李彦宏 stated that Wenxin's large model 4.0 is the most powerful Wenxin's large model to date, achieving a comprehensive upgrade of the basic model, with significant improvements in understanding, generation, logic, and memory capabilities, and a comprehensive level that is "no less than GPT-4." Zhipu AI has completed a financing of over 2.5 billion RMB, with participation from the Social Security Fund, Meituan, Ant Group, Alibaba, Tencent, Xiaomi, and Kingsoft, among others. Baichuan Intelligence has raised a $300 million Series A1 financing, with participation from Alibaba, Tencent, and Xiaomi.
Image source: Generated by Wujie AI
Financing
Zhipu AI has completed a financing of over 2.5 billion RMB
According to Babite, Beijing Zhipu Huazhang Technology Co., Ltd. has received a total of over 2.5 billion RMB in financing this year, with major participants including the Social Security Fund Zhongguancun Independent Innovation Fund (Junlian Capital as the fund manager), Meituan, Ant Group, Alibaba, Tencent, Xiaomi, Kingsoft, Shunwei, Boss Zhipin, TAL Education, Sequoia Capital, and Hillhouse, as well as some existing shareholders including Junlian Capital. The above financing will be used for further research and development of the base large model, better supporting the industry ecosystem, and high-speed development with partners.
Baichuan Intelligence raised $300 million in Series A1 financing, with participation from Alibaba, Tencent, and Xiaomi
According to First Financial, on October 17, the AI company Baichuan Intelligence, founded by former Sogou CEO Wang Xiaochuan, disclosed that it has completed a Series A1 strategic financing, with a financing amount of $300 million, with technology giants such as Alibaba, Tencent, and Xiaomi, as well as many top investment institutions participating in this round of financing. With the addition of the $50 million angel round, Baichuan Intelligence's total financing amount has reached $350 million.
"Ji Space" completed a B round financing of over 100 million RMB, with funds mainly used to expand the AI research and development team
According to 36Kr, on October 19, "Ji Space" announced the completion of a B round financing of over 100 million RMB, led by Anke Innovation and followed by the previous investor Challenger Capital. Ji Space founder Yuan Bin stated that the funds from this round of financing will be mainly used to expand the research and development team, especially the AI research and development scale, and expand product markets and overseas layout.
AI company Mind Foundry completed a $22 million financing
According to Webmaster's Home, on October 19, the artificial intelligence startup company Mind Foundry Ltd. recently successfully raised $22 million in financing, aiming to help insurance companies decide which drivers should be provided with insurance coverage. The startup's artificial intelligence tools are used to detect cognitive decline in elderly Japanese drivers, to assist the Asian insurance giant Aioi Nissay Dowa Insurance in predicting and preventing accidents. It is reported that Aioi, together with Parkwalk Advisors and the University of Oxford, invested in this round of financing. This round of Series B financing has raised the total financing amount of the startup to $44 million.
AI music generation application Riffusion completed a $4 million financing
According to TechCrunch, on October 17, the AI music generation application Riffusion completed a $4 million seed round of financing, led by Greycroft, with participation from South Park Commons and Sky9.
It is reported that Riffusion was originally created as a free open-source tool by founders Seth Forsgren and Hayk Martiros, and this year launched a new free application that allows users to generate "riffs" by describing lyrics and music styles, and share them publicly or with friends. Currently, the well-known music group The Chainsmokers is providing advisory services to Riffusion.
AI content detection company Reality Defender completed a $15 million Series A financing
According to TechCrunch, Reality Defender is one of several startups developing tools to detect deepfakes and other artificially generated content. The company announced today that it has raised $15 million in a Series A financing led by DCVC, with participation from Comcast, Ex/ante, Parameter Ventures, and Nat Friedman's AI Grant.
Large Models
Li Yanhong: Wenxin's large model 4.0 officially released, with a comprehensive level no less than GPT-4
According to The Paper, on October 17, Baidu's founder, chairman, and CEO Li Yanhong announced the official release of Wenxin's large model 4.0, starting with invitation testing. Li Yanhong stated that Wenxin's large model 4.0 is the most powerful Wenxin's large model to date, achieving a comprehensive upgrade of the basic model, with significant improvements in understanding, generation, logic, and memory capabilities, and a comprehensive level that is "no less than GPT-4."
Chinese Academy of Sciences National Astronomical Observatory open-sourced the StarWhisper astronomical large model
According to Webmaster's Home, the open-source StarWhisper astronomical large model of the Chinese Academy of Sciences National Astronomical Observatory achieved the second overall ranking in the CG-Eval released by the Oracle AI Research Institute and LanguageX AI Lab. The model is close to or surpasses GPT3.5 Turbo in mathematical reasoning and astronomical capabilities.
