What is Generative AI good for? One of the reasons that question is a bit hard to answer is because AI is a general-purpose technology. Unlike a lot of technologies like a car which is good for transportation or microwave oven good for heating food, AI isn't useful just for one thing. It's useful for a lot of things and that almost makes it harder to talk about. But let's take a look at what a general purpose technology really means. Similar to electricity, AI is useful for many tasks. If we were to ask you, what is electricity good for or what is the Internet good for? These are other general-purpose technologies and it's almost difficult to think what is electricity good for because it's so pervasive and it's used around us for so many different things. In fact, as you saw earlier, supervised learning is useful for many tasks like spam filtering, advertising, speech recognition, and many others. Generative AI is like this too. In the last video, you saw a few of the tasks that an LLM can carry out, answering certain questions and hoping with writing for example. Let's discuss more broadly a framework for what tasks LLMs can do. First off, Generative AI generates texts. Not surprisingly, perhaps it's useful for writing. I routinely use Generative AI tools as a brainstorming companion. If you're trying to name a product, you can ask it to brainstorm some names and it comes up with some creative suggestions. LLMs can also be good at answering questions, and if you give them access to information specific to a company, they can help members of your team find information that they need. In this case, about the availability of parking at the office. In addition to writing, Generative AI is also good for what I'm going to call reading task, where you're going to give it a relatively long piece of information and have it generate a short output. For example, if you run an online shopping e-commerce company and you get a lot of different customer emails, Generative AI can read the customer emails and help you very quickly figure out is this email a complaint or not, which can be used for helping to route complaints to the appropriate department to be handled quickly. Given, I love my new llama t-shirt, the fabric is so soft, that's not a complaint. But if someone emails, I wore my Ilama t-shirt to friend's wedding and now they're mad at me for stealing the show. Well, maybe that is a complaint. But Generative AI can help you route emails to the right departments. I call this a reading task because it's looking at a relatively long piece of text, that is a customer email, and then generating a relatively short output, just yes or no, is this a complaint or not? While supervised learning can also be used for this particular task, we'll see later that Generative AI is allowing this source of reading tasks and other examples that we'll see later this week to be built much more quickly and inexpensively. Lastly, Generative AI is also used for many chatbot types of tasks. Whereas ChatGPT, and Bard, and Bing chat are general-purpose chatbots. Generative AI technology and large language models is also enabling many special-purpose chat bots to be built. In this example, here's what a chatbot might be like for taking online orders where a user can say, I like a cheeseburger for delivery and the chatbot acknowledges and puts the order through for the user. Now in talking through these tasks, I find that it's sometimes useful to distinguish between two different types of LLM-based applications. One is examples like this brainstorming one, where it could be quite natural for you to type a prompt like this into ChatGPT, or Bard, or Bing chat, or one of the other free or paid large language models on the Internet and get a result back. I'm going to call an application like this a web interface-based application. In contrast, in the example of recognizing if an email is a customer complaint, this fits more into a company's email routing workflow, and it doesn't really make sense for anyone to cut and paste customer emails one at a time into a web interface to get back answers as to which ones are actually complaint emails. This is an example of an LLM that would make sense when it's built into a larger software automation, that in this case helps with a company's automated email routing. I'm going to call this a LLM-based software application, the second writing example of answering HR questions. It turns out this also will make more sense as a software-based LLM application because it'll need access to information about your specific company's parking policy for employees, whereas a general large language model on the Internet probably doesn't have that information. We'll talk more later in this course about how this technology is built and most of the specialized chatbots will also be software-based LLM applications. In the rest of this course, I'm going to use these two symbols to distinguish between web interface use cases and software-based LLM applications. For many people, it may be easier to get started with some of the web interface use cases because you can just go to a website like ChatGPT, or Bard, or Bing and type in a prompt, and get the result back. But I think both the web interface-based applications and the software-based LLM applications are important and will be very useful for individuals and for companies. I found that the framework of writing, reading, and chatting as a useful way to think about the many different tasks that LLM a large language model can do. In the next three videos, we'll dive more deeply into many different examples of writing, reading, and chatting tasks. I hope that you'll find some of them potentially useful for your own work. I look forward to seeing you in the next video, we'll talk more about writing, and until then, I look forward to enjoy my burger.