Glossary

Written By Odassia Wine

AI Cloud

AI Cloud is a specialization of cloud computing. It allows hosting and distributing AI models, which are more powerful and require much heavier infrastructure.

AI Model

An AI model is the result of AI training: the “brain” that has learned from data and is then able to analyze, predict, or create content. Think of it as a trained machine:

  • Data = the raw material

  • Algorithm = the recipe that explains how to learn

  • Model = the final product, capable of generating answers or content

Example: GPT-4, the LLM powering ChatGPT.

AI Overview (AIO)

AI Overview (formerly Google SGE) is a recent Google Search feature that uses generative AI to display a summary answer directly at the top of search results. Instead of clicking multiple links, the user gets an immediate summary written by AI, sometimes with cited sources.
Example: searching for “best surf spots in Portugal” might generate a box listing and describing the locations without opening the suggested sites.

AI Product

An AI product is a concrete application built from one or several AI models. Unlike the model (the raw technology), the product is designed for direct use by people or businesses.
Example: ChatGPT is an AI product built on the GPT-4 model.

AI Search Engine

An AI search engine (e.g., Perplexity, Bing Copilot, Google AI Overview) combines web indexing with generative AI to provide contextualized answers, often with cited sources.
Unlike standalone AI models (e.g., GPT, Gemini), which rely on training data, AI search engines actively pull, rank, and cite information from the web.

Algorithm

An algorithm is a sequence of logical instructions a machine follows to solve a problem or complete a task, based on input data. Algorithms are what fuel AI.

Aleph Alpha

A German startup specializing in generative AI, founded in 2019. It develops LLMs and positions itself as a European alternative to OpenAI, Anthropic, or Google. Aleph Alpha emphasizes technological sovereignty, transparency, and interpretability, with applications for businesses and public institutions.

Autonomous Agent

An autonomous agent is a more advanced form of AI. Instead of just responding, it can make decisions, plan actions, and execute tasks on its own.
Example: “Find me the three best flights to New York and book the cheapest one.” The agent searches, compares, chooses, and acts without step-by-step guidance.

Azure AI

The suite of AI services within Microsoft Azure. It allows hosting, training, and deploying AI models. Thanks to Azure AI, tools like ChatGPT can be securely and scalably hosted and distributed in businesses.

Chatbot

A chatbot or conversational agent is a program designed to answer specific questions or perform simple tasks. It usually works reactively, responding to what it is asked.
Example: a customer support bot that provides opening hours or tracks orders.

ChatGPT

ChatGPT is a conversational agent (chatbot) developed by OpenAI, based on LLMs such as GPT-4. It can understand natural language questions and generate text-based answers (explanations, summaries, ideas, etc.).
It’s one of the most well-known applications of generative AI, since it allows anyone to interact directly with a powerful AI model without technical expertise.
Examples of use cases: drafting emails, creating code, generating marketing content, answering questions, or explaining complex concepts.

Claude

Claude is an AI developed by Anthropic, an American company focused on AI. It’s an advanced chatbot similar to ChatGPT, capable of understanding and generating text, analyzing long documents, and assisting with writing or reasoning. It is designed with strong emphasis on safety and ethical controls.

Cloud

Traditional cloud computing provides IT resources (servers, storage, databases) accessible remotely. It saves users and companies from having to manage their own infrastructure.

Cognitive Resilience to AI in Society

The ability to maintain critical thinking, adaptability, and judgment in a world where generative AI—and, in the future, quantum AI—are deeply transforming how we live, work, and make decisions. It means being able to coexist with AI while understanding, questioning, and correcting its outputs.

Copilot

Copilot is Microsoft’s AI assistant, built directly into its tools (Word, Excel, Outlook, Teams, PowerPoint, etc.). It is powered by OpenAI models (like GPT-4), hosted on Microsoft Azure, and helps with writing, analyzing, summarizing, creating presentations, or automating tasks.

