Introduction

[00:00:00]
>> Steve Kinney: Hello everyone, my name is Steve, and we're gonna talk about AI today. But we're gonna talk about AI a little bit differently than I think a lot of people, including myself a few days ago, are talking about AI, where a lot of the zeitgeist around artificial intelligence right now involves going into one's favorite chatbot and being impressed.

[00:00:25]
Or maybe pointing said chatbot at their own code base or maybe going in there and saying, draw me a picture of a very, very angry hamburger or something along those lines. And that is all very good, and I do it and you do it, we all do it.

[00:00:42]
What we're going to do today is actually kind of go down to the fundamentals and start figuring out how all this stuff works kind of at the lower levels, we won't say the lowest, but lower levels where we'll do stuff like some very kind of basic text transformations.

[00:01:01]
We'll fine tune GPT2 to make it do something that we want. We will go look at how to generate our own images and then we'll look at how can we fine tune that to have it specifically generate images of, I mean, your dog. My dog. I'm going to generate pictures of my dog.

[00:01:20]
What you choose to generate is up to you and kind of actually figure out all this stuff works because there are a lot of very powerful practical applications, applications of just some of the smaller pieces. And I think what's really cool about some of this stuff is the fact that you can, there are given features.

[00:01:38]
You want to pull down an open source model and work with it to do that feature that you want as just part of your code base without all of the extra API key dancing, so on and so forth. We're going to do it today predominantly, almost exclusively in the Python programming language.

[00:01:59]
That said, for all of the text parts, we use this SDK service, Hugging Face. They have both a typescript and a Python SDK for all of the text stuff. When we get to the part of the show that is working with graphics cards and generated images, Python is the predominant game in town.

[00:02:27]
So we're gonna do it all today in Python. That said, if you're like, I don't know Python, you're gonna be okay. If you know JavaScript and you squint at Python, it's not that bad. So yeah, lots of really, really cool things that I'm really excited about today. And like I said along the way, it's not just about necessarily how do I use yet another SDK to do yet another thing but we will also kind of use these tools to find out, like, okay, I hear about Transformers more than meets the eye, but how does that work?

[00:03:03]
I know these words tokens and embeddings, I've heard these words before. I don't really know what that means. So we will by the end of this, and it'll be pretty cool. General flow of what we're gonna do today is, some of this stuff, obviously, if you're new to all of this, there are terms and words and stuff like that.

[00:03:24]
So the kind of rhythm and flow that we're gonna do is I'm gonna kind of explain a bunch of concepts conceptually with words. And then we'll do a second pass through them with actual code. That way you can kind of, like, put all the mental roadmap in place of what we're going to talk about, and then we will see the code.

[00:03:48]
Because sometimes both reading code and understanding concepts at the same time is dreadful. So we'll do it kind of two passes.