Profile - To The Moon

漫想无边

Posts · 10Posts · 10
Following · 0Following · 0
Followers · 0Followers · 0

- 漫想无边漫想无边
  ·2023-02-12
  √√√
  More than 10 major manufacturers are competing to be ChatGPT. Who has the strength and who is gaining popularity?
  大厂扎堆涌入，但有实力的玩家并不多。
  More than 10 major manufacturers are competing to be ChatGPT. Who has the strength and who is gaining popularity?
  1.02K Comment
  Like
  Share
  Report

Most Discussed

{"i18n":{"language":"en_US"},"isCurrentUser":false,"userPageInfo":{"id":"3498099723896079","uuid":"3498099723896079","gmtCreate":1577527184934,"gmtModify":1704447507598,"name":"漫想无边","pinyin":"mxwbmanxiangwubian","introduction":"","introductionEn":"","signature":"","avatar":"https://static.tigerbbs.com/2b4cb59025b29af53a2e56509f535f1a","hat":null,"hatId":null,"hatName":null,"vip":1,"status":2,"fanSize":2,"headSize":14,"tweetSize":10,"questionSize":0,"limitLevel":999,"accountStatus":4,"level":{"id":1,"name":"萌萌虎","nameTw":"萌萌虎","represent":"呱呱坠地","factor":"评论帖子3次或发布1条主帖（非转发）","iconColor":"3C9E83","bgColor":"A2F1D9"},"themeCounts":0,"badgeCounts":0,"badges":[],"moderator":false,"superModerator":false,"manageSymbols":null,"badgeLevel":null,"boolIsFan":false,"boolIsHead":false,"favoriteSize":3,"symbols":null,"coverImage":null,"realNameVerified":"success","userBadges":[{"badgeId":"44212b71d0be4ec88898348dbe882e03-1","templateUuid":"44212b71d0be4ec88898348dbe882e03","name":"Boss Tiger","description":"The transaction amount of the securities account reaches $100,000","bigImgUrl":"https://static.tigerbbs.com/c8dfc27c1ee0e25db1c93e9d0b641101","smallImgUrl":"https://static.tigerbbs.com/f43908c142f8a33c78f5bdf0e2897488","grayImgUrl":"https://static.tigerbbs.com/82165ff19cb8a786e8919f92acee5213","redirectLinkEnabled":0,"redirectLink":null,"hasAllocated":1,"isWearing":0,"stamp":null,"stampPosition":0,"hasStamp":0,"allocationCount":1,"allocatedDate":"2025.03.25","exceedPercentage":"60.35%","individualDisplayEnabled":0,"backgroundColor":null,"fontColor":null,"individualDisplaySort":0,"categoryType":1101},{"badgeId":"a83d7582f45846ffbccbce770ce65d84-1","templateUuid":"a83d7582f45846ffbccbce770ce65d84","name":"Real Trader","description":"Completed a transaction","bigImgUrl":"https://static.tigerbbs.com/2e08a1cc2087a1de93402c2c290fa65b","smallImgUrl":"https://static.tigerbbs.com/4504a6397ce1137932d56e5f4ce27166","grayImgUrl":"https://static.tigerbbs.com/4b22c79415b4cd6e3d8ebc4a0fa32604","redirectLinkEnabled":0,"redirectLink":null,"hasAllocated":1,"isWearing":0,"stamp":null,"stampPosition":0,"hasStamp":0,"allocationCount":1,"allocatedDate":"2021.12.21","exceedPercentage":null,"individualDisplayEnabled":0,"backgroundColor":null,"fontColor":null,"individualDisplaySort":0,"categoryType":1100}],"userBadgeCount":2,"currentWearingBadge":null,"individualDisplayBadges":null,"crmLevel":5,"crmLevelSwitch":0,"location":null,"starInvestorFollowerNum":0,"starInvestorFlag":false,"starInvestorOrderShareNum":0,"subscribeStarInvestorNum":0,"ror":null,"winRationPercentage":null,"showRor":false,"investmentPhilosophy":null,"starInvestorSubscribeFlag":false},"page":1,"watchlist":null,"tweetList":[{"id":625367166,"gmtCreate":1676201470957,"gmtModify":1676201470957,"author":{"id":"3498099723896079","authorId":"3498099723896079","name":"漫想无边","avatar":"https://static.tigerbbs.com/2b4cb59025b29af53a2e56509f535f1a","crmLevel":5,"crmLevelSwitch":0,"followedFlag":false,"idStr":"3498099723896079","authorIdStr":"3498099723896079"},"themes":[],"htmlText":"√√√","listText":"√√√","text":"√√√","images":[],"top":1,"highlighted":1,"essential":1,"paper":1,"likeSize":0,"commentSize":0,"repostSize":0,"link":"https://ttm.financial/post/625367166","repostId":"1162170218","repostType":2,"repost":{"id":"1162170218","kind":"news","pubTimestamp":1676190965,"share":"https://ttm.financial/m/news/1162170218?lang=en_US&edition=fundamental","pubTime":"2023-02-12 16:36","market":"hk","language":"zh","title":"More than 10 major manufacturers are competing to be ChatGPT. Who has the strength and who is gaining popularity?","url":"https://stock-news.laohu8.com/highlight/detail?id=1162170218","media":"Tech星球","summary":"大厂扎堆涌入，但有实力的玩家并不多。","content":"<p><html><head></head><body>Recently, Zhou Hongyi, founder of 360, commented on the Internet circle: The United States is ahead of us in the field of ChatGPT, and Chinese companies and scientific researchers must catch up.</p><p>ChatGPT does make domestic and foreign Internet companies a little angry. On February 8, in order to resist the impact brought by ChatGPT, Google hastily launched the artificial intelligence chatbot (10.940,-0.04,-0.36%) Bard. Domestically, Baidu announced the launch of \"Ernie Bot\" in early March, and other Internet companies are also scrambling.</p><p>As of February 10, major domestic manufacturers that have announced that they have ChatGPT-like technology reserves, are ready to release, and have released ChatGPT-like products and services include: Alibaba, Tencent, Huawei, ByteDance, JD.com, 360, iFlytek (47.910, 1.51, 3.25%), Kuaishou, NetEase Youdao, etc., more than 10 major Internet companies have entered the hot ChatGPT.</p><p>Under the wind, sediment and sand mixed. The first Internet company in China to officially announce its entry into ChatGPT is actually Secoo, a second-hand luxury goods trading platform. ChatGPT concept stocks caused its stock price to rise by 120%. Many netizens complained that it was not Secoo's initiative to speak out, and they still couldn't remember Secoo's arrears to consumers.