In recent years, one of the realizations in machine learning, at least to an outsider to the subject like myself, is just how much progress was possible just by increasing the volume of training data by several orders of magnitude. That is, even if the algorithms themselves aren’t that different than they were a few years ago, just by training them on a much larger datasets, AI researchers have achieved breakthroughs like GPT-3 (which temporarily gave tech Twitter a tantric orgasm). [...]
TikTok fascinates me because it is an example of a modern app whose design, whether by accident or, uhh, design, is optimized to feed its algorithm as much useful signal as possible. It is an exemplar of what I call algorithm-friendly design. [...]
The default UI of our largest social networks today is the infinite vertically scrolling feed (I could have easily used a screenshot of Facebook above, for example). Instead of serving you one story at a time, these apps display multiple items on screen at once. As you scroll up and past many stories, the algorithm can’t “see” which story your eyes rest on. Even if it could, if the user doesn’t press any of the feedback buttons like the Like button, is their sentiment towards that story positive or negative? The signal of user sentiment isn’t clean.
This is a very long post, and it links to some incredible long posts and a few entire books. Here's the big payoff: ML algorithms become more accurate with larger, more detailed data sets. TikTok, by placing a single post on the screen at once, is able to capture granular data about each user's interaction with the post, increasing the power of it's algorithm compared to other social media apps, which offer an infinite feed and therefore aren't always sure which post you're interacting with.
Very insightful. TikTok is a big deal whether we like it or not. (I do.)