South Korea's telecom giant SK collaborates with Deutsche Telekom to develop AI large language models
According to Webmaster's Home, on October 23, South Korea's telecom giant SK telecom announced a collaboration with global telecom company Deutsche Telekom to jointly develop a large language model (LLM) aimed at achieving AI vision. This collaboration plan is scheduled to release a specialized LLM for the telecom field in the first quarter of 2023.
Stanford University releases "Transparency Index" for AI basic models, Llama 2 ranks first with 54% but "fails"
According to IT Home, Stanford University recently released a "Transparency Index" for AI basic models, which shows that Meta's Llama 2 has the highest index, but the related "transparency" is only 54%, so researchers believe that almost all AI models on the market "lack transparency."
Alibaba DAMO Academy releases remote sensing AI large model
According to 36Kr, on October 20, Alibaba DAMO Academy released the industry's first remote sensing AI large model. The model can identify farmland, crops, buildings, etc., improving the analysis efficiency of remote sensing applications such as disaster prevention and control, natural resource management, and agricultural yield estimation. Currently, the model is available for use on the AI Earth Earth Science Cloud Platform.
AI company Adept releases multimodal large model Fuyu-8B, with 8 billion parameters
According to Webmaster's Home, AI company Adept has released the multimodal large model Fuyu-8B, which has 8 billion parameters and can understand various types of images, including photos, charts, PDFs, and UI interfaces. Fuyu-8B adopts a pure decoder Transformer architecture without using image encoders, and outperforms other large models in multiple tasks.
Peking University open-sources the 7 billion parameter code large model CodeShell, claiming to be the "strongest code base of the same scale"
According to IT Home, on October 19, the Knowledge Computing Laboratory of the National Engineering Research Center for Software Engineering at Peking University, in collaboration with the AI Laboratory of Sichuan Tianfu Bank, officially open-sourced their 7 billion parameter code large model CodeShell, claiming to be the "strongest code base of the same scale."
Applications
Tmall Genie announces "Genie Large Model" has been fully launched
According to Webmaster's Home on October 23, the "Genie Large Model" of Tmall Genie has been fully launched, bringing four major upgraded skills, including optimization of open-domain dialogue and AIGC creation functions, solving the problem of nearby wake-up, and supporting voice switching.
Baidu Baijiahao is testing "Baijiahao AI Assistant"
According to Webmaster's Home on October 20, Baidu's Baijiahao announced that the Baijiahao AI Assistant is currently in testing, aiming to significantly improve the efficiency of creators with technology. The Baijiahao AI Assistant is currently in the testing phase, and creators can sign up to participate. The upgrade includes features and changes such as the Baijiahao homepage permanently featuring the "Baijiahao AI Assistant" on the right, hot topics being written, and direct conversation being turned into images or dynamic content.
Kunlun Wanwei: Tiantool AI Assistant App is in testing, Tiantool large model inference capability leads GPT-3.5
According to Jinshi on October 19, Kunlun Wanwei announced on the interactive platform that the Tiantool AI Assistant's mobile app is in testing. Recently, Kunlun Wanwei's Tiantool large model ranked first in a multi-modal large language model evaluation conducted by Tencent YouTu Laboratory in collaboration with Xiamen University; in the authoritative reasoning benchmark GSM8 K test, the Tiantool large model achieved an 80% accuracy rate, surpassing GPT-3.5 (57.1%) and LLaMA2-70 B (56.8%), reaching a leading global level of inference capability.
Kugou Music releases version 12.0, adds "AI Smart Image Matching" and "AI Music Magic" features
According to 36Kr on October 17, Kugou Music released version 12.0, covering Android and iOS clients. This version introduces several new features, including "AI Smart Image Matching," which can generate images based on lyrics, and "AI Music Magic," which can transform a song into eight different versions.
Vivo's self-developed AI large model exposed: Landing on OriginOS 4, intelligent assistant named Jovi Copilot
According to Babite on October 17, Weibo blogger @Digital Chat Station revealed some information about Vivo's self-developed AI large model. He stated that the large model will land on OriginOS 4 and will be presented in the form of a floating window that can be called out from the sidebar, similar to Microsoft Copilot. The AI data volume and system integration are well done, creating a fully intelligent system.
US AI unicorn company Jasper AI launches new AI marketing tool
According to VentureBeat on October 19, following layoffs earlier this year, US AI unicorn company Jasper AI announced the release of a new "end-to-end AI-assisted tool to achieve better marketing results." The new features include new performance analysis for optimizing content, a company intelligence center to keep information consistent with brand strategy, and marketing campaign tools to expedite the review cycle. These features will begin testing in November and additional features are planned for release in the first quarter of 2024. Jasper's new AI tool can generate custom content for each channel by pre-determining campaign parameters and can also schedule, distribute, and track the entire campaign. Most importantly, the tool can achieve closed-loop feedback through integrated performance analysis.
