Vox, in collaboration with The Pudding, looked at what happens when a song goes viral on TikTok. It heads down the TikTok-to-Spotify pipeline, which signals money to be made and draws labels to take advantage.
Tag Archives: TikTok
Ben Smith for The New York Times got an internal document that outlines TikTok’s recommendation system. This quote caught my eye:
Julian McAuley, a professor of computer science at the University of California San Diego, who also reviewed the document, said in an email that the paper was short on detail about how exactly TikTok does its predictions, but that the description of its recommendation engine is “totally reasonable, but traditional stuff.” The company’s edge, he said, comes from combining machine learning with “fantastic volumes of data, highly engaged users, and a setting where users are amenable to consuming algorithmically recommended content (think how few other settings have all of these characteristics!). Not some algorithmic magic.”
The Wall Street Journal tested out the TikTok algorithm with bots to see how quickly the app converged towards a user’s pre-specified interests. As viewing time of videos as the main signal, and to nobody’s surprise (I think), it only took a couple of hours for TikTok to narrow down interests.
This is how most social services work these days? The concerning part is that almost all TikTok videos are served based on the algorithm, which makes it easy to fall into terrible rabbit holes.
Chris Ume, with the help of Tom Cruise impersonator Miles Fisher, created highly believable deepfakes of Tom Cruise and posted the videos to TikTok. Ume showed the breakdown of the arduous process of training the A.I. model and editing each frame.
The Verge talked to Ume more about the process:
“You can’t do it by just pressing a button,” says Ume. “That’s important, that’s a message I want to tell people.” Each clip took weeks of work, he says, using the open-source DeepFaceLab algorithm as well as established video editing tools. “By combining traditional CGI and VFX with deepfakes, it makes it better. I make sure you don’t see any of the glitches.”
The results are both entertaining and worrisome.
Eugene Wei looks closer at the algorithms that drive TikTok and how its design provided an effective feedback loop:
But for TikTok (or Douyin, its Chinese clone), who needed an algorithm that would excel at recommending short videos to viewers, no such massive publicly available training dataset existed. Where could you find short videos of memes, kids dancing and lip synching, pets looking adorable, influencers pushing brands, soldiers running through obstacle courses, kids impersonating brands, and on and on? Even if you had such videos, where could you find comparable data on how the general population felt about such videos? Outside of Musical.ly’s dataset, which consisted mostly of teen girls in the U.S. lip synching to each other, such data didn’t exist.
In a unique sort of chicken and egg problem, the very types of video that TikTok’s algorithm needed to train on weren’t easy to create without the app’s camera tools and filters, licensed music clips, etc.
At first I was confused by TikTok. I’m still confused by TikTok. But one thing that is for sure is that the system knows how to serve up videos that one might find interesting. Whether that’s good in the long run is anyone’s guess.