Wu said that Grok's self-evaluation indicates that the new algorithm recommendation system X is still in its adolescence, but it is one of the most 'AI-native' social recommendation systems currently available. It directly learns user attention relationships and historical interactions to generate profiles using large models, subsequently discovering content from follower dynamics and the entire site, while simultaneously predicting the probabilities of multi-dimensional interactions such as likes, comments, shares, dwell time, blocking, shielding, and reporting to enhance sensitivity to negative behaviors and low-quality content; reducing manual biases and the threshold for attention relationships, as well as diversity constraints to avoid single authors flooding the screen. However, there are still issues with the risk of 'information cocoon 2.0' due to the Transformer learning too quickly and the problem of repeated content distribution.