iiinova — where your digital self keeps evolving

Chase Million @millionconcepts.com · 28/07/2026

I just recently submitted this related abstract to NASA's Mission Software for the Science Mission Directorate Workshop 2026. (science.data.nasa.gov/features-eve...)

Development and support of core software libraries and tools is more critical in an era of LLM- and "AI"-assisted coding, and will yield a better return on investment to NASA. Large language models are an undeniably useful tool for software development and aid to scientific data exploration. In constrained budget environments, there may therefore be a temptation to pull back on direct support for development and maintenance of core software libraries due to a mistaken belief that "AI" can seamlessly cover for any gaps or lost capability. However, LLMs perform far more effectively on software development tasks when they have robust, well-documented, and interoperable software components to build upon. They are also definitionally less effective in niche or emerging domains, where there is less training data. "Niche" and "emerging" are precisely where NASA operates, and incurs the most risk, when working to "achieve the near impossible." As an LLM extrapolates farther afield from trusted, reliable, core tools, it becomes both more costly and more prone to errors. There is further risk of proliferation of parallel solutions to the same problem. Aside from representing unnecessarily duplication of effort, the cost of validation and verification will grow.

jonny (leachate composite form)@jo.nny.rip

If you want to "just do biology and never worry about programming" the answer looks like adequate funding for research software engineering so we can make validated, shared tools that reduce the need for new code to none. Not everyone generating analysis on demand that is impossible to validate.

030