tick tock It’s a trending app. Every week (or every day) new trends emerge and are joined by thousands of users of the platform. Wes Anderson A movie, another movie about “low vibration” things, or another movie that rejuvenates us (through a filter) and shows us our teenage selves.
And by “looks,” the creators of artificial intelligence are well aware of these fads. Why? Because, thanks to these fads, the developer can “train” his AI. from facial expressions and movements people’s appear in their videos.
If delivered today, news site vox (not to be confused with the Spanish political party) analyzed in a recent video how AI developers used Hundreds of TikTok videos of users dancing or popular mannequin challenge (a trend that became very famous a few years ago), when artificial intelligence learned to seeI’ll show you how.
How do you train your AI to look like it?
When training an artificial intelligence, it needs to assimilate thousands of different datasets. Depending on the type of artificial language model, Input from a specific type of dataset. For example, the text AI Chat GPT The AI is fed all kinds of text, and the image AI is fed various images.
A similar thing happens with video. Their training requires analyzing dozens, hundreds or even thousands of videos. find a set of “midpoints” AI detects patterns and allows processing power to better “understand” this aspect of our reality. But not everything is as easy as it sounds.
A video is a two-dimensional element starting from a three-dimensional reality. Humans are able to interpret these dimensions when watching a video, but AI lacks this basic ability and must start with some basic concepts such as: Interpret (physically) what a human being is and understand the space he or she is in.
Yasamin JafarianA researcher at the University of Minnesota. render peopleprovides valuable data to power AI. interpreting human spaces.
But despite the great value of this data, AI needs to be trained in a wide variety of contexts. For videos, AI requires different backgrounds, different movements of the people in it, different poses, etc. And there’s no better place to find them all than TikTok.
600 TikTok videos Used by researchers to train AI. These videos allow us to see a huge variety of people with completely different clothing, lighting, backgrounds, movements and shapes, providing artificial intelligence with a very diverse data set.
However, in order to teach AI the depth of three-dimensional space, researchers need not only people in motion, Also made use of completely static peopleRemember that internet trend that had everyone sitting still? Yes, they also used the famous Mannequin Challenge video.
from how much 2,000 completely different videos of “Frozen People” doing the Mannequin ChallengeAI developers create large datasets of videos containing static people with cameras moving around them, giving the AI the data it needs to “triangulate” their positions, Allow them to know the three-dimensional space they are in.
AI, which has the ability to “see” and know the space in which the elements of an image are arranged, You can start to “guess” what the frames of the video will look like If a person moves a few centimeters, jumps or lies on the ground.
Obviously, this is just the beginning and these types of AI require more extensive training, but their usefulness is amazing. Elements as mundane and ordinary as the videos you upload to TikTokMany consider it the great revolution of the last decade.
Some of the links added to articles are part of affiliate campaigns and may represent Softonic’s interests.
