Comment by JCharante

Chinese languages have a higher information density which makes me hypothesize that tokenizing works better than in Romance languages which can lead to more efficient LLMs.