laitimes

General Artificial Intelligence AGI Weekly Report (6.26-7.2)

author:Oppa Freedom

Essentials:

1. The inaugural meeting of GPT Industry Alliance was held in Beijing

2. Beijing issued the Action Plan for Innovation and Development of Robot Industry (2023-2025)

3. Byte Volcano Engine releases the Volcano Ark to be a water seller on the gold rush road of the big model

4. Meituan announced the acquisition of a large-model startup Lightyear, Wang Huiwen was hospitalized, and his brother Wang Xing was in the end

5. Khan Academy has announced that it will use GPT-4 to power the artificial intelligence assistant Khanmigo

6. A basket of big models of the financial industry has sprung up recently

Detailed interpretation:

Industry news

1. The inaugural meeting of GPT Industry Alliance was held in Beijing

On June 28, the "GPT Industry Alliance Inaugural Conference" jointly hosted by the Internet Society of China, the Cloud Institute of the China Academy of Information and Communications Technology, and 360 (hereinafter referred to as "360") was held in Beijing. In the future, the Alliance will develop 100 industry partners and work with 1,000 GPT application ecological partners to provide GPT one-stop services for millions of customers and accelerate the development of the GPT industry.

At the meeting, 360 Group, together with Zhipu AI, HUAWEI CLOUD, SenseTime, Dark Horse of Entrepreneurship, Mobvoi, Yunzhisheng, Lanzhou Technology, Zhongguancun Longmen Accelerator and other industry chain enterprises, announced the official establishment of the GPT Industry Alliance.

General Artificial Intelligence AGI Weekly Report (6.26-7.2)

Zhou Hongyi, founder of 360 (601360.SH, hereinafter referred to as "360") Group, released the 360 enterprise-level AI big model strategy at the inaugural meeting of the GPT Industry Alliance, and announced the official opening of the 360 intelligent brain API platform, which will take the lead in nearly 20 industries such as security, media, cultural tourism, government affairs, and energy, and build "five intelligent brains" for enterprise-level users, including "office writing brain, decision analysis brain, knowledge management brain, customer service brain, and cultural tourism investment brain".

Weekly Hongyi judged that the future landing form of large models in enterprises must be a combination of multiple vertical and small-scale large models. Enterprises need to take multiple steps from digitalization to intelligence, from loose coupling to deep integration and multi-step iteration. The first step is to start from the office scene, which can achieve "loose coupling", without destroying the existing business system, and seamlessly integrating with the existing system. He believes that the big model should not only serve the leaders of the enterprise, but also serve the employees of the enterprise, and truly play the role of solving the pain points and just needs, so that everyone from top to bottom can use it to achieve AI inclusiveness.

Related links: https://ggjd.cnstock.com/company/scp_ggjd/tjd_bbdj/202306/5082790.htm

2. Beijing issued the Action Plan for Innovation and Development of Robot Industry (2023-2025)

Main objectives: By 2025, the innovation ability of the city's robot industry will be greatly improved, 100 high-tech and high-value-added robot products and 100 application scenarios with national promotion value will be cultivated, and the ownership of 10,000 robots will reach the world's leading level, forming a development ecology of innovation factor agglomeration and active innovation and entrepreneurship. The income of the city's core robot industry has reached more than 30 billion yuan, creating a leading domestic and internationally advanced robot industry cluster.

High-end leading, focusing on the development of robot "1+4" product system: focusing on the world's cutting-edge technology and future strategic needs, stepping up the layout of humanoid robots, driving the development of four types of advantageous robot products in medical health, collaboration, special and logistics, implementing hundreds of new robot product projects, and creating an innovative product system with intelligent drive, integration of production and research, and open and leading.

Strengthen the support of artificial intelligence large models: Develop and continuously improve the robot general artificial intelligence large model, mine application scenario resources, provide diversified scenario data support for model pre-training, and improve the versatility and practicality of the model. Break through the key technology of large-model multi-mode data fusion, and develop large-model systems for image, text, speech, and multi-mode sensing data fusion processing such as force, heat, electricity, and magnetism. According to the characteristics of various robot technologies and application scenarios, develop efficient fine-tuning algorithms for large models to promote the deepening application of large models in the field of robotics. Build an open source platform for model optimization algorithms, create a large model ecology with wide participation and interactive optimization in the whole industry, and promote the iterative improvement of model performance.

