Proposing an "AI-Model Network" for Collaborative AI Systems

The rapid growth of large language models (LLMs) has highlighted their immense potential, but also their significant challenges, including high training costs and complex deployment. This has led to a shift towards more specialized, lightweight, and private models. However, as these diverse models proliferate, a critical bottleneck emerges: how to enable them to interact and collaborate effectively. Drawing parallels with the internet's evolution from isolated computers to a network of shared resources, this paper proposes the "AI-Model Network" (AI-ModelNet). AI-ModelNet envisions a system where different AI models can connect, share capabilities, and engage in collaborative reasoning. This paradigm aims to unlock greater value from individual models by allowing them to work together, much like the internet empowered computers through sharing and collaboration. The paper outlines a systemic vision and a hierarchical architecture for this network. The authors validate the feasibility of AI-ModelNet through a prototype system and various application cases, demonstrating how models can establish pathways for interconnection and mutual enhancement. The research also discusses key future directions, emphasizing the potential for this network to drive the next wave of AI development by fostering a more distributed, collaborative, and efficient ecosystem for AI models.