AI, LLMs & OpenAI

[00:00:00]
>> We're going to talk about ChatGPT. ChatGPT is actually a web app, right? It's a web app that it's based on or is using something known as a GPT model. GPT stands for generative pre-trained transformer. Okay, what's that? Well, that's actually one specific model of a set of models known as large language models, LLMs.

[00:00:29]
So every time we see LLMs, we are talking about large language models. So these are AI models that were trained for natural language processing. The idea is they were trained with text, but trillions of text. And they know how text work. Again, I'm not going to explain here how these models work, but the basic idea to take this for something really, really, really basic.

[00:01:03]
Think about when you have a smart keyboard on your phone, that we have a prediction of the next word. I mean, you're typing, hello, how, and your phone will say R, it will appear as a recommendation, as a suggestion, and then you, how are you? Because that's the most probably two words that comes after hello, how.

[00:01:28]
LLMs are kind of that with a million asteroids. That's kind of the idea. So LLMs knows, statistically, which words comes after other words. That's kind of the idea. That's the whole idea behind this. The problem is that, not the problem, the situation is that we the community, even the developers, the scientists working with GPT, they realize that it's not just words, that at one point, it feels like behind the scenes, there is some kind of reasoning.

[00:02:05]
I'm not going to discuss here if it's true or not, but we feel when we talk with ChatGPT, that sometimes it feels its reasoning. So there is, I don't wanna say intelligence, but we have some sparks of something. And that's why everyone is excited today about LLMs and ChatGPT.

[00:02:26]
And by the way, we have LLMs today in the market from other companies as well. We have LLaMA from Meta. We have from Google. We have Lambda, that's the one that we don't see, it's an internal. But then we have Google Bard that is a product using Lambda, and many other therapy are in from other companies as well.

[00:02:48]
But ChatGPT was the first public one, the first one that was available to the public. It seems to be still the best one out there. So LLMs are actually part of the deep learning umbrella. And the deep learning umbrella is part of the machine learning umbrella. That is actually what we typically say or typically know today as AI or artificial intelligence.

[00:03:13]
Just to mention different concepts and how they apply here. So AI, we have one kind of AI known as machine learning. And there is one kind of machine learning known as deep learning. And within deep learning we have LLMs and we have GPTs, and then we have ChatGPT as one product of that.

[00:03:33]
And ChatGPT is actually a product, a website, an app from a company known as OpenAI. OpenAI is not really open, these days, at least. It's the company behind ChatGPT. And it's not the only product they have. They have other things. For example, if you have heard about DALL-E, this was also one of the first public artificial intelligence image generator, in this case, DALL-E is not an LLM.

[00:04:06]
It's a different kind of machine learning model. And OpenAI is also offering to the community, the development community, and app community, APIs. We will focus there actually, not really over ChatGPT but on APIs, APIs that are also using GPT, so they're using the same GPT model as ChatGPT.

[00:04:33]
So when you go to ChatGPT, you're using a GPT model, and we're going to use an API from the same company that is also using the GPT model. So that's how we are going to connect to ChatGPT most of the time. And just to finish that empty part of the slide that we have there, OpenAI has partnered with Microsoft.

[00:04:59]
So Microsoft is not owning OpenAI, but there is a really strong partnership between both companies. And Microsoft is also offering Bing Chat that it's now integrated in many places on bing.com but also on Microsoft Edge as a tab. And also, Microsoft is integrating LLMs in many softwares such as Office or Visual Studio.

[00:05:26]
And Microsoft is also offering Azure OpenAI APIs. So if you're used to a Microsoft APIs on Azure, you can also use GPT, so the same source, the same model as ChatGPT, but using the Microsoft infrastructure and using the Microsoft business contract over there. Today, we will be focusing here.

[00:05:51]
So we are going to mostly use APIs that are using the GPT model from OpenAI. That is the same model that is using ChatGPT. Makes sense? That's kind of the idea. That's the architecture. So that's where we're going to focus. Azure OpenAI APIs are also the same thing.

[00:06:12]
So if you learn how to use directly API from OpenAI, you can move to Azure at any time. But as we are doing this video, Azure OpenAI APIs are still in beta, and closed, so you need to apply to get access to them. So we're going to use the one from OpenAI that they are open to everyone.

[00:06:33]
Also, we will talk a little bit about something on top of ChatGPT. It's known as plugins. Plugins for ChatGPT are also something that is not yet available to every user. But it's another way that we have as web developers, or as service providers, or content providers, to integrate into the ChatGPT ecosystem.

[00:06:59]
So large language models can be used for several tasks without changing or training models. So before LLMs were actually so popular, in the deep learning environment, we have to select one particular model that we need for something in particular. We wanna categorize products, we had a model for that.

[00:07:23]
We wanna translate, we had a model. And most of the time we were able to train our own models, okay? That's part of deep learning. But now we realize that we can use LLMs such as GPT for a lot of tasks without the need to change the model or to train any model.

[00:07:45]
That's why everyone is excited about LLMs. We found that it's very useful for a lot of tasks.