NAiOS Wiki

Artificial General Intelligence, capable of any human intellectual task

Systems that execute multi-step tasks without constant supervision

Aligning AI goals with human values

Acute cognitive fatigue from excessive use or supervision of AI tools. Term coined by researchers from BCG and the University of California (Riverside) in 2026.

AI Assistant integrated into work tools

Regulatory and management framework for the responsible use of AI

AI-generated summaries in search results

Informal term (not a recognized clinical diagnosis) for psychotic-like symptoms—delusions, paranoia—associated with the intensive use of AI chatbots.

Discipline focused on ensuring that AI is safe

Application that simply wraps an existing model without adding real value

AI capable of planning, deciding, and acting autonomously

Workflows where AI agents act autonomously

When AI reproduces or amplifies biases present in the data

Technique where AI thinks step-by-step before responding

Amount of text a model can remember in a conversation

When evaluation data leaks into training, biasing benchmarks

Competitive advantage of a company thanks to its unique data

Using AI to improve business decision-making

Hyper-realistic AI-generated fake content

Run AI directly on the device, without relying on the cloud

Numerical representation of the meaning of texts or images

Retraining a model with specific data for a use case

The most advanced and powerful models at all times

The new SEO, optimizing content to be cited by generative AI

Artificial intelligence that creates new content (text, image, video, audio)

Grounding AI responses in real and verifiable data

Restrictions and limits imposed on AI to prevent harm

When AI generates false information with total confidence

Framework for testing and evaluating agent performance

Dedicate more compute capacity during the response, not training

Techniques for evading model security restrictions

Large language model, the foundation of today's chatbots

Techniques for adapting models with limited resources

Running AI models on your own hardware

Designing the autonomous loops that guide code agents (2026 term)

Architecture that activates only parts of the model based on the task

Standard protocol for models to connect with external tools

Compressing a large model into a small one while preserving quality

Ethical debate on whether advanced models deserve some form of consideration

Operational management of the AI model lifecycle

Coordinate multiple AI agents to solve complex tasks together

Models that understand and generate text, image, audio and video

A chip specialized in accelerating neural network inference with maximum energy efficiency

AI tools that require no programming to use

When a prompt that worked stops delivering good results over time

Design and optimization of prompts for better AI results

Attacks that manipulate AI by inserting malicious instructions

Combine real data search with text generation

Reinforcement Learning from Human Feedback

Models designed specifically to reason, not just predict text

Testing an AI by attempting to break it to find vulnerabilities

The code structure that connects and orchestrates an AI agent

Unauthorized use of AI tools within organizations

Lightweight language models, designed to run on devices

Strategic priority for countries to have their own AI infrastructures

AI that surpasses human intellectual capacity in all fields

Artificially generated data for training models

Media (image, audio, video) artificially created by AI

Unit that measures the raw performance of an AI accelerator: trillions of operations per second

Parameter that controls AI creativity vs. precision

Generate video from a text description

Informative label for the phase of mass and cross-sector AI adoption across all industries. It is not an academic term or a quote from MIT.

Process of dividing text into units (tokens) that the model can process

Maximizing a model's context window. Colloquially, also: using many tokens as a (debatable) productivity metric.

The base architecture of almost all current language models

Program by describing what you want and letting the AI write the code

Clone a human voice with AI from a few seconds of audio

Models that learn the dynamics of an environment, not just text

Ability of a model to perform tasks without being specifically trained