A Primer on Generative AI

Generative AI, an increasingly popular topic, remains challenging for many. As I delve into emerging technologies, I see that generative AI stands out as an exciting new technology. This article serves as a primer and introduction to help those new to generative AI, including most people.

The popularity of Generative AI has surged, underscoring its growing significance. ChatGPT boasts approximately 180.5 million users globally and has achieved rapid growth, hitting 1 million users in just five days post-launch. OpenAI reports that ChatGPT sees 100 million weekly users.

Generative AI involves artificial intelligence that generates fresh content like text, images, music, or virtual worlds. Unlike traditional AI, which focuses on pattern recognition and predictions, generative AI fosters novelty. Deep learning models power this capability, trained on extensive data to mimic human creativity nuances. These models yield remarkably realistic and creative outputs, proving invaluable in entertainment, education, and art.

Large language models (LLMs), a subset of generative AI, specialize in text generation. Models like GPT-4 are trained on vast datasets from books, articles, and websites, employing transformer neural network architecture for coherent and contextually relevant text generation. LLMs excel in human-like text understanding and production, enabling tasks like essay writing, answering queries, and engaging in conversations.

Training these models involves feeding them extensive data and employing supervised learning, where the model forecasts the next word in a sentence given the preceding words. This iterative process tunes the model's internal parameters to minimize prediction deviations from actual data. The more data and computational power available, the better the model's text understanding and generation capabilities. Despite the computational intensity and hardware requirements, this training method yields highly precision models adept at various language-related tasks.

Generative AI and large language models hold immense potential to transform multiple industries. They expedite high-quality text generation for writers, journalists, and marketers in content creation. In customer service, they furnish prompt and accurate responses to inquiries. In education, they deliver personalized tutoring and support to students. Additionally, these models aid research by summarizing vast data volumes and providing insights that might elude human analysts.

Generative AI and large language models signify a significant leap in artificial intelligence. Their capacity to produce and comprehend human-like text presents vast opportunities for innovation and efficiency across diverse sectors. With ongoing enhancements, these models are poised to become even more integral to daily life, offering valuable support and augmenting human capabilities in novel and impactful ways.

Let’s start by summarizing the primary generative AI platforms/services. Then, we will contrast generative AI with two other well-known AI applications: Google Search and virtual assistants Siri and Alexa. Finally, we will briefly examine the range of applications for generative AI.

Subsequent articles will delve into more detailed examples of several generative eAI application areas.

Most Popular AI Platforms/Services

Although generative AI has emerged quickly and surprisingly, it has been under development for many years. There are a variety of Generative AI platforms and services, but here are the five most popular.

ChatGPT - Developed by OpenAI, ChatGPT is a large language model that can understand and generate human-like text. It is based on the GPT (Generative Pre-trained Transformer) architecture. ChatGPT is highly versatile and used in various applications, including customer support, content creation, tutoring, and more. It has strong context retention in conversations, making it suitable for interactive applications.

Claude - Anthropic developed Claude, a language model designed to generate human-like text. It focuses on being safer and more aligned with human values. Claude emphasizes ethical considerations and user safety, incorporating measures to reduce harmful outputs. It is often positioned as a model with enhanced user alignment and responsiveness.

Copilot - Copilot, developed by Microsoft/GitHub in collaboration with OpenAI, has several different applications. Microsoft primarily incorporates Copilot into its ecosystem by integrating development tools and platforms, enhancing the developer experience by leveraging AI capabilities. GitHub Copilot is an AI-powered code completion tool integrated directly into GitHub’s code editors, such as Visual Studio Code. Microsoft has also introduced Copilot functionalities into its Microsoft 365 suite (formerly Office 365), embedding AI capabilities into applications like Word, Excel, and PowerPoint. This integration helps users generate text, analyze data, and create content more efficiently using AI suggestions.

Gemini - Google DeepMind's Gemini is an AI system designed to be a versatile assistant capable of performing various tasks beyond text generation, such as image recognition. Unlike other models focused solely on text, Gemini aims to integrate multiple modalities, including text, images, and possibly more. It is leveraged within Google's ecosystem of services and products.

Perplexity - Perplexity AI is a search engine and AI assistant that uses language models to provide answers and generate human-like text based on queries. It is designed to function as a conversational search engine, providing direct answers to user queries.

Although generally similar, there are some differences among them. ChatGPT and Claude are general-purpose language models. Copilot is specialized for coding assistance. Gemini aims for multimodal capabilities. Perplexity focuses on integrating search and AI assistance.

