Progressive Determinization in Software Development (novian.works)

A new approach called "progressive determinization" has emerged in software development, leveraging the inherent weaknesses of large language models (LLMs) to create more reliable systems. Rather than treating LLMs solely as coding assistants, this method starts with a single prompt that encompasses the entire task, allowing the model to function as the software itself. As developers observe the model's errors—like incorrect arithmetic or slow responses—they extract these problematic areas into deterministic code, gradually refining the system while reducing reliance on the model. This iterative process encourages real-world testing to inform further refinements, enabling developers to identify and formalize the parts of their systems that require accuracy and efficiency. This shift has significant implications for the AI/ML community, moving the focus from extensive upfront design to dynamic, observational coding. By allowing models to handle a wider range of scenarios initially, developers can more effectively determine what needs to be extracted based on real use cases, leading to smarter resource allocation. The integration of behavior-driven testing into this workflow ensures that as code replaces model functions, functionality remains intact. Ultimately, progressive determinization represents a paradigm shift, emphasizing an evolving understanding of problem spaces as systems are developed and refined, rather than relying on pre-conceived architectures.