Founded in 2019, Zhipu AI has become the first domestic unicorn in the large model field with a valuation of over RMB 10 billion. Tencent, which "takes out the light bulbs a month earlier, doesn't matter", eventually had to follow the crowd. Tencent's Hunyuan model was officially launched, but users still had to queue up to apply. Ant Financial's big model appeared at the Bund Conference, claiming to solve the real proposition of the industry.
Image source: Generated by Unbounded AI
01
Financing
Zhipu AI’s latest valuation exceeded 10 billion yuan
According to exclusive information from AI Technology Review, the latest valuation of Zhipu AI has exceeded 10 billion yuan, and may reach as high as 15 billion yuan. It is one of the fastest domestic companies with a valuation exceeding 10 billion yuan. Zhipu AI was established in June 2019. It was transformed from the technical achievements of the Knowledge Engineering Laboratory of the Department of Computer Science of Tsinghua University. The core members of the team have participated in the research and development of the "Enlightenment" project, a cooperation project between Tsinghua University and Zhiyuan Research Institute.
AI smart hardware company "Future Intelligence" completed tens of millions of yuan in Pre-A round of financing
According to 36Kr, the AI intelligent hardware company "Future Intelligence" recently announced the completion of tens of millions of yuan in Pre-A round of financing, led by Tianji Capital. This round of financing will be used for new product launches, market expansion and other purposes. It is said that Future Intelligence is targeting office scenarios and has launched a variety of conference and office headsets such as iFLYBUDS Pro, iFLYBUDS Air, and iFLYBUDS Nano +.
Zhongtiao Technology completed tens of millions of yuan in A2 and A3 rounds of financing. The funds were mainly used for product iterations, large model upgrades, etc.
According to the Science and Technology Innovation Board Daily report on September 11, Shanghai Zhongdiao Information Technology Co., Ltd. recently announced the completion of the A2 and A3 rounds of financing invested by Yunqi Capital and Oriental Chase respectively, with a total amount of tens of millions of yuan, led by Shi Yu Capital acted as the exclusive financial advisor for this round. The funds are mainly used for product iteration, large model upgrades and expanding sales scale. Founded in 2015, Zhongtiao Technology focuses on the integration of automotive big data, intelligent algorithms, and intelligent empowerment.
AI startup Imbue received investment from Nvidia and others, with a valuation of over US$1 billion
According to a report by IT House on September 8, Imbue is a start-up company that builds large-scale language models to drive artificial intelligence tools. The company has currently raised US$200 million from investors such as Nvidia, and its valuation exceeds US$1 billion (currently approximately RMB 7.33 billion).
Silicon Valley AI startup d-Matrix completed US$110 million in Series B financing, led by Temasek
According to the Science and Technology Innovation Board Daily citing Reuters, Silicon Valley generative artificial intelligence chip startup d-Matrix recently completed a $110 million Series B round of financing. This round of financing was led by Singaporean investment company Temasek, with participation from California venture capital firm Playround Global, Microsoft's M12 Venture Fund, Nautilus Venture Partners, Entrada Ventures, etc. The funds raised will be used for the recruitment and commercialization of the company's Corsair platform.
American startup Delphi completes US$2.7 million in financing and launches AI digital cloning service
According to a report by Webmaster Home on September 7, American startup Delph raised US$2.7 million and launched a new AI digital cloning service. It is understood that the core function of this service is to allow users to generate AI chatbots that imitate the user's personality, writing or speaking style by uploading communication records, including emails, chat records, YouTube videos, etc. Through a partnership with voice cloning startup ElevenLabs, Delphi's AI clones can even mimic the user's voice.
AI chip developer "Zhonghao Xinying" completed tens of millions of yuan in Pre-B+ round of financing
According to the public account "Zhonghao Xinying Technology", Hangzhou Zhonghao Xinying recently completed tens of millions of yuan in Pre-B+ round of financing, with Hangzhou Huaxia Hengtian Capital participating in the investment. The funds will be used by the company to promote the implementation of AI computing cluster business.
Gleen AI receives $4.9 million in funding to use enterprise data to solve AI hallucination problem
According to a report by the Science and Technology Innovation Board Daily on September 6, the startup Gleen AI recently announced that it has received US$4.9 million in financing. Investors include Slow Ventures, 6th Man Ventures, South Park Commons, Spartan Group and other venture capital companies and Angel investors, including former Facebook/Meta platform VP of product management Sam Lessin, are used to continue developing anti-hallucination data layer software for enterprises by sifting through the enterprise's own internal data, converting it into a vector database, and using this data to improve Quality of AI model answers.