Generative AI Model Training

Training generative AI services, extensive language models (LLMs), involves several key steps and processes. The purpose of training these models is to enable them to understand and generate human-like text by learning patterns, structures, and nuances of the language from large datasets. Here’s a detailed breakdown:

1. Data Collection: The first step is to gather vast amounts of text data. This data comes from various sources, including books, articles, websites, and other written materials. The more diverse and extensive the dataset, the better the model can understand different contexts and nuances of the language.
2. Preprocessing: The collected data is then cleaned and preprocessed. This step involves removing irrelevant information, correcting errors, and converting the text into a format suitable for training. It may also include tokenization, which breaks the text into smaller units, such as words or subwords.
3. Training: The actual training involves using supervised learning designed for language models. During training, the model is fed sequences of text and learns to predict the next word in a sequence based on the previous words. This process involves adjusting the model's internal parameters to minimize the difference between its predictions and the actual text in the training data. This step is computationally intensive and requires powerful hardware.
4. Fine-Tuning: After the initial training, the model can be fine-tuned on more specific datasets to specialize in certain tasks or domains. Fine-tuning helps the model improve its performance on particular types of text or specific use cases.

The training data often includes text scraped from the internet, which provides a rich and diverse source of language data. However, the training process is conducted on powerful servers and does not continuously interact with the internet. Once the model is trained, it does not require internet access to generate responses. However, in specific applications, the model can be integrated with real-time internet access to fetch the latest information, enhance its responses, and stay updated. For example, a chatbot might use internet access to provide current news updates or weather information.

Post-training, models can be updated and fine-tuned periodically with new data to improve their accuracy and relevance. This supplementary training helps the models adapt to new language trends, emerging topics, and evolving user needs. Continuous learning and updating are crucial to maintaining the model’s performance and usefulness.

In summary, training generative AI and large language models is a complex, resource-intensive process designed to enable these models to understand and generate human-like text. The Internet plays a significant role in providing training data and can also be used post-training to keep the models updated and relevant.

The Differences Between Generative AI, Google Search, and Virtual Assistants (Siri and Alexa)

Generative AI, Google Search, and virtual assistants like Siri and Alexa are distinct technologies with unique functionalities and use cases. Here’s an explanation of their differences, along with examples to illustrate them:

Google Search

Google Search is a search engine designed to retrieve information from the web. It indexes vast web content and uses algorithms to rank and display the most relevant results based on user queries. If you search for "best Italian restaurants in New York," Google Search will list relevant restaurants, reviews, and maps. Most likely, this will include many website links to a wide range of articles and many advertisements with links. For queries like "What is the capital of France?" Google Search provides a direct answer ("Paris") at the top of the search results, and there are many links about Paris. Searching for "climate change effects" will yield a variety of articles, studies, videos, and news on the topic.

Google Search provides links to web pages, direct answers, and snippets from existing content. Google Search is optimized for finding and retrieving existing information by listing many websites with links. Google Search relies on typed or spoken queries to deliver relevant search results and direct answers.

Virtual Assistants (Siri, Alexa)

Virtual assistants like Siri (Apple) and Alexa (Amazon) are AI-powered services that assist users with various tasks using voice commands. They combine information retrieval, generative AI, and voice recognition elements to provide hands-free assistance. You can ask Siri or Alexa to set reminders, send messages, or make phone calls. For instance, "Hey Siri, remind me to call Mom at 5 PM." Virtual assistants can provide weather updates, play music, answer trivia questions, or tell jokes. For example, "Alexa, what’s the weather like today?" will provide the current weather forecast.

Virtual Assistants are designed to perform tasks and provide information based on voice commands, often integrating with other services and devices. Virtual Assistants perform actions, provide spoken responses, and interact with smart devices.

Generative AI

Generative AI is designed to create new content, such as text, images, music, or other forms of media, by learning patterns and structures from large datasets. It can produce original, coherent, and contextually relevant outputs based on its input. A generative AI model like GPT-4 can write an essay on a given topic, generate creative stories, or produce a news article based on a few keywords. Tools like DALL-E can create images from textual descriptions, such as developing a picture of "a cat playing the piano in a jazz club. A chatbot powered by generative AI can engage in human-like conversations, providing detailed and contextually appropriate responses.

Generative AI focuses on creating new content and engaging in complex, context-aware interactions. Generative AI can engage in extended, coherent conversations and produce detailed content based on prompts.

Examples

The best way to understand the differences is through a couple of examples:

Example 1: Request: What restaurants are within walking distance of the Camps- Elysée Mariotte in Paris?

• Siri provides general information on the hotel and a link to Trip Advisor. It is generally useless for this type of inquiry.? It would be helpful if you asked for restaurants near your current location.
• Google lists hundreds of website links about Parisian restaurants, including some of the best restaurants in Paris that are not located nearby. Some of these are paid advertisements. It includes a Trip Advisor look that could be helpful, but it takes work to narrow down the locations.
• Chat GPT Is by far the most useful: “Here are some great restaurants within walking distance of the Camps-élysées Mariotte in Paris: Bus Toqué?- This unique dining experience offers a five-course French dinner while being driven around Paris in a tour bus. It's an enjoyable way to see the city and savor gourmet food. The pick-up location is on the Champs-élysées, making it very convenient for a special night out?(Tripadvisor )?(The Tour Guy ). Origines?- Located on Rue de Ponthieu, this restaurant focuses on fresh ingredients and offers an intimate dining experience with a noteworthy cheese plate and tasting menu. It's a great spot for enjoying refined French cuisine in a relaxed setting?(Tripadvisor ). L'Orangerie?- This elegant restaurant features creative fish dishes and Japanese-inspired cuisine, complemented by vegetable plates. The setting is adorned with Monet paintings, enhancing the dining experience. It's located within walking distance on Avenue des Champs-élysées?(Tripadvisor ). Chez Gabrielle?- A small, cozy restaurant managed by a husband-and-wife team, Chez Gabrielle is known for its high-quality French cuisine, including standout dishes like flambéed scallops. It offers a welcoming ambiance and is perfect for a calm lunch or dinner?(The Tour Guy ). Le Gabriel?- Set within a classic hotel, this establishment serves creative French dishes in an opulent setting. The menu is concise but focuses on unique and hearty flavors, making it an excellent choice for a special occasion(Tripadvisor ). These restaurants offer a variety of dining experiences, from unique and casual to refined and elegant, all within walking distance of the Champs-élysées hotel. Enjoy your culinary adventures in Paris!”

