Domain Grounding als Designprinzip für LLM-taugliche DSLs · Lösen wir mit KI echte Probleme oder erfinden wir neue? · Atti giudiziari creati con l’AI: sanzioni e nuove regole negli USA

Inference

The moment an already trained model is used to answer a request. Every question to a chatbot is an inference: it uses far less than training, but it happens billions of times.

In depth: Why does artificial intelligence use so much water and energy?

In the news: AI’s hidden storage bill: What inference leaves behind · Microsoft plans to scale its own Maia AI chips to gigawatt levels by 2027 · Microsoft’s new AI chip makes it less reliant on Nvidia’s GPUs

LLM (large language model)

An AI model trained on huge amounts of text to predict the next word. That ability is what produces the answers, summaries and translations of chatbots.

In depth: What is artificial intelligence and how does it work? A simple explanation

In the news: Discord now formats Markdown previews right · Anthropic's Andrej Karpathy says "the external world has seen the preview" of what AI can actually do · The maker of non-text AI model Jev valued at $7.5B just weeks after launch

Machine learning

The method by which a program learns from examples instead of following rules written one by one. It is the basis of almost all of today's artificial intelligence.

In depth: What is artificial intelligence and how does it work? A simple explanation

In the news: Microsoft has a new machine learning tool to improve weather forecasts · Machine learning pioneer Michael I. Jordan dismantles AI doom rhetoric as "grand moralizing performance" · Il Mit rivoluziona le previsioni meteorologiche con il modello MicroDuet e il machine learning

Multimodal model

A model that can work with several kinds of content together: text, images, audio, video. It can, for example, describe a photo or answer by voice.

In depth: What is artificial intelligence and how does it work? A simple explanation

In the news: Nvidia's Nemotron 3.5 Super VL is an open model built to run enterprise AI agents nonstop · Welcome Muse Spark Open: open-weight frontier multimodal mixture-of-experts model from Meta · Meta e Apple tornano a lavorare assieme: merito dell'AI

Neural network

A mathematical model loosely inspired by the neurons of the brain: many connected nodes whose connections are adjusted during training until they produce the desired result.

In depth: What is artificial intelligence and how does it work? A simple explanation

In the news: Is AI interpretability close to a breakthrough? A new essay has researchers divided · Ben Affleck is an AI nerd, and the internet is impressed · Schneller als die Weltmeister: Diese KI schlägt die besten Drohnenpiloten

Open source and open weights

A model has open weights when anyone can download it and run it on their own computers. Open source, in the full sense, means the code and the terms for modifying and reusing it are public too.

In the news: Anthropic's Claude now writes code and proves it safe automatically · Hugging Face’s New AI Biology Model Lets You Chat With DNA · Anthropic launches free AI security scans for open-source projects

Parameters

The internal numbers of a model, set during training, that determine how it behaves. The largest models have billions; more parameters does not automatically mean better answers.

In depth: What is artificial intelligence and how does it work? A simple explanation

In the news: Muse Spark Open is Meta's answer to Qwen, GLM, and Kimi in the open-weight race · Nvidia's Nemotron 3.5 Super VL is an open model built to run enterprise AI agents nonstop · 30 Millionen Parameter: Lokaler KI-Sprachmodell für den ESP32

Prompt

The request you give an AI: a question, an instruction, a text to work on. The clearer the prompt and the more context it carries, the better the answer usually is.

In depth: How to use artificial intelligence: a guide for complete beginners

In the news: AI models hide mistakes in their own work reports, but one sentence in your prompt can help · Google wants Gemini 4 Argon to be the only AI model robots need · The eyes have it: Reform politician’s ‘smart glasses’ prompt calls for ban in Westminster

RAG

A technique that has the model look up information in documents or databases before answering, so the answer rests on specific, up-to-date sources. RAG stands for retrieval-augmented generation.

In depth: Is artificial intelligence reliable? When it can get things wrong

Token

The small piece of text, a word or part of one, that a language model reads and writes with. Limits and prices of AI services are often measured in tokens.

In depth: Is artificial intelligence free or paid?

In the news: Muse Spark Open is Meta's answer to Qwen, GLM, and Kimi in the open-weight race · Anthropic says it cut Claude reasoning token theft by up to 99% in fight against industrial-scale distillation · Microsoft plans to scale its own Maia AI chips to gigawatt levels by 2027

Training

The phase in which a model learns, by analysing large amounts of data and gradually correcting its own errors. It takes a great deal of computing power and is the most expensive part of building an AI.

In depth: What is artificial intelligence and how does it work? A simple explanation

In the news: OpenAI now trains its frontier AI models with no internet access · Meta delays new business terms after publisher pushback on AI scraping · Securing our frontier model training environment