</p><p>In addition, there are also participants in Internet vertical track companies. TAL will introduce ChatGPT-like technology in the future, and is currently preparing to develop ChatGPT educational projects. SenseTime is also exploring the construction of a general-purpose AIGC model. The project is led by SenseTime Research Institute.</p><p>Small and medium-sized technology companies are gaining popularity, and large companies are not easy to control \"ChatGPT\". For example, when Google Bard was released, a factual mistake was made in the demonstration video, and the first show \"overturned\", causing Google's stock price to plummet, and its current market value has dropped by US $170 billion.</p><p>Looking at domestic Internet companies, who has more scientific research capabilities to do ChatGPT? Who is investing in research and development of real knives and guns, and who is chasing the trend and rubbing hot concepts?</p><p><b>Get together to enter the ChatGPT competition</b></p><p>Baidu, which has the deepest accumulation in the fields of search and AI, naturally has high hopes from the market.</p><p>Baidu plans to launch an artificial intelligence chatbot service similar to ChatGPT in March this year. The name of the project is determined to be \"ERNIE Bot\", and the English name is ERNIE Bot. After completing the internal test in March, it will be open to the public. The difference is that the initial version will be embedded in its search service. This tool will allow users to get conversational search results.</p><p>In September last year, Baidu CEO Li Yanhong judged that the development of artificial intelligence has \"changed directionally at both the technical level and the commercial application level.\" It is speculated that Baidu started to do \"Ernie Bot\" at that time, and according to the rhythm of Google and Microsoft, the open internal beta of \"Ernie Bot\" may be advanced.</p><p>Although it did not take the lead in releasing ChatGPT, in terms of related technologies, Baidu has the computing power, algorithms and data required for artificial intelligence at the same time. It has a layout in the four-layer technology stack of chips, frameworks, models, and applications. It is one of the few domestic companies with full-stack AI. capable enterprises.</p><p><img src=\"https://static.tigerbbs.com/0e0daa170662f45ea545fe493b640d1d\" tg-width=\"550\" tg-height=\"673\" referrerpolicy=\"no-referrer\"/></p><p>Compared with Baidu's technology research and development reserves, Ali and ByteDance may be the other two players with earlier layout.</p><p>On February 9, the Dharma Academy version of ChatGPT in Alibaba's internal beta was exposed in advance. The functions that the new product can realize include knowledge question and answer, AI painting, code generation, novel continuation, copywriting, poetry and lyrics writing, etc. From the perspective of manual question and answer level, the Dharma Academy version of ChatGPT also achieves relatively real answers, and also achieves multiple rounds of question and answer.</p><p>In fact, in May 2022, after the release of the ultra-large-scale multi-modal pre-training model M6, Ali announced the corresponding capabilities. In terms of AIGC, the generated real-life pictures are already very realistic.</p><p><img src=\"https://static.tigerbbs.com/039d0964d8a6724b2b53f5493b7fd881\" tg-width=\"550\" tg-height=\"550\" referrerpolicy=\"no-referrer\"/></p><p>According to the introduction, M6 has expanded the scale of model parameters to ten trillion, and an average single card can carry 20 billion parameters. According to data released by Ali, the energy consumption of M6 at the same parameter scale is only 1% of the previous industry benchmark, which greatly lowers the threshold for large model training.</p><p>On the road of developing ChatGPT, although Byte's full-stack technology is not very prominent, it is relatively advanced in commercial use. Tech Planet learned that since the end of last year, Byte has been developing ChatGPT-like projects. It has currently launched a test plug-in internally, led by AI Lab, and is still being explored. And before that, Byte's AML team internally developed the Volcano Engine large model training framework veGiantModel. This model is similar to the GPT-3. 5 series of models that support ChatGPT.</p><p>Before GPT-3. 5, a former developer of Volcano Engine told Tech Planet that Byte has been researching training models and products based on GPT2 and GPT3 very early, and Byte's Volcano Engine is also working on NLP (natural language processing) There is research accumulation in applications.</p><p>At present, there is a lot of research on byte dialogue robots, machine translation, and machine writing, especially for content creation. Before ChatGPT was used to write articles, ByteDance developed the Xiaomingbot automatic writing platform. Since its launch in 2016, it has written a total of 600,000 articles, covering 17 sports events and supporting 6 languages.</p><p>Recently, Kuaishou also stated that it is carrying out research on large-scale language models (LLM Large Language Model) and has launched corresponding special projects, covering fields such as LLM model training, automatic copywriting creation and generation, and dialogue system development.</p><p>In addition, JD Yunyanxi also announced that it will launch an industrial version of ChatGPT: ChatJD, with an estimated parameter volume of hundreds of billions, which is mainly used in two fields: retail and finance.