Generative Artificial Intelligence (AI) refers to deep learning models that learn from representations of data and model artifacts to generate new artifacts. The defining feature of Generative AI is that it can learn from existing artifacts to generate new content that reflect the characteristics of their training data. (Read more)
In November 2022, OpenAI released its newest chatbot (ChatGPT-3), sparking the current interest we are seeing in AI-powered chatbots.
Here are some key terms and concepts to know:
Algorithm: Broadly defined, an algorithm is a set of instructions or a procedure for solving a problem. Like the recipe for a dish.
Artificial Intelligence: Artificial Intelligence is both a field of study and a type of technology. Artificial Intelligence or AI refers to the capacity of computers or other machines to exhibit or simulate human behavior (OED). (Read More)
Artificial General Intelligence (AGI): AGI, also known as Strong AI or Human-Level Machine Intelligence (HLML) refers to systems capable of conducting a complete range of intellectual tasks at a level equaling that of the best-performing human beings. This type of Artificial Intelligence is still entirely theoretical and continues to be researched. (Read More)
Closed Generative AI Models protect either/both their models and underlying source code by making them inaccessible to users outside of the organization. Similarly, the training data and the process of collecting it are kept private. In a closed model, users will not be able to evaluate the training data or understand how the model is trained, but they may have greater control over how their data is used, depending on the privacy and data protection policies of that product. Closed does not necessarily mean "more secure" -- it depends on the policies of the product.
Copyrighted Data: Copyrighted Data is information that is copyrighted by the owner or creator. Copyright gives owners the exclusive right to reproduce, adapt, publish, perform, and/or display their work. Examples include articles and reports acquired via licensed databases, books and images (that are not in the public domain), music, and more. You should never enter copyrighted data or works into any Generative AI model, unless you are the copyright owner.
Deep Learning: Deep Learning is a subset of Machine Learning that uses Deep Neural Networks to simulate the complex decision-making of the animal brain.
Large Language Models: Large Language Models (LLMs) are a type of Foundation Models that are trained on massive amounts of unlabeled data using significant computing power. They are important because of their flexibility – one LLM can perform a large variety of text-based tasks. They can be adapted to perform other tasks, as well, such as image generation.
Open Generative AI Models make their models and underlying source code available as open source software for researchers and developers to learn from and build upon. The training data is also available for developers and users alike to inspect, including user input and queries. The transparency makes it possible to evaluate the training data, but also means that users have less control over how their data and interactions with the model are used.
Machine Learning: The use of data and algorithms to imitate the way that humans learn – recognizing patterns and learning from past experiences to improve performance.
Neural Networks: Neural Networks try to recognize patterns and relationships in a set of data through a process that mimics the way the animal brain operates, by providing different weights/strength to each connection.
Regulated Data: Regulated Data is information that is protected by statute or regulation as personal, sensitive data. Examples include your Social Security Number (SSN), health information, credit card and financial account numbers, and student academic records. You should never enter regulated data into any Generative AI model.
Restricted Data: Restricted Data is data that is unavailable to the public or otherwise deemed confidential due to proprietary reasons. Examples will vary depending on the organization but can include internal information (e.g., trade secrets) or any information that could cause adverse impacts to the organization. You should never enter restricted use data into any Generative AI model, unless is it closed and secure.
This guide is licensed for reuse with attribution.
