Thread by @WillOremus

We hear a lot about the magic of TikTok's "for you" algorithm. This brilliantly done @WSJ video is a great introduction to how it actually works. Like most algorithms, the answer involves a lot less magic than it might seem. wsj.com/video/series/i…

@WSJ A few quick thoughts: First, the algorithm isn't quite as "secretive" as @wsj makes it out to be. In fact, a lot of the mechanics addressed in the video were laid out by TikTok itself in a blog post last year: newsroom.tiktok.com/en-us/how-tikt…

@WSJ Second, even as @wsj successfully deconstructs and demystifies the algorithm, its focus on depression-related content risks making it seem a bit more sinister than it typically is. While there are caveats, one could come away thinking TikTok is mostly MAGA and suicide content.

That said, I think the WSJ is right to use the video to highlight the rabbit-hole problem--which is common to many ML-based recommendation algorithms--and in particular, how the content becomes less moderated the deeper you go down a given hole.

All in all, a really good primer and highly recommended--maybe especially as a companion piece and corrective to the TikTok blog post, bc it calls into question TikTok's assurances that its algo effectively disrupts repetitive viewing patterns and guides users to diverse content.

This gets at probably my main concern with the (again, overall quite worthwhile) WSJ investigation. Is it possible that it found such deep “rabbit holes” not just bc of TikTok’s algo, but because the WSJ’s own bots were programmed to be singleminded rabbits?

https://twitter.com/stokel/status/1417857524657688578

Some additional good thoughts here: Rabbit holes can be sinister, but they can also be part of TikTok's fun & value--both in facilitating niche interest-based or identity-based communities, and in encouraging meme culture.