Benchmarking and comprehensive implementation of "robot +" application demonstration: Facing the key needs of medical, manufacturing, construction, trade logistics, pension, emergency, agriculture and other fields, implement hundreds of application scenario demonstration projects, carry out a number of "robot +" application demonstrations, and accelerate the formation of landmark scenes, iconic services, iconic models and iconic formats.

Related links: https://www.beijing.gov.cn/zhengce/zfwj/202306/t20230628_3148572.html

Product Information

3. Byte Volcano Engine releases the Volcano Ark to be a water seller on the gold rush road of the big model

Large models have become the market outlet since this year. According to the "Research Report on the Map of Chinese and Industrial Intelligence" released by the China Institute of Science and Technology Information and other institutions, as of May this year, 79 large models have been publicly disclosed in China.

Previously, Baidu, Ali, Tencent, SenseTime, 360 and other domestic manufacturers have successively announced their entry into the market, most of the practices are to release a general large model or several industry large model bases, industry customers can based on these basic models and their own industry data fine-tuned, to create an AI application serving their own business.

Now, ByteDance has cautiously taken the first step into the big model track. On June 28, at the "V-Tech Experience Innovation Technology Summit" hosted by Volcano Engine and co-organized by NVIDIA, Volcano Engine released the large model service platform "Volcano Ark", which provides enterprises with a full range of platform services (MaaS, that is, Model-as-a-Service) such as model fine-tuning, evaluation, and reasoning.

General Artificial Intelligence AGI Weekly Report (6.26-7.2)

But the way bytes cut in is obviously different from other big manufacturers. Volcano did not release its own general large model or industry large model, Volcano Ark aggregated a number of large model bases developed by third-party manufacturers, and helped enterprises choose the right model for different scenarios, including large model fine-tuning, evaluation (effect indicators), reasoning, intervention (content security, quality evaluation) and other all-round professional support; Support customers to systematically perceive model performance and provide a data basis for subsequent decision-making.

At present, "Volcano Ark" integrates the large models of many AI technology companies and research institutes such as Baichuan Intelligence, Mos, Moss of Fudan University, IDEA Research Institute, Lanzhou Technology, MiniMax, Zhipu AI (in alphabetical order by pinyin), etc., and has launched the invitation for testing.

The business model of the Volcano Ark: that is, the computing power resources are provided to the large model supplier, and the supplier will superimpose the reasonable profit after removing the cost of the IaaS level, and finally quote on the Volcano Ark. Downstream customers currently only need to pay the model service fee, and the platform service fee will be set to zero for a long time.

Related links: https://new.qq.com/rain/a/20230628A058ZX00

Investment and financing

4. Meituan announces acquisition of big-model startup Lightyears Beyond

On June 29, Meituan announced on the Hong Kong Stock Exchange that it had reached a 100% equity acquisition agreement with domestic and foreign entities outside Lightyear, an artificial intelligence company founded by former co-founder Wang Huiwen, for a purchase price of about 2.058 billion yuan.

Meituan said that Lightyears away is China's leading AGI innovator, founded and controlled by Mr. Wang Huiwen, co-founder of Meituan, former director and related person of the Company. As at the date of this announcement, the Group's total net cash position outside Lightyear is approximately US$285,035,563 (approximately $2,062 million).

It can be seen that Meituan obtained 2.062 billion yuan in cash through a consideration of 2.058 billion yuan, which is equivalent to getting the team and technology outside of light-years for free, and almost did not spend money, but this still rescued Wang Huiwen, who was in a predicament of illness, dealt with the pending projects, and also settled the funds of investors outside of light-years, reducing the potential investment risk in the future to zero.

Upon closing, Meituan will hold a 100% interest in Lightyear's Group.

At the beginning of 2023, ChatGPT was born, and the wave of AI began to sweep the world. On February 11, Wang Huiwen, the co-founder of Meituan, who is almost "retired", reappeared and announced his entry into the AI field with a high profile.

