VIDEO TRANSCRIPT
There's a lot of information out there about AI, but if you just stay with me for this slide, you'll have a mental model that will make it easy for you to interpret everything that comes up. There's so much noise on LinkedIn and a lot of hype and a lot of terms, but if you just hang on here, you should have a good mental model for everything.
So the first layer here is the type of content. It's text, code, images, video, audio, 3D, and others. Others are things like biology, different types of molecular processing, a whole new world in there as well, very specialized. And the rest are getting much better, particularly with video, which saw a massive surge recently. Not so relevant for the typical business, but definitely for multimedia production.
So you have your type of content. The next is the models themselves, and this is the secret source. This is what makes all of this possible. These are the brains of making AI happen. So we have the most common one known as GPT models, so GPT-3.5 and 4. But we've also got Mentors, EleutherAI, Gemini Pro, and many others.
We have that on tech code, their variances on images, video, and audio. And then what each of these models has is the underlying content that they've been trained on. So GPT-4 is an example. It's a very large web scrape. So if it's on the internet, and it may or may not be under copyright, there's a lot of things in there that are actually in the courts at the moment, saying it's a breach of copyright.
That's the first part. The second part is books out of copyright. So all the books out of copyright are part of this big scrape that GPT has. But then there's also licensed content as well. So they're in license deals around specific trials of content. And so all of these models have that underlying data.
But all of this is actually very useless without an application layer.
So the application layer is where the user uses it. So I don't use the underlying model. That's just a whole big complicated series of vectors, vector databases.
I use an application, and the application most of us think of is ChatT, which is really just a chatbot on top of this underlying model.
Microsoft CoPilot, Grammarly.
But also we'll start seeing it embedded into emails, into Canva audios.
Chat and email support marketing copy.
Things like GitHub CoPilot for coding, MidJourney, Dolby 3 for image generation. And so while we might see this explosion of applications, and there literally are hundreds of AI startups in Silicon Valley right now, they're usually using the same set of underlying models, which is typically GPT-4, things like Codex, and a couple of others.
But that's fundamentally how they all work. Now there's one secret sauce that will make anybody win this race, which is actually quite commodified. So these underlying models we're all using and you might embed it as a different feature in the system, but there's one thing that can give you a superpower that none of the competitors have. You might have your own data.
So you might have your own knowledge base. You might have your own codebase, your own style guide, your own design stock, all your call recordings, your CAD files, your financial models. If you're a pet shop and you have a whole lot of pet sales information, and you also have a whole lot of information about what content people like interacting with on your website.
You combine that with the large language models. All of a sudden, you have a pet personalization engine that somebody else wouldn't be able to do without your unique information.
This is one of the arguments for Google being a powerhouse, incumbent for AI. Because even though we've been talking about ChatGPT and OpenAI, Google has Gmail.
It has your calendar. It has your web browsing activity.
It has ads. It has all of these different footprints, as well as a lot of Android information as well, what it can get access to. So the idea that you're able to fly somewhere, and it's automatically reserving restaurants for you based on where you've been in the past. It has all your map information as well. It's very, absolutely within the technical realms that will be able to take so much action and give you so much personalization based on that wealth of information. And it's one of the reasons why many are thinking that Google's stack is basically going to supersede OpenAI's very quickly and be a major competitor also for Microsoft, which is very aligned to OpenAI.