AI Dictionary

What does artificial general intelligence (AGI) mean?

Artificial general intelligence (AGI) means an advanced AI system capable of learning and performing any cognitive task as well as or better than a human. Unlike today’s narrow AI systems, AGI would be able to apply what it has learned flexibly across situations and use knowledge creatively in new contexts. AGI remains theoretical, and it carries both great possibilities and significant ethical and safety risks.

What does the AI Act mean?

The AI Act is the European Union’s regulatory framework setting common requirements for developing, deploying, and using AI systems in the EU. It is the world’s first comprehensive AI legislation, designed to balance the promotion of innovation with the protection of fundamental rights.

The AI Act classifies AI applications into categories by risk level, with different degrees of requirement applying to each. Higher requirements on transparency, documentation, human oversight, and risk management are set for high-risk systems, such as AI in critical infrastructure, healthcare, and recruitment. Certain AI applications that violate EU values or human rights, such as social scoring, are prohibited entirely.

The regulation sets standards for AI reliability, avoiding bias, transparency, and accountability, and imposes significant penalties for breaches. The EU AI Act entered into force in 2024 and applies in stages over the following years.

What does AI ethics mean?

AI ethics is the field that studies the moral questions involved in developing, deploying, and using AI. It considers the ethical principles for designing, building, and using AI systems, and how to ensure AI works fairly, transparently, and for the good of humanity. Central questions include privacy, non-discrimination, accountability, and safety.

What does AI transparency mean?

AI transparency means how well an AI system’s operation, decision-making, and effects can be understood and explained. Transparency matters for building trust and ensuring accountability. It makes it possible to identify and correct a system’s errors and helps assess whether the AI is working fairly and as designed.

What does algorithm mean?

An algorithm is a precisely defined series of rules or instructions by which a computer performs a task. In AI and machine learning, algorithms process data, recognise patterns, and make decisions. They can be simple if-then rules or complex neural networks.

What is Anthropic?

Anthropic is a US AI research company developing safe and reliable AI. It is known in particular for its Claude language models, which compete with those of OpenAI and Google DeepMind. Anthropic emphasises AI ethics and model transparency.

• www.anthropic.com

• www.claude.ai

What does API mean?

An API (application programming interface) enables communication and data exchange between different pieces of software and applications. In the context of AI, APIs let developers integrate AI services (such as language models, image recognition, or speech synthesis) into their own applications without having to develop or maintain AI models themselves.

With APIs, companies and developers can build AI features into their applications by sending requests to a provider’s servers and receiving the results in a standardised format such as JSON or XML.

What does attention mechanism mean?

An attention mechanism is a component of AI that allows a model to focus on particular parts of an input and weight them more heavily than others. It is a central part of transformer-based models such as GPT and BERT. The attention mechanism makes it possible for a model to handle long sequential inputs and understand the relationships between words or elements in context. It is a significant advance that enabled the development of today’s large language models.

What does AI bias mean?

AI bias means systematic errors in AI systems that lead to unfair or inaccurate results. These biases can arise from shortcomings in the training data, from the design of the algorithm, or from societal prejudices being reflected in the data. Bias can lead to discrimination and inequality in AI use — in recruitment, lending decisions, or facial recognition, for example.

What does big data mean?

Big data means very large, complex, and rapidly growing volumes of data that cannot be handled with traditional database software. AI uses big data to learn and to improve its predictions. These datasets are analysed with machine learning algorithms and cloud-based systems, among other tools.

What does chatbot (AI bot) mean?

A chatbot or AI bot is an AI program designed to hold a conversation with a person in natural language. Modern AI bots use large language models (LLMs) and deep learning to understand users’ questions and produce relevant answers. AI bots are used in customer service, information retrieval, and assistive systems.

What is ChatGPT?

ChatGPT is a conversational AI model developed by OpenAI, based on the GPT (Generative Pre-trained Transformer) architecture. ChatGPT can hold natural-feeling conversations, answer questions, produce text, and help with a wide range of tasks. It is one of the world’s most used AI services, available in both a free version and a paid one that gives access to the newest models and features. Tools such as Deep Research and DALL-E, OpenAI’s image generator, also run inside ChatGPT. ChatGPT’s models are in addition very capable coding assistants and code generators.

• ChatGPT

• OpenAI

What is Claude?

Claude is a conversational AI model developed by Anthropic, designed to be a helpful, safe, and honest assistant. Claude can handle long texts, answer questions, and produce many kinds of content. It is available through an API and through a web interface aimed at consumers.

Claude is particularly strong at understanding and producing text, but it also offers significant help with programming. It can write, explain, fix, and optimise code in several programming languages, including Python, JavaScript, SQL, Ruby, C++, and Go. Developers use Claude for designing algorithms, fixing bugs, following good practice, and writing documentation.

• www.claude.ai

What is GitHub Copilot?

