It’s one of the most common little annoyances of modern life: You collapse on the couch at the end of the day, hoping to finally get a few minutes to watch one of the dozens of great shows and movies that were made accessible to you thanks to the heyday of TV and the advent of streaming, and you start scrolling. Instead of actually watching something, you open an app and spend an endless night aimlessly scrolling through endless rows of identical tiles. Eventually you give up and end up rewatching The Office.
In this episode of The Vergecast, we explore why TV and movie recommendations are so complicated, and whether AI can do it better. If Spotify can create an infinite playlist of your favorite music, and YouTube and TikTok always have the perfect one, why can’t Netflix, Hulu, or Max do it better?
As it turns out, AI can help at least a little bit. Models from OpenAI, Google, and others are ingesting vast amounts of information about movies and shows (not just titles and genres, but also summaries, reviews, summaries, and more, collected from across the web) and can then synthesize that information to find connections between titles that were previously hard to find. And as the context window gets larger, these models can actually ingest and understand an entire movie at once, opening up entirely new ways of understanding movies.
But at the end of the day, recommendations are a human problem. Because, you know, we’re all human. What you want to watch, and why you like what you like, is much more complex and much more diverse than even the best models can understand. As a result, the idea that you can sit down, open Netflix, and have a perfect title pop up right away isn’t going to happen anytime soon. So instead of hoping for the best, investigate how you can use AI tools right now to get to your content at least a little faster. Because, you know, watching movies is great, but scrolling through too many movies is really overrated.
If you want to learn more about what we discussed in this episode, start with these links:
