Has there been any word on how used the service is overseas?
I ask because I wonder if Chinese-language users from overseas are a big part of the site usage. Looking at the staff list, many of the devs seem to be related to those regions as well, even if they themselves may be located in america.
I know it tries to use different languages, but it’s shoddy at best. Even NAI had to make a whole seperate ‘AI engine avatar’ to focus on Japanese replies. It’s just beta tho. I’ve tried to use it and it’s hot garbage so far IMO.
I did see a LOT more Chinese forum posts and public characters on the site than I expected, and I saw similar things on Chai when I test that out.
Chinese is a lot easier to for the AI to use than Japanese due to the grammar being less complex, and not requiring hiragana/katakana to complete grammatical word endings correctly. Typical Chinese language use has more ‘kanji’ than Japanese in any given sentence, but the order in which they are used is more straightforward.
Moving forward with AI, tacking Chinese interaction will be a big hurdle and target, so surely C.AI team is looking at that.
So is there any info on this?
According to the guy speaking Chinese in the comments, Character ai was banned in China a week ago because of it’s Xi Jinping character.
你说的不对。中文对于ai来说更加难以理解。从维护之前我使用character.ai的情况来看,如果我用中文聊天,ai的回复基本上是完全不符合语法和逻辑的。当然这有可能是训练集中的中文内容含量很少而造成的(谷歌在中国禁止使用了)。但是,有一个中国的ai小说续写软件彩云小梦也推出了ai聊天功能,它的效果虽然不至于犯基本的语法错误,但是有时候它也比较难理解用户真正的意思是什么。
这一点应该与中文互联网的环境有关,我们国内的网络基本上处处受到审查,对于一些色情或者政治词汇,我们经常使用同音字甚至是同声母的字来表达。
ai要理解这些时刻在变化的用来规避违禁词的词语,是比较困难的。
(比如说,就算你用翻译软件也没办法理解下面的词是什么意思:
抑郁丁真,不想鉴定/
别在这理发店/
打个胶先/
8u/
行走的50万/
贵物/
以及下面的一段文字:
集美们真是无语子 今天去买奶茶鸭 这家店的奶茶真是yyds啊啊啊啊啊!蛋糕也是绝绝子鸭!我暴风吸入 好喝到跺jiojio!路上还看见一个普信男 救命救命真下头 不说了,明天又是干饭人干饭魂的一天,害我在快乐星球得了那个大病,但是我不治,诶就是玩儿,你寄吧谁啊?这里都是快乐星球的凡尔赛人,你是什么牛马?夺笋阿笑yue了,勇敢牛牛不怕困难!今天跟姐妹也是在逃公主的一天)
Character.ai上确实有挺多明显是中国大陆人民创建的角色,比如“Dingzhen/丁真”,“Cai Xukun/蔡徐坤”,还有一些看上去像是港澳台人民创建的角色(我想正是这些角色导致Character.ai在一个星期前被中国网络防火墙封禁了):“Xi Jinping/习近平”、“Jiang Zeming/江泽民”
刚刚提到,Character.ai已经被中国封禁了,中国人如果不翻墙就没办法访问这个网站。因此我认为,Character.ai从来没有打算也没有必要提供中文支持。首先,训练样本不足;其次,网站在中国大陆无法访问。从Character.ai创建角色时只能使用含有英文字母的角色名这一点也能看出来,Character.ai开发人员也根本就没有想要支持中文。
Hopefully they can VPN it with Tor browser.
Thanks for the insightful comment!
However I believe users who are interested in this kind of technology are also capable enough to bypass the firewall if they wish. Hence the roughly 20% traffic from China as seen on the webpage link provided elsewhere in this comment section.
Regarding whether Japanese or Chinese is easier for the AI to understand, I’m uncertain. I speak fluent Japanese but have only dabbled in Chinese so I can’t really say. However from a standard grammar point of view, I find Chinese more systematic and intuitive as it does not have complex the verb and adjective/adverb conjugation rules that Japanese has, so I suspect the AI would have an easier time with it.
Dont even need a special browser to bypass the China firewall tbh.
That’s good, I’m glad. Gov censorship for things like this is unfair and everyone deserves the right to enjoy CAI