GitHub Copilot is an AI-based coding assistant developed jointly by Microsoft-owned GitHub and OpenAI, which helps developers write code by suggesting whole lines or functions. Copilot was trained on millions of open-source GitHub projects and can understand programming context and a developer’s intentions. The tool integrates directly into several development environments such as Visual Studio Code, Visual Studio, and JetBrains products.

• GitHub Copilot

• GitHub

What is Cursor?

Cursor is an AI-assisted code editor designed to make developers’ work more effective. Cursor is based on Visual Studio Code but integrates AI models that can create, explain, and edit code through conversation. Cursor understands context and the structure of your project, which makes it possible to automate even complex programming tasks.

Cursor supports several programming languages and has a built-in chat interface through which a developer can communicate with the AI in natural language. It has quickly become one of the most popular tools among developers for supporting software work.

• https://www.cursor.com/ (tool)

What does data science mean?

Data science is an interdisciplinary field that uses scientific methods, algorithms, and AI to analyse and make use of large volumes of data. Data scientists combine statistics, machine learning, and computer science to find valuable information in data and turn it into usable predictions. Data science forms the basis for many AI applications.

What does deep learning mean?

Deep learning is the area of machine learning that uses multi-layered neural networks to process information and learn. These networks consist of several layers of artificial neurons that can gradually learn increasingly abstract and complex features from data, without rules defined separately by a human.

Deep learning has transformed AI by enabling breakthroughs in many fields, such as image recognition, speech recognition, natural language processing, and game-playing AI. With deep learning methods, AI can recognise faces in images, translate text between languages, or beat humans at complex strategy games.

The power of deep learning rests on its ability to learn automatically to recognise complex patterns and structures in large volumes of data. Unlike traditional algorithms, deep learning does not require a programmer to specify which features the algorithm should attend to; the neural network learns to find the relevant features from the data itself.

What is DeepL?

DeepL is a machine translation service developed by the German company DeepL GmbH, which uses deep learning to offer accurate and natural translations. The service supports over 30 languages and can translate both text and whole documents, such as Microsoft Word, PDF, and PowerPoint files, while preserving the original formatting.

DeepL is known for its intuitive interface and seamless integration on Windows and iOS. Its ability to understand a text’s context and nuances makes it a popular choice for individuals and companies that need reliable translation.

• deepl.com

What is DeepSeek?

DeepSeek is a Chinese AI company developing large language models and AI solutions. It is known for its DeepSeek LLM series, which includes both open and commercial language models. DeepSeek models compete on performance with other leading language models, particularly in technical problem-solving and programming.

DeepSeek has released several open-source models, such as DeepSeek-Coder, which specialises in programming, and DeepSeek-7B, -67B, and others for general use. The company focuses in particular on complex reasoning tasks and natural language understanding.

What does diffusion model mean?

A diffusion model is generative AI that creates new content such as images or audio through a gradual “denoising” process. Diffusion models work by first adding random noise to data and then learning to reverse that process. They are the basis of modern image generators such as Stable Diffusion, DALL-E, and Midjourney. Diffusion models have transformed the quality and controllability of generative AI compared with earlier methods.

What is ElevenLabs?

ElevenLabs is an AI-based audio technology company offering a range of tools for synthesising natural speech. Its services include text-to-speech conversion, voice cloning, and multilingual dubbing. Its technology is widely used in audiobooks, video narration, and chatbots. Tekoälyfoorumi’s own audio content is produced with ElevenLabs.

What does few-shot learning mean?

Few-shot learning means AI’s ability to learn and perform new tasks with only a handful of examples. Unlike traditional machine learning methods, which need hundreds or thousands of examples, few-shot learning allows a model to adapt quickly to new situations using a limited number of examples.

The approach is used particularly with language models (LLMs), where a user can give the model a few examples of the desired answer style before the actual question or task. A user might, for example, show the model a couple of examples of poems written in a particular style before asking it to create a similar poem on a new subject.

Few-shot learning is a significant advance in AI’s flexibility and usability, because it allows models to be used in situations where data is limited or where you want to adapt quickly to new tasks without extensive retraining.

What does fine-tuning mean?

Fine-tuning means the process of adapting a pre-trained AI model, such as a language model, to a particular task or domain using a smaller, specialised dataset. Fine-tuning is a cost-effective way to create specialised models, because it draws on the general capability already built into the model during pre-training. It is used to tailor language models for medical, legal, or technical use, for example.

What is Gemini?

Gemini is a multimodal AI model developed by Google DeepMind, capable of handling and producing different content formats such as text, images, audio, and video. Gemini models are available in different sizes (Ultra, Pro, Nano) for different purposes and device platforms. They power Google’s conversational assistant and features on Android devices. Gemini is designed to compete directly with OpenAI’s GPT models and Anthropic’s Claude.

