Jibàn Launches Shared-Experience AI Social

🔑 Enhanced Key Takeaways

• •Jibàn is developed by the startup 'Jibàn AI,' which secured seed funding from prominent Chinese venture capital firms focusing on generative AI and social entertainment.
• •The platform utilizes a proprietary 'Nested Cognition' architecture that separates long-term episodic memory from real-time sensory processing to reduce hallucination in complex roleplay scenarios.
• •The product targets the 'Gen Z' demographic in China, specifically aiming to capture market share from traditional text-based AI companion apps by emphasizing visual-first, collaborative world-building.

🛠️ Technical Deep Dive

• •Nested Cognition Architecture: A multi-agent framework where a 'Director' agent manages world state and environmental consistency, while 'Character' agents manage individual personality and emotional state.
• •Three-Layer Memory System: Implemented via a vector database (likely Milvus or Pinecone) that categorizes data into: 1) Event Log (factual history), 2) Behavioral Patterns (habitual responses), and 3) Interaction Sentiment (emotional weight).
• •World Generation: Integrates a fine-tuned diffusion model for map/asset generation triggered by natural language prompts, coupled with a procedural generation engine for NPC dialogue trees.
• •Latency Optimization: Uses a streaming inference pipeline to minimize the time between user input and world-state updates, targeting sub-500ms response times for interactive elements.

🔮 Future ImplicationsAI analysis grounded in cited sources

⏳ Timeline