AI productivity platform Kindo raises US$7 million in financing
According to a report by Webmaster Home on September 6, artificial intelligence startup Kindo recently announced that it has received US$7 million in seed round financing to launch a secure AI productivity platform for enterprises.
Atomicwork raises $11 million to integrate AI into employee workflows
According to a report by Forbes on September 5, San Francisco-based startup Atomicwork announced that it has received $11 million in seed round financing to seek to commercialize the software-as-a-service solution it developed. Atomicwork’s work assistant, Atom, is powered by generative AI and is designed to allow employees to interact with it in a more natural way than traditional software.
02
large model
People familiar with the matter: Meta is developing a new large language model and plans to rival GPT-4
According to a report by the Wall Street Journal on September 10, people familiar with the matter revealed that Meta is developing a new artificial intelligence system whose functions will be comparable to the most advanced model GPT-4 provided by OpenAI. Meta's goal is that its new artificial intelligence model will be several times more powerful than the model it released two months ago called Llama 2. The planned system will help other companies build services capable of generating complex text, analytics and other outputs.
Baichuan Intelligent releases Baichuan 2 large model
According to the Science and Technology Innovation Board Daily report on September 6, Baichuan Intelligent announced that it has officially open sourced the fine-tuned Baichuan 2-7B, Baichuan 2-13B, Baichuan 2-13B-Chat and their 4-bit quantized versions, and they are all free and commercially available.
Tencent officially releases Hunyuan large language model, with parameter scale exceeding 100 billion
On the morning of September 7, at the Tencent Global Digital Ecosystem Conference, Tang Daosheng, senior executive vice president of Tencent Group and CEO of the Cloud and Smart Industry Group, released the Tencent Hunyuan large model, which is Tencent's full-link self-developed large language model. , the parameter scale exceeds 100 billion, the pre-training corpus exceeds 2 trillion tokens (large language model text unit), and it has Chinese creation capabilities, task execution capabilities and logical reasoning capabilities in complex contexts.
Baidu Smart Cloud announces the comprehensive upgrade and release of Qianfan Large Model Platform 2.0
According to reports from Webmaster Home, at the 2023 Baidu Cloud Intelligence Conference, Baidu Smart Cloud announced the comprehensive upgrade and release of Qianfan Large Model Platform 2.0, providing the most complete tool chain, the best computing performance, the most large models and data sets, and Enterprise-level security helps enterprises bridge the gap from large models to productivity.
Ant Financial large model officially released
Babbitt News, on September 8, at the Bund Conference in Shanghai, Ant Group officially released a large financial model. It is understood that the Ant Financial large model is based on Ant's self-developed basic large model and is deeply customized for the financial industry. The underlying computing power cluster reaches the scale of 10,000 cards. This large model focuses on the needs of real financial scenarios and performs outstandingly in 28 types of financial exclusive tasks in the five dimensions of "cognition, generation, professional knowledge, professional logic, and compliance". He has reached the level of industry experts in many fields such as "Financial Event Reasoning". Currently, the Ant Financial large model has been fully tested on Ant Group’s wealth and insurance platforms.
The iFlytek Spark model is now open to the public
Babbitt News, on September 5, iFlytek officially announced that starting today, the iFlytek Spark Cognitive Large Model will be open to the public. Users can download it from major app stores and register to use it directly. It is reported that iFlytek Spark is a new generation of cognitive intelligence model launched by iFlytek. It has cross-domain knowledge and language understanding capabilities, and can understand and perform tasks based on natural dialogue. iFlytek Spark was released on May 6 this year and has been upgraded to V2.0.
360 brain model is open to the public today
According to a report from the Associated Press on September 5, the 360 Smart Brain model will be open to the public from now on. Open carriers include five major platforms, including the 360 Zhi Nao official website, App application store, or upgrade 360 Security Guard, 360 Secure Browser, and 360 Search to the latest version, and log in to experience the large model service.
Leo Shares: Its subsidiary Leo Digital launches a large model in the marketing field
According to Jiemian News, on September 11, Leo Digital, a subsidiary of Leo Holdings, announced through its official Weibo account that it had strategically invested in the artificial intelligence start-up Hangzhou Guiyi Intelligent Technology Co., Ltd. and launched "Leo Guiyi", a large model in the marketing field. According to reports, enterprises can achieve intelligent execution and strategic optimization of advertising through the AI pitcher Agent "Guiyi Miaozhi" in the "Leo Guiyi" large-scale model in the marketing field.
