Core Viewpoint - The recent announcement by X.com regarding the open-sourcing of its new recommendation algorithm has generated significant attention, indicating a shift towards an AI-driven content recommendation system that removes many manual rules previously in place [1][3]. Group 1: Algorithm Overview - The new recommendation algorithm utilizes a Transformer architecture similar to that of xAI's Grok model, marking a departure from the previous system that relied heavily on manually designed features and rules [1][3]. - The algorithm operates through two main systems: Thunder, which ensures users do not miss updates from followed accounts, and Phoenix, which identifies content that users are likely to engage with, regardless of follower count [3][4]. Group 2: User Interaction and Content Value - The algorithm predicts user behavior based on potential interactions with content, assessing 15 different actions, including likes, replies, and shares, to determine the content's value [6][7]. - Content is filtered through nine criteria before being considered for recommendation, ensuring only relevant and unique posts are promoted [7]. Group 3: Implications for Creators - Ordinary creators are encouraged to adapt their content strategies to align with AI logic, emphasizing the importance of engaging with comments and producing high-quality, information-dense content [9][10]. - The algorithm's focus on user engagement metrics means that even accounts with fewer followers can gain visibility if their content resonates with users, while larger accounts can suffer if they generate negative feedback [9][10]. Group 4: Challenges and Considerations - The algorithm's design may inadvertently promote controversial or emotionally charged content, raising concerns about the potential for creators to prioritize engagement over authenticity [10][11]. - The ongoing updates to the algorithm every four weeks will require creators to continuously adapt their strategies, creating a dynamic and potentially volatile content environment [10].
马斯克又一大动作，AI教你制造爆款
3 6 Ke·2026-01-23 12:06