</p><p><b>Why did major domestic manufacturers fail to seize the opportunity?</b></p><p>Big manufacturers are pouring in, but there are not many powerful players.</p><p>Some studies estimate that training the 175 billion-parameter language large model GPT-3 requires tens of thousands of CPUs/GPUs to input data 24 hours a day. The required energy consumption is equivalent to driving back and forth between the earth and the moon, and one operation costs 4.5 million dollars. (approximately 30 million yuan).</p><p>High R&D costs mean that mainstream large models can only be mastered by large technology companies or a few research institutions.</p><p>In terms of the number of parameters, Baidu's Wenxin large model has 260 billion parameters, which is much higher than the 170 billion parameters of GPT-3. However, the number of parameters is not representative of the model results. There is no shortage of data in China. Companies such as Baidu and ByteDance have accumulated massive amounts of data, but subsequent data cleaning, annotation, model structure design, training, reasoning, etc. will take time to accumulate. This may be an important reason why the performance results of domestic large models are not as amazing as those of foreign countries.</p><p>The key time point to widen the gap is in mid-2020, when GPT3.0 first came out. A circulated expert minutes shows that at that time, many people may think that GPT3.0 is not only a technology, but also embodies the new concept of AI model development, because OpenAI, the developer of ChatGPT, has unique insights into this area of technology and is also ahead of foreign Google., Google is probably half a year to one year behind OpenAI, and domestic companies are two years behind.</p><p>This is also related to the fact that many technology companies at home and abroad believe that ChatGPT is not a revolutionary technology. ChatGPT is a conversational robot developed by OpenAI after fine-tuning its GPT-3 model released in 2020. Strictly speaking, ChatGPT has not achieved a significant breakthrough in the underlying model. It cleverly adopts the combination of understanding, generation and interaction. Intensive training based on human feedback gives people a sense of intelligence in experience.</p><p>Yann LeCun, chief artificial intelligence scientist at Meta, expressed his views on ChatGPT at a small online gathering of media and executives not long ago: \"As far as the underlying technology is concerned, ChatGPT is not particularly innovative, nor is it revolutionary. Things. Many research labs are using the same technology and doing the same work.\"</p><p>A Baidu employee told Tech Planet that in fact, many major domestic manufacturers have been doing it for a long time, but they are not in the direction of GPT, but in other industrial directions. In 2022, many developers will make applications based on Baidu Wenxin, mostly in scenarios such as psychological counseling, medical advice, and college entrance examination volunteers. Ali's' Tongyi 'large model also covers 200 + service scenarios, including AIGC, and intelligent dialogue is also available, but it is not very strong.</p><p>I didn't realize that ChatGPT disrupted the existing search model, and didn't accurately predict the relationship that ChatGPT would replace mental workers. The crux of the problem is that many Internet companies think that \"chat\" is a very shallow application. From the current scenario, it is a stronger chatbot with certain question-and-answer attributes, and there is no commercial point in sight for the time being.</p><p>A byte NPL algorithm engineer who shares the same view as above told Tech Planet that ChatGPT is a project with high uncertainty, and there is no successful precedent in China. If All-in is really done, the above may not necessarily provide resources, let alone prioritize it. In addition, domestic technology is indeed limited in capabilities. For example, if you want to make a large model, you lack not only funds, but also manpower and basic models. For commercial companies such as major Internet companies, it is relatively difficult.</p><p>There are not only cognitive limitations, but also technical limitations.</p><p>Currently, ChatGPT's information is as of 2021, which means that if it is used as a search tool, it is obviously impossible to obtain real-time information. A Baidu employee said that Baidu is now solving the problem of real-time information. At the same time, ChatGPT is already preparing to launch iterative products based on GPT-4, and domestic companies have new goals to catch up with.</p><p><b>In the craze of following the trend, who is trying to rub the hot spots?</b></p><p>\"Companies with large models can do ChatGPT, but those that don't are rubbing hot spots.\"</p><p>Yang Hongxia, the former head of Alibaba Cloud's large model, sharply commented on the current situation of Internet companies flocking to do ChatGPT. She believes: \"Only when large models reach the level above 100B can they have GOT capabilities (or provide answers to more difficult reasoning questions).