Tencent Health releases large medical model
According to Jinshi News, on September 8, Tencent Health announced a large medical model and released a multi-scenario AI product matrix such as intelligent question and answer, family doctor assistant, and digital intelligent medical imaging platform. The drug discovery platform "iDrug" revealed a new Protein structure prediction algorithm framework tFold. At present, Tencent's medical large-scale models include scenario large-scale models such as copywriting generation, intelligent question and answer, medical record structuring and retrieval, imaging reports, and auxiliary diagnosis, which can be embedded in the entire medical process.
03
application
Tencent Conference connects to Hunyuan large model and releases AI assistant
According to the "Kechuangban Daily" news on September 9, Tencent Conference announced the access to Tencent Hunyuan large model and released the Tencent Conference AI assistant. Based on Tencent Conference AI Assistant, participants can complete various complex tasks such as information extraction, content analysis, conference management and control through simple and natural instructions. Tencent Conference also announced commercialization data: the number of users exceeded 400 million, and the number of paying users increased 5 times year-on-year.
China’s first large-scale smart financial assistant unveiled, Ant Fortune releases Zhi Xiaobao 2.0
Babbitt News, on September 8, at the "Bund Conference" held in Shanghai, Ant Wealth Platform released Zhi Xiaobao 2.0, China's first intelligent financial assistant using large model technology. Zhi Xiaobao version 2.0 is developed based on the large financial model just released by Ant Group. It has been greatly improved in terms of knowledge, professionalism, language and security. It can help financial institutions provide users with high-quality market analysis and position diagnosis. , asset allocation and investment education companionship and other professional services, the knowledge level is not inferior to that of real industry experts. According to reports, Zhi Xiaobao 2.0 is currently in internal testing and will be launched after completing relevant filing work.
Taobao launches AI large model application "Taobao Wenqi" and conducts internal testing
According to 36Kr’s report on September 8, Taobao launched the AI large-model application “Taobao Wenqi” for internal testing. It is understood that Taobao Wenqi can understand consumers' needs across categories and fields, and can also realize needs and scenarios in information search, creative content generation, etc. Currently, the application can be submitted for internal testing by searching "Taobao Wenqi" on the Taobao App, or participating in the test through an invitation code. The official has not disclosed the time when the application will be fully launched.
Morgan Stanley will launch an AI chatbot jointly developed with OpenAI this month to provide investment insights
According to a report by Sina Finance on September 7, after months of testing with 1,000 financial advisors, Morgan Stanley will launch a generative artificial intelligence (AI) chatbot later this month. This artificial intelligence chatbot was jointly developed by Morgan Stanley and ChatGPT developer OpenAI.
Alimama releases two new AI delivery products, "Wanxiangtai Unbounded Edition" and "Alimama Bailing"
According to 36Kr, on September 6, Alimama released two new AI advertising products at the 2023 m summit: the one-stop brand management platform "Alimama Bailing" and the one-stop marketing delivery system "Wanxiangtai Unbounded Edition". Create a "dual engine" of growth for merchants that combines business, product and performance.
Kingsoft Office: WPS AI is officially open to the public
Babbitt News, on September 5, Kingsoft Office officially announced that WPS AI is officially open to the public. AI capabilities are first applied to WPS smart documents, and more WPS AI capabilities will be opened in other components. Users can experience WPS smart documents in the latest version of WPS client/App, Kingsoft Document applet/official website.
It is said that WPS AI, as the first generative AI application implemented in China's collaborative office track, has been connected to WPS text, presentation, form, PDF, Kingsoft documents and other product lines to solve users' problems in content generation, content understanding, command operation, etc. daily office problems. Kingsoft Office positions WPS AI as the application of large language models, and anchors its future development in three strategic directions: AIGC (content creation), Copilot (smart assistant), and Insight (knowledge insight).
04
Supervision
It was revealed that many domestic small and medium-sized AI painting websites established using AIGC technology have been shut down.
According to a report by the public account "Shizhi Future" on September 7, there are many small and medium-sized AI painting websites in China established using AIGC technology, such as Sibisan (4B3.com), Libu Libu (liblibai.com), Tu company (tusiart.com) and others were shut down without warning, or transferred to overseas domain names, and at the same time became inaccessible in China.
