AI Glossary

A narrowly focused piece of artificial intelligence that can use language models and other tools to perform simple tasks and achieve goals based on user instructions. Agents can create and escalate support tickets, provide product recommendations, act as content moderators, and more.

The process by which personally identifiable information is removed from a dataset. This helps protect confidentiality in the event of data breaches, data leakage, and other threats to data privacy.

Skewed or inaccurate responses produced by an AI system. Bias is often the result of a flaw in the company's training data and/or training practices. It can be mitigated through a focus on data quality, bias testing, and AI explainability.

The uncertain nature of the internal mechanisms and processes an AI system uses to arrive at its results. The black box makes it difficult to explain how AI reaches certain conclusions and to verify that it is following user instructions throughout the process.

An LLM prompting technique that simulates human reasoning by breaking large, complex tasks into smaller steps. In addition to improving explainability, CoT prompting helps improve the accuracy of LLM responses by forcing the model to follow a logical thought process.

The inadvertent exposure of sensitive information within an LLM response. Leakage may occur unprompted or be the result of malicious prompting techniques designed to extract sensitive data from a system.

The process of identifying patterns and extracting insights from large data sets using machine learning algorithms. Through data mining, businesses can improve decision-making and optimize performance across functions.

A type of machine learning that simulates human thought processes using multilayered neural networks. A deep learning model can analyze text and images, recognize patterns, predict outcomes, and more.

A deep learning model with the ability to create, analyze, and refine content in response to user prompts. Depending on the user's instructions, generative AI may be used to write emails, create images, make predictions, answer questions by pulling information from a knowledge base, and more.

An AI-generated response that is inaccurate or misleading, often caused by erroneous training data, insufficient training data, or technological flaws. Frequent hallucinations can erode user trust in an AI model, and may be damaging to users who rely on models for critical business functions.

A popular form of AI that processes and generates text in human languages based on user-provided context. Training data is stored within the LLM, and the system pulls from that data to formulate relevant responses to user prompts.

A method by which AI decision-making is improved, using algorithms to detect patterns and obtain relevant insights from datasets. Through machine learning, AI models can expand their knowledge and capabilities without direct human intervention.

The technology that allows AI systems to process and generate human language. NLP can be used for text translation, sentiment analysis, summarization, keyword extractions, and more.

A machine learning model that simulates the thought processes of the human brain to improve AI accuracy and decision-making. Neural networks can be deployed for complex tasks, such as pattern recognition and text analysis.

The process of structuring requests in a way that promotes an optimal response from an AI system. Well-engineered prompts make it easier for the system to understand the user's intention, resulting in more accurate and relevant responses.

A set of principles that promote ethical AI development and usage, with a focus on areas such as security, sustainability, and accountability. Responsible AI has already been embraced by a number of big tech companies, as well as governments, research institutes, and other organizations.

An AI framework that pulls data from a knowledge base or vector database to formulate a response, eliminating the need for users to store data within their AI models. Because RAG models can access external data, the accuracy and relevance of their responses are often improved.

Sets of content or information used to improve the functioning and accuracy of an AI system. For example, a company may use CRM data on customers and their purchase history to teach its AI model to make relevant product recommendations.