\"</p><p>From this perspective, Baidu has a flying paddle platform with more than 200,000 enterprise users, as well as a Wenxin large model, and is expected to make conversation products closest to the level of ChatGPT; Ali has a \"Tongyi\" large model, which has achieved trillions of parameter training, and has rich application scenarios for M6 development on the industrial side; Huawei began to plan for large models in 2020. In 2021, it released the Pengcheng Pangu large model, which is the industry's first 100-billion-level NLP large model to generate and understand Chinese.</p><p>In contrast, some Internet companies that have announced their entry into ChatGPT are in the second echelon in technology research and development and progress. For example, 360 has the deep learning platform Xlearning, but there is no related large model. NetEase Youdao, which also announced its entry into ChatGPT, does not have AI chips, deep learning frameworks, and large AI models. It only has online education of application scenarios. This is similar to Xueersi, which is rumored to participate in ChatGPT. Enterprises only have learning machine applications scenarios, but there are no relevant technical reserves.</p><p>Other companies are considered to be chasing the concept of the trend. Hanwang Technology (33.380,-2.04,-5.76%) benefited from ChatGPT concept stocks, with seven daily limits starting on January 30. According to the 2022 semi-annual report, its text big data business revenue close to chatGPT is 77.9795 million, accounting for only 14% of the total revenue. Not only does the revenue account for a relatively small proportion, but it also falls by 7.14% year-on-year. The development of the enterprise is difficult, but with the help of the ChatGPT concept, the stock price has soared.</p><p>This is not the most striking. On February 6, Secoo announced that it would explore the application of AIGC and ChatGPT technologies to luxury goods-related businesses. On that day, Secoo's share price rose 124.4% per share. In fact, this company has been included in the list of \"Lao Lai\". Many suppliers' arrears and consumer refunds have not been settled immediately, and it has been listed on the list of dishonesty by the State Administration for Market Regulation many times.</p><p>Unlike these popular companies that actively use ChatGPT, there are also companies that have done nothing and win with the help of ChatGPT. On February 8, Microsoft's new version of Bing integrating ChatGPT technology was launched. Many of the answers in the answers came from Zhihu. Zhihu's stock price once rose by more than 56% that afternoon. Since ChatGPT may subvert community Q&A, Zhihu has remained silent on ChatGPT-related matters. But it was also welcomed by the market and put it on the train of stock price growth.</p><p>Of course, some industry insiders said that many small and medium-sized enterprises do not have technical strength and can also use the open source Transformer large model to realize ChatGPT functional services to a certain extent.</p><p>However, experts from the Dharma Academy told Tech Planet that although the current large-scale pre-trained Transformer-based model can well solve tasks related to perceptual intelligence, it lacks a well-designed pure Transformer model and basically cannot complete the attempt to realize cognitive intelligence.</p><p>As in the case of ByteDance, Tech Planet asked Volcano Engine's pre-sales support staff whether Volcano Engine has ChatGPT-like products? The other party said that the platform does not have this product, but it can combine its own products through the \"machine learning platform\", and after secondary development, it can achieve ChatGPT-like effects.</p><p>Tech Planet also discovered that after the explosion of ChatGPT, Wang Huiwen, the former \"No. 2 person\" of Meituan, was suspected of making a comeback, claiming to bring US $50 million into the artificial intelligence game. At the same time, positions related to NPL, AIGC, etc. have attracted the attention of many people, and algorithm positions have become a new hot job search. In Maimai's job popularity list, 8 of the top ten positions are algorithm-related positions.</p><p><img src=\"https://static.tigerbbs.com/fee122729017f1a35992123485884e48\" tg-width=\"550\" tg-height=\"717\" referrerpolicy=\"no-referrer\"/></p><p>Caption: Hot list of Maimai positions.</p><p>Internet companies, including Tencent, Alibaba, Byte, Xiaohongshu, Cainiao, etc., have added requirements for practitioners to be familiar with GPT-3 and other related models to related recruitment positions such as NPL algorithm engineers. The monthly salary of recruitment is mostly around 40,000 yuan, up to 100,000 yuan.</p><p>In 2023, the fire ignited by ChatGPT can temper real gold companies and talents, and it can also evaporate false concepts.</p><p></body></html></p>","source":"techxq","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>More than 10 major manufacturers are competing to be ChatGPT. Who has the strength and who is gaining popularity?</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\">\nMore than 10 major manufacturers are competing to be ChatGPT. Who has the strength and who is gaining popularity?\n</h2>\n<h4 class=\"meta\">\n<p class=\"head\">\n<strong class=\"h-name small\">Tech星球</strong><span class=\"h-time small\">2023-02-12 16:36</span>\n</p>\n</h4>\n</header>\n<article>\n<p><html><head></head><body>Recently, Zhou Hongyi, founder of 360, commented on the Internet circle: The United States is ahead of us in the field of ChatGPT, and Chinese companies and scientific researchers must catch up.