Abstract:
Social networking site

However, the platform’s open source recommendation algorithm does mention that it will detect AI-generated content. For accounts that frequently use AI to generate posts or comments, the platform may internally mark and limit the flow of relevant accounts.
How to handle accounts and single posts:
System judgment: llm_slop_user Account level tag: SpamHighRecall Tag validity: 7 days Current limiting measures: Posts are no longer recommended to non-followers
System judgment: llm_slop_post Post-level tag: RiskyHighVizReply Tag validity: 30 days Current limiting measures: no longer send reply or mention notifications to users who do not follow the author
System judgment: gibberish_post Post-level tag: SpamHighRecall Tag validity: 30 days Current limiting measures: Posts are no longer recommended to non-followers
System judgment: fast_reply_spam_post Post-level tag: SpamHighRecall Tag validity: 30 days Current limiting measures: Posts are no longer recommended to non-followers
The current limit mainly affects those who have not followed:
The transparency tool of
However, even if it is flagged, users who have already followed it can still see these posts, and the account itself will not be automatically demoted due to this. Although shadow shielding is not mentioned in the algorithm description, this processing method of not notifying users and recommending algorithm current limiting is in line with the performance of shadow shielding.
High-reputation accounts can bypass inspections:
It is worth noting that the system whitelist, number of fans and account reputation score will also be checked before the rules are executed. Accounts that have reached a certain number of fans or reputation level may skip these pre-checks. However, the specific whitelist mechanism, account reputation score and fan number threshold are confidential and are not provided in the public algorithm.
It should be emphasized that these algorithm tags were not added recently. According to the warehouse records, they were added as early as September 4. However, they have recently attracted attention on X, and the X team also mentioned that they will manage more and more AI-manipulated accounts.
Comments