In April 2023, only 2 months after its birth, the valuation beyond light-years has exceeded $1 billion, ranking among the unicorns. In the same month, Lightyear completed the acquisition of OneFlow Technology, a Beijing AI model company founded by Tsinghua Dr. Yuan Jinhui in 2017.

On April 6, Wang Huiwen took Ma Zhankai, Liu Jiang, Yuan Jinhui and other members of the entrepreneurial team to move to a new office in the Sohu Network Building light-years away. On the day of the groundbreaking ceremony, several people present also ate cake together to celebrate this new beginning.

The speed of development outside the light years led by Wang Huiwen is very fast, and it can be said that it is a wild advance. However, after entering May, the situation began to take a sharp turn, and the problem was Wang's personal health problems. Until June, it was rumored that Wang Huiwen was hospitalized due to his aggravated illness, which eventually led to Meituan's wholly-owned acquisition of light-years away.

Related links: https://baijiahao.baidu.com/s?id=1770090441299849173

Industry applications

5. Khan Academy has announced that it will use GPT-4 to power the artificial intelligence assistant Khanmigo

Recently, Khan Academy announced that it will use GPT-4 to support Khanmigo, an artificial intelligence assistant that can serve as both a virtual tutor for students and a classroom assistant for teachers.

General Artificial Intelligence AGI Weekly Report (6.26-7.2)

Khan Academy is a non-profit organization with a mission to provide free, world-class education to anyone, anywhere. The organization offers thousands of math, science, and humanities courses for students of all ages.

Khanmigo is the latest project of Khan Academy, which uses the latest artificial intelligence technology to provide students and teachers with more accurate testing and course evaluation. With Khanmigo, students can quickly understand their level of learning and get more personalized study advice. Teachers can better understand students' learning and provide them with more effective teaching support.

Khan Academy used GPT4 to create a virtual Socrates that can engage in conversations with students, asking questions and guiding like a mentor. For example, when a student asks Khanmigo an answer to a math question, Khanmigo does not give the answer directly, but asks the student how they think and solve the problem themselves, and gives hints and suggestions.

It can provide personalized guidance to each student according to their level and progress, so that students can not only do the questions, but also understand the concepts and principles behind the questions. It can also provide students with more relevant knowledge and application scenarios according to their interests and needs. It is like a private tutor, accompanying students anytime, anywhere, helping them learn and grow.

Related links: https://www.khanacademy.org/khan-labs

6. A basket of big models of the financial industry has sprung up recently

Bloomberg launched BloombergGPT, a large-language model in the financial industry, which benefits from the training of large-scale text datasets such as documents, industry news, and social media in the financial vertical field, which can understand the market "sentiment" behind financial news and assist financial information writing, which solves the pain points of the application of general NPL models in the financial field.

Recently, Columbia University and NYU Shanghai launched a new large-model product FinGPT, which is a large-model product for the financial field.

Agricultural Bank of China launched ChatGPT-like large-model application ChatABC, and Industrial and Commercial Bank of China released a general model for the financial industry based on Ascend AI.

Recently, Baidu's financial technology company Du Xiaoman officially opened the first 100-billion-level Chinese financial model in China - "Xuanyuan". The Xuanyuan model is trained on the basis of the 176 billion parameter Bloom model, and the effect is greatly improved compared with the general large model in tasks such as financial term understanding, financial market commentary, financial data analysis and financial news understanding, showing obvious advantages in the financial field. At present, the 100-billion-level Xuanyuan model can be applied for download in Huggingface and is open to all financial institutions.

On the afternoon of June 28, Hang Seng Electronics released LightGPT, a large model in the financial industry, which will complete a new round of financial capability upgrades at the end of September and officially open the trial interface. Bai Shuo, Chief Scientist of Hang Seng Electronics, said that LightGPT uses more than 400 billion tokens of financial field data and more than 40 billion tokens of language enhancement data, and uses it as the secondary pre-training corpus of large models, supporting more than 80+ financial exclusive task instruction fine-tuning, which can provide underlying AI capability support for financial business scenarios such as investment advisory, customer service, investment research, operation, risk control, compliance, and R&D.

Read on