</p><p>ChatGPT does make domestic and foreign Internet companies a little angry. On February 8, in order to resist the impact brought by ChatGPT, Google hastily launched the artificial intelligence chatbot (10.940,-0.04,-0.36%) Bard. Domestically, Baidu announced the launch of \"Ernie Bot\" in early March, and other Internet companies are also scrambling.</p><p>As of February 10, major domestic manufacturers that have announced that they have ChatGPT-like technology reserves, are ready to release, and have released ChatGPT-like products and services include: Alibaba, Tencent, Huawei, ByteDance, JD.com, 360, iFlytek (47.910, 1.51, 3.25%), Kuaishou, NetEase Youdao, etc., more than 10 major Internet companies have entered the hot ChatGPT.</p><p>Under the wind, sediment and sand mixed. The first Internet company in China to officially announce its entry into ChatGPT is actually Secoo, a second-hand luxury goods trading platform. ChatGPT concept stocks caused its stock price to rise by 120%. Many netizens complained that it was not Secoo's initiative to speak out, and they still couldn't remember Secoo's arrears to consumers.</p><p>In addition, there are also participants in Internet vertical track companies. TAL will introduce ChatGPT-like technology in the future, and is currently preparing to develop ChatGPT educational projects. SenseTime is also exploring the construction of a general-purpose AIGC model. The project is led by SenseTime Research Institute.</p><p>Small and medium-sized technology companies are gaining popularity, and large companies are not easy to control \"ChatGPT\". For example, when Google Bard was released, a factual mistake was made in the demonstration video, and the first show \"overturned\", causing Google's stock price to plummet, and its current market value has dropped by US $170 billion.</p><p>Looking at domestic Internet companies, who has more scientific research capabilities to do ChatGPT? Who is investing in research and development of real knives and guns, and who is chasing the trend and rubbing hot concepts?</p><p><b>Get together to enter the ChatGPT competition</b></p><p>Baidu, which has the deepest accumulation in the fields of search and AI, naturally has high hopes from the market.</p><p>Baidu plans to launch an artificial intelligence chatbot service similar to ChatGPT in March this year. The name of the project is determined to be \"ERNIE Bot\", and the English name is ERNIE Bot. After completing the internal test in March, it will be open to the public. The difference is that the initial version will be embedded in its search service. This tool will allow users to get conversational search results.</p><p>In September last year, Baidu CEO Li Yanhong judged that the development of artificial intelligence has \"changed directionally at both the technical level and the commercial application level.\" It is speculated that Baidu started to do \"Ernie Bot\" at that time, and according to the rhythm of Google and Microsoft, the open internal beta of \"Ernie Bot\" may be advanced.</p><p>Although it did not take the lead in releasing ChatGPT, in terms of related technologies, Baidu has the computing power, algorithms and data required for artificial intelligence at the same time. It has a layout in the four-layer technology stack of chips, frameworks, models, and applications. It is one of the few domestic companies with full-stack AI. capable enterprises.</p><p><img src=\"https://static.tigerbbs.com/0e0daa170662f45ea545fe493b640d1d\" tg-width=\"550\" tg-height=\"673\" referrerpolicy=\"no-referrer\"/></p><p>Compared with Baidu's technology research and development reserves, Ali and ByteDance may be the other two players with earlier layout.</p><p>On February 9, the Dharma Academy version of ChatGPT in Alibaba's internal beta was exposed in advance. The functions that the new product can realize include knowledge question and answer, AI painting, code generation, novel continuation, copywriting, poetry and lyrics writing, etc. From the perspective of manual question and answer level, the Dharma Academy version of ChatGPT also achieves relatively real answers, and also achieves multiple rounds of question and answer.</p><p>In fact, in May 2022, after the release of the ultra-large-scale multi-modal pre-training model M6, Ali announced the corresponding capabilities. In terms of AIGC, the generated real-life pictures are already very realistic.</p><p><img src=\"https://static.tigerbbs.com/039d0964d8a6724b2b53f5493b7fd881\" tg-width=\"550\" tg-height=\"550\" referrerpolicy=\"no-referrer\"/></p><p>According to the introduction, M6 has expanded the scale of model parameters to ten trillion, and an average single card can carry 20 billion parameters. According to data released by Ali, the energy consumption of M6 at the same parameter scale is only 1% of the previous industry benchmark, which greatly lowers the threshold for large model training.</p><p>On the road of developing ChatGPT, although Byte's full-stack technology is not very prominent, it is relatively advanced in commercial use. Tech Planet learned that since the end of last year, Byte has been developing ChatGPT-like projects. It has currently launched a test plug-in internally, led by AI Lab, and is still being explored. And before that, Byte's AML team internally developed the Volcano Engine large model training framework veGiantModel. This model is similar to the GPT-3. 5 series of models that support ChatGPT.</p><p>Before GPT-3. 