20.43: Notes on Tiktok

ℹ️ This is not a newsletter, nor is it a weekly update one can expect to be delivered every Sunday. These are just random thoughts I came across while preparing for the episode 8 of Safareig.

TikTok’s background

The history got started with Musical.ly in 2014, in Shanghai. Alex Zhu and Luyu “Louis” Yang had launched an educational short-form video app that hadn’t gotten much traction. Then they pivoted to lip-synch music videos and launched in the United States market, where it became very popular. Mostly among teenagers that started using the app for creating and sharing song covers.

Musical.ly may as well be the first Chinese app to break the Western social and cultural barrier. However, there are only so many teenage girls in the United States. When they saturated that market, usage, and growth flatlined.

Later ByteDance cloned the idea in China with an app called Douyin. Ironically, it became more popular than its American competitor — which ended up acquiring. Under ByteDance’s ownership, Musical.ly got rebranded as TikTok.

The algorithm

TikTok is not your typical western social network. It can be considered a true mobile-first, user-generated platform. However, its approach looks more like YouTube, rather than Facebook.

Social graphs — the ones Facebook or Twitter are built on top of — are more rigid because of its path-dependency on your friends or followers. They fail to quickly iterate and refine recommendations as your tastes evolve.

It is its fundamental structure that sets TikTok apart. It is not based (and by extension not limited by) on who you follow or who your friends are. Instead of being friend-dependent, the app learns as you mindlessly scroll and engage — no need to search, friend, or share.

Looking at it from a pure algorithmic design perspective, TikTok’s will be thought of as an explore algorithm — one that tries to broaden your exposure to more than just what you’ve shown you like. While YouTube’s (exploit algorithm) will give you more of what you like.

Yet any AI researcher would agree that TikTok’s algorithm has no special “sauce”. What makes TikTok special is the ability to combine its algorithm with the data on which it is being trained. There is not much data available about user preferences, but TikTok’s algorithm can quickly create it and act on it.

The algorithm excels at incentivizing the user to create its own training data through a closed-loop. Unlike western social networks (overwhelmed by a myriad of CTAs) TikTok is presenting one video at a time — and learning from it. It is tracking each interaction with each video, which gives an unambiguous data point to nurture the algorithm.

TikTok helps its algorithm (through human-assisted tagging and user feedback) “see” the video.

The creator

TikTok makes, what used to be pro workflows, very accessible and easy to use. Things like effects, editing music… are effortlessly available to creators. A lot of people ask why YouTube hasn’t already tackled this opportunity. This is something we are quite familiar with at Gamestry. YouTube has become so vast that it can’t create vertical experiences that fit each use case on the platform.

Network effects on creativity: how each creator makes others more creative.

• Bypassing the blank canvas: start by remixing others’ ideas, make duets, re-use other videos as “components”.
• Distribution: instead of creating content “in the dark”, know how to be promoted through featured challenges or trending hashtags.

Update: I wrote an entire post about how to retain and help creators thrive within your product.