White Papers

The dominant paradigm in artificial intelligence research has treated capability as fundamentally a function of scale: larger models, larger datasets, larger training compute. This paper challenges that assumption by introducing Intelligence Density — a formal framework for measuring the concentration of general-purpose capability relative to computational resources consumed. We propose the Intelligence Density Quotient (IDQ) as a composite metric spanning task performance, parameter efficiency, inference cost, and benchmark breadth. We demonstrate that IDQ and related density measures reveal a dimension of model quality currently underserved by existing benchmarks, and argue that maximising intelligence density may represent a more tractable and scientifically productive research objective for a substantial class of applications. This paper establishes the theoretical foundations and measurement methodology underpinning the CHISEI research programme at SHINGETSU.