5, a former developer of Volcano Engine told Tech Planet that Byte has been researching training models and products based on GPT2 and GPT3 very early, and Byte's Volcano Engine is also working on NLP (natural language processing) There is research accumulation in applications.</p><p>At present, there is a lot of research on byte dialogue robots, machine translation, and machine writing, especially for content creation. Before ChatGPT was used to write articles, ByteDance developed the Xiaomingbot automatic writing platform. Since its launch in 2016, it has written a total of 600,000 articles, covering 17 sports events and supporting 6 languages.</p><p>Recently, Kuaishou also stated that it is carrying out research on large-scale language models (LLM Large Language Model) and has launched corresponding special projects, covering fields such as LLM model training, automatic copywriting creation and generation, and dialogue system development.</p><p>In addition, JD Yunyanxi also announced that it will launch an industrial version of ChatGPT: ChatJD, with an estimated parameter volume of hundreds of billions, which is mainly used in two fields: retail and finance.</p><p><b>Why did major domestic manufacturers fail to seize the opportunity?</b></p><p>Big manufacturers are pouring in, but there are not many powerful players.</p><p>Some studies estimate that training the 175 billion-parameter language large model GPT-3 requires tens of thousands of CPUs/GPUs to input data 24 hours a day. The required energy consumption is equivalent to driving back and forth between the earth and the moon, and one operation costs 4.5 million dollars. (approximately 30 million yuan).</p><p>High R&D costs mean that mainstream large models can only be mastered by large technology companies or a few research institutions.</p><p>In terms of the number of parameters, Baidu's Wenxin large model has 260 billion parameters, which is much higher than the 170 billion parameters of GPT-3. However, the number of parameters is not representative of the model results. There is no shortage of data in China. Companies such as Baidu and ByteDance have accumulated massive amounts of data, but subsequent data cleaning, annotation, model structure design, training, reasoning, etc. will take time to accumulate. This may be an important reason why the performance results of domestic large models are not as amazing as those of foreign countries.</p><p>The key time point to widen the gap is in mid-2020, when GPT3.0 first came out. A circulated expert minutes shows that at that time, many people may think that GPT3.0 is not only a technology, but also embodies the new concept of AI model development, because OpenAI, the developer of ChatGPT, has unique insights into this area of technology and is also ahead of foreign Google., Google is probably half a year to one year behind OpenAI, and domestic companies are two years behind.</p><p>This is also related to the fact that many technology companies at home and abroad believe that ChatGPT is not a revolutionary technology. ChatGPT is a conversational robot developed by OpenAI after fine-tuning its GPT-3 model released in 2020. Strictly speaking, ChatGPT has not achieved a significant breakthrough in the underlying model. It cleverly adopts the combination of understanding, generation and interaction. Intensive training based on human feedback gives people a sense of intelligence in experience.</p><p>Yann LeCun, chief artificial intelligence scientist at Meta, expressed his views on ChatGPT at a small online gathering of media and executives not long ago: \"As far as the underlying technology is concerned, ChatGPT is not particularly innovative, nor is it revolutionary. Things. Many research labs are using the same technology and doing the same work.\"</p><p>A Baidu employee told Tech Planet that in fact, many major domestic manufacturers have been doing it for a long time, but they are not in the direction of GPT, but in other industrial directions. In 2022, many developers will make applications based on Baidu Wenxin, mostly in scenarios such as psychological counseling, medical advice, and college entrance examination volunteers. Ali's' Tongyi 'large model also covers 200 + service scenarios, including AIGC, and intelligent dialogue is also available, but it is not very strong.</p><p>I didn't realize that ChatGPT disrupted the existing search model, and didn't accurately predict the relationship that ChatGPT would replace mental workers. The crux of the problem is that many Internet companies think that \"chat\" is a very shallow application. From the current scenario, it is a stronger chatbot with certain question-and-answer attributes, and there is no commercial point in sight for the time being.</p><p>A byte NPL algorithm engineer who shares the same view as above told Tech Planet that ChatGPT is a project with high uncertainty, and there is no successful precedent in China. If All-in is really done, the above may not necessarily provide resources, let alone prioritize it. In addition, domestic technology is indeed limited in capabilities. For example, if you want to make a large model, you lack not only funds, but also manpower and basic models. For commercial companies such as major Internet companies, it is relatively difficult.</p><p>There are not only cognitive limitations, but also technical limitations.