Learn the essential terms and concepts you need to know to navigate the rapidly evolving world of generative AI
Generative AI is a fascinating and rapidly evolving field that has the potential to transform the way we interact with technology. However, with so much buzz and hype surrounding this topic, making sense of it all can be challenging. In this article, we’ll cut through the noise and gain a clear understanding of the essential terminology you need to know to navigate the world of generative AI.
According to a variety of sources, including McKinsey & Company and Vox Media, the critical difference between generative AI and other emerging technologies is that millions of people can — and already are — using it to create new content, such as text, photos, video, code, and 3D renderings, from data it is trained on. Recent breakthroughs in the field have the potential to drastically change the way we approach content creation. This has led to widespread excitement and some understandable apprehension about the potential for generative AI to impact virtually every aspect of society and disrupt industries, including media and entertainment, healthcare and life sciences, education, advertising, legal services, and finance.
Even if your current role is not in technology, it is highly likely that generative AI will have a direct impact on both your personal and professional life.Gary Stafford
Even if your current role is not in technology, it is highly likely that generative AI will have a direct impact on both your personal and professional life. Familiarizing yourself with basic terminology related to generative AI can help you better comprehend the discussions on social media and in the news.
Let’s explore the following terminology (in alphabetical order):
- Artificial General Intelligence (AGI)
- Artificial Intelligence (AI)
- Deep Learning (DL)
- Generative AI
- Generative Pre-trained Transformer (GPT)
- Intelligence Amplification (IA)
- Large Language Model (LLM)
- Machine Learning (ML)
- Neural Network
- Prompt Engineering
- Reinforcement Learning with Human Feedback (RLHF).
Below is a knowledge graph, created with OpenAI ChatGPT, showing the approximate relationships between the post’s terms.
Artificial General Intelligence (AGI)
According to the all-new Bing Chat, based on ChatGPT, artificial general intelligence (AGI) is the ability of an intelligent agent to understand or learn any intellectual task that human beings or other animals can. It is a primary goal of some artificial intelligence research and a common topic in science fiction and Futurism. According to Forbes, which also prompted ChatGPT, Artificial General Intelligence (AGI) refers to a theoretical type of artificial intelligence that possesses human-like cognitive abilities, such as the ability to learn, reason, solve problems, and communicate in natural language.
Eliezer Yudkowsky is an American researcher, writer, and philosopher on the topic of AI. The podcast Eliezer Yudkowsky: Dangers of AI and the End of Human Civilization, by prominent MIT Research Scientist Lex Fridman, explores various aspects of artificial general intelligence against the backdrop of the recent release of OpenAI’s GPT-4.
Artificial Intelligence (AI)
According to the Brookings Institute, AI is generally thought to refer to machines that respond to stimulation consistent with traditional responses from humans, given the human capacity for contemplation, judgment, and intention. Similarly, according to the U.S. Department of State, the term artificial intelligence refers to a machine-based system that can, for a given set of human-defined objectives, make predictions, recommendations, or decisions influencing real or virtual environments.
ChatGPT, according to ChatGPT, is a large language model developed by OpenAI. I was trained on a massive dataset of human-written text using a deep neural network (DNN) architecture called GPT (Generative Pre-trained Transformer). Its purpose is to generate human-like responses to questions and prompts, engage in conversations, and perform various language-related tasks. It is a virtual assistant capable of understanding and generating natural language.
According to Wikipedia, DALL·E is a deep learning model developed by OpenAI to generate digital images from natural language descriptions, called prompts. DALL·E is a portmanteau of the names of the animated robot Pixar character WALL-E and the Spanish surrealist artist Salvador Dalí. According to OpenAI, DALL·E is an AI system that can create realistic images and art from a description in natural language. OpenAI introduced DALL·E in January 2021. One year later, in April 2022, they announced their newest system, DALL·E 2, which generates more realistic and accurate images with 4x greater resolution. DALL·E 2 can create original, realistic images and art from a text description. It can combine concepts, attributes, and styles.
Deep Learning (DL)
According to IBM, deep learning is a subset of machine learning (ML), which is essentially a neural network with three or more layers. These neural networks attempt to simulate the behavior of the human brain — albeit far from matching its ability — allowing it to “learn” from large amounts of data. While a neural network with a single layer can still make approximate predictions, additional hidden layers can help to optimize and refine for accuracy.
According to Wikipedia, generative artificial intelligence (AI), aka generative AI, is a type of AI system capable of generating text, images, or other media in response to prompts. Generative AI systems use generative models such as large language models (LLMs) to statistically sample new data based on the training data set used to create them.
Generative Pre-trained Transformer (GPT)
According to ChatGPT, Generative Pre-trained Transformer (GPT) is a deep learning architecture used for natural language processing (NLP) tasks, such as text generation, summarization, and question-answering. It uses a transformer neural network architecture with a self-attention mechanism, allowing the model to understand each word’s context in a sentence or text. The success of GPT models lies in their ability to generate natural-sounding and coherent text similar to human-written language. The term “pre-trained” refers to the fact that the model is trained on large amounts of unlabeled text data, such as books or web pages, to learn general language patterns and features, before being fine-tuned on smaller labeled datasets for specific tasks.
