Want to sound smart (and nerdy) about AI tech?

Want to sound smart (and nerdy) about AI tech? If you're a massive nerd likes talking and working around AI technologies, or if you're a total new to entire space and just be able to say something makes you sound smart, I you I've been working on that is making the entire office into a big master debate what one thing versus what one thing versus another. excuse me in advance because I'm about to nerd out for a little bit. Let's say you want to AI system that's looking videos and trying to understand what's in them. Let's say, for example, some sort of media company. And in said company, you have millions of hours of footage that either you film or license, you buy from someone. But in some kind of way, you want to be able to find exactly what you want very, very quickly at any given moment. There are two ways to go about this. The first is just plain metadata What I mean by that is, say we have this video and it's got two people inside it. And if I ran an AI model of this, I don't even I just wrote, on top of this to say what is in the video, it will then tell me there are people in this video. It might just say two. But the important thing is how it gets delivered. And that is a piece of information that tells this along the side of that original piece of content. Which means if the video ever gets moved from one house, I'm going to quickly little house with a big window, from one house to another, it is very easy to take piece of information and it from one to the next. Because it's just plain English or in whatever translated. But that means when you to find this exact video, you have to now find the specific keywords that are written in this document almost word for word. On the other hand, if I have a media company that's there's color in this, Let's say on the other have a media company focuses on nature documentaries. I'm going to try and draw a nice elephant here with a big body, four legs, and they have tails, right? So there's my nice little elephant documentary. If I did it in the original way, it would have said elephant it would have probably said things like nature documentary filmed in this country, would be signed and written in this extra piece of information. What's very interesting here, that's kind of smudgy, want to do something like, I want to say animals. Now animals might not be written in this piece of information, this document that comes along with video. But this other kind of that's been used to what is in the video is actually trained to something called fuzzy searching. Fuzzy searching is done process called vectorization, where you're taking each frame and you're almost marking out where features are in a frame. Then each of these pieces of information called vectors, database. It gets really like, in, yeah. When you're now looking this specific video, you're not trying to match up word word what you've written down on an electronic piece of paper. If you look for dog, if I'm going to write this backwards, might still point you elephant because if you had another video of a dog, terrible drawing right there. They might look similar enough that the vectors in this one this one kind of match vectors in this one. And you might say, Why would I want to find a dog look for an elephant? And it's a good point, use the other way, you not actually hit the words that are written in here in order to find the you're looking for. So commonly, some kind of interesting balance between the two will be most helpful for any kind of system. But here's the drawback. With vectorization, you the model that's doing same work here to do the exact decoding or reading. If you type in animal here, you need the same model to able to point you back video. If you use one model for and then trying to use a different model for the search, it's like you're talking in two completely different languages. Unlike here, where is kept in human language so it can go from house to house. Here, if you start with it's usually quite an expensive process, you're stuck you have to be 100% sure this is the model that want to keep using for longer term. Otherwise, if you try to house, you'll have to do entire process all over again. And in media companies today, that is a massive debate when it comes to AI technologies.