What Can AI Do For You?

So what can AI do for you specifically? It can help make you more productive. It can help make you more creative. This is actually sometimes the opposite of what people think.

They're like, oh, it's going to take all the jobs from creative people. Well, it's also going to empower creative people as well. And it can enable you to do things that are beyond what you are capable of doing because it has knowledge that you don't.

But maybe you know enough that you can work with it. So think of this as your assistant. Think of it as your digital assistant who will work as much as you ask it to work.

And your subscription is kind of way cheaper than a physical person would be. This is your digital person. It's called ChatGPT.

And in fact, they just got chat.com. And when you go to chat.com, it redirects to ChatGPT.com. So I don't know in the future if they'll switch it over to redirect because they just paid like millions of dollars or something for chat.com. We'll see. They might just actually end up switching to chat.com. But the chat part is you're chatting with a digital assistant. So think of it as you're having a conversation.

And that's the way we want to start to think about this. So first, I want to take a look at some of the technology that is behind the scenes because I think by understanding that, you can actually understand how ChatGPT works. And so you can work with it better and you understand why it has certain limitations.

And then we're going to dive in and this would be very practical, hands-on class where we can see how all of this works, of course. We also want to be familiar with companies and terminology and things in the AI industry. So ChatGPT is the product.

OpenAI is the company. So this company called OpenAI created this product called ChatGPT. And it's just this kind of virtual assistant who you're going to chat with.

It's a chat bot. So what is the GPT part of this? That's the confusing part. The chat part makes sense.

So I'm going to chat. If you've ever done a text chat with somebody, you probably texted somebody, you've had a text chat with a person, but now you're going to be text chatting with an AI, a digital person, essentially. The GPT stands for Generative Pre-trained Transformer.

Generative. So it's going to generate text responses, right? So you could send a message to a computer, but in the past, it didn't message you back. So you could talk to your computer, but it wouldn't talk back.

Well, now with AI, you can actually talk to your computer and AI will talk back. And actually, it's not just text chatting. You can also do voice chat.

If you want to speak to it instead of typing, you can also do that as well. The pre-trained transformer part comes in because it's being pre-trained. So just imagine a digital person.

And so it's like this digital person, they love to read and they love to write. So how do you get knowledge? You read. So you read books, you read articles, you read websites.

That's the pre-training. So they put a lot of knowledge in and they said, let's train you in advance so that you come with some knowledge. And then when people talk to you, you can use that knowledge in responding to them.

And so that's the pre-trained part. The generative, and the reason we call this class the intro to generative AI is because that's the new thing is generative AI, where there was artificial intelligence in some ways before, like a calculator is intelligent in some ways, like you can type in mathematical equations and it can answer them. That's a form of intelligence.

Now it's not what I would call artificial intelligence per se, but where does artificial intelligence begin and end? Anything that's somewhat intelligent, like, hey, this thing can do math, that's intelligence. But the generative part of where you can talk to it and they can generate responses, it actually behind the scenes, it kind of came out from, you know, when you're typing on your keyboard, on your phone, and there's some auto suggestions, sometimes they're right and sometimes they're wrong. So they were just trying to work on making that better.

And imagine if it keeps generating word after word after word after word, and it just keeps doing that, in theory, you could speak. Everything you say is just putting words after each other, right? So that's kind of what this is doing. It's just looking at a lot of information and looking at that from a predictive standpoint of what words come after what words, right? Like, you know, they probably don't say, like, yummy ketchup chocolate, right? That's not something they've probably seen before, but, you know, they could say, like, chocolate covered strawberry, right? They could predict better based on all of that training.

So really it's guessing what comes next. In some ways, does it really know anything? Not really, in a way, because they've just looked at a vast knowledge of stuff to say, look for patterns, try to understand this stuff, and then try to regurgitate it. But it's interesting, because after I started learning about this stuff, my daughter's young, she's eight years old, but maybe a couple years ago, it was like one or two years ago, she was saying a sentence.

And for her, she loves to talk. She was very gifted with early speech. And so she was, but she was using a word that for, even for her, I'm like, that's kind of an adult level word.

