Generative Artificial Intelligence (GAI or GenAI)

 

What are GenAI Tools?

Main Points

• GenAI tools like ChatGPT are designed to generate new content based on user inputs, leveraging vast datasets to mimic human-like text responses.
• The technology underlying GenAI involves complex pattern recognition, driven by machine learning algorithms that improve through human evaluation and training.
• Despite the advanced capabilities of GenAI, these tools do not possess real intelligence; they predict and mimic based on learned data patterns without genuine understanding or reasoning.

Video Transcript

You’ve been hearing a lot in the news about ChatGPT and other generative artificial intelligence tools, often referred to as GenAI tools. You may understand what these tools can do, but do you understand how they work?

Artificial intelligence has been around us for some time. You use AI tools all the time and don’t realize it. An AI tool is an agent that simulates human intelligence in some way. For example, weather apps use AI to determine your location and obtain forecasts for that location. Your iPhone is using AI coding to determine if your fingerprint is you. Tools like ChatGPT are generative AI…they create something new from a user’s input, or prompt.

Generative AI is not new. What is new is how useful the tools have become, coupled with free or inexpensive access. There are all kinds of tools out there. Dall-E and Stable Diffusion can create images. InVideo and HeyGen can create human-like videos. Every single day there are more tools.

ChatGPT 3.5 was the first powerful generative AI text tool made available to the public for free. It was followed by many others. In fact, we could reshoot this video almost every week! Google, another first in the Gen AI space, now has Gemini. Microsoft implemented Copilot (based off of ChatGPT), and Anthropic is making waves with Claude. There are smaller AIs with interesting characteristics like Character.ai, which pretends to be a particular person or a historical figure. Or Hume.ai, that reacts to your voice depending on the mood you are in.

So what exactly is Generative AI? There is confusion about Generative AI tools and what they are. First of all, we often refer to them as GenAI tools, AI tools, content generation tools, or just AI. Some people are spreading ideas that these tools are going to take over the world, like in the Terminator or Matrix movies. While there’s quite a bit we don’t know about AI, and we should be concerned about AI in general, ChatGPT isn’t going to bring a horde of evil robots crashing down on our planet.

So How do they work? They almost seem magical. You type a question or request and out comes an often thoughtful, well-written work tailored to what you asked for. But tools like ChatGPT currently can’t accurately reason or understand. They are simply a well-trained mimic. A mimic is something good at imitating or copying things; in this case simulating human-like text based on patterns it learned from a vast dataset of existing text.

Text-based GenAI tools absorb an incredible amount of information from the internet; articles, social media, documents, and so on. Then, they start to make patterns. Once data is consumed, the model is trained by humans. The AI begins to put together words in context based on the data it has consumed. Humans evaluate the responses and provide better responses. ChatGPT learns and remembers.

ChatGPT is taught basic syntax and then predicts words and phrases. Essentially, tools like ChatGPT are looking at the context of words it has already generated or provided in a prompt and predict what the next word or short phrase should be. For instance, if you typed a prompt that said “pharmacy,” there are words that have a strong connection to pharmacy: medicine, prescription, technician, CVS, etc. Words like "frog" or "Harry Potter" don’t have much meaning to a pharmacy, so ChatGPT would avoid using those words. ChatGPT knows which words are typically associated with other words because of all this data it absorbed. Human training adds an extra touch by fine-tuning the responses.

Since ChatGPT runs on powerful computers, it can come up with thousands of potential responses. It can compare those responses to its database and present the best-looking ones to the human trainers. Over time -- thousands and thousands of tries -- ChatGPT began to have a good idea of how to “write.” But remember, it’s not a real intelligence like you and me. It is mimicking the way humans communicate.

There was once a story that if you put a typewriter in a cage with a bunch of monkeys, given an unlimited amount of time, you would get Shakespeare. ChatGPT is like that cage of monkeys, except we don’t need forever. The computers are fast enough to write and rewrite what would take one of us years in a few minutes. Over and over, it keeps learning how words go together. Eventually, the model gets pretty good. But remember, it’s not thinking…it’s predicting and mimicking.

Consider simple reasoning. You and I both know that 2 plus 3 is 5. You can ask ChatGPT what 2 plus 3 is, but depending on how you ask (the context), you might get the right answer...or the wrong answer. Of course, ChatGPT gets basic addition right most of the time. But, can you be sure it is accurate with a more complex request?

Keeping in mind that generative AI, with all of its computing and machine learning is simply a tool that produces responses based on its training and data sets. The real expert is you.