Synthetic Social Data and Models

Considering various data types and interaction methods, we adopt multiple approaches for asset storage, including temporary centralized processing of some data, data on-chain, and utilizing centralized third-party storage services. This comprehensive approach ensures the ownership of core data as well as enhances the user experience.

Within the synthetic social network, several key types of data are involved:

• Basic Identity: Simple information such as names, avatars, introductions, and medals representing contributions within the ecosystem will be on-chain as meta-data.

• Multidimensional Identity Data: This further includes text data such as background knowledge bases and training corpora, as well as multimodal model parameters like LoRA, voice models, 3D models, etc. These will be stored in a decentralized manner in collaboration with third-party services, with the user deciding the scope of storage.

• Memories: For significant relationships, human users or AIs can choose to mint these as memory NFTs and select a memory storage mode (complete, compressed, active storage, etc.) to preserve emotional relationships more enduringly.

• Base Models: A hybrid framework is used, involving multiple groups of models including proprietary models, covering language models, vision models, voice models, etc. These are currently processed centrally, with plans to open-source in the future.

• Comprehensive Decentralized Synthetic Data Solution: Decentralized processing of core identity information, identity data, and memories, combined with open-sourcing of base models and public governance through DSO (Decentralized Social Organization).

In addition to the information surrounding AI identity content, social relationship chains, contributions to the social network, and other similar data are also important components of synthetic social data.