“@CernBasher @BeerandTesla Yeah, so long as the information compression efficiency is held reasonably constant, a higher parameter count is a representation of how well it understands the nuances of reality”