I wouldn't think that would come out of your mouth. It wasn't a bad word or anything. I forget what the word was, but it was, it was beyond her years.

I'm like, but she used it perfectly. And I asked her, I said, because I was surprised she said that word. I'm like, how does she know that word? Like, you know, and I was like, what does that word mean? She goes, I don't know.

And I was like, how did you know how to use it properly? I don't know. Why did you say it? I don't know. The response was always, I don't know.

But yet she used the perfect word in the perfect way. Obviously she heard it somewhere, right. But you know, so, but that's a human doing that.

And so in a way she heard something, she said it because she's heard it. And how many of us repeat things we've heard without really understanding what we're talking about? There's a lot of people who do that, right? So like it gets very meta of like, well, how intelligent are we? How intelligent are computers? What's, you know, and, and the thing is lots of human people can think they're right because they've regurgitating something that they've heard and the computers are kind of the same way. So can ChatGPT be wrong? Of course, it's just, it's pre-trained on some stuff and trying to guess kind of what word comes next.

If it was trained on bad data, it could, and if it didn't understand it, because it doesn't really understand it, you know, it could, it could put out false things. We can't trust it for being a hundred percent correct, but also it has read a lot of information. It does have a lot of knowledge, so it can also be very good.

And it can know; it has vastly more memory than we as humans have. So that's, so it can know in theory a lot more than us. So behind the scenes, basically ChatGPT will learn from reading.

They can put a lot of stuff. So it can't generate any responses until, first of all, they go through this big library of text. They start to understand how words, sentences are put together.

Also, one of the things it does is there's something called context. So with context, it tries to understand the whole thing. So it doesn't just only predict the next word.

It predicts the next words based on kind of context situation. Like they always say, don't take what I said out of context, right? Because the meaning can change. So it has to understand in what context are you talking about that? And also when you think about context, context is like given a certain situation, the more you know about the situation, the better your response can be, right? If it knows more and it can remember more stuff, then it can give a better response.

So there are memory limitations as far as how much context or how much information we can feed it. And so depending on your account, I'll talk more about this later, some depending on the level of what you pay for, you can have bigger context or smaller context. So how much information can you give it? The more you can give it, the more it can do more accurately for you.

And so it looks at all that information and tries to predict language patterns, trying to understand what's being said, but it's kind of really predicting the next word over and over again. It also needs feedback because it doesn't really know what it's doing per se, but through feedback, it can say, oh, well, okay, you tend to like this or you tend to like this. We can give it directions.

We can both do that ourselves in ChatGPT. We can give it feedback and it can learn from that. But also the makers of ChatGPT can also give it feedback.

So during the training, they can train it to say these are good responses and it can learn over time with that feedback. So in short, it's fed a lot of information. It reads a lot of information.

It tries to understand the context of all of what you're saying. It starts predicting words and putting them together to give you a response and generates a response. So that's what it's doing.

You might have heard or might hear of LLMs or large language models. Just think about what we talked about, large language models. So they take a large amount of text and they train on this and they create a model that is how they're going to predict and how they're going to understand and how they're going to generate these responses.

So models are the training and the way that that training affects the artificial intelligence. And this generative AI is trying to mimic human language and human communication because that's the way it would work. But the more they train on, the more it can know.

So the interesting thing about these large language models and the innovation that was there is that they could feed a lot of this stuff in and they basically made algorithms where it would train itself. So they were making connections. We weren't just telling it how to do everything.

It wasn't like, well, if it's this, do this. We were just feeding a lot of information and they would use probability to make connections. And so being able to recognize things by itself without our need for human guidance.

Now, can we guide it? Can we give it feedback? Sure. But a lot of this training can be done without humans actually doing the training where we just give it a lot of stuff, let it go through, let it crunch the stuff and where it can learn in a way by what we give it. That was the big innovation in AI.

So OpenAI is what created or the company that created ChatGPT, but Microsoft invested into OpenAI. And so they can commercially license their products and create their own products based on OpenAI technology. So you might've heard of Microsoft Copilot.

