Ever since researchers began noticing a slowdown in improvements to large language models using traditional training methods, ...