“Analysis by Constantine at François Challet's company indicates that the "outer refinement loop" is the primary reason for the high performance of Hierarchical Reasoning Models (HRM).”