Hacker News Reader

Comment by notyourwork | Hacker News Reader

Their algo is super reactive at times with the right threshold of tuning for expanding/exploring new content. It’s also great at leaning back into familiar stuff.

Overall it’s a really well done ML model.

View on HN

I actually worked briefly with an engineer who claimed to have seen the inner workings of the TikTok algorithm when she was consulting for them (she was a software engineer and mathematician), and I was very surprised to hear her say that there was no ML involved in the algorithm! Of course, I'm not sure I believe her, but it definitely raises the question of whether an algorithm of TikTok's caliber is even possible without ML.

Hand crafted recommendation algorithms are on par with current ML offerings if not better - but they take much much longer to develop.

Each video is node. Figure out where to add edges to existing content, graph theory some shit and tune for similar content maximizing viewing time of short video clips. No ML necessary. Of course, given the magic ML sometimes seems to be, they could certainly be using it but the inputs and outputs are remarkably simple if you get the data structures right. To be clear, I'm not saying it's easy, just that ML's not necessary.

It's all if-then statements, like "If this video has a bird in it, then show it to people who like birds." No ML needed /s

You jest, but at this point YouTube’s recommendation system sees even less complicated than this. Just random videos from years ago will one day become blessed by the algorithm and receive millions of views.

YTs algorithm now just seems to amplify very, very small fluctuations in view counts. 3 people just happened to watch the same video at the same time? It must be trending! Shove it in front of everyone. It’s like Google has used all of machine learning to recreate taking a derivative.

Reply on news.ycombinator.com