</p><p>Currently, ChatGPT's information is as of 2021, which means that if it is used as a search tool, it is obviously impossible to obtain real-time information. A Baidu employee said that Baidu is now solving the problem of real-time information. At the same time, ChatGPT is already preparing to launch iterative products based on GPT-4, and domestic companies have new goals to catch up with.</p><p><b>In the craze of following the trend, who is trying to rub the hot spots?</b></p><p>\"Companies with large models can do ChatGPT, but those that don't are rubbing hot spots.\"</p><p>Yang Hongxia, the former head of Alibaba Cloud's large model, sharply commented on the current situation of Internet companies flocking to do ChatGPT. She believes: \"Only when large models reach the level above 100B can they have GOT capabilities (or provide answers to more difficult reasoning questions).\"</p><p>From this perspective, Baidu has a flying paddle platform with more than 200,000 enterprise users, as well as a Wenxin large model, and is expected to make conversation products closest to the level of ChatGPT; Ali has a \"Tongyi\" large model, which has achieved trillions of parameter training, and has rich application scenarios for M6 development on the industrial side; Huawei began to plan for large models in 2020. In 2021, it released the Pengcheng Pangu large model, which is the industry's first 100-billion-level NLP large model to generate and understand Chinese.</p><p>In contrast, some Internet companies that have announced their entry into ChatGPT are in the second echelon in technology research and development and progress. For example, 360 has the deep learning platform Xlearning, but there is no related large model. NetEase Youdao, which also announced its entry into ChatGPT, does not have AI chips, deep learning frameworks, and large AI models. It only has online education of application scenarios. This is similar to Xueersi, which is rumored to participate in ChatGPT. Enterprises only have learning machine applications scenarios, but there are no relevant technical reserves.</p><p>Other companies are considered to be chasing the concept of the trend. Hanwang Technology (33.380,-2.04,-5.76%) benefited from ChatGPT concept stocks, with seven daily limits starting on January 30. According to the 2022 semi-annual report, its text big data business revenue close to chatGPT is 77.9795 million, accounting for only 14% of the total revenue. Not only does the revenue account for a relatively small proportion, but it also falls by 7.14% year-on-year. The development of the enterprise is difficult, but with the help of the ChatGPT concept, the stock price has soared.</p><p>This is not the most striking. On February 6, Secoo announced that it would explore the application of AIGC and ChatGPT technologies to luxury goods-related businesses. On that day, Secoo's share price rose 124.4% per share. In fact, this company has been included in the list of \"Lao Lai\". Many suppliers' arrears and consumer refunds have not been settled immediately, and it has been listed on the list of dishonesty by the State Administration for Market Regulation many times.</p><p>Unlike these popular companies that actively use ChatGPT, there are also companies that have done nothing and win with the help of ChatGPT. On February 8, Microsoft's new version of Bing integrating ChatGPT technology was launched. Many of the answers in the answers came from Zhihu. Zhihu's stock price once rose by more than 56% that afternoon. Since ChatGPT may subvert community Q&A, Zhihu has remained silent on ChatGPT-related matters. But it was also welcomed by the market and put it on the train of stock price growth.</p><p>Of course, some industry insiders said that many small and medium-sized enterprises do not have technical strength and can also use the open source Transformer large model to realize ChatGPT functional services to a certain extent.</p><p>However, experts from the Dharma Academy told Tech Planet that although the current large-scale pre-trained Transformer-based model can well solve tasks related to perceptual intelligence, it lacks a well-designed pure Transformer model and basically cannot complete the attempt to realize cognitive intelligence.</p><p>As in the case of ByteDance, Tech Planet asked Volcano Engine's pre-sales support staff whether Volcano Engine has ChatGPT-like products? The other party said that the platform does not have this product, but it can combine its own products through the \"machine learning platform\", and after secondary development, it can achieve ChatGPT-like effects.</p><p>Tech Planet also discovered that after the explosion of ChatGPT, Wang Huiwen, the former \"No. 2 person\" of Meituan, was suspected of making a comeback, claiming to bring US $50 million into the artificial intelligence game. At the same time, positions related to NPL, AIGC, etc. have attracted the attention of many people, and algorithm positions have become a new hot job search. In Maimai's job popularity list, 8 of the top ten positions are algorithm-related positions.</p><p><img src=\"https://static.tigerbbs.com/fee122729017f1a35992123485884e48\" tg-width=\"550\" tg-height=\"717\" referrerpolicy=\"no-referrer\"/></p><p>Caption: Hot list of Maimai positions.</p><p>Internet companies, including Tencent, Alibaba, Byte, Xiaohongshu, Cainiao, etc., have added requirements for practitioners to be familiar with GPT-3 and other related models to related recruitment positions such as NPL algorithm engineers. The monthly salary of recruitment is mostly around 40,000 yuan, up to 100,000 yuan.