Reddit has implemented the use of Large Language Models (LLMs) in its moderation systems to detect and prevent approximately 23 million daily spam posts from reaching communities. This update aims to combat fake accounts, organized campaigns, and other violations, such as hate speech, acting preventively rather than relying solely on reports or manual removals after the material has spread.
Improvements in User Experience
With these new tools, the platform reported a 20% decrease in user exposure to spam. Furthermore, exposure to harmful posts was reduced by 40%, and the content removal process now occurs in less than five seconds. There was also a 40% drop in false positives, meaning instances where legitimate content was incorrectly removed.
How the Moderation System Works
Reddit's moderation system maintains a hybrid approach, combining artificial intelligence with human intervention, relying on administrators, reviewers, and volunteer moderators in each community. Users themselves actively participate through voting mechanisms (upvote and downvote), which help decrease the visibility of low-quality posts—an aspect also monitored by automated systems.