It is reported that similar website closures are accompanied by supporting measures such as interviewing the person in charge of the website by the Cyberspace Administration of China. This type of interview was conducted in accordance with the relevant provisions of the "Administrative Law Enforcement Procedure Regulations of the Cyberspace Administration" and the "Interim Measures for the Management of Generative Artificial Intelligence Services", and the main reason was "the use of overseas ChatGPT products without evaluation" The interface provides functions such as intelligent dialogue and question answering, which poses great security risks."
As previously reported, the Cyberspace Administration of China, in conjunction with the National Development and Reform Commission, the Ministry of Education, the Ministry of Science and Technology, the Ministry of Industry and Information Technology, the Ministry of Public Security, and the State Administration of Radio, Film and Television, issued the "Interim Measures for the Management of Generative Artificial Intelligence Services" on July 10. Effective from today.
05
big company
NVIDIA releases TensorRT-LLM, which can improve H100 inference performance by up to 8 times
According to a report by IT House on September 9, NVIDIA recently announced the launch of TensorRT-LLM, a deeply optimized open source library that can accelerate the inference performance of all large language models on AI GPUs such as Hopper. In the performance test, NVIDIA used A100 as the basis and compared H100 and H100 with TensorRT-LLM enabled. In GPT-J 6B inference, the inference performance of H100 was 4 times higher than that of A100, while the performance of H100 with TensorRT-LLM enabled was better than A100. 8 times.
Media: ARM may be oversubscribed five times
ARM's U.S. IPO may be more than five times oversubscribed, according to bankers participating in ARM's roadshow, as the British chip design company expects the artificial intelligence (AI) boom to drive accelerated revenue growth in the future. ARM CEO Rene Haas said in a promotional video: "Artificial intelligence will be everywhere, and all AI will run on ARM's software. Our opportunities are unlimited. With the advent of the artificial intelligence era, global computing needs are Insatiable.”
06
View
Robin Li: It makes no sense for entrepreneurs to roll up large models, but there are greater opportunities for roll-out applications.
According to 36Kr’s report on September 5, Baidu founder, chairman and CEO Robin Li said at the Baidu Cloud Intelligence Conference that currently more than 70 large models with more than 1 billion parameters in China have been released, but the models themselves are not Applications developed based on basic large models that directly generate value are the meaning of the existence of these large models. "So for entrepreneurs, it is meaningless to roll up large models, and there are greater opportunities to roll up applications, just like the operating system in the mobile Internet era only has Android and iOS, and there are many successful apps.”
Managing partner of Jinshajiang Venture Capital: AIGC is expected to bring economic benefits of US$2.6 trillion to US$4.4 trillion to corporate business
According to the Shanghai Securities News, the 2023 Inclusion Bund Conference was held in Shanghai from September 7th to 9th. At the "New Wave of AI Frontier Technology Development and Application" sub-forum, Zhang Yutong, managing partner of Jinshajiang Venture Capital, predicted that AIGC will release huge value in key functions of the enterprise. It is expected that among the 16 business functions of the enterprise, AIGC can bring The total value of future economic benefits is US$2.6 trillion to US$4.4 trillion, with the impact concentrated on four functions, including customer operations, marketing and sales, software engineering, and product research and development, accounting for 75% of the total value.
Mobvoi CEO Li Zhifei: Computing power itself does not constitute a barrier to competition in large models
According to a report by the Science and Technology Innovation Board Daily on September 8, at the 2023 Tencent Global Digital Ecology Conference held today, Li Zhifei, CEO of Mobvoi, a Chengshi AI and voice interactive AI startup, said that the computing power of large models is a consumable. , does not constitute a barrier in itself, and investment needs to be rational and moderate. Data, especially the data accumulated by users in their own products, is more critical. The data flywheel is difficult, but once established, it is a more reliable barrier than models and computing power.
Tencent Tang Daosheng: It is too early to talk about the specific business prospects of large models. Competition in the large model market is still in the first mile.
According to The Paper, on September 7, Tang Daosheng, senior executive vice president of Tencent Group and CEO of the cloud and smart industry group, said that it is too early to talk about the specific business prospects of large models, and the competition in the large model market is still in the first place. kilometer. Different scenarios require different models, and it is not necessarily necessary to use only one solution to meet the needs. Tencent hopes to choose the most appropriate tools for different users.