Unlike ChatGPT or Claude, Microsoft Copilot is not a single chatbot but a brand of integrated AI assistants across products:

  • Microsoft 365 Copilot → embedded in Word, Excel, Outlook, PowerPoint, Teams, etc.

  • Windows Copilot → built into the Windows operating system.

  • GitHub Copilot → a coding assistant for developers.

  • Copilot in Bing → a chatbot within Bing (formerly Bing Chat), powered by OpenAI’s GPT-4.

DeepMind

DeepMind is a British AI company, founded in 2010 and acquired by Google in 2014 (now Google DeepMind). It is known for major AI breakthroughs such as AlphaGo and AlphaFold, as well as fundamental research that contributed to the development of LLMs.

Generative AI

Generative Artificial Intelligence is a branch of AI capable of producing original content (texts, images, sounds, videos, code, etc.) from training data.
Unlike traditional AI that is limited to recognizing, analyzing, or predicting information, generative AI creates new outputs.

Generative AI Cloud

Generative AI Cloud goes further, as it hosts and distributes so-called generative models like LLMs, whose computing power would be far too heavy to run internally. Without generative AI cloud, we wouldn’t be able to use ChatGPT on our computers.
Examples: Microsoft Azure, Google Vertex AI.

Generative Engine Optimization (GEO)

A new SEO approach adapted to AI search engines (ChatGPT, Gemini, Google AI Overview). The goal isn’t traffic but rather to be cited by AIs and convert that visibility into trust and conversions.

Graphic Processing Unit (GPU)

A GPU (Graphics Processing Unit) is an electronic chip originally designed to accelerate graphics and animations. Today, it is essential for AI because it can process millions of calculations in parallel (much faster than a CPU), speed up model training (like LLMs), and reduce the time needed to analyze or generate content.
Without GPUs, training or running modern AI would take weeks—or be impossible.

GPT (Generative Pre-trained Transformer)

A Large Language Model (LLM) developed by OpenAI, capable of holding natural language conversations. The most advanced version, GPT-4o (released in 2024), is optimized to be faster, more relevant, and multimodal (text, voice, and images).

Hugging Face

An American company founded in 2016, now a global reference in open-source AI. It provides a platform and libraries (like Transformers) to share, train, and deploy language, vision, or multimodal models. Hugging Face is now a central hub for AI research and innovation.

Hyperscalers

Hyperscalers are the largest tech companies capable of delivering IT resources on a global scale thanks to massive cloud infrastructures. They provide storage, computing, AI, and networking services to millions of users and businesses.
The main hyperscalers are: Amazon Web Services (AWS), Microsoft Azure, and Google Cloud (GCP).

Impressions

In digital marketing, an impression is the number of times content (a link, ad, post, etc.) is displayed to a user, whether they click it or not.
Example: if a site appears 1,000 times in Google results, that equals 1,000 impressions, even if no one clicks.
Impressions measure visibility, not engagement.

Large Language Model (LLM)

An LLM is a generative AI model specialized in language. It is trained on massive amounts of text to learn how to understand and generate natural language.
In practice, an LLM can answer questions, translate, write, summarize, or hold conversations fluently.
Examples: GPT (OpenAI), Gemini (Google), Claude (Anthropic).

Machine Learning (ML)

Machine Learning (ML) is a branch of AI that involves training algorithms to automatically learn from data, instead of just following programmed rules.
Example: an ML model can learn to recognize spam emails by analyzing thousands of examples.

Machine Learning Operations (MLOps)

MLOps combines Machine Learning and DevOps. It’s the set of methods, tools, and best practices for deploying, monitoring, and maintaining ML models in production reliably and at scale. Thanks to MLOps, companies can ensure their models remain effective, even after years of use.

Mistral AI

A French startup founded in 2023, specializing in open-source LLMs (Large Language Models). It develops European alternatives to OpenAI or Google, focusing on transparency, performance, and digital sovereignty.

Multimodal AI

AI capable of understanding and generating multiple types of content (text, image, sound, video, code, etc.) and linking them together. GPT-4 and Gemini are multimodal AI systems.

