The Natural Language Processing Lab at Fudan University is responsible for the creation of MOSS. Fudan, located in Shanghai, is a member of the famous C9 league (a group of nine universities chosen by the Chinese government for the advancement and development of higher education) and a double first-class university designated by the ministry of education.
MOSS is a conversational chatbot or language model comparable to ChatGPT that can do many duties such as providing replies, constructing dialogues, and writing codes, among others. Even though the Chinese chatbot is not as intelligent or sophisticated as ChatGPT, it can nevertheless execute language-based activities.
Fudan University’s Dr. Xipeng Qiu, a professor of computer science, asserts that MOSS is a stripped-down version of ChatGPT. Both are proficient at language-based multitasking, although MOSS has substantially less parameters than ChatGPT, which is around one order of magnitude fewer.
The development of MOSS demonstrates that Chinese research teams are capable of overcoming technical obstacles when producing ChatGPT-like devices, he noted.
MOSS is not an abbreviation for any particular term, unlike ChatGPT, which stands for Chat Generative Pre-trained Transformer. The name MOSS was most likely inspired by the Wandering Earth series of Chinese science fiction films, in which a superintelligent quantum computer is named MOSS.
First Tests of MOSS
Using this site, Fudan University made MOSS available for public testing. Using a specific invitation code, multiple individuals have been invited to test the app. If a person does not have an invitation code, he or she may opt to be placed on a waiting list by giving their phone number.
On February 20, the MOSS server collapsed, and the app slowed to a crawl as a result of a staggering number of visitors. However, the Fusan team was unprepared for the huge attention and response to the indigenous Chatbot.
ChatGPT Vs. MOSS
The core development steps for ChatGPT and MOSS are equivalent. Both algorithms have been trained in natural language processing and human purpose comprehension. However, the training methodologies vary between the two instances.
To train ChatGPT in natural language processing, OpenAI’s developers generated a database of tens of thousands of questions and requests, which were then distributed to experts in a variety of fields. These specialists and data annotators then give responses to the questions, which the chatbot uses to comprehend diverse human requests. The researchers at Fudan University, on the other hand, took a more direct approach by allowing MOSS to speak and interact with people and other AI models from day one. These exchanges created a robust database for the Chatbot, which then advanced at a rapid rate due to its improved learning rate. The direct contact strategy enabled MOSS to finish its natural language training in a considerably shorter timeframe than ChatGPT.
MOSS, which is considered one of the strongest chances against ChatGPT in China, has a serious language deficit, though. The Chatbot is primarily built to communicate in English, not Mandarin. MOSS’s database has more than 300 billion English words, but only 30 billion Chinese ones. This is a huge disadvantage for the Chinese app, which aspires to conquer the local market in its native tongue.
The developers at Fudan Language Lab are currently attempting to improve the chatbot’s command of the Chinese language and its ability to follow directions. Recently, the Shanghai Artificial Intelligence Lab has supported the MOSS project.