“The quarterly release of improved AI models by major labs is primarily driven by scaling compute and, more importantly, adding new data targeted at the weaknesses of the previous model version.”