How TikTok’s Next Video Just Got Way Smarter

If you’ve ever scrolled TikTok or Instagram Reels and felt the app somehow just ‘gets you,’ thank a quiet arms race among AI researchers. A new paper out this week proposes a smarter way for apps to guess which short video you’ll watch next.

The trick is breaking every clip into two pieces: the things that don’t move (a sunset background, a brand logo) and the things that do (a dancer’s twirl, a product unboxing). Think of it like shining light through a prism—white light splits into colors. Here, the AI splits video frames into stable “static” parts and shifting “dynamic” parts. Then it weighs those parts against your past likes and skips, so the next video feels tailor-made instead of random.

The team tested their system—called PrismRec—on four real-world datasets from two short-video platforms. It beat the current best AI by up to 22 % while also running faster and using less memory. In plain words, that means crisper recommendations without your phone slowing down.

So why should you care? Because the apps you use every day are about to feel less like slot machines and more like a friend who just knows what you’ll enjoy next.