</p><p>In 2023, the fire ignited by ChatGPT can temper real gold companies and talents, and it can also evaporate false concepts.</p><p></body></html></p>\n<div class=\"bt-text\">\n\n\n<p> source：<a href=\"https://finance.sina.com.cn/tech/internet/2023-02-12/doc-imyfmpit6023637.shtml\">Tech星球</a></p>\n\n\n</div>\n</article>\n</div>\n</body>\n</html>\n","type":0,"thumbnail":"https://static.tigerbbs.com/760056d3ed9e8c1a595c0d9f2094cc4d","relate_stocks":{},"source_url":"https://finance.sina.com.cn/tech/internet/2023-02-12/doc-imyfmpit6023637.shtml","is_english":false,"share_image_url":"https://static.laohu8.com/e9f99090a1c2ed51c021029395664489","article_id":"1162170218","content_text":"最近，360创始人周鸿祎又点评了互联网圈：在ChatGPT领域美国领先了我们，中国公司和科研人员要奋起直追。ChatGPT确实让国内外互联网公司有点上火。2月8日，为了抵御ChatGPT带来的冲击，谷歌仓促推出人工智能聊天机器人(10.940,-0.04,-0.36%)Bard。国内这边，百度宣布在3月初推出“文心一言”，其他家互联网公司也争先恐后。截止2月10日，宣布有类ChatGPT技术储备、准备发布和已经发布类ChatGPT产品服务的国内大厂有：阿里巴巴、腾讯、华为、字节跳动、京东、360、科大讯飞(47.910,1.51,3.25%)、快手、网易有道等，超过10余家互联网大厂进军火热的ChatGPT。风口之下，泥沙混杂。国内最早官宣进军ChatGPT的互联网企业，其实是二手奢侈品交易平台寺库，ChatGPT概念股让其股价应声上涨120%。很多网友吐槽，不是寺库主动发声，还想不起来寺库对消费者的欠款。此外，互联网垂类赛道企业也有参与者，好未来将在未来引入类ChatGPT技术，目前正着手筹备研发ChatGPT的教育项目。商汤也正在探索搭建通用型AIGC模型，该项目由商汤研究院牵头。中小科技公司在蹭热度，大公司也并非轻松掌控“ChatGPT”。诸如谷歌Bard发布时，演示视频中犯了一个事实性错误，首秀大“翻车”，致使谷歌股价大跌，目前市值已经跌去了1700亿美元。放眼国内互联网企业，谁更有科研实力做ChatGPT？谁在真刀真枪的投入研发，谁在追风口、蹭热点概念？扎堆入局ChatGPT竞赛在搜索和AI领域积累最深的百度，自然被市场寄予厚望。百度计划在今年3月推出与ChatGPT类似的人工智能聊天机器人服务，该项目名字确定为“文心一言”，英文名ERNIE Bot，3月份完成内测后，将面向公众开放。不同的是，最初版本将嵌入其搜索服务中。这项工具将允许用户获得对话式的搜索结果。去年9月，百度CEO李彦宏判断人工智能发展在“技术层面和商业应用层面，都有方向性改变”。据推测，百度那时候就开始做“文心一言”，另按照谷歌和微软节奏，“文心一言”开放内测还有可能提前。尽管没有率先发布ChatGPT，但在相关技术方面，百度同时具备人工智能需要的算力、算法和数据，在芯片、框架、模型、应用四层技术栈均有布局，是国内少数的具备AI全栈能力的企业。相比百度的技术研发储备，阿里和字节跳动可能是另外两个较早布局的玩家。2月9日，阿里内测中的达摩院版ChatGPT提前曝光。新产品可实现的功能包括知识问答、AI绘画、代码生成、小说续写、文案撰写、写诗作词等。从人工问答的水平看，达摩院版ChatGPT也实现了相对真实的回答，而且也实现了多轮问答。实际上，在2022年5月份，在超大规模多模态预训练模型M6的发布之后，阿里就曾公布过相应的能力，在AIGC方面，生成的实景图片已经十分逼真。根据介绍，M6将模型参数规模扩展到了十万亿，平均单卡可承载200亿参数。据阿里方面公布的数据显示，M6在同等参数规模能耗仅为此前业界标杆的1%，极大降低大模型训练门槛。字节在研发ChatGPT的道路上，虽然全栈技术不是很突出，但在商用方面走的比较靠前。Tech星球了解到，从去年年底开始，字节就在研发类ChatGPT项目，目前已在内部上线了测试插件，由AI Lab主导，当前还在探索当中。并且在此前，字节的AML团队内部开发了火山引擎大模型训练框架veGiantModel。这一模型类似于为ChatGPT提供支持的GPT-3.5系列模型。而在GPT-3.5之前，一位火山引擎的前开发人员告诉Tech星球，字节很早就着手研究基于GPT2、GPT3的训练模型和产品，并且字节旗下的火山引擎也在NLP（自然语言处理）应用方面有研究积累。目前，字节对话机器人、机器翻译、机器写作上有不少研究，特别是针对内容创作这一块。在ChatGPT被用作写文章前，字节跳动就研发了Xiaomingbot自动写稿平台，自2016年上线以来，已经累计写了60万篇文章，覆盖了17项的体育赛事，支持6种语言。近日，快手也对外表示，正在开展大规模语言模型（LLM Large Language Model）相关研究，并启动了相应专项，覆盖LLM模型训练、文案自动创作与生成、对话系统开发等领域。此外，京东云言犀也宣布将推出产业版ChatGPT：ChatJD ，预计参数量为千亿级，主要应用在零售和金融2个领域。国内大厂为何没能抢得先机？大厂扎堆涌入，但有实力的玩家并不多。有研究估测，训练1750亿参数语言大模型GPT-3，需要有上万个 CPU/GPU24小时不间输入数据，所需能耗相当于开车往返地球和月球，且一次运算就要花费450万美元（约合3000万元）。高昂的研发成本意味着，主流的大模型只能由大型科技公司或是少数研究机构掌握。从参数数量来说，百度的文心大模型有2600亿参数，远高于GPT-3的1700亿参数。但参数数量并不代表模型结果。国内并不缺数据，百度、字节跳动这样的公司都有海量的数据积累，不过后续包括数据清洗、标注以及模型的结构设计，训练、推理等都需要时间积累。这可能是国内大模型表现结果不如国外惊艳的重要原因。拉开差距的关键时间点是在2020年年中，也就是GPT3.0刚出来时。一份流传的专家纪要显示，当时可能很多人觉得GPT3.0不仅是技术，体现了AI模型的发展新理念，因为ChatGPT的开发商OpenAI对这一块有技术的独到见解，也领先了国外的Google，大概Google比OpenAI差了半年到一年的时间，国内公司差了两年的时间。这与国内外很多科技企业认为，ChatGPT并不是革命性的技术也有关系。ChatGPT是OpenAI对其2020年发布的GPT-3模型微调后开发出的对话机器人，严格来说，ChatGPT并未实现底层模型的显著突破，它巧妙地采用了理解、生成和交互相结合的方式，基于人类反馈进行强化训练，在体验上带给人智能的感觉。Meta首席人工智能科学家Yann LeCun，不久前在一次小型媒体和高管在线聚会上发表了自己对ChatGPT的看法：“就底层技术而言，ChatGPT 并没有什么特别的创新，也不是什么革命性的东西。许多研究实验室正在使用同样的技术，开展同样的工作。”一位百度员工告诉Tech星球，其实国内很多大厂做了很久，但都不是GPT方向，而是其他产业方向。2022年很多开发者基于百度文心做的应用，多是心理咨询、寻医问药、高考志愿等场景。阿里‘通义’大模型也覆盖200+服务场景，包括AIGC，智能对话也有，但不是很强。没有意识到ChatGPT对现有搜索模式的颠覆，没有准确预测到ChatGPT对脑力劳动者产生替代的关系。问题的关键在于，很多互联网企业认为“chat”这是一个很浅的应用，从现在的场景上，就是一个更强的聊天机器人，有一定的问答属性，暂时看不到商用的点。有着上述相同看法的一位字节NPL算法工程师告诉Tech星球，ChatGPT是一种不确定性极高的项目，国内没有成功先例，如果真的All-in去做，上面并不一定会提供资源，更别说提上优先级。另外，国内的技术确实是能力有限，比如想做一个大模型，缺少的不仅仅是资金，还有人力和基础模型，对于互联网大厂这种商业公司而言，有着较大的难度。不仅认知层面存在局限，技术上的局限也有。目前，ChatGPT的信息截止到了2021年，这意味着如果把它当做搜索工具使用，显然无法获得实时信息。一位百度员工表示，百度现在正在解决信息实时性的问题。同时，ChatGPT已经在准备推出基于GPT-4的迭代产品，国内企业又有新的目标需要追赶。跟风热潮中，谁在硬蹭热点？“有大模型的企业能做ChatGPT，没有的则是在蹭热点。”前阿里云大模型负责人杨红霞犀利点评了当下互联网公司，蜂拥扎堆做ChatGPT的现状。她认为：“大模型只有达到了100B以上的水平，才有GOT能力（或者提供较难推理问题的回答）。”从这方面看，百度有20多万企业用户的飞桨平台，也有文心大模型，预计能做出最接近ChatGPT水平的对话产品；阿里有“通义”大模型，目前已经实现万亿参数训练，在产业端M6开发的应用场景也比较丰富；华为则是在2020年对大模型开始有布局，2021年发布了鹏城盘古大模型，是业界首个千亿级生成和理解中文NLP大模型。相比之下，目前宣布进军ChatGPT的一些互联网企业，技术研发和进度则处在第二梯队。比如，360有深度学习平台Xlearning，但没有相关大模型。同样宣布进军ChatGPT的网易有道，是AI芯片、深度学习框架、AI大模型都没有，只是有应用场景在线教育，这与传出要参与ChatGPT的学而思类似，企业只是有学习机的应用场景，但没有相关技术储备。还有一些企业则被认为是追风口概念。汉王科技(33.380,-2.04,-5.76%)受益于ChatGPT概念股，在1月30日开始的七个涨停板。根据2022年半年报显示，其接近chatGPT的文本大数据业务营收为7797.95万，仅占总营收的14%，不仅营收占比较小，还同比下滑7.14%。企业发展困难，却借助ChatGPT概念实现了股价一飞冲天。这还不算最引人注目的，2月6日，寺库宣布将探索将AIGC和ChatGPT技术应用于奢侈品相关业务。当天，寺库每股股价上涨124.4%。实际这家企业已经被纳入“老赖”名单，很多供应商欠款和消费者退款都没即时结算，多次被市场监管总局登上失信名单。与这些主动蹭ChatGPT的热度企业不同，也有什么都没做的企业，借助ChatGPT躺赢。2月8日，微软集成ChatGPT技术的新版必应上线，回答内容中有不少答案来自知乎，当天午后知乎股价一度涨超56%。由于ChatGPT可能会对社区问答产生颠覆，所以知乎对ChatGPT相关事宜一直保持沉默。但也被市场热捧，送上了股价增长列车。当然，有行业人士表示，很多中小企业没有技术实力，也可借助开源的 Transformer大模型，一定程度实现ChatGPT功能服务。不过，达摩院专家告诉Tech星球，现在大规模预训练的 Transformer-based 模型虽然能够很好地解决感知智能相关任务，但缺乏精心设计的纯 Transformer 模型，基本无法完成对于实现认知智能的尝试。正如字节跳动的情况，Tech星球向火山引擎的售前支持人员咨询火山引擎是否有类ChatGPT的产品？对方表示，平台并没有该产品，但是可以通过“机器学习平台”结合自家产品，经过二次开发，能够实现类ChatGPT的效果。Tech星球还发现，在ChatGPT爆火后，前美团“2号人物”王慧文疑似复出，宣称要带资5000万美元入局人工智能。同时，与NPL、AIGC等相关的岗位被不少人关注，算法岗成为了新的求职热门。在脉脉的职位热度榜单中，前十位中，有8个是和算法相关的岗位。图注：脉脉职位热榜。包括腾讯、阿里、字节、小红书、菜鸟等互联网公司，已经在NPL算法工程师等相关招聘岗位中，加入了从业者需了解熟悉GPT-3等相关模型的要求。而招聘的月薪多在4万元左右，最高可达10万元。2023年，ChatGPT点燃的这把火，可以淬炼出真金的企业和人才，也可以蒸发掉虚妄的概念。","news_type":1,"symbols_score_info":{}},"isVote":1,"tweetType":1,"viewCount":1022,"authorTweetTopStatus":1,"verified":2,"comments":[],"imageCount":0,"langContent":"EN","totalScore":0}],"defaultTab":"posts","isTTM":true}