@vrose.eth

The more I work with the Farcaster code base the more convinced I am that that top reason was a bad spam model. This model marks most people, including me, spam while a select, mostly hand-picked, few are designated "high quality" and were boosted in the feed. The boosting of those select few contributed to the "cabal effect" that many users could sense on the timeline. Over the past week, I've replaced the use of the old spam label with the Neynar score in the home feed algos, bit by bit. As of today, home feed algos no longer use the spam label at all. Engagement and diversity of authorship in the home feed have grown as a clear result of swapping the model.

23 replies

12 recasts

79 reactions