An Amorphous Model: Semantic Vector Search and RecursiveSelf-Improvement

This research note proposes an amorphous machine learning model whose realisations span common model families and their meta-learners. The vector space of all semantically distinct objects contains all computational architectures (implying self-reference). The space must contain a sub-region whose traversal means transitioning between different intelligent systems. This note proposes a seed improver that, via recursive self-improvement, searches the subregion for increasingly capable architectures. The note offers formalisms (axioms, definitions, falsifiable conjecture, and theory-agnostic empirical tests.)