According to ZDNET, GPT-4, announced on March 14, 2023, is the newest version of OpenAI’s language model systems. Its previous version, GPT 3.5, powered the company’s wildly popular ChatGPT chatbot when it launched in November 2022. According to OpenAI, GPT-4 is the latest milestone in OpenAI’s effort to scale up deep learning. GPT-4 is a large multimodal model (accepting image and text inputs, emitting text outputs) that, while less capable than humans in many real-world scenarios, exhibits human-level performance on various professional and academic benchmarks.
According to Wikipedia, intelligence amplification (IA) (aka cognitive augmentation, machine-augmented intelligence, or enhanced intelligence) refers to the effective use of information technology in augmenting human intelligence. Similarly, Harvard Business Review describes intelligence amplification as the use of technology to augment human intelligence. And a paradigm shift is on the horizon, where new devices will offer less intrusive, more intuitive ways to amplify our intelligence.
In his latest book, Impromptu: Amplifying Our Humanity Through AI, co-authored by ChatGPT-4, Greylock general partner Reid Hoffman discusses the subject of intelligence amplification and AI’s ability to amplify human ability. The topic was also explored in Hoffman’s interview with OpenAI CEO Sam Altman on Greylock’s podcast series AI Field Notes.
Large Language Model (LLM)
According to Wikipedia, a large language model (LLM) is a language model consisting of a neural network with many parameters (typically billions of weights or more), trained on large quantities of unlabelled text using self-supervised learning. Though the term large language model has no formal definition, it often refers to deep learning models having a parameter count on the order of billions or more.
Machine Learning (ML)
According to MIT, machine learning (ML) is a subfield of artificial intelligence, which is broadly defined as the capability of a machine to imitate intelligent human behavior. Artificial intelligence systems are used to perform complex tasks in a way that is similar to how humans solve problems. The function of a machine learning system can be descriptive, meaning that the system uses the data to explain what happened; predictive, meaning the system uses the data to predict what will happen; or prescriptive, meaning the system will use the data to make suggestions about what action to take.
According to MathWorks, a neural network (aka artificial neural network or ANN) is an adaptive system that learns by using interconnected nodes or neurons in a layered structure that resembles a human brain. A neural network can learn from data to be trained to recognize patterns, classify data, and forecast future events. Similarly, according to AWS, a neural network is a method in artificial intelligence that teaches computers to process data in a way that is inspired by the human brain. It is a type of machine learning process called deep learning that uses interconnected nodes or neurons in a layered structure that resembles the human brain. It creates an adaptive system that computers use to learn from their mistakes and improve continuously.
Types of Neural Networks
Deep neural networks (DNNs) are improved versions of conventional artificial neural networks (ANNs) with multiple layers. While ANNs consist of one or two hidden layers to process data, DNNs contain multiple layers between the input and output layers. Convolutional neural networks (CNNs) are another kind of DNN. CNNs have a convolution layer, which uses filters to convolve an area of input data into a smaller area, detecting important or specific parts within the area. Recurrent neural networks (RNNs) can be considered a type of DNN. DNNs are neural networks with multiple layers between the input and output layers. RNNs can have multiple layers and can be used to process sequential data, making them a type of DNN.
OpenAI is a San Francisco-based AI research and deployment company whose mission is to “ensure that artificial general intelligence benefits all of humanity.” According to Wikipedia, OpenAI was founded in 2015 by current CEO Sam Altman, Greylock general partner Reid Hoffman, Y Combinator founding partner Jessica Livingston, Elon Musk, Ilya Sutskever, Peter Thiel, and others. OpenAI’s current products include GPT-4, DALL·E 2, Whisper, ChatGPT, and OpenAI Codex.
According to Cohere, prompting (aka prompt engineering) is at the heart of working with LLMs. The prompt provides context for the text we want the model to generate. The prompts we create can be anything from simple instructions to more complex pieces of text, and they are used to encourage the model to produce a specific type of output. Cohere’s Generative AI with Cohere blog post series is an excellent resource on the topic of Prompting. Similarly, according to Dataconomy, using prompts to get the desired result from an AI tool is known as AI prompt engineering. A prompt can be a statement or a block of code, but it can also just be a string of words. Similar to how you may prompt a person as a starting point for writing an essay, you can use prompts to teach an AI model to produce the desired results when given a specific task.
Reinforcement Learning with Human Feedback (RLHF)
According to Scale AI in their blog, Why is ChatGPT so good?, instead of simply predicting the next word(s), large language models (LLMs) can now follow human instructions and provide useful responses. These advancements are made possible by fine-tuning them with specialized instruction datasets and a technique called reinforcement learning with human feedback (RLHF). Similarly, according to Hugging Face, RLHF (aka RL from human preferences) uses methods from reinforcement learning to directly optimize LLMs with human feedback. RLHF has enabled language models to begin to align a model trained on a geprogneral corpus of text data to that of complex human values.
Ready for More?
Mastered all the terminology, ready for more? Here are some additional generative AI terms for you to learn:
- Alignment (AI Alignment, Aligned AI)
- Attention (Self-attention)
- Foundational Model
- Next-Token Predictors
- Reinforcement Learning (RL)
- Self-Supervised Learning (SSL)
🔔 To keep up with future content, follow Gary Stafford on LinkedIn.
This blog represents my viewpoints and not those of my employer, Amazon Web Services (AWS). All product names, logos, and brands are the property of their respective owners.