Social RAG group assistant AI, Matrix virtual social network AI comment like

Previously, AI applications in the field of pan-entertainment were all emotional chat and AI emotional companionship. Recently, I have seen Social RAG and Matrix AI, which are more valuable in production and practical. The former can be used as a group assistant to organize members' discussions, and also focuses on combining chat history and each user's preferences and emotions to make personalized replies. The latter will immediately provide comments and likes when posting on the virtual X platform, collect ideas and feedback, simulate social group behavior, and prepare for real-world releases in advance.

Social RAG is a paper project, while Matrix AI already has a test website.https://matrix.eigent.ai/x, the open source version is called Oasis https://github.com/camel-ai/oasis

The actual implementation of these two projects is definitely too early, so they are sorted out here only as a reference for ideas and directions

01

Social RAG is an AI system designed for group interaction. It generates more natural messages that are more in line with social contexts by retrieving and utilizing the social history and interaction data of the group. Currently, it mainly uses Coze to customize robot personality, plug-ins and workflows as group chat and private chat robots. This system is still lacking in social emotional contexts, group history records, and personalization.

The Social RAG paper uses "PaperPing" as an example. PaperPing is designed specifically for paper recommendation and discussion in academic research groups.

PaperPing's workflow is divided into four main steps:

Collect data from group interactions and build a social knowledge base; extract relevant social signals and metadata based on current discussions; use LLM to transform retrieved signals into interpretive messages; send to group channels and analyze reactions

Social RAG Core AI Features

Social context and emotion extraction: The system can understand the social dynamics of the group, grasp the relationships and interests between members; analyze the reactions, comments and preferences of group members

Group history retrieval and personalized generation: Use RAG technology to retrieve relevant information from the group's past interactions and use LLM to generate messages that match the group's style.

Transparency design: AI explains why it recommends specific content, increasing credibility

Application scenarios of AI group assistant:

Academic Research Group: Helps researchers find relevant papers and establish collaborations

Enterprise team collaboration: recommend relevant resources and promote knowledge sharing in work groups

Online learning communities: recommend relevant learning materials to learners and explain their relevance

Professional network: recommend content and contacts based on members’ interests and backgrounds. This should be combined with a recommendation system.

02

The Oasis paper currently says that there are millions of agents on X and Reddit, but not on other social platforms like Instagram and Xiaohongshu, and Matrix cannot test it.

These virtual agents will post, comment, and like on their own, and in chronological order.

This environment is quite complex. It requires not only organizing the profiles of existing users and AI-generated users, but also determining the likes and social relationships of each agent in advance in the Environment Server, as well as the Recommendation System. Under these two major systems, the LLM-driven Agent starts to like and comment, thereby updating the previous Environment Server.

Because the system is large and complex, Matrix and Oasis are currently in the toy stage, focusing on ideas, and the actual effect is certainly limited. Although the agent itself combines multiple profiles, it is still not personalized enough. In addition, the recommendation system is already so complex, and the number of agents is still in the millions.

I tested the Matrix website online (only 2 or 3 free tests), and the post was about a trip to a park to enjoy the flowers. The AI ​​registered various people’s identities to comment, such as the virtual IDG Capital or Elon Musk.

But the mood of these virtual agent accounts is not good, and the comments are all negative, jealous and sarcastic, such as "while the rest of us are buried in our cubicles" and "Did you manage to leave the dog at home while you created this serene Saturday Insta-worthy moment?"

I don’t know why there are so many unfriendly people even among virtual agents. Even though the agent will generate virtual agent names and avatars based on the information of real social platforms, the specific comment styles of these virtual agents are far from those of real users.

Moreover, these virtualized real-life information can only be used as games. Such products will infringe on privacy.

Before clicking on a post, the random interface that appears is a post posted by each virtual agent. After all, it is an X platform, so it mainly contains technology and Internet information.

Epilogue

Social RAG and Oasis are both very futuristic, with virtual groups, virtual WeChat robots, and virtual X accounts in the future, which are a bit like science fiction movies.

Social RAG is limited to group AI papers and data assistants, which are closer to the present and can help organize recommended content and do something helpful to human production.