• Google Gemini

What does generative AI mean?

Generative AI means AI systems capable of creating new content such as text, images, music, or video on the basis of existing data. These systems learn to recognise patterns and regularities in data, on the basis of which they can produce new material resembling the original data. Well-known examples are GPT language models and DALL-E image generators.

What does GPT mean?

GPT (Generative Pre-trained Transformer) is a family of language models developed by OpenAI, based on the transformer architecture. GPT models are pre-trained on vast amounts of text data and can produce coherent, context-dependent text, hold conversations, write code, and perform many other language tasks. The GPT series is updated continuously and new models are always more capable than the last. GPT models power ChatGPT, among other services.

• OpenAI

What is Grok?

Grok is a conversational AI model developed by xAI, the company founded by Elon Musk, competing with other leading language models such as ChatGPT, Claude, and Gemini. Grok is designed to be a humorous and permissive AI that answers questions other AI systems might not. It is integrated into the X platform (formerly Twitter) and available to Premium+ subscribers.

Grok is based on a model trained on publicly available data. Its distinguishing features include the ability to use real-time information from X and a more direct, less restricted answering style than many other general-purpose AI systems.

What does image generator mean?

An image generator is an AI system capable of creating new images from text descriptions, existing images, or other inputs. Modern image generators are often based on diffusion models or GANs (generative adversarial networks) and can produce strikingly realistic or stylised images. Well-known examples are DALL-E, Midjourney, Imagen, and Stable Diffusion.

What is Imagen?

Imagen is a text-to-image AI model developed by Google that creates high-quality images from text descriptions. Imagen is based on diffusion models and is optimised to produce photorealistic images that closely match the text prompt. Google has since integrated Imagen technology into several of its products, such as Google Workspace and Google Cloud.

Imagen competes with other image generators such as DALL-E, Midjourney, and Stable Diffusion, and its particular strength is the ability to understand complex text descriptions and create matching images with high fidelity.

What does language model mean?

A language model (large language model, LLM) is an AI system trained on vast amounts of text data to understand and produce natural language. Language models can answer questions, write different kinds of text, translate between languages, and even write code. They work by predicting the most likely next word or sentence on the basis of the preceding text context. Well-known language models include GPT-4, Claude, and BERT.

What does machine learning mean?

Machine learning is the area of AI in which computer programs improve their performance through experience, without being explicitly programmed for every situation. Machine learning algorithms use data to detect patterns and make decisions with minimal human guidance. Machine learning is used in recommendation systems, search engines, and image recognition, among other things.

What is Midjourney?

Midjourney is an AI-based image generator that creates high-quality images from text descriptions. It is known particularly for aesthetically striking artistic results and can produce a wide range of styles. The service was previously used through Discord, but now runs on Midjourney’s own website. Midjourney is popular with artists and companies alike for creating visual content.

What is Mistral AI?

Mistral AI is a European AI company developing open-source language models and other AI solutions. Its aim is to offer effective, scalable AI transparently and in a community-led way. Mistral AI is known for lightweight but high-performing models that compete with those of OpenAI and Anthropic.

Recent releases:

• Le Chat: in February 2025 Mistral AI released Le Chat, a generative AI assistant. It can answer at up to 1,000 words per second, making it one of the fastest AI assistants on the market.

• mistral.ai

What does narrow AI mean?

Narrow AI means AI designed and trained to perform only one or a few precisely defined tasks excellently. Unlike artificial general intelligence (AGI), narrow AI cannot apply what it has learned outside its task area and has no human-like understanding or capacity to generalise. Every AI application in use today is a form of narrow AI — spam filters, image recognition, search engines, recommendation systems, and even advanced language models.

What does natural language processing mean?

Natural language processing (NLP) is the area of AI focused on the interaction between computers and human language. NLP technologies allow computers to understand, interpret, and produce human language in a meaningful way. Applications include translation, text analytics, speech recognition, and chatbots.

What does neural network mean?

A neural network is a machine learning model loosely modelled on how the human brain works. Neural networks consist of interconnected artificial neurons that pass signals to one another. They learn to recognise patterns in large volumes of data by adjusting the connections between neurons. Neural networks form the basis of deep learning and make complex AI applications possible.

What is NotebookLM?

NotebookLM is an AI-based research and note-taking tool developed by Google, powered by the Gemini language model. It lets users upload files such as PDFs, Google Docs, and web pages, and get summaries, explanations, and answers to their questions from that content.

The “Audio Overviews” feature, released in September 2024, allows the content of documents to be listened to in a podcast-style format. NotebookLM is designed to work as a personal research assistant that helps users absorb and organise information more effectively.

What is Perplexity?

Perplexity is an AI-based search engine and information retrieval tool developed by Perplexity Labs, which answers users’ questions in natural language and provides sources to back up its answers. Perplexity combines language model technology with up-to-date information from the internet, so the user gets both a conversational interface and fact-based, cited answers.