Wang Xiaohang, Vice President of Ant Group: All key links in the financial field are worthy of transformation using large model technology
Babbitt News, on September 8, at the Bund Conference held in Shanghai, Wang Xiaohang, vice president of Ant Group and head of financial large models, said that all key links in the financial field are worthy of transformation using large model technology. Financial large models must ensure the rigor of domain knowledge and professional logic as well as the controllability of content generation in order to truly bring industrial value. This is also the real proposition of the industry that the financial model needs to solve.
Wang Minghui of Spinj Capital: In the end, one or two companies may dominate the large To-C model
According to a report by Metaverse NEWS on September 7, Wang Minghui, founding partner and chairman of Spinj Capital, believes that large models on the To C side face dual challenges of regulatory rules and technological breakthroughs. From a business perspective, only one or two companies may succeed in the end. ; For the B-side large model, it mainly serves B-side users, and is optimistic about its ability to empower the real industry and improve industry efficiency. However, B-side customer needs are more rigorous, and large models need to become trustworthy scene tools. Wang Minghui said that the vigorous development of large models this year will definitely improve the digitalization level of enterprises. In fact, data elements, large models and the metaverse are an organic whole.
Yili Ni, Chairman of McKinsey China: The positive impact of generative AI on the global economy can reach up to US$7.9 trillion
On September 7, at the 2023 Tencent Global Digital Ecology Conference, McKinsey China Chairman Yili Ni said that the positive impact of generative AI on the global economy can reach up to 7.9 trillion US dollars, but the overall broad-based AI may reach 25 trillion US dollars. In addition, Ni Yili said that due to the influence of generative AI, the time when half of the current work activities will be replaced by automation has been pushed forward by ten years compared with previous predictions.
AMD Su Zifeng: The data center AI accelerator market is expected to reach US$150 billion by 2027
According to a report by Science and Technology Innovation Board Daily on September 6, AMD CEO Su Zifeng recently stated that the demand for artificial intelligence AI semiconductors is still strong, and AMD’s data center business will be very strong in the second half of the year. She reiterated her expectation that the data center AI accelerator market will reach US$150 billion by 2027, and confirmed that AMD's upcoming MI300 AI accelerator product will be available in the fourth quarter.
Microsoft: AIGC tools will be the company’s fastest software business to reach $10 billion in revenue
According to a report by Science and Technology Innovation Board Daily on September 8, Microsoft Chief Financial Officer Amy Hood recently stated that the generative artificial intelligence tool suite will be the company’s fastest software business to reach US$10 billion in revenue. “The opportunity is there, and AI enterprise software can attract customers,” she said. Microsoft has released a series of AI-driven products, including the 365 Copilot service for $30 per person per month. Hood said customers have responded well to 365 Copilot and feel it has changed the way they work. She added that the target market for the 365 Copilot service is particularly large because Microsoft 365 has 160 million users.
07
research report
The report states that generative artificial intelligence will replace 2.4 million jobs in the United States by 2030, mainly affecting white-collar workers.
According to a report by IT House on September 10, according to the latest report from the analysis company Forrester, by 2030, generative artificial intelligence will replace 2.4 million jobs in the United States and affect 11 million jobs. These jobs are mainly white-collar jobs, including technical writing, social science research, proofreading, copywriting, etc. Generative artificial intelligence refers to artificial intelligence technology that can automatically generate text, images and other content, such as OpenAI's GPT-4 and ChatGPT.
Microsoft and OpenAI supercomputers consume huge amounts of water in Iowa
In its latest environmental report, Microsoft revealed that its global water consumption increased by 34% from 2021 to 2022, reaching nearly 1.7 billion gallons (equivalent to more than 2,500 Olympic-sized swimming pools), which is up from previous years. The ratio is a significant increase. Shaolei Ren, a researcher at the University of California, Riverside, attributes much of the growth to artificial intelligence, including significant investment in generative artificial intelligence and a partnership with OpenAI.
As the data center cluster for Microsoft's cloud computing services, West Des Monines, Iowa, bears the main source of water consumption. In addition, the supercomputing data center used by OpenaAI to train GPT-4 is also located there. External researchers estimate that each time 5 to 50 questions are asked using ChatGPT, approximately 500 ml of water is consumed.
Babbitt Park is open for cooperation!
Chinese Twitter: https://twitter.com/8BTC_OFFICIAL
English Twitter: https://twitter.com/btcinchina
Discord community: https://discord.gg/defidao
Telegram channel: https://t.me/Mute_8btc
Telegram community: https://t.me/news_8btc
