Three Cobblers, One Zhuge Liang: Making Cheaper Models Work Together (markhuang.ai)

In a recent exploration of AI architectural design, a novel approach emphasizes the synergy of multiple smaller models over reliance on a single, more powerful model. The principle, inspired by a Chinese proverb, suggests that by coordinating weaker AI systems—akin to "three humble cobblers"—it’s possible to achieve results that rival more complex models like GPT. The key is to segment tasks and responsibilities, utilizing a structured system prompt to define each model's specific role. This reduces cognitive overload and helps smaller models to focus on narrower tasks, which leads to better performance by minimizing the risk of important details being overlooked. This paradigm shift is significant for the AI/ML community as it allows for more efficient and cost-effective processing of AI workflows. By implementing a "hub-and-spoke" architecture, where a master orchestrator coordinates specialized models for tasks like requirements checking or validation, workflows can dynamically adjust to complexity and interconnectedness in real-world applications. Furthermore, the approach advocates for adaptive temperature settings in models—low for tasks requiring consistency and high for creative endeavors—ensuring that the workflow remains robust yet flexible, thus fundamentally transforming how AI systems collaborate and execute tasks.