Skip to main content

Meta employee sharing how #Threads algo works reminds me of an old dril tweet
Screenshot of Thread post and screenshot of a Twitter post below: Threads: In Large-scale Recommender Systems like Threads and Instagram, Multi-task Multi-label modeling is used predict the likelihood that you will click, reply, or share. Call it p(click), p(reply), and p(repost). At the end of this, you have something called a value model where you can adjust the weights of these probabilities to rank posts. These weights are the knobs you would use to change what people see on their feed. They always come with complex tradeoffs. Twitter: turning a big dial taht says "Racism" on it and constantly looking back at the audience for approval like a contestant on the price is right
#threads
⇧