Perplexity differs from traditional search engines by offering direct answers to questions rather than collections of links. The service is particularly useful in research, information retrieval, and working through complex questions. It offers both a free version and a paid Pro version with more features and more advanced models.

What is Pika Labs?

Pika Labs is an AI-based video generator that turns text descriptions or images into short videos. Pika is known particularly for enabling fast video creation with simple commands. The tool can create animations, visual effects, and varied video content.

What does prompt engineering mean?

Prompt engineering means the skill of formulating the text prompts given to an AI effectively, so as to get the kind of answers you want. It is a method of steering an AI model to produce precise, relevant, high-quality results through carefully written instructions.

Good prompt design takes account of how AI works and includes clear goals, the necessary context, and often examples of the desired output.

Techniques include step-by-step instruction, role prompting, and guiding through examples. Prompt engineering has become a central skill in the age of AI, and can be regarded as a new kind of digital literacy.

What does RAG mean?

RAG (retrieval-augmented generation) is an AI method in which a language model is combined with an information retrieval system, so that it can fetch up-to-date or specialised information from external sources while forming an answer. RAG improves the reliability and factual accuracy of AI answers by giving the model access to knowledge outside its own parameters.

In a typical RAG process, the system first retrieves the relevant documents or pieces of information, which are then fed to the language model as context, after which the model produces an answer grounded in that information.

What does reinforcement learning mean?

Reinforcement learning, or trial-and-error learning, is a machine learning method in which an AI agent learns in its environment through rewards and penalties. The agent makes decisions and gets feedback on the consequences of its actions, seeking to maximise long-term reward. The method is used in game-playing AI, robotics, and self-driving vehicles, among other things.

What does responsible AI mean?

Responsible AI means an approach to developing, deploying, and governing AI that takes account of ethical principles, societal effects, and human wellbeing. Responsible AI seeks to ensure AI systems work transparently, fairly, safely, and in a human-centred way. That includes designing AI so that it respects human rights, privacy, diversity, and different cultural values.

The principles of responsible AI include identifying and reducing algorithmic bias, explainability and transparency, information security and privacy, and assessing the consequences of AI use and taking responsibility for any harm. It also matters that people are kept in the decision-making process, so that AI assists rather than replaces human judgement in critical decisions.

What is Runway?

Runway is a versatile AI platform for creative work, offering tools for creating video, images, and visual effects. Runway Gen-3 is a well-known video generator that can create new video from text descriptions or edit existing video. Runway is one of the most used and most advanced tools in the creative industries.

What is Sora?

Sora is a text-to-video model developed by OpenAI that can create realistic and creative video from a text description alone. Sora can generate high-quality videos up to a minute long, with complex movement, detailed characters, and accurate physics. Its distinguishing feature is an ability to understand the physical world and create convincing human movement and intricate camera moves. Sora is still in development and available only to a limited group of users.

• https://openai.com/sora/ (more information)

• https://sora.com/ (tool)

What does supervised learning mean?

Supervised learning is a machine learning method in which an algorithm is trained using inputs and their corresponding predefined outputs. The algorithm learns to make predictions about future inputs by comparing them with earlier examples. Typical applications of supervised learning include email spam filters and facial recognition.

What is Synthesia?

Synthesia is an AI-based video generator that lets you create video with AI-generated virtual presenters (AI avatars), without cameras or actors. Synthesia uses deep learning to synchronise the presenter’s lip movements with the audio track and produce natural-looking video. The service is used particularly in corporate communications, training, marketing, and multilingual video production.

• https://www.synthesia.io/ (try it free)

What does synthetic data mean?

Synthetic data means artificially created information that imitates real data but contains no genuine information. Synthetic data is produced with algorithms and simulations, and is used to train AI models particularly where real data is hard to obtain, expensive, or subject to privacy restrictions. Its advantages are privacy protection, adjustability, and scalability, which make it valuable in healthcare, finance, and autonomous vehicle development, among other fields.

What does unsupervised learning mean?

Unsupervised learning is a machine learning method in which an algorithm looks for patterns and structures in data without predefined classes or outcomes. The method suits situations where the meaningful patterns or groupings are not known in advance. Common applications are customer segmentation, anomaly detection, and recommendation systems.

What does zero-shot prompting mean?

Zero-shot prompting means using AI by giving the model a task or question without prior examples or fine-tuning for that task. The AI can perform the task on the basis of its general training, transferring its knowledge to a new context. This is one of the most significant capabilities of modern large language models (LLMs) — answering questions and performing tasks they were not specifically trained for.

Zero-shot prompting is the simplest form of prompt design, in which the user asks a direct question or requests a task without special instructions or examples. It differs from few-shot prompting, in which the model is given a few examples of the desired answer style before the actual question.