Example 2: Request: What were George Clooney’s three most successful movies?

• Siri - Briefly lists several of his movies with no ranking or qualification.
• Google - Lists websites and links to many of his movies, such as his top 17 and 34, but it doesn’t specifically answer the question. You need to go to some number of these websites to determine the answer on your own.
• Chat GPT - Here again, Chat GPT was most specific in answering the question: “George Clooney's three most successful movies in terms of box office revenue are: 1.???? Gravity (2013)?- This science fiction thriller, co-starring Sandra Bullock, grossed approximately $723 million worldwide. The film's gripping story and impressive visual effects contributed to its success. 2.???? Ocean's Eleven (2001)?- A heist film featuring an ensemble cast, including Brad Pitt and Matt Damon. It earned about $450 million globally and became a cornerstone of a popular franchise. 3.???? Batman & Robin (1997)?- Despite being critically panned, this superhero movie grossed around $238 million worldwide, making it one of Clooney's top-grossing films. These movies highlight Clooney's versatility and ability to draw audiences across different genres.” Gemini responded with the same three movies in a table format:

Example 3: Explaining the June 6th Normandy Invasion

• Siri - a brief sentence regarding it.
• Google - Links to hundreds of websites:
• GPT-4o - Explain the June 6 Normandy Invasion, how it unfolded, initial progress, and implications:

The Normandy Invasion, commonly known as D-Day, took place on June 6, 1944, and was a pivotal operation during World War II. It marked the beginning of the Allied forces' effort to liberate Western Europe from Nazi occupation. Here’s an overview of how the invasion unfolded, its initial progress, and its implications

Planning and Preparation

The Normandy coastline was divided into five sectors:

Utah Beach (American sector)

Early Morning Hours

Establishment of beachheads to secure landing zones.

Implications

The Normandy Invasion demonstrated the Allies' ability to coordinate a complex and massive military operation. It underscored the importance of joint military efforts, planning, and cooperation among Allied nations, ultimately contributing significantly to the defeat of Nazi Germany in World War II.

Applications of Generative AI

The applications of generative AI are extensive, opening entirely new ways of learning, thinking, analyzing, working, and creating.? I can’t underestimate the product change this can have for people who learn how to harness the power of generative AI.

I’ll outline some of these here for brevity, but in future articles, I’ll focus on how to use generative AI in specific categories.

1. Writing - You can use it to write articles, essays, short stories, emails, speeches, prayers, chapters, meeting minutes, and almost anything else you want.
2. Editing -- You can use it to edit anything you write or have it rewrite anything to make it clearer or better.
3. Research - If you want to know about anything, you can ask it to do your research.
4. Investing - You can get investment advice, such as identifying the best dividend stocks and getting a complete list of analyst ratings and price targets for a specific stock.
5. Medical Advice – You can Explain the June 6 Normandy Invasion, how it unfolded, initial progress, and implications. and instantly learn about any medical condition, medications, symptoms, etc.
6. Summarization – It can summarize an article or website.
7. Image generation – It can create realistic images, artwork, and graphics.
8. Image interpretation – It can interpret an image you attach and provide informative explanations.
9. Language learning - It provides a broad set of language learning capabilities, such as translation, conversation generation, and grammatical explanations.
10. Education – The applications for education are almost endless, including preparing teaching and instructional materials, lesson plans, tests and quizzes, and complete curricula.
11. Student learning – It can provide learning tools and explanations to help students learn and solve problems.
12. Write poetry or music – It can write a customized poem or song based on what you want it to include.

Critics

Generative AI has its critics. They paint a picture of potential abuses and risks. While some of the risks have merit, they are usually driven by a failure to understand how the benefit of new technologies requires a change in thinking.

For example, critics will claim that it will enable students to cheat on writing essays. Still, they ignore that generative AI can propel learning to more advanced levels than ever anticipated.

Hopefully, this primer will get you started to explore generative AI and how its benefits.

Summary

Generative AI is a transformative technology that will profoundly affect all of us. This article is intended as a primer, and subsequent articles will provide more in-depth applications and examples.