汉字的信息密度更高，适用于和 LLM 通信

[king-bcolor]

可以提取一些抽象规范，用于规定汉字和大模型沟通，打造最节省 token 的语言
比如一段中文
晨,小明食椒,腹痛而奔,求水于厨人,不遂而亡
chatGPT翻译成英文:
In the morning, Xiao Ming ate chili, suffered stomach pain, ran away, asked the cook for water, but failed to get it, and died.

到 https://platform.openai.com/tokenizer 网站中计算 token
最后中文 22 token ，英文 28 token。
在文言文中，"小明"如果再次出现往往简写成"明"来代指，
如果是字符 "Xiao Ming" 占 3 token，如果是"明"只占 1 token

[king-bcolor]

如果扩写一些
晨,小明食椒,腹痛而奔,求水于厨人,不遂而亡.父锄而归,见之,痛而嚎,问邻人,我子何故至此.答曰:汝子晨醒而食,椒燃胃中,求水不得,故死
英文
In the morning, Xiao Ming ate chili peppers. His stomach burned with pain, and he ran in distress, seeking water from the cook. However, he failed to obtain it and perished. His father returned from hoeing the fields and saw him. Overcome with grief, he wailed and asked the neighbors, "Why has my son ended up like this?" The neighbor replied, "Your son woke up in the morning and ate. The chili burned in his stomach, and he could not get water, so he died."

同一段意思，汉字只消耗70 token，英文消耗 107 token

[nobodxbodon]

不妨从用中文命名标识符开始积累代码作语料库。现有ai写码助手已支持中文标识符：https://www.zhihu.com/question/484193394/answer/2205177360