That is OpenAI ChatGPT branded as Microsoft Copilot. Microsoft and OpenAI are the companies, ChatGPT and Copilot are the products. So there's pros and cons about using Copilot, but the big difference here is that when you're using Copilot, you don't just only have the chat bot, but you also have it integrated into all the Microsoft apps.

When you're in Excel, you can be in Excel chatting to the chat bot in Excel to tell it to do things in Excel. You can be in Outlook writing an email and saying, write my email in Outlook. You don't have to do it separately in ChatGPT and then go to Outlook.

Also it can look at all your emails if you want to say, okay, look at these emails and summarize these emails. And you're in Outlook doing it instead of having to copy and paste into ChatGPT. Now, if that might sound like, oh, I should definitely use Copilot.

It depends. There are some pros and cons. With OpenAI, you're getting the latest OpenAI technology right away in ChatGPT.

So there's models that are only included in ChatGPT that are not yet made available to Copilot. Microsoft could take all of those things and bring them into Copilot, but just think OpenAI creates the technology first. They then have to be later integrated into Copilot if Microsoft wants to, but Microsoft has different approaches to things.

They want more integration into Office. If you're a deep Office user with the Microsoft 365 apps, Word, Excel, PowerPoint, Teams, all that sort of stuff, then maybe Copilot is a better choice. And what we learned fundamentally by learning how to use ChatGPT, you can apply it equally well to Copilot.

When you're learning one, it's kind of like you're learning the other one on a conceptual level. But if you want the latest technology, ChatGPT is going to give you the latest technology because it's from the people who are making the technology. But Microsoft, I've realized their vision because they're integrating Copilot into everything they offer.

They're trying to create, like they always say Apple has a walled garden around their products. Like once you use one, you want to use another, you want to use another. Microsoft's walled garden is AI.

It's with Copilot because you have to use it to get the full advantage of Copilot. You have to use Word and Excel and PowerPoint and Outlook and Teams. For example, you can't use Gmail.

You'd have to use their Outlook. You'd have to use Teams instead of Slack. You'd have to use OneDrive by Microsoft instead of Dropbox.

So they're trying to say, hey, don't use these other services. Use ours because you get Copilot in. So they're trying to get more people to use their services because by embedding Copilot into all their services, they switch from Slack to Teams, switch from Zoom to Teams.

So don't use these other third parties. Use ours. More Copilot for you.

Come over to our side. So that's how they're trying to create their walled garden of stuff. Microsoft does not own OpenAI.

They did invest, but they are a separate company that operates independently. There have been some rumors that they might at some point try to acquire them. That would be interesting to see, but who knows exactly what's happening.

The term OpenAI was originally, it was supposed to be this open thing, like open source, benefit the world thing. And now they realize, oh, there's lots of money. So they've kind of abandoned that side of things.

And some people have said that that's bad for them to do, but you know, it's now a commercial enterprise basically. So the open part of OpenAI is kind of closed now. But Apple actually struck a deal with them so that in the new macOS and iOS, now you can also throw stuff over to ChatGPT if you want.

So Apple's building an integration that if you want to talk to Siri, for example, they have their Apple intelligence, their Apple AI. But if it doesn't do what you want, you can say, oh, send this over to ChatGPT. And then they'll kick you over to ChatGPT.

You can actually even sign in with your paid account into macOS or iOS. So Apple's also integrating ChatGPT into their operating system too. We'll see, they're just starting to do that now.

We're going to see. Everybody is building AI into all of their products. For example, this was the first time in a long time that Microsoft added a button to the keyboard for Copilot.

Dedicated, now that's only on the new computers, like new laptops that we don't have it here. But the new laptops, I guess, new keyboards you buy, they're going to have a Copilot key. There's also a Copilot button on the latest version of Windows.

So in the task bar at the bottom. So they're infusing Copilot into everything they do everywhere, their web browser, their search engine, everywhere it is, basically with Microsoft putting that in there. And they want to make their $10 billion outlay back.

So they invested and they got the use to be able to make commercial products based on OpenAI technology. We're only going to be seeing more and more and more of this. So it's just a matter of what can you do with it?