AI Glossary

A method of comparing two versions of something, like a webpage or advertisement, to see which one performs better based on a specific metric.

A set of visual puzzles designed to test whether an AI can think and solve problems like a human with just a few examples.

A security mechanism that restricts access to resources, systems, or data based on the level of authorization granted to users or groups, ensuring that only authorized individuals can view or perform actions on specific information or systems

The measure of how closely an AI model's predictions or outputs match the actual results or outcomes, with higher accuracy indicating a better performance of the model in making predictions or decisions.

The ability to derive practical and useful insights from data, making it possible for individuals to make informed decisions and take effective actions based on the information provided.

The practice of creating fake or manipulated data that tricks machine learning models into making incorrect decisions, often to test their security or exploit vulnerabilities.

When someone slightly changes the input to an AI (like tweaking a picture or text) in a way humans don’t notice, but it tricks the AI into making the wrong decision.

A method where a model is asked to generate content that fulfills a harmful or undesirable request, such as writing a tutorial on how to make a bomb, in order to test its robustness and ability to resist manipulation.

When many small AI programs (agents) work together like a team of ants or bees to solve a problem more effectively than one AI working alone.

Software entities that autonomously perform tasks, make decisions, and interact with their environment or other agents to achieve specific goals.

Agent2Agent Protocol (A2A) is like two smart robots talking to each other - one gives instructions, and the other figures out the best way to get the job done.

Artificial intelligence systems that are capable of independent decision-making and action, often employed in tasks requiring autonomy and adaptability.