AI programming language Mojo lands on the Mac platform
According to IT Home on October 20, the Mojo programming language recently landed on the Mac platform, providing AI developers with a programming experience similar to Python. Mojo is a high-performance "Python++" programming language for computation, targeting AI developers, and is expected to evolve into a superset of Python over time. Currently, Mojo seamlessly integrates with any Python code and has an extensible programming model for performance-critical systems, including accelerators commonly found in artificial intelligence, such as GPUs.
Regulation
EU plans to classify AI models into three categories for regulation, with stricter rules for the most powerful models
According to Bloomberg on October 18, the EU plans to classify AI models into three categories for regulation. According to documents seen by Bloomberg, the EU also plans to subject the most powerful AI generative models to additional external testing. Negotiators hope to finalize this legislation at the next meeting on October 25, with the goal of completion by the end of this year. Under the proposed rules, the first category will include all basic models. The "very capable" systems in the second category will be distinguished by the computing power used to train their large language models. According to the proposal, these models may "exceed current technical levels and may not yet be fully understood." Finally, the third category, called large-scale general artificial intelligence systems, will include the most popular AI tools and will be measured by the number of users.
Major Companies
Apple plans to invest $10 billion annually to apply generative AI to a range of products and services including iOS and Siri
According to IT Home on October 23, Apple is currently preparing a response and plans to develop AI features for its entire range of devices. An insider stated that Apple executives have been caught off guard by the sudden surge in the AI industry and have been working to make up for missed opportunities since the end of last year. In July, Apple established its own large language model, Ajax, and internally launched a chatbot called "Apple GPT." The next key step for the company is to determine whether this technology can compete with its rivals and how Apple will apply it to its products.
Intel launches its first AI PC acceleration program, aiming to provide AI features for over 1 million PCs by 2025
According to IT Home on October 20, Intel announced the official launch of its first AI PC acceleration program during the "Intel on Technology Innovation Conference." The program aims to bring AI features to over 1 million PCs by 2025, and will be driven by the Intel Core Ultra processors set to be released on December 14.
Meta AI releases new artificial intelligence system capable of decoding images from brain activity
According to Webmaster's Home, Meta AI announced on October 18 through a blog post the launch of a new artificial intelligence (AI) system designed to decode images from human brain waves. This system combines a non-invasive brain scanning method called magnetoencephalography (MEG).
Opinions
Yu Minhong: AI is not very beneficial for educating children, Li Yanhong refutes: This is a misunderstanding
According to Pear Video, on October 21, at the New Oriental Family Education Summit, Yu Minhong stated that AI is not very beneficial for educating children. Children just use Wenxin to write essays and then make some changes. Li Yanhong immediately refuted, "I think this is a misunderstanding." Yu Minhong said, anyway, if a boy writes a particularly beautiful love letter to a girl, don't believe that the boy wrote it himself, it may have been written by artificial intelligence. Li Yanhong responded, "It still depends on the boy's level of writing prompts."
Early investor in OpenAI: 80% of jobs will be eliminated by artificial intelligence in the next ten years
According to Fortune magazine, on Tuesday, Silicon Valley billionaire and technology venture capital giant Vinod Khosla stated at a Wall Street Journal live technology conference that "80% of jobs" will be eliminated by artificial intelligence in the next ten years, including professions in high-paying fields such as accounting and medicine.
OpenAI CEO: Future AI devices will not be as ubiquitous as smartphones
According to the Science and Technology Innovation Board Daily on October 19, there have been reports that OpenAI may be collaborating with other tech giants to develop AI devices. Recently, OpenAI CEO Sam Altman stated at the Wall Street Journal's Tech Live event that while he is interested in AI-based consumer electronic devices, he is unsure of what these devices will look like and believes that future AI devices will not be as ubiquitous as smartphones.
Research Reports
Report: Despite the proliferation of AI tools, demand for human writing remains strong
According to Webmaster's Home on October 19, the rise of artificial intelligence content generation tools has had an impact on the writing industry, but new data shows that the demand for writers remains strong. The quarterly report from the freelance website Freelancer.com shows that in the third quarter of 2023, jobs related to writing, content creation, and marketing became the fastest-growing areas in the freelance market. Compared to the second quarter of this year, copy typing jobs grew by 28.7%, Microsoft Word projects grew by 24.7%, search engine marketing grew by 24.1%, copywriting grew by 23.8%, and ghostwriting jobs grew by 23.1%.
IDC predicts: Generative AI spending to reach $143 billion by 2027
According to Webmaster's Home on October 17, according to the latest forecast from IDC (International Data Corporation), spending on Generative AI is expected to reach $143 billion in 2027, indicating rapid growth in this field. The forecast indicates that the growth rate of spending on Generative AI will far exceed that of comprehensive AI spending, and will be 13 times higher than the compound annual growth rate (CAGR) of global IT spending during the same period.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