Nurturing

A set of actions designed to guide a prospect along the buying journey by sending relevant, educational, and progressive content.
The goal isn’t immediate sales but to build trust and prepare for conversion.

Open Source

A software development model where the code is freely accessible, modifiable, and shareable by anyone. It fosters transparency, collaboration, and collective innovation. Of course, there are methods to protect the code against hacking or malicious use.

OpenAI

OpenAI is an American company founded in 2015 by Sam Altman, Elon Musk (who later left), and other entrepreneurs. It is behind well-known AI models and products such as ChatGPT (text), DALL·E (images), and Whisper (voice), also integrated into Microsoft solutions (Copilot).
Since 2019, Microsoft has been OpenAI’s main investor and technology partner, with OpenAI models primarily hosted and distributed through the Azure cloud.

Perplexity

Perplexity is an AI-powered search engine. Unlike Google, which lists links, Perplexity provides a direct natural language answer with reliable sources. It uses RAG technology to search in real time and generate clear, sourced answers.

Quantum AI

Quantum AI refers to using quantum computing to boost AI performance. While traditional AI runs on classical computers, quantum AI uses the principles of quantum physics to process massive data faster and solve complex problems.
The goal: enable AI to learn and compute much faster than today’s machines.
It’s still in development but could revolutionize fields such as healthcare, finance, and scientific research.

Re-skilling

The process of training someone in new skills so they can transition to a new job or role.

Retrieval-Augmented Generation (RAG)

RAG (Retrieval-Augmented Generation) is a technology used by some AI systems: before answering, the AI retrieves information from reliable external sources (retrieval), then generates a natural language response (generation).
This allows AI to:

  • Be more up to date

  • Cite sources

  • Provide more accurate and contextual answers

In short, RAG connects AI to live knowledge bases to avoid outdated or fabricated answers.

Retrieval Engine Optimization (REO)

Just as SEO helps you appear on Google, REO ensures you are present in answers generated by retrieval-based AIs (Perplexity, Bing Copilot, You.com), which pull live data from the web and cite their sources. It’s part of GEO (Generative Engine Optimization).

Retargeting

A marketing technique targeting people who have already interacted with a brand (visited a site, added to cart, clicked content, etc.) to encourage them to return and convert.
The goal isn’t necessarily to build a relationship but rather to drive short-term actions.

Search Engine Optimization (SEO)

SEO (Search Engine Optimization) refers to all techniques used to improve a website’s visibility in search engine results like Google.
Concretely, SEO involves:

  • Optimizing content (text, images, videos)

  • Improving site technicals (speed, structure, HTML tags)

  • Boosting site authority via backlinks and citations

The goal is to rank as high as possible on SERPs, generating more traffic without paid ads.

SEO Off-Site

All actions taken outside the website to strengthen credibility and authority: backlinks, social media mentions, citations in articles, reputation building.
Goal: show SERPs and AIs that the site is reliable, recognized, and trustworthy.

SEO On-Site

All optimizations made directly on a website to improve visibility: content quality and structure, keyword choices, tags (titles, meta descriptions), internal linking, etc.
Goal: make the site clearer and more attractive to SERPs and AI.

SERP

The SERP (Search Engine Results Page) is the search results page from engines like Google or Bing, shown after a user enters a query. It can include:

  • Classic blue links

  • Paid ads (SEA)

  • Rich elements like images, maps, or AI Overviews

Transformer

The Transformer is an AI technology invented by Google in 2017 that revolutionized natural language processing. Based on an architecture that understands text context by analyzing relationships between words, it powers LLMs like GPT, Claude, or Gemini.

Vertex AI

Google Cloud’s AI platform, enabling companies to develop, train, host, and deploy AI models, including generative ones (like Gemini). Vertex AI offers ready-to-use infrastructure so businesses can harness AI power without managing heavy technical resources themselves.

Odassia Wine
Previous

Fundamentals of Generative AI