Comment by marrone12

I asked the same question to chatgpt, claude, and kimi and kimi gave the most sane answer. Part of me wonders if it has anything to do with having more chinese language structure in their dataset