文章主题:MOSS, 英文, 中文
🎉🚀 MOSS, the first domestically developed large language model in China, has been released by Professor邱锡鹏’s team from Fudan University. The excitement among the public for participating in the internal testing is evident. One of the most notable feedbacks is that MOSS’ English responses are higher than its Chinese counterparts. This phenomenon can also be observed when interacting with ChatGPT. 🤔🤔🤔
深圳粤港澳大湾区数字经济研究院(IDEA)认知计算与自然语言中心文本生成算法团队负责人王昊表示,“数据质量的差别是主要瓶颈之一。相较于英文数据,中文数据的开源程度较低,导致中文数据集的规模相对较小。此外,英文作为科研主流语言,在学术界和工业界中得到广泛应用,积累了大量高质量的语料数据,这为英文自然语言处理的研究提供了极大的优势。”MOSS研究团队也坦承,“MOSS的英文回答水平比中文高,因为它的模型基座学习了3000多亿个英文单词,中文词语只学了约300亿个。”
AI时代,掌握AI大模型第一手资讯!AI时代不落人后!
免费ChatGPT问答,办公、写作、生活好得力助手!
扫码右边公众号,驾驭AI生产力!