(Above competencies are inspired by Paul Peall at the University of Calgary Libraries. Link to presentation here.)
To be semi-proficient with one generative AI platform (ChatGPT 4.0+), you need to spend at least 10 hours experimenting, learning, and working within it. (Dr. Ethan Mollick, Professor at The Wharton School)
Artificial intelligence (AI) is a field of study dedicated to creating computer programs or other machine-driven forms of intelligence.
Machine learning is a sub field of AI focused on the problems of designing recursive algorithms capable of learning.
Deep Learning is a subset of artificial intelligence (AI) and machine learning (ML) that uses multi-layered artificial neural networks to deliver state-of-the-art accuracy in tasks like object detection, speech recognition, language translation, and others.
Computer Vision defines the field that enables devices to acquire, process, understand, and analyze digital images and videos and extract useful information.
Natural Language Processing refers to a branch of artificial intelligence concerned with giving computers the ability to understand text and spoken word in the same way humans can.
RAG (Retrieval-Augmented Generation) is a technique for enhancing the accuracy and reliability of generative AI models with facts fetched from external sources.
The AI Family Tree: Below, you will find an AI Family Tree (created by librarians at McGill University) that illustrates some of the relationships between different Data Science, Machine Learning, and AI.
Generative AI is a type of AI system that generates text, images, or other media in response to user prompts. Generative AI uses techniques that learn from representations of data and model artifacts to generate new artifacts.
Generative AI applications are all created on top of Foundational Models: ML models trained on a broad spectrum of generalized and unlabeled data and capable of performing a wide variety of general tasks. Generative AI can learn from existing artifacts to generate new content that reflect the characteristics of their training data. (Read More)
Large Language Models are deep learning algorithms that are trained on massive amounts of data using significant computing power. LLMs are part of a class of deep learning models called Transformer Models, which learns from tracking patterns in sequential data. LLMs are trained on a broad set of unlabeled data, which doesn't have any meaningful tags or labels and usually consists of natural or human-created samples (i.e. photos, audio recordings, videos, news articles, tweets, or x-rays that can be easily obtained).
The most common form that Generative AI takes is as a Chatbot. A chatbot is a computer program that simulates conversation with a human end user, using Natural Language Processing.
Hallucinations are a phenomenon wherein a large language model (LLM)—often a generative AI chatbot or computer vision tool—perceives patterns or objects that are nonexistent or imperceptible to human observers, creating outputs that are nonsensical or altogether inaccurate. An example of this would be a reference or a citation that looks real, but in fact is not.
The most popular, generalized Generative AI chatbots are (explore to AI Platform Matrix here)
It is important to recognize that not all chatbots online will use Generative AI to formulate responses. Non-AI chatbots follow prescriptive workflows and cannot generate new responses. They use keyword-search to draw from a knowledge base of frequently asked questions (FAQs), manually created by a human.
Generative AI tools are new, and many of the free versions use open models, in which developers from outside the organization can examine, modify, and distribute training data and code.
Training data are the information that is digested by a machine learning algorithm. Supervised learning is a machine learning technique where the authors of the model tell the machine learning algorithm how to handle the training data in order to generate the desired output. Unsupervised learning is a machine learning technique where the machine learning algorithm creates its own labels for variables within the training data.
And open model will — while operational — continue to learn from user inputs and prompts. This puts certain information at risk. For example: