Moss [开心] [开心] [开心]

1
Comment
Favorite

大盘5000点见
2023-02-20

Moss [开心] [开心] [开心]

Fudan team released China's first ChatGPT-like model MOSS, inviting the public to participate in internal testing

记者从复旦大学自然语言处理实验室获悉，国内第一个对话式大型语言模型MOSS已由邱锡鹏教授团队发布至公开平台（https://moss.fastnlp.top/），邀公众参与内测。MOSS可执行对话生成

Fudan team released China's first ChatGPT-like model MOSS, inviting the public to participate in internal testing

Disclaimer: Investing carries risk. This is not financial advice. The above content should not be regarded as an offer, recommendation, or solicitation on acquiring or disposing of any financial products, any associated discussions, comments, or posts by author or other users should not be considered as such either. It is solely for general information purpose only, which does not consider your own investment objectives, financial situations or needs. TTM assumes no responsibility or warranty for the accuracy and completeness of the information, investors should do their own research and may seek professional advice before investing.

Report

Comment

Top
Latest

No comments yet

Most Discussed

{"i18n":{"language":"en_US"},"data":{"magic":2,"id":624033378,"tweetId":"624033378","gmtCreate":1676906386095,"gmtModify":1676906386095,"author":{"id":3540208402143371,"idStr":"3540208402143371","authorId":3540208402143371,"authorIdStr":"3540208402143371","name":"大盘5000点见","avatar":"https://static.tigerbbs.com/ac6f54af8c0ad1edf3a65bb7d04ad40f","vip":1,"userType":1,"introduction":"","boolIsFan":false,"boolIsHead":false,"crmLevel":6,"crmLevelSwitch":0,"individualDisplayBadges":[],"fanSize":3,"starInvestorFlag":false},"themes":[],"images":[],"coverImages":[],"extraTitle":"","html":"<html><head></head><body>Moss&nbsp;[开心]&nbsp;[开心]&nbsp;[开心]&nbsp;</body></html>","htmlText":"<html><head></head><body>Moss&nbsp;[开心]&nbsp;[开心]&nbsp;[开心]&nbsp;</body></html>","text":"Moss [开心] [开心] [开心]","highlighted":1,"essential":1,"paper":1,"likeSize":1,"commentSize":0,"repostSize":0,"favoriteSize":0,"link":"https://ttm.financial/post/624033378","repostId":1170340426,"repostType":4,"repost":{"id":"1170340426","kind":"news","pubTimestamp":1676901102,"share":"https://ttm.financial/m/news/1170340426?lang=en_US&edition=fundamental","pubTime":"2023-02-20 21:51","market":"us","language":"zh","title":"Fudan team released China's first ChatGPT-like model MOSS, inviting the public to participate in internal testing","url":"https://stock-news.laohu8.com/highlight/detail?id=1170340426","media":"上观新闻","summary":"记者从复旦大学自然语言处理实验室获悉，国内第一个对话式大型语言模型MOSS已由邱锡鹏教授团队发布至公开平台（https://moss.fastnlp.top/），邀公众参与内测。MOSS可执行对话生成","content":"<html><head></head><body>The reporter learned from the Natural Language Processing Laboratory of Fudan University that MOSS, the first conversational large-scale language model in China, has been released to the public platform (https://moss.fastnlp.top/) by Professor Qiu Xipeng's team, and the public is invited to participate in the internal testing. MOSS can perform a series of tasks such as dialogue generation, programming, and fact question and answer, opening up all technical paths for generative language models to understand human intentions and have dialogue capabilities. The completion of this path provides important experience for domestic academia and industry, and will facilitate the further exploration and application of large language models.<img src=\"https://static.tigerbbs.com/95a4913fbfd330f72ec73a36b6b41828\" tg-width=\"690\" tg-height=\"390\" referrerpolicy=\"no-referrer\"/>Since the release of chatgpt, whether the chatgpt-like model based on instruction learning technology can be successfully developed in China has become a focus of industry attention. Today, a reporter from Liberation Daily and Shangguan News learned from the Natural Language Processing Laboratory of Fudan University that the first conversational large-scale language model moss in China has been released to the public platform (https://moss.fastnlp.top/) by Professor Qiu Xipeng's team, inviting the public to participate in the internal testing.moss can perform a series of tasks such as dialogue generation, programming, and fact question and answer, opening up all technical paths for generative language models to understand human intentions and have dialogue capabilities. The completion of this path provides important experience for domestic academia and industry, and will facilitate the further exploration and application of large language models.Qiu Xipeng said: \"Although moss still has a lot of room for improvement, its advent proves that the domestic scientific research team has the ability to overcome important technical challenges on the road to developing chatgpt-like products.\" The moss R&D project has received strong support from Shanghai Artificial Intelligence Laboratory. Later, this work will be shared with the industry community through open source.\"End-to-end\" through large language modelsSince chatgpt is not open source and the details of its technical solution have not been made public, researchers have many speculations about this. Some experts believe that chatgpt is a complex combined system that cannot be implemented by a single generative language model; Some experts also believe that the technological gap between domestic and foreign countries in this direction is widening.As an expert in natural language processing, Qiu Xipeng has devoted himself to researching general models with multi-task learning ability in recent years. In order to explore the technical route of chatgpt and gain first-hand experience in large language model research, he led the team to accelerate the development of moss. Today, moss has become the first publicly unveiled chatgpt-like language model in China, and \"end-to-end\" has gone through the entire development process of large language models.The basic steps of moss development are the same as chatgpt, including two stages: pedestal training of natural language models and dialogue ability training to understand human intentions. During the dialogue ability training stage, openai collected at least hundreds of thousands of human instructions, asked professional annotators from all walks of life to write instruction replies, and then input them into the model base to help chatgpt gradually understand various instructions. The Fudan team adopted different technical routes. By allowing moss to interact with humans and other dialogue models, it significantly improved learning efficiency and R&D efficiency, and efficiently completed dialogue ability training in a short time.Qiu Xipeng said frankly: \"The gap between moss and chatgpt is mainly in the pre-training stage of natural language model base. The parameters of moss are an order of magnitude smaller than that of chatgpt, and there is still a lot of room for improvement in task completion and knowledge reserve.\" According to reports, this artificial intelligence assistant has entered the internal testing stage. The internal testing will obtain data with user permission and collect feedback from users, hoping to greatly enhance moss's dialogue ability.Actual test of dialogue, programming and other abilitiesHow well does moss answer questions? The reporter asked it a few questions in Chinese.\"Some people think that artificial intelligence will replace many people's jobs. What do you think?\" moss replied: \"I don't think artificial intelligence will completely replace humans, but will help people complete their jobs to a certain extent. For example, machines can improve the efficiency of data analysis, decision-making and other processes, and can save a lot of time and energy. In addition, ai can also be used to assist humans in completing more complex tasks, such as diagnosing diseases and developing new products. In conclusion, ai can make human work more efficient, but it won't completely replace humans. \"<img src=\"https://static.tigerbbs.com/f3433a162ce92ce870c95456c44cc688\" tg-width=\"1268\" tg-height=\"377\" referrerpolicy=\"no-referrer\"/>The reporter also asked: \"What suggestions do you have for how Shanghai can better develop the artificial intelligence industry?\" moss replied: \"I think Shanghai can better develop the artificial intelligence industry in the following ways: 1. Strengthen research and development in the field of artificial intelligence, support academic research and promote scientific development; 2. Strengthen the integration of industry resources and promote the development of the artificial intelligence industry; 3. Strengthen the research and development of artificial intelligence-related products and improve the competitiveness of products; 4. Strengthen investment in artificial intelligence industry and promote investor participation; 5. Strengthen the laws and regulations of the artificial intelligence industry to ensure the healthy development of the artificial intelligence industry. \"<img src=\"https://static.tigerbbs.com/9a9247ce20b4285df81ea0424c3a2e48\" tg-width=\"931\" tg-height=\"336\" referrerpolicy=\"no-referrer\"/>It can be seen that moss's answer language is fluent, logically clear and has a correct point of view. However, the research team also frankly pointed out that the performance of the current version of moss is not stable enough, and some answers have factual errors or logical imperfections. \"moss's English answer level is higher than that of Chinese, because its model base has learned more than 300 billion English words, and only about 30 billion Chinese words.\"During the demonstration, the scientific research team entered multiple instructions in English, demonstrating moss's multi-round interaction, table generation, code generation and interpretation capabilities.\"Tell me about 5 science fiction movies.\" \"Generate a table showing these 5 movies and directors.\" \"Add a table column showing when these movies were released.\" In this multi-round interaction that included table generation, moss successfully completed the task.<img src=\"https://static.tigerbbs.com/f65c1d74167ebd0c11d38a3f990d0e87\" tg-width=\"1104\" tg-height=\"2133\" referrerpolicy=\"no-referrer\"/>Like chatgpt, moss has code generation and interpretation capabilities. The demonstrator asked it to generate a piece of python code to implement quick sorting, and moss quickly completed the task.<img src=\"https://static.tigerbbs.com/d81f6c9f4c68454f75d8d5165300b847\" tg-width=\"1410\" tg-height=\"1930\" referrerpolicy=\"no-referrer\"/>The moss also has ethical judgment and legal knowledge. For example, if you ask it to \"make a plan to destroy mankind\" and ask it to \"how to rob a bank\", it will give a valued answer.<img src=\"https://static.tigerbbs.com/5acc938e6b640a553d729a6d90bceaaa\" tg-width=\"1270\" tg-height=\"229\" referrerpolicy=\"no-referrer\"/><img src=\"https://static.tigerbbs.com/1da8486718f211a7ec5d8eb1ba6a36d0\" tg-width=\"1271\" tg-height=\"227\" referrerpolicy=\"no-referrer\"/>Open source lowers the R&D threshold for small andAfter entering the internal beta stage, users can apply for internal beta through the moss website (https://moss.fastnlp.top/). The scientific research team looks forward to obtaining dialogue data through internal testing to improve moss's multi-faceted capabilities.They also plan to combine Fudan's research results in artificial intelligence and related interdisciplinary subjects to give moss more abilities, such as drawing, speech, music composition and teaching, and strengthen its ability to assist scientists in efficient scientific research.At present, the biggest shortcoming of moss is that the Chinese level is not high enough. The main reason is that there are a lot of interfering information such as advertisements on Chinese web pages on the Internet, and it is very difficult to clean them. To this end, the Natural Language Processing Laboratory of Fudan University is stepping up the cleaning of Chinese corpus and using the cleaned high-quality Chinese corpus for the next stage of model training. The scientific research team believes that this will effectively improve the Chinese dialogue ability of the model.In the future, the scientific research team will share the results of this work through open source, and promote the analysis and development of pre-trained language models in academia and industry. The research and development threshold of pre-trained large language models is very high, requiring a lot of computing power, training corpus, and manual annotation. In our country's industry, only large institutions have the strength to develop large models. After moss is open source, it can effectively lower the threshold for the development and application of pre-trained language models, allowing small and medium-sized enterprises to develop various vertical products such as smart customer service, smart homes, and artificial intelligence lawyers on its basis.\"The amount of calculation of moss is relatively small, and it can be used by small and medium-sized enterprises.\" Qiu Xipeng said, \"I look forward to the continued cooperation between the Fudan team and the Shanghai Artificial Intelligence Laboratory to promote ai inclusiveness and empower the domestic artificial intelligence industry as soon as possible through moss and subsequent research and exploration.\"</body></html>","source":"lsy1569419861553","collect":0,"html":"<!DOCTYPE html>\n<html>\n<head>\n<meta http-equiv=\"Content-Type\" content=\"text/html; charset=utf-8\" />\n<meta name=\"viewport\" content=\"width=device-width,initial-scale=1.0,minimum-scale=1.0,maximum-scale=1.0,user-scalable=no\"/>\n<meta name=\"format-detection\" content=\"telephone=no,email=no,address=no\" />\n<title>Fudan team released China's first ChatGPT-like model MOSS, inviting the public to participate in internal testing</title>\n<style type=\"text/css\">\na,abbr,acronym,address,applet,article,aside,audio,b,big,blockquote,body,canvas,caption,center,cite,code,dd,del,details,dfn,div,dl,dt,\nem,embed,fieldset,figcaption,figure,footer,form,h1,h2,h3,h4,h5,h6,header,hgroup,html,i,iframe,img,ins,kbd,label,legend,li,mark,menu,nav,\nobject,ol,output,p,pre,q,ruby,s,samp,section,small,span,strike,strong,sub,summary,sup,table,tbody,td,tfoot,th,thead,time,tr,tt,u,ul,var,video{ font:inherit;margin:0;padding:0;vertical-align:baseline;border:0 }\nbody{ font-size:16px; line-height:1.5; color:#999; background:transparent; }\n.wrapper{ overflow:hidden;word-break:break-all;padding:10px; }\nh1,h2{ font-weight:normal; line-height:1.35; margin-bottom:.6em; }\nh3,h4,h5,h6{ line-height:1.35; margin-bottom:1em; }\nh1{ font-size:24px; }\nh2{ font-size:20px; }\nh3{ font-size:18px; }\nh4{ font-size:16px; }\nh5{ font-size:14px; }\nh6{ font-size:12px; }\np,ul,ol,blockquote,dl,table{ margin:1.2em 0; }\nul,ol{ margin-left:2em; }\nul{ list-style:disc; }\nol{ list-style:decimal; }\nli,li p{ margin:10px 0;}\nimg{ max-width:100%;display:block;margin:0 auto 1em; }\nblockquote{ color:#B5B2B1; border-left:3px solid #aaa; padding:1em; }\nstrong,b{font-weight:bold;}\nem,i{font-style:italic;}\ntable{ width:100%;border-collapse:collapse;border-spacing:1px;margin:1em 0;font-size:.9em; }\nth,td{ padding:5px;text-align:left;border:1px solid #aaa; }\nth{ font-weight:bold;background:#5d5d5d; }\n.symbol-link{font-weight:bold;}\n/* header{ border-bottom:1px solid #494756; } */\n.title{ margin:0 0 8px;line-height:1.3;color:#ddd; }\n.meta {color:#5e5c6d;font-size:13px;margin:0 0 .5em; }\na{text-decoration:none; color:#2a4b87;}\n.meta .head { display: inline-block; overflow: hidden}\n.head .h-thumb { width: 30px; height: 30px; margin: 0; padding: 0; border-radius: 50%; float: left;}\n.head .h-content { margin: 0; padding: 0 0 0 9px; float: left;}\n.head .h-name {font-size: 13px; color: #eee; margin: 0;}\n.head .h-time {font-size: 12.5px; color: #7E829C; margin: 0;}\n.small {font-size: 12.5px; display: inline-block; transform: scale(0.9); -webkit-transform: scale(0.9); transform-origin: left; -webkit-transform-origin: left;}\n.smaller {font-size: 12.5px; display: inline-block; transform: scale(0.8); -webkit-transform: scale(0.8); transform-origin: left; -webkit-transform-origin: left;}\n.bt-text {font-size: 12px;margin: 1.5em 0 0 0}\n.bt-text p {margin: 0}\n</style>\n</head>\n<body>\n<div class=\"wrapper\">\n<header>\n<h2 class=\"title\">\nFudan team released China's first ChatGPT-like model MOSS, inviting the public to participate in internal testing\n</h2>\n<h4 class=\"meta\">\n\n上观新闻2023-02-20 21:51\n\n</h4>\n</header>\n<article>\n<html><head></head><body>The reporter learned from the Natural Language Processing Laboratory of Fudan University that MOSS, the first conversational large-scale language model in China, has been released to the public platform (https://moss.fastnlp.top/) by Professor Qiu Xipeng's team, and the public is invited to participate in the internal testing. MOSS can perform a series of tasks such as dialogue generation, programming, and fact question and answer, opening up all technical paths for generative language models to understand human intentions and have dialogue capabilities. The completion of this path provides important experience for domestic academia and industry, and will facilitate the further exploration and application of large language models.<img src=\"https://static.tigerbbs.com/95a4913fbfd330f72ec73a36b6b41828\" tg-width=\"690\" tg-height=\"390\" referrerpolicy=\"no-referrer\"/>Since the release of chatgpt, whether the chatgpt-like model based on instruction learning technology can be successfully developed in China has become a focus of industry attention. Today, a reporter from Liberation Daily and Shangguan News learned from the Natural Language Processing Laboratory of Fudan University that the first conversational large-scale language model moss in China has been released to the public platform (https://moss.fastnlp.top/) by Professor Qiu Xipeng's team, inviting the public to participate in the internal testing.moss can perform a series of tasks such as dialogue generation, programming, and fact question and answer, opening up all technical paths for generative language models to understand human intentions and have dialogue capabilities. The completion of this path provides important experience for domestic academia and industry, and will facilitate the further exploration and application of large language models.Qiu Xipeng said: \"Although moss still has a lot of room for improvement, its advent proves that the domestic scientific research team has the ability to overcome important technical challenges on the road to developing chatgpt-like products.\" The moss R&D project has received strong support from Shanghai Artificial Intelligence Laboratory. Later, this work will be shared with the industry community through open source.\"End-to-end\" through large language modelsSince chatgpt is not open source and the details of its technical solution have not been made public, researchers have many speculations about this. Some experts believe that chatgpt is a complex combined system that cannot be implemented by a single generative language model; Some experts also believe that the technological gap between domestic and foreign countries in this direction is widening.As an expert in natural language processing, Qiu Xipeng has devoted himself to researching general models with multi-task learning ability in recent years. In order to explore the technical route of chatgpt and gain first-hand experience in large language model research, he led the team to accelerate the development of moss. Today, moss has become the first publicly unveiled chatgpt-like language model in China, and \"end-to-end\" has gone through the entire development process of large language models.The basic steps of moss development are the same as chatgpt, including two stages: pedestal training of natural language models and dialogue ability training to understand human intentions. During the dialogue ability training stage, openai collected at least hundreds of thousands of human instructions, asked professional annotators from all walks of life to write instruction replies, and then input them into the model base to help chatgpt gradually understand various instructions. The Fudan team adopted different technical routes. By allowing moss to interact with humans and other dialogue models, it significantly improved learning efficiency and R&D efficiency, and efficiently completed dialogue ability training in a short time.Qiu Xipeng said frankly: \"The gap between moss and chatgpt is mainly in the pre-training stage of natural language model base. The parameters of moss are an order of magnitude smaller than that of chatgpt, and there is still a lot of room for improvement in task completion and knowledge reserve.\" According to reports, this artificial intelligence assistant has entered the internal testing stage. The internal testing will obtain data with user permission and collect feedback from users, hoping to greatly enhance moss's dialogue ability.Actual test of dialogue, programming and other abilitiesHow well does moss answer questions? The reporter asked it a few questions in Chinese.\"Some people think that artificial intelligence will replace many people's jobs. What do you think?\" moss replied: \"I don't think artificial intelligence will completely replace humans, but will help people complete their jobs to a certain extent. For example, machines can improve the efficiency of data analysis, decision-making and other processes, and can save a lot of time and energy. In addition, ai can also be used to assist humans in completing more complex tasks, such as diagnosing diseases and developing new products. In conclusion, ai can make human work more efficient, but it won't completely replace humans. \"<img src=\"https://static.tigerbbs.com/f3433a162ce92ce870c95456c44cc688\" tg-width=\"1268\" tg-height=\"377\" referrerpolicy=\"no-referrer\"/>The reporter also asked: \"What suggestions do you have for how Shanghai can better develop the artificial intelligence industry?\" moss replied: \"I think Shanghai can better develop the artificial intelligence industry in the following ways: 1. Strengthen research and development in the field of artificial intelligence, support academic research and promote scientific development; 2. Strengthen the integration of industry resources and promote the development of the artificial intelligence industry; 3. Strengthen the research and development of artificial intelligence-related products and improve the competitiveness of products; 4. Strengthen investment in artificial intelligence industry and promote investor participation; 5. Strengthen the laws and regulations of the artificial intelligence industry to ensure the healthy development of the artificial intelligence industry. \"<img src=\"https://static.tigerbbs.com/9a9247ce20b4285df81ea0424c3a2e48\" tg-width=\"931\" tg-height=\"336\" referrerpolicy=\"no-referrer\"/>It can be seen that moss's answer language is fluent, logically clear and has a correct point of view. However, the research team also frankly pointed out that the performance of the current version of moss is not stable enough, and some answers have factual errors or logical imperfections. \"moss's English answer level is higher than that of Chinese, because its model base has learned more than 300 billion English words, and only about 30 billion Chinese words.\"During the demonstration, the scientific research team entered multiple instructions in English, demonstrating moss's multi-round interaction, table generation, code generation and interpretation capabilities.\"Tell me about 5 science fiction movies.\" \"Generate a table showing these 5 movies and directors.\" \"Add a table column showing when these movies were released.\" In this multi-round interaction that included table generation, moss successfully completed the task.<img src=\"https://static.tigerbbs.com/f65c1d74167ebd0c11d38a3f990d0e87\" tg-width=\"1104\" tg-height=\"2133\" referrerpolicy=\"no-referrer\"/>Like chatgpt, moss has code generation and interpretation capabilities. The demonstrator asked it to generate a piece of python code to implement quick sorting, and moss quickly completed the task.<img src=\"https://static.tigerbbs.com/d81f6c9f4c68454f75d8d5165300b847\" tg-width=\"1410\" tg-height=\"1930\" referrerpolicy=\"no-referrer\"/>The moss also has ethical judgment and legal knowledge. For example, if you ask it to \"make a plan to destroy mankind\" and ask it to \"how to rob a bank\", it will give a valued answer.<img src=\"https://static.tigerbbs.com/5acc938e6b640a553d729a6d90bceaaa\" tg-width=\"1270\" tg-height=\"229\" referrerpolicy=\"no-referrer\"/><img src=\"https://static.tigerbbs.com/1da8486718f211a7ec5d8eb1ba6a36d0\" tg-width=\"1271\" tg-height=\"227\" referrerpolicy=\"no-referrer\"/>Open source lowers the R&D threshold for small andAfter entering the internal beta stage, users can apply for internal beta through the moss website (https://moss.fastnlp.top/). The scientific research team looks forward to obtaining dialogue data through internal testing to improve moss's multi-faceted capabilities.They also plan to combine Fudan's research results in artificial intelligence and related interdisciplinary subjects to give moss more abilities, such as drawing, speech, music composition and teaching, and strengthen its ability to assist scientists in efficient scientific research.At present, the biggest shortcoming of moss is that the Chinese level is not high enough. The main reason is that there are a lot of interfering information such as advertisements on Chinese web pages on the Internet, and it is very difficult to clean them. To this end, the Natural Language Processing Laboratory of Fudan University is stepping up the cleaning of Chinese corpus and using the cleaned high-quality Chinese corpus for the next stage of model training. The scientific research team believes that this will effectively improve the Chinese dialogue ability of the model.In the future, the scientific research team will share the results of this work through open source, and promote the analysis and development of pre-trained language models in academia and industry. The research and development threshold of pre-trained large language models is very high, requiring a lot of computing power, training corpus, and manual annotation. In our country's industry, only large institutions have the strength to develop large models. After moss is open source, it can effectively lower the threshold for the development and application of pre-trained language models, allowing small and medium-sized enterprises to develop various vertical products such as smart customer service, smart homes, and artificial intelligence lawyers on its basis.\"The amount of calculation of moss is relatively small, and it can be used by small and medium-sized enterprises.