I always found the frontier AI labs' focus on trained knowledge a bit confusing. From my perspective, training creates a reasoning engine. Beyond that, models reason about data from more reliable sources. I wonder why this is not a distinction they make?