Grok has introduced a new algorithmic recommendation system for platform X, which it describes as one of the most "AI-native" social recommendation systems available. The system utilizes large models to directly learn user engagement patterns and historical interactions, creating user profiles. It predicts various interaction probabilities, such as likes, comments, shares, and reports, to improve sensitivity to negative behaviors and low-quality content. This approach aims to reduce human bias and lower the threshold for engagement, while diversity constraints prevent content saturation by single authors. However, the system faces challenges with rapid Transformer learning, which could lead to "information cocoon 2.0" risks and repetitive content distribution.