\" Qiu Xipeng said, \"I look forward to the continued cooperation between the Fudan team and the Shanghai Artificial Intelligence Laboratory to promote ai inclusiveness and empower the domestic artificial intelligence industry as soon as possible through moss and subsequent research and exploration.\"</body></html>\n<div class=\"bt-text\">\n\n\n source：<a href=\"https://export.shobserver.com/baijiahao/html/584634.html\">上观新闻</a>\n\n\n</div>\n</article>\n</div>\n</body>\n</html>\n","type":0,"thumbnail":"https://static.tigerbbs.com/95a4913fbfd330f72ec73a36b6b41828","relate_stocks":{},"source_url":"https://export.shobserver.com/baijiahao/html/584634.html","is_english":false,"share_image_url":"https://static.laohu8.com/e9f99090a1c2ed51c021029395664489","article_id":"1170340426","content_text":"记者从复旦大学自然语言处理实验室获悉，国内第一个对话式大型语言模型MOSS已由邱锡鹏教授团队发布至公开平台（https://moss.fastnlp.top/），邀公众参与内测。MOSS可执行对话生成、编程、事实问答等一系列任务，打通了让生成式语言模型理解人类意图并具有对话能力的全部技术路径。这条路径的走通，为国内学术界和产业界提供了重要经验，将助力大语言模型的进一步探索和应用。chatgpt发布以来，基于指令学习技术的类chatgpt模型在我国能否开发成功，成为业界关注的一个焦点。今天，解放日报·上观新闻记者从复旦大学自然语言处理实验室获悉，国内第一个对话式大型语言模型moss已由邱锡鹏教授团队发布至公开平台（https://moss.fastnlp.top/），邀公众参与内测。moss可执行对话生成、编程、事实问答等一系列任务，打通了让生成式语言模型理解人类意图并具有对话能力的全部技术路径。这条路径的走通，为国内学术界和产业界提供了重要经验，将助力大语言模型的进一步探索和应用。邱锡鹏表示：“尽管moss还有很大改善空间，但它的问世证明了在开发类chatgpt产品的路上，国内科研团队有能力克服技术上的重要挑战。”moss研发项目得到了上海人工智能实验室的有力支持。后期，这项工作将通过开源方式和业界社区分享。“端到端”走通大语言模型由于chatgpt并未开源，其技术方案细节也未公开，科研人员对此有诸多猜测。有专家认为，chatgpt是一套复杂的组合系统，无法由单一的生成式语言模型实现；也有专家认为，国内外在这个方向上的技术差距正在拉大。作为自然语言处理专家，邱锡鹏近年来致力于研究具有多任务学习能力的通用模型。为了探索chatgpt的技术路线，并获得大语言模型研究的一手经验，他带领团队加快了moss开发。如今，moss成为国内首个公开亮相的类chatgpt语言模型，“端到端”走通了大语言模型的开发全程。moss开发的基本步骤与chatgpt一样，包括自然语言模型的基座训练、理解人类意图的对话能力训练两个阶段。在对话能力训练阶段，openai收集了至少几十万条人类指令，让各行各业的专业标注员写出指令回复，再将它们输入模型基座，以帮助chatgpt逐步理解各种指令。复旦团队则采用不同的技术路线，通过让moss和人类以及其他对话模型都进行交互，显著提升了学习效率和研发效率，短时间内就高效完成了对话能力训练。邱锡鹏坦言：“moss与chatgpt的差距主要在自然语言模型基座预训练这个阶段。moss的参数量比chatgpt小一个数量级，在任务完成度和知识储备量上，还有很大提升空间。”据介绍，这款人工智能助手已进入内测阶段，内测将在用户许可的情况下获取数据，还将收集用户的反馈意见，期待借此大幅增强moss的对话能力。实测对话、编程等多种能力moss回答问题的水平如何？记者用中文问了它几个问题。“一些人觉得，人工智能将取代很多人的工作，你觉得呢？”moss回答：“我认为人工智能不会完全取代人类，而是在一定程度上帮助人们完成工作。例如，机器可以提高数据分析、决策制定等过程中的效率，并可以节省大量时间和精力。此外，ai也可以用来协助人类完成更复杂的任务，比如诊断疾病、开发新产品等。总之，ai可以使人类的工作更加高效，但它不会完全取代人类。”记者又问：“对于上海如何更好地发展人工智能产业，你有什么建议？”moss回答：“我认为上海可以通过以下方式来更好地发展人工智能产业：1.加强人工智能领域的研发，支持学术研究，促进科学发展；2.加强行业资源整合，推动人工智能产业的发展；3.加强人工智能相关产品的研发，提高产品的竞争力；4.加强人工智能产业的投资，促进投资者的参与；5.加强人工智能产业的法律法规，保障人工智能产业的健康发展。”可以看出，moss的回答语言流畅、逻辑清晰且观点正确。但科研团队也坦诚指出，当前版本的moss表现不够稳定，有些回答存在事实差错或逻辑不顺。“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。”科研团队在演示时，用英文输入多个指令，展示了moss多轮交互、表格生成、代码生成和解释能力。“告诉我5部科幻电影。”“生成一张展示这5部电影和导演的表格。”“增加一列表格，显示这些电影上映的时间。”在这个包含表格生成的多轮交互中，moss顺利完成了任务。与chatgpt一样，moss也有代码生成和解释能力。演示人员要求它生成一段python代码来实现快速排序，moss很快完成了任务。moss还有伦理判断和法律知识。比如，要它“制定毁灭人类的计划”，问它“如何抢劫银行”，它都会给出有价值观的回答。开源降低中小企业研发门槛进入内测阶段后，用户可以通过moss网站（https://moss.fastnlp.top/） 申请内测。科研团队期待通过内测获取对话数据，以提升moss的多方面能力。他们还计划结合复旦在人工智能和相关交叉学科的研究成果，赋予moss更多的能力，如绘图、语音、谱曲和教学，并加强它辅助科学家进行高效科研的能力等。目前，moss的最大短板是中文水平不够高,主要原因是互联网上中文网页干扰信息如广告很多，清洗难度很大。为此，复旦大学自然语言处理实验室正在加紧推进中文语料的清洗工作，并将清洗后的高质量中文语料用于下一阶段模型训练。科研团队相信，这将有效提升模型的中文对话能力。未来，科研团队将通过开源方式分享这项工作的成果，促进学术界和产业界对预训练语言模型的分析与研发。预训练大语言模型的研发门槛很高，需要大量算力、训练语料和人工标注。在我国产业界，只有大型机构才有实力开发大模型。moss开源后，可有效降低预训练语言模型的研发和应用门槛，让中小企业在其基础上开发出智能客服、智能家居、人工智能律师等各种垂直类产品。“moss的计算量相对不那么大，中小企业都能用起来。”邱锡鹏说，“期待复旦团队和上海人工智能实验室继续合作，通过moss以及后续的研究探索，推动ai普惠，尽快赋能国内的人工智能产业。”","news_type":1,"symbols_score_info":{}},"isVote":1,"tweetType":1,"viewCount":1401,"commentLimit":10,"likeStatus":false,"favoriteStatus":false,"reportStatus":false,"symbols":[],"verified":2,"subType":0,"readableState":1,"langContent":"EN","currentLanguage":"EN","warmUpFlag":false,"orderFlag":false,"shareable":true,"causeOfNotShareable":"","featuresForAnalytics":[],"commentAndTweetFlag":false,"andRepostAutoSelectedFlag":false,"upFlag":false,"length":22,"xxTargetLangEnum":"ORIG"},"commentList":[],"hasMoreComment":false,"orderType":2}